Datasets:

TaylorAI

/

pubmed_commercial

like 12

PMC005xxxxxx/PMC5003388.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757120510.1371/journal.pone.0161367PONE-D-16-12687Research ArticleMedicine and Health SciencesNeurologyBrain DamageMedicine and Health SciencesCritical Care and Emergency MedicineTrauma MedicineBrain DamageMedicine and Health SciencesHealth CareHealth Care FacilitiesHospitalsIntensive Care UnitsMedicine and Health SciencesCritical Care and Emergency MedicineResearch and Analysis MethodsResearch DesignSurvey ResearchQuestionnairesResearch and Analysis MethodsImaging TechniquesNeuroimagingComputed Axial TomographyBiology and Life SciencesNeuroscienceNeuroimagingComputed Axial TomographyMedicine and Health SciencesDiagnostic MedicineDiagnostic RadiologyTomographyComputed Axial TomographyResearch and Analysis MethodsImaging TechniquesDiagnostic RadiologyTomographyComputed Axial TomographyMedicine and Health SciencesRadiology and ImagingDiagnostic RadiologyTomographyComputed Axial TomographyMedicine and Health SciencesHealth CareHealth Care PolicyTreatment GuidelinesPeople and PlacesGeographical LocationsEuropeMedicine and Health SciencesHealth CareHealth Care ProvidersMedical DoctorsPhysiciansPeople and PlacesPopulation GroupingsProfessionsMedical DoctorsPhysiciansVariation in Structure and Process of Care in Traumatic Brain Injury: Provider Profiles of European Neurotrauma Centers Participating in the CENTER-TBI Study Structure and Process of TBI Care in European Neurotrauma CentersCnossen Maryse C. 1Polinder Suzanne 1Lingsma Hester F. 1*Maas Andrew I. R. 2Menon David 3Steyerberg Ewout W. 1CENTER-TBI Investigators and Participants ¶1 Center for Medical Decision Sciences, Department of Public Health, Erasmus Medical Center, Rotterdam, the Netherlands2 Department of Neurosurgery, Antwerp University Hospital and University of Antwerp, Edegem, Belgium3 Division of Anaesthesia, University of Cambridge/Addenbrooke’s Hospital, Cambridge, United KingdomLazzeri Chiara EditorAzienda Ospedaliero Universitaria Careggi, ITALYCompeting Interests: The authors have declared that no competing interests exist. Conceptualization: EWS AIRM DM. Data curation: MCC. Formal analysis: MCC. Funding acquisition: AIRM DM. Investigation: MCC SP HFL EWS AIRM DM. Methodology: EWS. Project administration: AIRM DM. Resources: AIRM DM. Supervision: SP HFL EWS. Writing – original draft: MCC. Writing – review & editing: SP HFL AIRM DM EWS. ¶ Membership of the CENTER-TBI Investigators and Participants is provided in the Acknowledgments. * E-mail: h.lingsma@erasmusmc.nl29 8 2016 2016 11 8 e016136729 3 2016 4 8 2016 © 2016 Cnossen et al2016Cnossen et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Introduction The strength of evidence underpinning care and treatment recommendations in traumatic brain injury (TBI) is low. Comparative effectiveness research (CER) has been proposed as a framework to provide evidence for optimal care for TBI patients. The first step in CER is to map the existing variation. The aim of current study is to quantify variation in general structural and process characteristics among centers participating in the Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) study. Methods We designed a set of 11 provider profiling questionnaires with 321 questions about various aspects of TBI care, chosen based on literature and expert opinion. After pilot testing, questionnaires were disseminated to 71 centers from 20 countries participating in the CENTER-TBI study. Reliability of questionnaires was estimated by calculating a concordance rate among 5% duplicate questions. Results All 71 centers completed the questionnaires. Median concordance rate among duplicate questions was 0.85. The majority of centers were academic hospitals (n = 65, 92%), designated as a level I trauma center (n = 48, 68%) and situated in an urban location (n = 70, 99%). The availability of facilities for neuro-trauma care varied across centers; e.g. 40 (57%) had a dedicated neuro-intensive care unit (ICU), 36 (51%) had an in-hospital rehabilitation unit and the organization of the ICU was closed in 64% (n = 45) of the centers. In addition, we found wide variation in processes of care, such as the ICU admission policy and intracranial pressure monitoring policy among centers. Conclusion Even among high-volume, specialized neurotrauma centers there is substantial variation in structures and processes of TBI care. This variation provides an opportunity to study effectiveness of specific aspects of TBI care and to identify best practices with CER approaches. European Commission FP7 Framework Program602150Data used in preparation of this manuscript were obtained in the context of CENTER-TBI, a large collaborative project with the support of the European Commission 7th Framework program (602150). The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Data AvailabilityThere are however legal constraints that prohibit us from making the data available. Since there are only a limited number of centers per country included in this study (for 2 countries only 1 center), data will be identifiable. Readers may contact Dr. Hester Lingsma (h.lingsma@erasmusmc.nl) for requests for the data.Data Availability There are however legal constraints that prohibit us from making the data available. Since there are only a limited number of centers per country included in this study (for 2 countries only 1 center), data will be identifiable. Readers may contact Dr. Hester Lingsma (h.lingsma@erasmusmc.nl) for requests for the data. ==== Body Introduction Traumatic Brain Injury (TBI) is an important threat to public health with a crude incidence rate of up to 849 per 100,000 people in European countries [1, 2]. TBI is emerging as one of the leading causes of death and disability worldwide resulting in huge personal suffering and far-reaching socioeconomic consequences [3, 4]. Different perspectives on various aspects of care exist, and the evidence underpinning guideline recommendations for treatment of patients with TBI is weak [3, 5]. There is growing realization that randomized clinical trials alone will not be able to provide the evidence base that is needed to address these knowledge gaps [6]. Comparative effectiveness research (CER) has been proposed as a good complementary approach to strengthen the evidence base. CER has been defined as “the generation and synthesis of evidence that compares the benefits and harms of alternative methods to prevent, diagnose, treat, and monitor a clinical condition or to improve the delivery of care” [7]. CER exploits between-center differences in patient management by comparing centers that perform a certain intervention routinely to others that do not. This approach is expected to be particularly suitable for TBI since large between-center differences in both patient management and outcomes have been previously reported [8, 9]. The Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) study is a large-scale observational multicenter study focusing on characterization and CER in TBI. The first step for CER is to provide an overview of variation in structures and processes of care in the participating centers (‘provider profiling’). Such an overview can be used to identify areas where large between-center variation exists, to guide future CER analyses. But it can also directly be used for CER. For example, treatment effectiveness of a certain intervention can be studied by comparing outcome in patients from centers that routinely perform the intervention to outcome in patients from centers that do not routinely perform the intervention. Therefore, the objective of the current study is to quantify variation in general structure and process characteristics among centers participating in the CENTER-TBI study and to identify topics for CER. Material and Methods CENTER-TBI study CENTER-TBI is a prospective longitudinal observational study conducted in 72 centers from 20 countries across Europe and Israel [3]. One of the global aims is to “identify the most effective clinical care and provide high-quality evidence in support of treatment recommendations and guidelines” [3]. This will be pursued by CER approaches. For more information, see also www.center-tbi.eu. Before the patient inclusion started, a detailed inventory of center characteristics was performed by distributing a set of questionnaires on structures and process of TBI care: The Provider Profiling (PP) questionnaires (S1 File). This set of questionnaires was distributed among 71 centers, since two CENTER-TBI centers represented different departments from the same hospital with similar structures and processes. Development process of the Provider Profiling Questionnaires The PP questionnaires went through a comprehensive developing process to warrant completeness and relevance of topics and face validity of questions. The neurotrauma evidencemap (http://neurotrauma.evidencemap.org/) was searched for gaps and inconsistencies in knowledge of optimal treatment and organization of TBI care, and used to define topics of interest. We included topics relevant for CER as well as topics relevant for descriptive analyses. Initial questions were formulated based on literature and suggestions from experts in the field. Available surveys and questionnaires in the field of TBI or critical care [10, 11] were searched for and used for the (re)formulation of (additional) questions. Questions related either to structures or processes of general or TBI-specific care. Structure refers to the conditions under which patient care is provided (e.g. the number of beds, trauma center designation, hospital facilities), and process refers to activities that constitute patient care (e.g. general hospital or department policies) [12]. Structural information could be extracted from hospital databases, annual reports and local registries. Process information refers to general policies rather than individual treatment preferences of responsible physicians. General policy was defined as ‘the way the large majority of patients (>75%) with a certain indication would be treated’, recognizing that there might be exceptions. We included open questions and multiple-choice questions. All questions were presented with text boxes that contained definitions and a short explanation about the interpretation and completion of the question. The definitions used in this paper are summarized in the Supplemental material (S2 File). Experts in the field provided feedback on the initial formulated questions and proposed new questions and topics in three subsequent phases. Consulted experts included neurosurgeons, (neuro)intensivists, neurologists, emergency department (ED) physicians, rehabilitation physicians, medical ethicists, health care economists and epidemiologists. Some of the consulted experts had previous experience with the design and conduct of surveys in the field of TBI or critical care. In a first phase, a small group of involved experts discussed the questionnaires during an email conversation and a group discussion. In a second phase, an international expert panel, consisting of 25 experts from 9 countries, was consulted per email. These experts provided feedback on one or more of the questionnaires. Decisions on proposed content and formulation were then made during a group discussion with a small group of involved experts. These draft PP questionnaires were then pilot-tested in 16 of the participating CENTER-TBI centers. Each center completed two or three questionnaires, such that each questionnaire was pilot-tested at least three times. All answers were checked for unexpected or missing values and ambiguous questions were subsequently reformulated or deleted. Pilot-testers additionally completed a form in which they were asked to provide feedback, which was incorporated accordingly. All these processes resulted in a final set of eleven questionnaires related to different phases of TBI care (see Table 1). In total, there were 321 questions included in the PP. 10.1371/journal.pone.0161367.t001Table 1 Characteristics of the Provider Profiling questionnaires. Questionnaire No. of questions Topics 1.General 41 Structural characteristics of the hospital, catchment area, volume, facilities, staffing characteristics, payment, equipment, costs 2.Medical ethics 17 Department of medical ethics, IRB approval, informed consent procedures 3. Prehospital trauma care 28 First aid initiatives, dispatch systems, emergency services, hospital reception and initial treatment 4. Emergency department 50 Structural characteristics of the ED, imaging, guidelines, ED overcrowding, treatment, admission policy, discharge policy, withdrawal of life support 5. Admission 22 Structural characteristics of the ward, admission policy, guidelines, observations, treatment policy, step down beds, discharge policy 6. Structural and organizational aspects of the ICU 27 Structural characteristics of the ICU(s), staffing characteristics, admission policy, ICU decision making 7. Treatment at the ICU 70 Protocol use, ICP- and CPP monitoring, sedation, non-surgical treatment of severe TBI patients, seizure prophylaxis, treatment of fever, DVT prophylaxis, mechanical ventilation 8. Ethical aspects of the ICU 20 Withdrawal of life support, age and ICU admission 9. Neurosurgery 21 Volume, staffing characteristics, decision making, protocols, surgical management of mass lesions 10. Rehabilitation 14 In-hospital rehabilitation facilities, referral to post-acute care 11. Country 11 Health care policy, dispatch systems, insurance Note. The provider profiling questionnaires consist of 11 separate questionnaires. Table shows number of questions and topics for each of the questionnaires. Abbreviations. IRB = institutional review board, ED = emergency department, ICU = intensive care unit, ICP = intracranial pressure, CPP = cerebral perfusion pressure, TBI = traumatic brain injury, DVT = deep venous thrombosis prophylaxis Distribution of the questionnaires During presentations and workshops at two consecutive CENTER-TBI investigators meetings, information on the PP questionnaires was provided. Local investigators, as the senior persons supervising the CENTER-TBI study in the centers, were extensively informed in person and per email about the aim of the study and we emphasized the confidentiality of their responses. Additionally, to achieve unequivocal responses, we instructed them on how to respond to the process questions. We emphasized that we were asking for general policies, rather than individual treatment preferences and stimulated discussions with colleagues to identify the general policy of their department/center. Questionnaires were completed using a web-based system (Quesgen Systems Inc.) An instruction video was made available and any questions from local investigators were answered per email. The local investigators in each center were responsible for the completion process in their center. Staff members with the appropriate expertise and knowledge needed to complete one or more questions or questionnaires. The local investigators were responsible for monitoring progress and checking face validity of all answers. The first author (MC) reminded local investigators regularly and answered any questions by email. We aimed to receive completed questionnaires before centers started recruiting patients. As CENTER-TBI had a phased start of the inclusion period, PP questionnaires were completed between December 2014 and April 2016. Questionnaire completion and data cleaning A questionnaire was considered completed by a center if > 90% of the questions had been answered. Data from participating centers were included in the current paper if the center had completed the first PP questionnaire (‘general’), since the first questionnaire provides the general structure information necessary for provider profiles. The first author (MC) screened the completed questionnaires for missing values and contacted local investigators if any missings were present. They were asked to complete the missing data if possible or provide a reason for missingness. Data were further screened for outliers and local investigators were contacted to confirm values that were considered out of range. Statistical analyses To estimate reliability of the questionnaires, we included 17 (5%) duplicate questions, including all question formats. We equally included structure and process questions in the duplicate questions. Concordance rates were estimated by calculating the percentage of overlap between duplicate questions, and presented as mean, median and range. For open questions (e.g. what is the number of intensivist in your center), a maximum difference of 10% was considered concordant. For all hospital characteristics in this paper, frequencies and percentages were presented for categorical variables and medians and interquartile ranges (IQR) were presented for continuous variables. For a more in-depth understanding of the variation among centers, we checked whether there were differences between relatively high- and middle-income countries versus relatively lower-income countries, and also if there were differences between countries from different geographic locations (North and West Europe versus South and East Europe and Israel). We used the Chi-square test, and if appropriate, Fisher’s exact test to examine whether differences between groups were statistically significant (p < .05). The designation into relatively lower-income countries was based on a 2007 report by the European Commission [13]. Bosnia Herzegovina, Bulgaria, Hungary, Latvia, Lithuania, Romania and Serbia were subsequently classified as relatively lower-income countries. The subdivision into geographic location was based on the classification by the United Nations. Austria, Belgium, Denmark, Finland, France, Germany, Lithuania, the Netherlands, Norway, Sweden and the United Kingdom (UK) were subsequently classified as countries from West and North Europe, while all other countries were classified as countries from South and East Europe and Israel. Analyses were performed using the Statistical Package for Social Sciences (SPSS) version 21. Results Completion process All 71 eligible centers completed the provider profiling questionnaire about general structural and process information. Questionnaires were completed by multiple persons per center, including neurologists, neurosurgeons, trauma surgeons, intensivists, research nurses and administrative staff members. The 71 centers were from 20 European countries (see Fig 1). Each country had 1 to 9 participating centers (median = 2.5). The United Kingdom (UK) had most centers participating (n = 9), while Serbia and Switzerland both had one participating center. Thirteen of the included centers were from relatively lower-income countries and 25 centers were from countries in South and East Europe (including Israel). 10.1371/journal.pone.0161367.g001Fig 1 Centers and countries included in the Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) study Note. Reprinted and updated from Maas et al. (2015). Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury: a prospective longitudinal observational study. Neurosurgery, 76:67–80, under a CC BY licence, with permission from professor A.I. Maas. Reliability of the questionnaires The median concordance rate between duplicate questions was 0.85 (mean: 0.81; range 0.44–0.97), meaning that 85% of the responses were similar. Concordance rates were lowest for questions about treatment policy (e.g. on what indications would you admit a patient with mild TBI to the ward) and for open questions (e.g. what is the number of intensivists working at your center). Most multiple-choice questions about structure had concordance rates above 0.90. General structural characteristics The participating centers were predominately academic centers (n = 65, 92%), designated as a level I or II trauma center (n = 54, n = 74%) and situated in an urban location (n = 70, 99%, see Table 2). The majority of participants indicated that they had access to a helicopter platform (n = 57, 80%) and an acute trauma team (n = 63, 89%). Around half of the centers (n = 40, 57%) had a dedicated neuro ICU. Centers from relatively high- and middle-income countries more often indicated that they have a dedicated neuro ICU (n = 35, 61%) than centers from relatively lower-income countries (n = 5, 39%, p = .13, S1 Table). The large majority of centers had participated previously in research about acute cerebral disorders. Fifty-one (72%) centers were involved in more than five neurotrauma research applications over the past five years (see Table 2). 10.1371/journal.pone.0161367.t002Table 2 General structural characteristics of the participating centers (n = 71). Characteristic N completed N (%)* Academic hospital (vs. non-Academic) 71 65 (92%) Trauma center designation 71     - Level I 48 (68%)     - Level II 4 (6%)     - Level III 1 (1%)     - No designation / NA 18 (25%) Urban location (vs. suburban and rural location) 71 70 (99%) Helicopter platform 71 57 (80%) Acute trauma team 71 63 (89%) The availability of a dedicated neuro ICU 70 40 (57%) Number of ICUs (median, IQR) 69 3 (2–5) The availability of an in-hospital rehabilitation unit 70 36 (51%) Neurotrauma research applications in the past 5 y 71     - > 5 51 (72%)     - 3–5 13 (18%)     - 1–2 4 (6%)     - 0 or unknown 3 (4%) Distance nearest trauma center that receives patients with severe TBI (km, median, IQR) 52 56 (17–100) Note. ICU = Intensive care unit; IQR = Interquartile Range * Table presents number and percentage of centers unless otherwise specified The median number of beds in the participating centers was 1000 (IQR 682–1395) of which 31 (IQR 22–44) were ICU beds (see Table 3 and S1 Fig). Centers had a median of 3 (IQR 2–6) resuscitation rooms at the ED and 24 (IQR 16–39) operating rooms. Three (IQR 2–4) of these were potentially available for TBI patients. The median number of annual ED visits was 53,428 (IQR 30,002–90,268). The median number of annual ICU admission was 1240 (IRQ 560–2019), of which 91 (IQR 52–160) were TBI patients. 10.1371/journal.pone.0161367.t003Table 3 Volume characteristics of the participating centers (n = 71). Characteristic N completed Median (IQR) Number of beds Number of ED observational beds 69 16 (7–32) Number of hospital beds 69 1000 (682–1395) Number of ICU beds 71 31 (22–44) Number of resuscitation and operating rooms Number of resuscitating rooms 69 3 (2–6) Number of operating rooms 70 24 (16–39) Number of operating rooms potentially available for TBI patientsA 69 3 (2–4) Number of patients Annual ED visits 63 53,428 (30,002–90,268) Annual ICU admissions 65 1240 (560–2019) Number of TBI patients Annual number of TBI patients at the ICU 63 91 (52–160) Annual neurosurgical procedures to evacuate contusion 59 9 (4–21) Annual decompressive craniectomies 56 13 (8–22) Note. IQR = interquartile range; ED = emergency department; ICU = intensive care unit; TBI = traumatic brain injury; SAH = subarachnoid hemorrhage A Operating rooms potentially available for TBI patients are the operating rooms that can be used for emergency and non-emergency TBI patients (e.g. trauma operating rooms, neurosurgical operating rooms etc). Rooms that are used for non-TBI surgery in TBI patients (e.g. orthopedic surgery in patients with multiple trauma) should be excluded here. Seventy-five per cent (n = 53) of the centers had separate 24/7 emergency operation rooms. The majority of centers indicated that they had an electronic patient system at the ward (n = 57, 80%) and the ICU (n = 56, 79%). There was variation in the organization of the ICU in the participating centers; i.e. 45 (64%) centers had a closed ICU organization, 3 (4%) an open ICU organization and the remainder (n = 22, 32%) a mixed ICU organization. Centers from relatively high- and middle-income countries more often reported that they had a closed ICU structure (n = 40, 70%) compared to centers from relatively lower-income countries (n = 5, 39%). Step down beds were available in 71% (n = 50) of the centers. Centers from North and West Europe more often reported that they had a step down bed facility than centers from South and East Europe and Israel (n = 36, 80% vs. n = 14, 56%, p = .03, S1 Table). Maximum laboratorium turnaround times, the possibility for in-hospital coma stimulation and the location of TBI relevant facilities also varied widely among the included centers (see Table 4). 10.1371/journal.pone.0161367.t004Table 4 Hospital facilities of the participating centers (n = 71). Characteristic N completed N (%) General Separate 24/7 emergency operation rooms 71 53 (75%) Electronic patient system     - Ward 71 57 (80%)     - ICU 71 56 (79%) Facility for overnight observation 69 54 (78%) Lab turnaround time A 68     - 0-30minutes 25 (36%)     - >30 minutes 26 (38%)     - NA. No lab SOP at the ED 17 (25%) Organization of the ICU 70     - Closed 45 (64%)     - Open 3 (4%)     - Mixed 22 (32%) Step down beds 70 50 (71%) In-hospital coma stimulation 70 34 (49%) TBI related Location TBI facilities 71     - Different buildings 20 (28%)     - Same building, different floors 45 (63%)     - Same building, same floors 6 (9%) Note. ICU = intensive care unit; NA = not applicable; SOP = Standard Operating Procedures; TBI = traumatic brain injury A The laboratory turnaround times that are record in the lab Standard Operating Procedures (SOP) at the emergency department for severely injured patients On average 14 neurologists, 10 neurosurgeons, 17 intensivists, 4 trauma surgeons and 10 ED physicians were working in the centers (see Table 5). Nearly all centers (n = 69, 97%) had at least one residency program for trainees towards becoming a specialist. The specialist most often in charge of TBI patients at respectively the ED, ward and ICU were predominately ED physicians, neurosurgeons and intensivists. Most centers had 24/7 in-house availability of OR personnel (n = 62, 87%) and CT technicians (n = 66, 93%). Median intensivist-to-patient ratio, and ICU nurse-to-patient ratio were 1: 5 (IQR 1:3 to 1:8) and 1:2 (IQR 1:1 to 1:3). Night coverage at the ICU was performed by a certified intensivist in two-third of the centers (n = 44, 65%) and by a trainee or fellow in the remainder of centers. Almost all centers from the relatively lower-income countries (n = 12, 92%) reported that night coverage was performed by a certified intensivist, in comparison to 58% of the centers from the relatively high- and middle-income countries. Also, more centers from South and East Europe (n = 22, 88%) had night coverage by a certified intensivist, compared to centers from North and West Europe (n = 22, 51%, S1 Table). 10.1371/journal.pone.0161367.t005Table 5 Staffing characteristics of the participating centers (n = 71). Characteristic N completed N (%)* Number of specialists (median, IQR) A     -    Neurologist 71 14 (8–21)     -    Neurosurgeon 68 10 (7–13)     -    Intensivist 68 17 (10–28)     -    Trauma surgeon 68 4 (0–10)     -    ED physician 69 10 (3–19) Residency programs     -    Neurologist 70 65 (93%)     -    Neurosurgeon 71 67 (94%)     -    Intensivist 71 64 (90%)     -    Trauma surgeon 71 36 (51%) Availability OR personnel 71     -    24/7 in-house availability 62 (87%)     -    On call within 30 minutes 9 (13%) Availability CT technicians 71     -    24/7 in-house availability 66 (93%)     -    On call within 30 minutes 5 (7%) Intensivist-to-patient ratio (median, IQR) 69 1: 5 (1: 3–1: 8) ICU nurse-to-patient ratio (median, IQR) 69 1: 2 (1: 1–1: 3) Night coverage ICU 68     -    Certified intensivist/ ICU physician 44 (65%)     -    Trainee (in residency training) 20 (29%)     -    Fellow in training for ICU 4 (6%) Note. IQR = interquartile range; ED = emergency department; OR = operating rooms; CT = computed tomography * Table presents number and percentage of centers unless otherwise specified A Number of specialists is displayed per 40-hour workweek. General process characteristics With regard to computed tomography (CT) scanning in patients with mild TBI at the ED, 79% of the centers (n = 54) indicated to use CT guidelines (see Table 6). In addition, seven centers (10%) from Austria, Denmark, France, Spain and Sweden routinely determine S100B as a prognostic biomarker for neurological deterioration at the ED. There was variation among centers in their ICU admission policy; i.e. 44 (64%) centers generally admit patients with moderate TBI (Glasgow Coma Scale (GCS) 9–12) and CT abnormalities to the ICU, while 25 (36%) centers only admit these patients to the ICU in the presence of other risk factors. This variation was also shown for moderate TBI patients without CT abnormalities and patients with mild TBI on anti-coagulant therapy. There was a trend towards a higher ICU admission rate in centers from relatively high- and middle-income countries than in centers from relatively lower-income countries (S2 Table). 10.1371/journal.pone.0161367.t006Table 6 General process information of the participating centers (n = 71). Characteristic N Completed N (%) Emergency department Use of CT scan guidelines at the ED 68 54 (79%) Routine use of S100B as prognostic biomarker at the ED 71 7 (10%) ICU admission policy Patients with moderate TBI (GCS 9–12) without CT abnormalities are admitted to the ICU 69     -    No or only in the presence of other risk factors 50 (72%)     -    General policy 19 (28%) Patients with moderate TBI (GCS 9–12) with CT abnormalities are admitted to the ICU 69     -    No or only in the presence of other risk factors 25 (36%)     -    General policy 44 (64%) Patients with mild TBI (GCS 13–15) using anti-coagulant therapy are admitted to the ICU 69     -    No or only in the presence of other risk factors 53 (77%)     -    General policy 16 (23%) ICP monitoring ICP monitoring is performed in patients with GCS<9 and CT abnormalities 67     -    No or only in the presence of other risk factors 6 (9%)     -    General policy 61 (91%) ICP monitoring is performed in patients with GCS<9 without CT abnormalities 67     -    No or only in the presence of other risk factors 52 (78%)     -    General policy 15 (22%) ICP monitoring is performed in patients with intraventricular hemorrhages 67     -    No or only in the presence of other risk factors 46 (69%)     -    General policy 21 (31%) ICP sensors that are used at the ICU: 67     -    Parenchymal 21 (31%)     -    Ventricular 6 (9%)     -    Both 40 (60%) Management of elevated ICP Threshold for medical management of elevated ICP 66     -    >15mmHg 3 (5%)     -    >20mmHg 57 (86%)     -    >25mmHg 6 (9%) Threshold for decompressive craniotomy in elevated ICP 61     -    >20mmHg 7 (12%)     -    >25mmHg 35 (57%)     -    >30mmHg 19 (31%) ICU policies Structural variation between (neuro)surgeons with regard to their decision to place an ICP sensor 69 33 (48%) General policy with regard to the management of extremity fractures in patients with sTBI 68     -    Damage control 58 (85%)     -    Definitive care 10 (15%) Note. CT = computed tomography; ED = emergency department; ICU = intensive care unit; ICP = intracranial pressure; BTF = Brain Trauma Foundation; GCS = Glasgow Coma Scale; sTBI = severe traumatic brain injury The large majority of participants (n = 61, 91%) indicated that their general policy is to insert intracranial pressure (ICP) monitors in patients with GCS <9 and CT abnormalities. However, centers vary in whether they would place an ICP monitor in patients with GCS <9 without CT abnormalities and patients with intraventricular haemorrhages. Variation in ICP monitoring is also reported within the centers, since half of the centers indicated that there is structural variation between (neuro)surgeons in their center with regard to the decision to place an ICP monitor. The threshold for medical management of elevated ICP was 20 mmHg in the large majority of centers (n = 57, 87%). However, centers varied widely in their threshold for decompressive craniotomy; i.e. in 12% (n = 7) the threshold was 20 mmHg, in 57% (n = 35) the threshold was 25 mmHg and in 31% (n = 19) the threshold was 30 mmHg. Insurance and payment systems In the majority of countries (n = 16, 80%), a health care insurance was compulsory for all inhabitants. In 45% of the countries (n = 9), patients nevertheless had to pay a part of the delivered care themselves via either a co-payment (5 countries) or a deductible (4 countries). Most centers were funded by the government (n = 60; 85%). Centers typically got reimbursed by all-in amounts per patient rather than by payment for individual interventions. Most doctors received a fixed monthly salary (n = 58, 82%). In 11% (n = 8) of the centers, doctors received an additional fee for services. Twenty-three (32%) centers received additional payment for the treatment of privately insured patients. Discussion We found considerable variation in general structure and process characteristics among 71 specialized neurotrauma centers participating in the CENTER-TBI study. Most of these centers were high-volume academic level I trauma centers situated in an urban location. Centers varied widely in their ICU organization, hospital facilities and admission- and treatment policies. The effectiveness of these structures and interventions can therefore adequately be studied with CER. Our provider profiling questionnaires have strengths and limitations. One of the strengths is the comprehensive development process, which consisted of several stages and involved many experts. As a consequence, the questionnaires address all aspects relevant to TBI care. Secondly, local investigators were extensively informed about the aim, procedures and practical issues during presentations, workshops and emails. This might explain the 100% response rate. The length of our questionnaires can be regarded as a limitation. Long questionnaires have been associated with lower data quality [14, 15], an effect that is often due to fatigue and boredom [15]. Since the questionnaires could be spread over time and over different persons, the negative effect of length was however confined. Another limitation of our study concerns the generalizability of our findings. The included centers comprise a group of neurotrauma centers participating in a European multicenter study. Our findings therefore cannot be generalized to all centers caring for neurotrauma patients in Europe. Furthermore, our study provides information on what centers reported rather than characteristics that were directly observed. Therefore, we cannot exclude that some of our findings provide a too optimistic picture. For example, almost all centers indicated that they would insert an ICP monitor in patients with severe TBI and CT abnormalities, which is recommended by Brain Trauma Foundation guidelines. However, a systematic review about guideline adherence reported that ICP monitoring guidelines were only followed in one-third of the patients [5]. Later, results from the ongoing CENTER-TBI study will provide insight into discrepancies between reported and actual policies in the participating centers. The concordance rate between duplicate questions (median: 0.85), indicates a certain degree of subjectivity in the responses. The concordance rate was especially low for process questions, which indicates that there might be differences in policy among wards and doctors, no clear policy at all or difficulties in understanding and interpreting the questions. It might also indicate that some of the doctors that completed the questionnaire might not be representative of their department or center. Although our concordance rate was very similar to a 2001 survey study among European countries [11], results on process characteristics should be interpreted with caution. The reported concordance rate does not account for chance concordance since no statistical measures are available that do account for chance and can also provide one figure for different outcomes (dichotomous, categorical and continuous) that we had in our questionnaire. When interpreting the concordance rate, it should however be acknowledged that some answers might be similar by chance. Finally, there were only 13 centers from a relatively lower-income country and 25 centers from South and East Europe (including Israel). We therefore had limited power to detect differences between centers from relatively high-and middle-income countries versus centers from relatively lower-income countries and centers from different geographic locations. Although we studied a sample of highly specialized centers, we found substantial differences in important structural and process characteristics. Largest differences were seen in the specialization and organization of the ICU, i.e. half of the centers indicated to have a dedicated neuro ICU and 64% indicated to have a closed ICU organization. Additionally, rehabilitation facilities varied widely, with half of the centers having an in-hospital rehabilitation unit and the possibility for coma stimulation. We also found large differences in the reported policies regarding ICU admission and ICP monitoring across centers. The variation in structure and process among specialized neurotrauma centers was in line with previous survey studies [11, 12]. Enblad and associates [11] included European centers with a particular interest in neuro ICU and brain monitoring in their survey study. They also found large between-center differences in structures of care (e.g. 76% had a separate NICU, 50% had a neurosurgeon as ICU director). Checkley and associates [12] reported similar findings. They conducted a survey in 69 centers participating in the United States critical illness and injury outcome study. The majority of their centers were teaching hospitals with critical care training. However, 58% of their centers had a closed ICU organization and their annual hospital admission rate ranged from 1,170 to 56,330, indicating large between-center differences in volume. Also there were large differences in the protocols available at their surveyed ICUs. Although in this study we only reported on general structure and process characteristics, it is clear that the between-center variation is substantial and provides an opportunity for CER. Variation among centers and countries comprises an important prerequisite for CER and enables between-center and between-country comparisons of effective structures and processes of care. We can for example study the influence of a dedicated neuro ICU on outcome in severe TBI patients by studying patients’ outcome in the 40 centers with a dedicated neuro ICU and in the 30 centers without a dedicated neuro ICU. This requires outcome data on patient level, which are currently collected in the CENTER-TBI study. In such a comparison it is important to correct for differences in other structural and process characteristics between these centers, which can potentially be accomplished with advanced statistical modelling. Other potential interesting topics for CER based on the current study include the effectiveness of an in-hospital rehabilitation unit, the effectiveness of high-volume vs. low-volume hospitals, the effectiveness of closed vs. mixed ICU organization, and the effectiveness of admission- and ICP monitoring policies. Conclusion Even among high-volume, specialized neurotrauma centers there is substantial variation in structures and processes of TBI care. This variation provides an opportunity to study effectiveness of specific aspects of TBI care and to identify best practices with CER approaches. Supporting Information S1 Fig Distribution of number of beds. (PDF) Click here for additional data file. S1 File The Provider Profiling Questionnaires. (PDF) Click here for additional data file. S2 File Definitions. Note. Table presents all definitions used in the paper in the order that they are used in the results section of the paper. TBI = traumatic brain injury; ICU = intensive care unit (PDF) Click here for additional data file. S1 Table Structural characteristics that show substantial variation among the participating centers. A P-value for the difference between high/middle and low income countries B P-value for the difference between North-West and South-East Europe and Israel (PDF) Click here for additional data file. S2 Table Process characteristics that show substantial variation among the participating centers. A P-value for the difference between high/middle and low income countries B P-value for the difference between North-West and South-East Europe and Israel (PDF) Click here for additional data file. The authors would like to thank all CENTER-TBI investigators and their staff, who are listed below, for completing the provider profiling questionnaires. Authors would further like to thank Nada Andelic, Sasha Brazinova, Ruben van der Brande, Peter Cameron, Guiseppe Citerio, Ari Ercole, Thomas van Essen, Mathieu van der Jagt, Erwin Kompanje, Fiona Lecky, Joukje van der Naalt, David Nelson, Wilco Peul, Jukka Ranta, Cecilia Roe, Gerard Ribbers, Nino Stochetti, Olli Tenovuo and Lindsay Wilson for their help with the development of the provider profiling questionnaires. CENTER-TBI investigators and participants Principal Investigators and contact information: Professor A.I. Maas: Andrew.Maas@uza.be Professor D. Menon: dkm13@wbic.cam.ac.uk Adams Hadie 1, Alessandro Masala 2, Allanson Judith 3, Amrein Krisztina 4, Andaluz Norberto 5, Andelic Nada 6, Andrea Nanni 2, Andreassen Lasse 7, Anke Audny 8, Antoni Anna 9, Ardon Hilko 10, Audibert Gérard 11, Auslands Kaspars 12, Azouvi Philippe 13, Baciu Camelia 14, Bacon Andrew 15, Badenes Rafael 16, Baglin Trevor 17, Bartels Ronald 18, Barzó Pál 19, Bauerfeind Ursula 20, Beer Ronny 21, Belda Francisco Javier 16, Bellander Bo-Michael 22, Belli Antonio 23, Bellier Rémy 24, Benali Habib 25, Benard Thierry 24, Berardino Maurizio 26, Beretta Luigi 27, Beynon Christopher 28, Bilotta Federico 16, Binder Harald 9, Biqiri Erta 14, Blaabjerg Morten 29, Borgen Lund Stine 30, Bouzat Pierre 31, Bragge Peter 32, Brazinova Alexandra 33, Brehar Felix 34, Brorsson Camilla 35, Buki Andras 36, Bullinger Monika 37, Bučková Veronika 33, Calappi Emiliana 38, Cameron Peter 39, Carbayo Lozano Guillermo 40, Carise Elsa 24, Carpenter K. 41, Castaño-León Ana M. 42, Causin Francesco 43, Chevallard Giorgio 14, Chieregato Arturo 14, Citerio Giuseppe 44, 45, Cnossen Maryse 46, Coburn Mark Coburn 47, Coles Jonathan 48, Cooper Jamie D. 49, Correia Marta 50, Covic Amra 51, Curry Nicola 52, Czeiter Endre 53, Czosnyka Marek 54, Dahyot-Fizelier Claire 24, Damas François 55, Damas Pierre 56, Dawes Helen 57, De Keyser Véronique 58, Della Corte Francesco 59, Depreitere Bart 60, Ding Shenghao 61, Dippel Diederik 62, Dizdarevic Kemal 63, Dulière Guy-Loup 55, Dzeko Adelaida 64, Eapen George 15, Engemann Heiko 51, Ercole Ari 65, Esser Patrick 57, Ezer Erzsébet 66, Fabricius Martin 67, Feigin Valery L. 68, Feng Junfeng 61, Foks Kelly 62, Fossi Francesca 14, Francony Gilles 31, Frantzén Janek 69, Freo Ulderico 70, Frisvold Shirin 71, Furmanov Alex 72, Gagliardo Pablo 73, Galanaud Damien 25, Gao Guoyi 74, Geleijns Karin 41, Ghuysen Alexandre 75, Giraud Benoit 24, Glocker Ben 76, Gomez Pedro A. 42, Grossi Francesca 59, Gruen Russell L. 77, Gupta Deepak 78, Haagsma Juanita A. 46, Hadzic Ermin 64, Haitsma Iain 79, Hartings Jed A. 80, Helbok Raimund 21, Helseth Eirik 81, Hertle Daniel 28, Hill Sean 82, Hoedemaekers Astrid 83, Hoefer Stefan 51, Hutchinson Peter J. 1, Håberg Asta Kristine 84, Jacobs Bram 85, Janciak Ivan 86, Janssens Koen 58, Jiang Ji-yao 74, Jones Kelly 87, Kalala Jean-Pierre 88, Kamnitsas Konstantinos 76, Karan Mladen 89, Karau Jana 20, Katila Ari 69, Kaukonen Maija 90, Keeling David 52, Kerforne Thomas 24, Ketharanathan Naomi 41, Kettunen Johannes 91, Kivisaari Riku 90, Kolias Angelos G. 1, Kolumbán Bálint 92, Kompanje Erwin 93, Kondziella Daniel 67, Koskinen Lars-Owe 35, Kovács Noémi 92, Kálovits Ferenc 94, Lagares Alfonso 42, Lanyon Linda 82, Laureys Steven 95, Lauritzen Martin 67, Lecky Fiona 96, Ledig Christian 76, Lefering Rolf 97, Legrand Valerie 98, Lei Jin 61, Levi Leon 99, Lightfoot Roger 100, Lingsma Hester 46, Loeckx Dirk 101, Lozano Angels 16, Luddington Roger 17, Luijten-Arts Chantal 83, Maas Andrew I.R. 58, MacDonald Stephen 17, MacFayden Charles 65, Maegele Marc 102, Majdan Marek 33, Major Sebastian 103, Manara Alex 104, Manhes Pauline 31, Manley Geoffrey 105, Martin Didier 106, Martino Costanza 2, Maruenda Armando 16, Maréchal Hugues 55, Mastelova Dagmara 86, Mattern Julia 28, McMahon Catherine 107, Melegh Béla 108, Menon David 65, Menovsky Tomas 58, Morganti-Kossmann Cristina 109, Mulazzi Davide 38, Mutschler Manuel 102, Mühlan Holger 110, Negru Ancuta 111, Nelson David 82, Neugebauer Eddy 102, Newcombe Virginia 65, Noirhomme Quentin 95, Nyirádi József 4, Oddo Mauro 112, Oldenbeuving Annemarie 113, Oresic Matej 114, Ortolano Fabrizio 38, Palotie Aarno 91, 115, 116, Parizel Paul M. 117, Patruno Adriana 118, Payen Jean-François 31, Perera Natascha 119, Perlbarg Vincent 25, Persona Paolo 120, Peul Wilco 121, Pichon Nicolas 122, Piilgaard Henning 67, Piippo Anna 90, Pili Floury Sébastien 123, Pirinen Matti 91, Ples Horia 111, Polinder Suzanne 46, Pomposo Inigo 40, Psota Marek 33, Pullens Pim 117, Puybasset Louis 124, Ragauskas Arminas 125, Raj Rahul 90, Rambadagalla Malinka 126, Rehorčíková Veronika 33, Rhodes Jonathan 127, Richardson Sylvia 128, Ripatti Samuli 91, Rocka Saulius 125, Rodier Nicolas 122, Roe Cecilie 129, Roise Olav 130, Roks Gerwin 131, Romegoux Pauline 31, Rosand Jonathan 132, Rosenfeld Jeffrey 109, Rosenlund Christina 133, Rosenthal Guy 72, Rossaint Rolf 47, Rossi Sandra 120, Rostalski Tim 110, Rueckert Daniel 76, Ruiz de Arcaute Felix 101, Rusnák Martin 86, Sacchi Marco 14, Sahakian Barbara 65, Sahuquillo Juan 134, Sakowitz Oliver 135, 136, Sala Francesca 118, Sanchez-Pena Paola 25, Sanchez-Porras Renan 28, 135, Sandor Janos 137, Santos Edgar 28, Sasse Nadine 51, Sasu Luminita 59, Savo Davide 118, Schipper Inger 138, Schlößer Barbara 20, Schmidt Silke 110, Schneider Annette 97, Schoechl Herbert 139, Schoonman Guus 131, Schou Rico Frederik 140, Schwendenwein Elisabeth 9, Schöll Michael 28, Sir Özcan 141, Skandsen Toril 142, Smakman Lidwien 143, Smeets Dirk 101, Smielewski Peter 54, Sorinola Abayomi 144, Stamatakis Emmanuel 65, Stanworth Simon 52, Stegemann Katrin 110, Steinbüchel Nicole 145, Stevens Robert 146, Stewart William 147, Steyerberg Ewout W. 46, Stocchetti Nino 148, Sundström Nina 35, Synnot Anneliese 149, 150, Szabó József 94, Söderberg Jeannette 82, Taccone Fabio Silvio 16, Tamás Viktória 144, Tanskanen Päivi 90, Tascu Alexandru 34, Taylor Mark Steven 33, Te Ao Braden 68, Tenovuo Olli 69, Teodorani Guido 151, Theadom Alice 68, Thomas Matt 104, Tibboel Dick 41, Tolias Christos 152, Tshibanda Jean-Flory Luaba 153, Tudora Cristina Maria 111, Vajkoczy Peter 154, Valeinis Egils 155, Van Hecke Wim 101, Van Praag Dominique 58, Van Roost Dirk 88, Van Vlierberghe Eline 101, Vande Vyvere Thijs 101, Vanhaudenhuyse Audrey 25, 95, Vargiolu Alessia 118, Vega Emmanuel 156, Verheyden Jan 101, Vespa Paul M. 157, Vik Anne 158, Vilcinis Rimantas 159, Vizzino Giacinta 14, Vleggeert-Lankamp Carmen 143, Volovici Victor 79, Vulekovic Peter 89, Vámos Zoltán 66, Wade Derick 57, Wang Kevin K.W. 160, Wang Lei 61, Wildschut Eno 41, Williams Guy 65, Willumsen Lisette 67, Wilson Adam 5, Wilson Lindsay 161, Winkler Maren K.L. 103, Ylén Peter 162, Younsi Alexander 28, Zaaroor Menashe 99, Zhang Zhiqun 163, Zheng Zelong 28, Zumbo Fabrizio 2, de Lange Stefanie 97, de Ruiter Godard C.W. 143, den Boogert Hugo 18, van Dijck Jeroen 164, van Essen Thomas A. 121, van Heugten Caroline 57, van der Jagt Mathieu 165, van der Naalt Joukje 85 1 Division of Neurosurgery, Department of Clinical Neurosciences, Addenbrooke’s Hospital & University of Cambridge, Cambridge, UK 2 Department of Anesthesia & Intensive Care,M. Bufalini Hospital, Cesena, Italy 3 Department of Clinical Neurosciences, Addenbrooke’s Hospital & University of Cambridge, Cambridge, UK 4 János Szentágothai Research Centre, University of Pécs, Pécs, Hungary 5 University of Cincinnati, Cincinnati, Ohio, United States 6 Division of Surgery and Clinical Neuroscience, Department of Physical Medicine and Rehabilitation, Oslo University Hospital and University of Oslo, Oslo, Norway 7 Department of Neurosurgery, University Hospital Northern Norway, Tromso, Norway 8 Department of Physical Medicine and Rehabilitation, University hospital Northern Norway 9 Trauma Surgery, Medical University Vienna, Vienna, Austria 10 Department of Neurosurgery, Elisabeth-Tweesteden Ziekenhuis, Tilburg, the Netherlands 11 Department of Anesthesiology & Intensive Care, University Hospital Nancy, Nancy, France 12 Riga Eastern Clinical University Hospital, Riga, Latvia 13 Raymond Poincare hospital, Assistance Publique–Hopitaux de Paris, Paris, France 14 NeuroIntensive Care, Niguarda Hospital 15 Neurointensive Care, Sheffield Teaching Hospitals NHS Foundation Trust, Sheffield, UK 16 Department Anesthesiology and Surgical-Trauma Intensive Care, Hospital Clinic Universitari de Valencia, Spain 17 Cambridge University Hospitals, Cambridge, UK 18 Department of Neurosurgery, Radboud University Medical Center 19 Department of Neurosurgery, University of Szeged, Szeged, Hungary 20 Institute for Transfusion Medicine (ITM), Witten/Herdecke University, Cologne, Germany 21 Department of Neurocritical care, Innsbruck Medical University, Innsbruck, Austria 22 Deparment of Neurosurgery & Anesthesia & intensive care medicine, Karolinska University Hospital, Stockholm, Sweden 23 NIHR Surgical Reconstruction and Microbiology Research Centre, Birmingham, UK 24 Intensive care Unit, CHU Poitiers, Poitiers, France 25 Anesthesie-Réanimation, Assistance Publique–Hopitaux de Paris, Paris, France 26 Department of Anesthesia & ICU, AOU Città della Salute e della Scienza di Torino—Orthopedic and Trauma Center, Torino, Italy 27 Department of Anesthesiology & Intensive Care, S Raffaele University Hospital, Milan, Italy 28 Department of Neurosurgery, University Hospital Heidelberg, Heidelberg, Germany 29 Department of Neurology, Odense University Hospital, Odense, denmark 30 Departments of Neuroscience and Nursing Science, Norwegian University of Science and Technology, Trondheim, Norway 31 Department of Anesthesiology & Intensive Care, University Hospital of Grenoble, Grenoble, France 32 BehaviourWorks Australia, Monash Sustainability Institute, Monash University, Victoria, Australia 33 Department of Public Health, Faculty of Health Sciences and Social Work, Trnava University, Trnava, Slovakia 34 Department of Neurosurgery, Bagdasar-Arseni Emergency Clinical Hospital, Bucharest, Romania 35 Department of Neurosurgery, Umea University Hospital, Umea, Sweden 36 Department of Neurosurgery, University of Pecs and MTA-PTE Clinical Neuroscience MR Research Group and Janos Szentagothai Research Centre, University of Pecs, Hungarian Brain Research Program, Pecs, Hungary 37 Department of Medical Psychology, Universitätsklinikum Hamburg-Eppendorf, Hamburg, Germany 38 Neuro ICU, Fondazione IRCCS Cà Granda Ospedale Maggiore Policlinico, Milan, Italy 39 Department of Epidemiology and Preventive Medicine, Monash University, Melbourne, Victoria, Australia 40 Department of Neurosurgery, Hospital of Cruces, Bilbao, Spain 41 Intensive Care and Department of Pediatric Surgery, Erasmus Medical Center, Sophia Children’s Hospital, Rotterdam, The Netherlands 42 Department of Neurosurgery, Hospital Universitario 12 de Octubre, Madrid, Spain 43 Department of Neuroscience, Azienda Ospedaliera Università di Padova, Padova, Italy 44 NeuroIntensive Care, Azienda Ospedaliera San Gerardo di Monza, Monza, Italy 45 School of Medicine and Surgery, Università Milano Bicocca, Milano, Italy 46 Department of Public Health, Erasmus Medical Center-University Medical Center, Rotterdam, The Netherlands 47 Department of Anaesthesiology, University Hospital of Aachen, Aachen, Germany 48 Department of Anesthesia & Neurointensive Care, Cambridge Universiyt Hospital NHS Foundation Trust, Cambridge, UK 49 School of Public Health & PM, Monash University and The Alfred Hospital, Melbourne, Victoria, Australia 50 Radiology/MRI department, MRC Cognition and Brain Sciences Unit, Cambridge, UK 51 Institute of Medical Psycholology and Medical Sociology, Universitätsmedizin Göttingen, Göttingen, Germany 52 Oxford University Hospitals NHS Trust, Oxford, UK 53 Department of Neurosurgery, University of Pecs and MTA-PTE Clinical Neuroscience MR Research Group and Janos Szentagothai Research Centre, University of Pecs, Hungarian Brain Research Program (Grant No. KTIA 13 NAP-A-II/8), Pecs, Hungary 54 Brain Physics Lab, Division of Neurosurgery, Dept of Clinical Neurosciences, University of Cambridge, Addenbrooke’s Hospital, Cambridge, UK 55 Intensive Care Unit, CHR Citadelle, Liège, Belgium 56 Intensive Care Unit, CHU, Liège, Belgium 57 Movement Science Group, Faculty of Health and Life Sciences, Oxford Brookes University, Oxford, UK 58 Department of Neurosurgery, Antwerp University Hospital and University of Antwerp, Edegem, Belgium 59 Department of Anesthesia & Intensive Care, Maggiore Della Carità Hospital, Novara, Italy 60 Department of Neurosurgery, University Hospitals Leuven, Leuven, Belgium 61 Department of Neurosurgery, Renji Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, China 62 Department of Neurology, Erasmus MC, Rotterdam, the Netherlands 63 Department of Neurosurgery, Medical Faculty and clinical center University of Sarajevo, Sarajevo, Bosnia Herzegovina 64 Department of Neurosurgery, Regional Medical Center dr Safet Mujić, Mostar, Bosnia Herzegovina 65 Division of Anaesthesia, University of Cambridge, Addenbrooke’s Hospital, Cambridge, UK 66 Department of Anaesthesiology and Intensive Therapy, University of Pécs, Pécs, Hungary 67 Departments of Neurology, Clinical Neurophysiology and Neuroanesthesiology, Region Hovedstaden Rigshospitalet, Copenhagen, Denmark 68 National Institute for Stroke and Applied Neurosciences, Faculty of Health and Environmental Studies, Auckland University of Technology, Auckland, New Zealand 69 Rehabilitation and Brain Trauma, Turku University Central Hospital and University of Turku, Turku, Finland 70 Department of Medicine, Azienda Ospedaliera Università di Padova, Padova, Italy 71 Department of Anesthesiology and Intensive care, University Hospital Northern Norway, Tromso, Norway 72 Department of Neurosurgery, Hadassah-hebrew University Medical center, Jerusalem, Israel 73 Fundación Instituto Valenciano de Neurorrehabilitación (FIVAN), Valencia, Spain 74 Department of Neurosurgery, Shanghai Renji hospital, Shanghai Jiaotong University/school of medicine, Shanghai, China 75 Emergency Department, CHU, Liège, Belgium 76 Department of Computing, Imperial College London, London, UK 77 Lee Kong Chian School of Medicine, Nanyang Technological University, Singapore; and Monash University, Australia 78 Department of Neurosurgery, Neurosciences Centre & JPN Apex trauma centre, All India Institute of Medical Sciences, New Delhi-110029, India 79 Department of Neurosurgery, Erasmus MC, Rotterdam, the Netherlands 80 Department of Neurosurgery, University of Cincinnati, Cincinnati, Ohio, USA 81 Department of Neurosurgery, Oslo University Hospital, Oslo, Norway 82 Karolinska Institutet, INCF International Neuroinformatics Coordinating Facility, Stockholm, Sweden 83 Department of Intensive Care Medicine, Radboud University Medical Center 84 Department of Medical Imaging, St. Olavs Hospital and Department of Neuroscience, Norwegian University of Science and Technology, Trondheim, Norway 85 Department of Neurology, University Medical Center Groningen, Groningen, Netherlands 86 International Neurotrauma Research Organisation, Vienna, Austria 87 National Institute for Stroke & Applied Neurosciences of the AUT University, Auckland, New Zealand 88 Department of Neurosurgery, UZ Gent, Gent, Belgium 89 Department of Neurosurgery, Clinical centre of Vojvodina, Novi Sad, Serbia 90 Helsinki University Central Hospital 91 Institute for Molecular Medicine Finland, University of Helsinki, Helsinki, Finland 92 Hungarian Brain Research Program—Grant No. KTIA 13 NAP-A-II/8, University of Pécs, Pécs, Hungary 93 Department of Intensive Care and Department of Ethics and Philosophy of Medicine, Erasmus Medical Center, Rotterdam, The Netherlands 94 Department of Neurological & Spinal Surgery, Markusovszky University Teaching Hospital, Szombathely, Hungary 95 Cyclotron Research Center, University of Liège, Liège, Belgium 96 Emergency Medicine Research in Sheffield, Health Services Research Section, School of Health and Related Research (ScHARR), University of Sheffield, Sheffield, UK 97 Institute of Research in Operative Medicine (IFOM), Witten/Herdecke University, Cologne, Germany 98 VP Global Project Management CNS, ICON, Paris, France 99 Department of Neurosurgery, Rambam Medical Center, Haifa, Israel 100 Department of Anesthesiology & Intensive Care, University Hospitals Southhampton NHS Trust, Southhampton, UK 101 icoMetrix NV, Leuven, Belgium 102 Cologne-Merheim Medical Center (CMMC), Department of Traumatology, Orthopedic Surgery and Sportmedicine, Witten/Herdecke University, Cologne, Germany 103 Centrum für Schlaganfallforschung, Charité–Universitätsmedizin Berlin, Berlin, Germany 104 Intensive Care Unit, Southmead Hospital, Bristol, Bristol, UK 105 Department of Neurological Surgery, University of California, San Francisco, California, USA 106 Department of Neurosurgery, CHU, Liège, Belgium 107 Department of Neurosurgery, The Walton centre NHS Foundation Trust, Liverpool, UK 108 Department of Medical Genetics, University of Pécs, Pécs, Hungary 109 National Trauma Research Institute, The Alfred Hospital, Monash University, Melbourne, Victoria, Australia 110 Department Health and Prevention, University Greifswald, Greifswald, Germany 111 Department of Neurosurgery, Emergency County Hospital Timisoara, Timisoara, Romania 112 Centre Hospitalier Universitaire Vaudois 113 Department of Intensive Care, Elisabeth-Tweesteden Ziekenhuis, Tilburg, the Netherlands 114 Department of Systems Medicine, Steno Diabetes Center, Gentofte, Denmark 115 Analytic and Translational Genetics Unit, Department of Medicine; Psychiatric & Neurodevelopmental Genetics Unit, Department of Psychiatry; Department of Neurology, Massachusetts General Hospital, Boston, MA, USA 116 Program in Medical and Population Genetics; The Stanley Center for Psychiatric Research, The Broad Institute of MIT and Harvard, Cambridge, MA, USA 117 Department of Radiology, Antwerp University Hospital and University of Antwerp, Edegem, Belgium 118 NeuroIntenisve Care Unit, Department of Anesthesia & Intensive Care Azienda Ospedaliera San Gerardo di Monza, Monza, Italy 119 International Projects Management, ARTTIC, Munchen, Germany 120 Department of Anesthesia & Intensive Care, Azienda Ospedaliera Università di Padova, Padova, Italy 121 Dept. of Neurosurgery, Leiden University Medical Center, Leiden, The Netherlands and Dept. of Neurosurgery, Medical Center Haaglanden, The Hague, The Netherlands 122 Intensive Care Unit, CHU Dupuytren, Limoges, France 123 Intensive Care Unit, CHRU de Besançon, Besançon, France 124 Department of Anesthesiology and Critical Care, Pitié -Salpêtrière Teaching Hospital, Assistance Publique, Hôpitaux de Paris and University Pierre et Marie Curie, Paris, France 125 Department of Neurosurgery, Kaunas University of technology and Vilnius University, Vilnius, Lithuania 126 Rezekne Hospital, Latvia 127 Department of Anaesthesia, Critical Care & Pain MedicineNHS Lothian & University of Edinburg, Edinburgh, UK 128 Director, MRC Biostatistics Unit, Cambridge Institute of Public Health, Cambridge, UK 129 Department of Physical Medicine and Rehabilitation, Oslo University Hospital/University of Oslo, Oslo, Norway 130 Division of Surgery and Clinical Neuroscience, Oslo University Hospital, Oslo, Norway 131 Department of Neurology, Elisabeth-TweeSteden Ziekenhuis, Tilburg, the Netherlands 132 Broad Institute, Cambridge MA Harvard Medical School, Boston MA, Massachusetts General Hospital, Boston MA, USA 133 Department of Neurosurgery, Odense University Hospital, Odense, Denmark 134 Department of Neurosurgery, Vall d'Hebron University Hospital, Barcelona, Spain 135 Klinik für Neurochirurgie, Klinikum Ludwigsburg, Ludwigsburg, Germany 136 University Hospital Heidelberg, Heidelberg, Germany 137 Division of Biostatistics and Epidemiology, Department of Preventive Medicine, University of Debrecen, Debrecen, Hungary 138 Department of Traumasurgery, Leiden University Medical Center, Leiden, The Netherlands 139 Department of Anaesthesiology and Intensive Care, AUVA Trauma Hospital, Salzburg, Austria 140 Department of Neuroanesthesia and Neurointensive Care, Odense University Hospital, Odense, Denmark 141 Department of Emergency Care Medicine, Radboud University Medical Center 142 Department of Physical Medicine and Rehabilitation, St.Olavs Hospital and and Department of Neuroscience, Norwegian University of Science and Technology, Trondheim, Norway 143 Neurosurgical Cooperative Holland, Department of Neurosurgery, Leiden University Medical Center and Medical Center Haaglanden, Leiden and The Hague, The Netherlands 144 Department of Neurosurgery, University of Pécs, Pécs, Hungary 145 Universitätsmedizin Göttingen, Göttingen, Germany 146 Division of Neuroscience Critical Care, John Hopkins University School of Medicine, Baltimore, USA 147 Department of Neuropathology, Queen Elizabeth University Hospital and University of Glasgow, Glasgow, UK 148 Department of Pathophysiology and Transplantation, Milan University, and Neuroscience ICU, Fondazione IRCCS Cà Granda Ospedale Maggiore Policlinico, Milano, Italy 149 Australian & New Zealand Intensive Care Research Centre, Department of Epidemiology and Preventive Medicine, School of Public Health and Preventive Medicine, Monash University, Melbourne, Australia 150 Cochrane Consumers and Communication Review Group, Centre for Health Communication and Participation, School of Psychology and Public Health, La Trobe University, Melbourne, Australia 151 Department of Reahabilitation, M. Bufalini Hospital, Cesena, Italy 152 Department of Neurosurgery, Kings college London, London, UK 153 Radiology/MRI Department, CHU, Liège, Belgium 154 Neurologie, Neurochirurgie und Psychiatrie, Charité–Universitätsmedizin Berlin, Berlin, Germany 155 Pauls Stradins Clinical University Hospital, Riga, Latvia 156 Department of Anesthesiology-Intensive Care, Lille University Hospital, Lille, France 157 Director of Neurocritical Care, University of California, Los Angeles, USA 158 Department of Neurosurgery, St.Olavs Hospital and Department of Neuroscience, Norwegian University of Science and Technology, Trondheim, Norway 159 Department of Neurosurgery, Kaunas University of Health Sciences, Kaunas, Lithuania 160 Department of Psychiatry, University of Florida, Gainesville, Florida, USA 161 Division of Psychology, University of Stirling, Stirling, UK 162 VTT Technical Research Centre, Tampere, Finland 163 University of Florida, Gainesville, Florida, USA 164 Department of Neurosurgery, The HAGA Hospital, The Hague, The Netherlands 165 Department of Intensive Care, Erasmus MC, Rotterdam, the Netherlands ==== Refs References 1 Peeters W , van den Brande R , Polinder S , Brazinova A , Steyerberg EW , Lingsma HF , et al Epidemiology of traumatic brain injury in Europe . Acta Neurochir (Wien) . 2015 ;157 (10 ):1683 –96 . .26269030 2 Brazinova A , Rehorcikova V , Taylor MS , Buckova V , Majdan M , Psota M , et al Epidemiology of traumatic brain injury in Europe: a living systematic review . J Neurotrauma . 2015 .26537996 3 Maas AI , Menon DK , Steyerberg EW , Citerio G , Lecky F , Manley GT , et al Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI): A Prospective Longitudinal Observational Study . Neurosurgery . 2015 ;76 (1 ):67 –80 . 10.1227/NEU.0000000000000575 25525693 4 Ad Hoc Committee on Health Research Relating to Future Intervention Options. Investing in health research and development. Geneva: World Health Organization, 1996 TDR/Gen/96.1. 5 Cnossen MC , Scholten AC , Lingsma H , Synnot A , Tavender E , Gantner D , et al Adherence to guidelines in adult patients with traumatic brain injury: A living systematic review . J Neurotrauma . 2015 .26431625 6 Maas AI , Menon DK , Lingsma HF , Pineda JA , Sandel ME , Manley GT . Re-orientation of clinical research in traumatic brain injury: report of an international workshop on comparative effectiveness research . J Neurotrauma . 2012 ;29 (1 ):32 –46 . 10.1089/neu.2010.1599 21545277 7 Io Medicine . Initial National Priorities for Comparative Effectiveness Research . Washington, DC : National Academies Press ; 2009 . 8 Alali AS , Fowler RA , Mainprize TG , Scales DC , Kiss A , de Mestral C , et al Intracranial pressure monitoring in severe traumatic brain injury: results from the American College of Surgeons Trauma Quality Improvement Program . J Neurotrauma . 2013 ;30 (20 ):1737 –46 . 10.1089/neu.2012.2802 23731257 9 Bulger EM , Nathens AB , Rivara FP , Moore M , MacKenzie EJ , Jurkovich GJ , et al Management of severe head injury: institutional variations in care and effect on outcome . Crit Care Med . 2002 ;30 (8 ):1870 –6 . .12163808 10 European Brain Injury Consortium. EBIC Center Survey [cited 2015 October, 12th]. Available from: http://www.ebic.nl/survey. 11 Enblad P , Nilsson P , Chambers I , Citerio G , Fiddes H , Howells T , et al R3-survey of traumatic brain injury management in European Brain IT centres year 2001 . Intensive Care Med . 2004 ;30 (6 ):1058 –65 . .15024565 12 Checkley W , Martin GS , Brown SM , Chang SY , Dabbagh O , Fremont RD , et al Structure, process, and annual ICU mortality across 69 centers: United States Critical Illness and Injury Trials Group Critical Illness Outcomes Study . Crit Care Med . 2014 ;42 (2 ):344 –56 . 10.1097/CCM.0b013e3182a275d7 24145833 13 European Comission . Remuneration of researchers in the public and private sectors Brussels : European Communities ; 2007 . 14 Kalantar JS , Talley NJ . The effects of lottery incentive and length of questionnaire on health survey response rates: a randomized study . J Clin Epidemiol . 1999 ;52 (11 ):1117 –22 . .10527007 15 Galesic M , & Bosnjak M . Effects of Questionnaire Length on Participation and Indicatiors of Response Quality in a Web Survey . Oxford Journals . 2009 ;73 (2 ):349 –60 .

PMC005xxxxxx/PMC5003389.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 10.1371/journal.pone.0161568PONE-D-16-01863Research ArticleEarth SciencesGeologyPetrologySedimentEarth SciencesGeologySedimentary GeologySedimentEarth SciencesMarine and Aquatic SciencesBodies of WaterLagoonsEarth SciencesGeomorphologyTopographyLandformsBeachesEarth SciencesMarine and Aquatic SciencesOceanographyPaleooceanographyBiology and Life SciencesPaleontologyPaleooceanographyEarth SciencesPaleontologyPaleooceanographyEarth SciencesMarine and Aquatic SciencesMarine GeologyPhysical SciencesChemistryChemical ElementsStrontiumPeople and placesGeographical locationsNorth AmericaMexicoPhysical SciencesChemistryChemical CompoundsCarbonatesRe-Evaluating the Geological Evidence for Late Holocene Marine Incursion Events along the Guerrero Seismic Gap on the Pacific Coast of Mexico Re-Evaluating Geological Evidence of Marine Incursions on the Pacific Coast of MexicoBianchette Thomas A. *McCloskey Terrence A. Liu Kam-biu Department of Oceanography and Coastal Sciences, College of the Coast and Environment, Louisiana State University, Baton Rouge, Louisiana, United States of AmericaZhu Liping EditorInstitute of Tibetan Plateau Research Chinese Academy of Sciences, CHINACompeting Interests: The authors have declared that no competing interests exist. Conceptualization: TAB TAM KBL. Formal analysis: TAB TAM KBL. Funding acquisition: TAB KBL. Investigation: TAB TAM. Methodology: TAB TAM KBL. Resources: KBL. Supervision: TAM KBL. Visualization: TAB. Writing – original draft: TAB. Writing – review & editing: TAB TAM KBL. * E-mail: tbianc1@lsu.edu29 8 2016 2016 11 8 e016156815 1 2016 27 7 2016 This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.Despite the large number of tsunamis that impact Mexico’s Pacific coast, stratigraphic studies focusing on geological impacts are scanty, making it difficult to assess the long-term risks for this vulnerable region. Surface samples and six cores were taken from Laguna Mitla near Acapulco to examine sedimentological and geochemical evidence for marine incursion events. Sediment cores collected from behind the beach barrier are dominated by intercalated layers of peat and inorganic sediments, mostly silt and clay, with little or no sand. Sand- and shell-rich clastic layers with high levels of sulfur, calcium, and strontium only occur adjacent to the relict beach ridge remnants near the center of the lagoon. With the exception of one thin fine sand layer, the absence of sand in the near-shore cores and the predominance of the terrigenous element titanium in the inorganic layers, evidently eroded from the surrounding hillslopes, suggests that these large-grained intervals do not represent episodic marine incursions, but rather were likely formed by the erosion and redeposition of older marine deposits derived from the beach ridge remnants when water levels were high. These results do not support the occurrence of a large tsunami event at Laguna Mitla during the Late Holocene. Inter-American Institute for Global Change ResearchIAI-CRN2050Liu Kam-biu http://dx.doi.org/10.13039/100000001National Science FoundationBCS-1003654Liu Kam-biu http://dx.doi.org/10.13039/100005720Geological Society of AmericaThomas Anthony Bianchettehttp://dx.doi.org/10.13039/100002634Association of American GeographersThomas Anthony BianchetteThis work was supported by the Inter-American Institute for Global Change Research (IAI) - Grant number: IAI-CRN2050 (URL: http://www.iai.int/) to KBL, the National Science Foundation (NSF) - Grant number: BCS-1003654 (URL: www.nsf.gov) to KBL and TAB, the Geological Society of America (URL: www.geosociety.org) to TAB, and the Association of American Geographers (URL: www.aag.org) to TAB. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Data AvailabilityAll excel files are available from the Dryad database. doi:10.5061/dryad.7ff11.Data Availability All excel files are available from the Dryad database. doi:10.5061/dryad.7ff11. ==== Body Introduction Mexico’s Pacific Coast is frequently affected by tsunamis, many of which can cause marine inundations reaching many kilometers inland and result in significant societal damages and geomorphological changes [1, 2]. Globally, many studies have examined sedimentary units deposited via historic or prehistoric tsunami run-ups in order to identify their sedimentological characteristics and spatial extent [3–7]. Studies along Mexico’s Pacific coast have used the presence of sand layers in sediments to identify marine intrusions from tsunamis [8, 9]. To develop accurate risk assessments for this economically vibrant region, additional work must be conducted to establish the long-term (multi-centennial to -millennial) record, with the aim of detecting geological imprints (e.g., sedimentological and/or geochemical signatures) of historical tsunamis to better understand the origins and magnitudes of older events in the sedimentary record. The necessity for this research is especially true for the coastal state of Guerrero, which has experienced at least 48 tsunamis since AD 1732 [1]. Located near the Guerrero Seismic Gap, which has not experienced a large earthquake since 1911 [10–12], this tectonically-active area is due for a major event, possibly of Mw 8.1–8.4 strength [12]. Recent studies from Laguna Mitla, a large backbarrier lagoon along the coast of Guerrero, suggest that a major tsunami impacted this coastal zone ~3400 years ago [12–14]. The geological evidence comes from a ~60 cm thick layer dominated by sand enriched in Na and Sr, overlaid by a ~45 cm thick layer of dark-blue to blue-grey laminated clayish silt occurring from a depth of 300–195 cm in a sediment core extracted from the landward, or back side, of the lagoon [12]. The sand interval was interpreted as having been deposited by an extreme marine incursion, while the overlying blue clay was interpreted as representing a marine-dominated setting [12]. The posited tsunamigenic origin of this deposit [13], if confirmed, would imply that the Guerrero coastal zone, including the major city of Acapulco, is highly vulnerable to the devastating impacts of tsunami activity. In this paper, we present geochemical and sedimentological data from six new cores taken from Laguna Mitla to reconstruct the Holocene depositional history of the site itself, with the objective of evaluating the evidence for marine incursions. Such a re-examination of prehistoric tsunami occurrence is vital for accurately assessing the hazard risks for this coastal region. Regional Setting and Study Site Guerrero is located along the southern section of Mexico’s Pacific coast (Fig 1). The Sierra Madre del Sur form southwestern Mexico’s ‘spine,’ with elevations surpassing 3,500 m in Guerrero’s interior. The rivers and ephemeral streams that drain the uplands often terminate in coastal lagoons, typically elongated coast-parallel lagoons trapped behind solid beach ridges [15]. When lagoon water level rises during wet periods, pressure along the tidal inlet opening can erode and eventually blowout the beach, partially draining the lagoon [15]. The climate along the coast is monsoonal, with the majority of the annual precipitation (~130 cm) falling during the wet season from May to November [16], which is also the active season for TCs [17]. 10.1371/journal.pone.0161568.g001Fig 1 Study site. Location of Guerrero, Laguna Mitla, and Acapulco (a.) along Mexico’s southern Pacific coast (inset). A satellite image (USGS National Map Viewer) of Laguna Mitla is shown (b). A ~60 cm thick, predominantly sand unit attributed to a marine event was discovered in a sediment core (ACA04-06) extracted 5 km from the coast at the location marked with a red star [12]. Sedimentary evidence of a marine intrusion was also reported from a core (ACA03-02) extracted 4 km from the coast at the location marked with a red square [14]. The six coring locations discussed in this paper are marked by white dots. Exposed relict beach ridge remnants Isla Magueyes (IM) and Isla El Conejo (IEC) are labeled. The Middle America Trench, which marks the subduction of the Cocos plate beneath the continental North American plate, lies ~70 km west-southwest of Laguna Mitla beyond a narrow continental shelf, resulting in significant earthquake and tsunami activity for the area [18]. The Guerrero Seismic Gap covers most of Guerrero’s coastline, extending offshore to the Middle America Trench. The seismic model for this area suggests periods of interseismic (long-term) subsidence interrupted by coseismic uplifts ([13] and references therein). Recent studies have also documented the existence of regional slow slip events [19]. The long-term subsidence, estimated at 0.3–0.4 cm per year, overwhelms the coseismic uplift, as shown by the lack of marine terraces and wave-cut notches along the coast [13]. Laguna Mitla, located 40 km west of Acapulco (Fig 1), is an elongated, shore-parallel lagoon, approximately 22 km long and 4 km wide, with water depths of ~0.5–1 m along the western end. An inactive tidal inlet is located along the lagoon’s southeast corner [20]. The lagoon is classified as oligohaline, with a salinity level of ~3.5 ppt [21]. Runoff from the surrounding hillside is the principal source of inorganic sediments [22]. A ~15 m wide ephemeral stream sourced in the adjacent highlands enters the lagoon’s northwestern edge, nearest our study sites. Lagoon currents are strong enough near the stream to resuspend surface sediments along the western side. Land use to the north of Laguna Mitla is primarily agricultural. Wetland vegetation dominated by Typha and mangroves surrounds the lagoon. An extensive salt marsh lies along the northwest edge of the site [12, 14] Laguna Mitla is morphologically categorized as a “barred inner shelf” lagoon, the most common classification along Mexico’s Pacific coast [15]. Formerly a depression along the continental shelf margin, it was formed during the mid-Holocene transgression when sea levels stabilized and coastal beach ridges formed [15]. The beach ridge plain is ~900 m wide with a maximum height of >8 m. A chain of linear islands, including Isla Magueyes (IM) and Isla El Conejo (IEC), located along the lagoon’s midline are likely relict beach ridge segments [12, 14] deposited during a Pleistocene highstand. Such relict beach ridges are a common feature along Mexico’s Pacific coast [15]. Salt pans dot the interior linear islands, and a plantation covers the eastern half of IM [12, 14]. Methods In December 2009, six composite cores were taken along two shore-parallel transects from the lagoon’s western end (Fig 1). No permits were required for the described study, which complied with all relevant regulations. Field studies did not involve endangered or protected species. The cores were extracted with a Russian peat borer in overlapping 50 cm segments. Sediments were transferred to pre-cut PVC tubes and wrapped to prevent moisture loss. Cores 1, 3, and 5 were collected along a ~3.3 km shore-parallel transect behind the beach (seaward transect), whereas cores 2, 4, and 6 comprise a parallel transect closer to the relict beach ridges along the mid-rib of the lagoon (landward transect). Core 2 was extracted 0.5 km landward of core 1 and near the unnamed ephemeral stream. Cores 4 and 6 were extracted 0.5 km and 3.3 km east of core 2, respectively, and near Isla Magueyes. Surface samples were collected from the beach and from lagoon bottom mud surrounding each core site to determine modern sedimentological characteristics. All material was transported to the Global Change and Coastal Paleoecology Laboratory, Louisiana State University, and stored in a refrigerated room (4°C). Cores were described lithologically and subjected to loss-on-ignition (LOI) analysis [23] at one cm resolution. Samples were dried overnight at 105°C, and burned for one hour at 550°C and 1000°C to determine water (% wet weight), organics (% dry weight), and carbonate (% dry weight) contents, respectively. Seventeen organic (plant/wood) samples were sent to the National Ocean Sciences Accelerator Mass Spectrometry Laboratory (NOSAMS) at the Woods Hole Oceanographic Institution (WHOI) for AMS radiocarbon dating. Results (Table 1) were converted to calendar years using Calib 7.0 [24] and the INTCAL 13 calibration curve [25]. Age models were developed for each core based on the median calendar age, as supplied by Calib 7.0, using linear interpolation between pairs and extrapolation to the core tops and bottoms. 10.1371/journal.pone.0161568.t001Table 1 Radiocarbon dating results for cores 2–6. Calib 7.0 [24] and Intcal 13 [25] were used for calibration. Core Depth Lab # Material 14C age Cal BP (2δ) Rel area under prob. distribution Med. prob 2 114 OS-90703 Plant/Wood 1990±30 1879–1997 1 1940 2 197 OS-93080 Plant/Wood 3850±55 4094–4123 0.038 4271 4144–4418 0.962 2 242 OS-90697 Plant/Wood 4080±40 4438–4488 0.118 4582 4497–4655 0.635 4667–4707 0.074 4756–4811 0.174 2 316 OS-93065 Plant/Wood 4430±30 4875–5068 0.765 5016 5109–5123 0.018 5169–5172 0.002 5180–5275 0.215 3 175 OS-91548 Plant/Wood 4110±30 4523–4711 0.747 4632 4753–4814 0.253 4 111 OS-92322 Plant/Wood 3840±65 4010–4029 0.012 4254 4082–4423 0.988 4 198 OS-93067 Plant/Wood 4110±30 4523–4711 0.747 4632 4753–4814 0.253 5 2 OS-90761 Plant/Wood >Modern N/A N/A N/A 5 34 OS-90770 Plant/Wood 1840±45 1629–1655 0.039 1775 1660–1666 0.005 1693–1880 0.955 5 61 OS-84452 Plant/Wood 2710±110 2491–2602 0.061 2836 2607–2642 0.019 2678–3083 0.896 3090–3145 0.022 3152–3155 0.001 5 125 OS-83891 Plant/Wood 3160±30 3273–3284 0.026 3388 3340–3450 0.974 5 207 OS-83892 Plant/Wood 4070±30 4440–4486 0.159 4558 4499–4644 0.704 4677–4692 0.015 4761–4800 0.122 5 290 OS-83942 Plant/Wood 4500±30 5046–5205 0.65 5167 5210–5296 0.35 5 324 OS-93066 Plant/Wood 4760±30 5333–5348 0.041 5521 5353–5371 0.029 5463–5587 0.93 5 414 OS-83941 Plant/Wood 6050±35 6795–6990 1 6903 5 475 OS-90763 Plant/Wood 6000±60 6678–6708 0.031 6842 6712–6984 0.969 6 127 OS-92903 Plant/Wood 2800±80 2756–3080 0.965 2917 3093–3113 0.019 3122–3141 0.016 Once unwrapped, the moist sediments were scanned at two cm resolution with a Delta Innov-X handheld X-ray fluorescence (XRF) equipped with a tantalum x-ray tube and a factory calibration (Compton Normalization) to determine the concentration of 27 chemical elements. Standards NIST 2702 and 2781 were scanned for validation. Here we present the results of four: strontium (Sr), sulfur (S), calcium (Ca), and titanium (Ti) for cores 1–6. Sr, S, and Ca are enriched in seawater or marine sediments and have been widely used in detecting marine intrusion events in coastal sediments [8, 26–30]; whereas Ti, a lithogenic element, is commonly used as an indicator of soil erosion from the watershed [31]. The surface samples were also XRF-scanned in the laboratory. Surface sample concentrations were averaged according to the corresponding coring locations: four samples from near core 1, five near core 2, three near core 3, three near core 4, five near core 5, and one near core 6, in addition to two collected from the coastal beach. Results Cores are generally dominated by inorganic (clay) sediments, marked by low water and organic values (LOI curve, Fig 2). Organic-rich sections consist of either black to dark brown decomposed peat, black to dark brown muddy peat, or dark brown organic clay. Sand is present in only a few cores, usually either as scattered grains or thin, episodic layers. Calibrated calendar dates are shown alongside the LOI curves (Fig 2). Of the seventeen dates, a single stratigraphic reversal occurred, near the bottom of core 5 (6050±35 14C years at 414 cm, 6000±60 14C years at 475 cm). The date at 414 cm was rejected as its inclusion would imply an improbably large change in the sedimentation rate. 10.1371/journal.pone.0161568.g002Fig 2 XRF results. Elemental concentrations are paired with LOI results for all cores. Note scales vary between cores. Seaward transect Core 1 (190 cm) was extracted near Laguna Mitla’s western, seaward edge from a water depth of ~50 cm. Two thin, dark brown organic-rich clay layers occur near the core bottom at 183–177 and 173–170 cm (Fig 2). The remaining sediment is mostly gray to brown clay, with low percentages of water (~35–40%), organic (6–12%), and carbonate (1–3%). Ti concentrations increase slightly up-core. With the exception of large spikes in S and Sr in the basal material and in Ca near the core top, elemental concentrations are fairly stable (Fig 2). Water (24–32%) and organic (~5–8%) percentages are depressed from 64–53 cm in the interval containing clay with fine sand. Small gastropods are scattered near the core top. Core 3 (218 cm) was retrieved from ~50 cm water depth ~0.5 km east of core 1. A thick peat section from 212–164 cm lies on top of 6 cm of clay. Clay, marked by fairly consistent percentages of water (~30–40%), organic (~7–24%), and carbonate (~1–4%) dominates the remainder of the core (Fig 2). S concentration decreases up-core, while Ti, Sr, and Ca concentrations increase up-core (Fig 2). Core 5 (479 cm), extracted ~3 km east of core 3 from ~1 m water depth, consists of gray sand throughout the bottom 8 cm. The remaining 471 cm is dominated by black peat with high water (~85%) and organic (~80%) values (Fig 2). The clay occurring from ~400–291 cm has a relatively high concentration of Ti, with the exception of four carbonate-rich layers at 294–292 (14%), 309–308 (8%), 327–325 (13%), and 394–392 cm (6%). These carbonate layers are marked by elevated concentrations of S, Sr, and Ca and low concentrations of Ti (Fig 2). Seven Ti-rich clay layers varying in thickness and composition occur within the otherwise homogenous peat section from 290–0 cm. Landward transect Core 2 (329 cm) was extracted 0.5 km landward of core 1 at ~50 cm water depth. The bottom section of the core (329–196 cm) is a dark, muddy peat containing thin gray clay layers with low water and organic values (Fig 2). The top 196 cm consists of brown to gray clay, with a dark, organic-rich clay section at 145–115 cm. Ti concentrations are elevated in the low-organic clay sections. At 145–144 cm, the concentration of Ti (5,428 ppm) is the transect maximum (Fig 2). Concentrations of S, Ca, and Sr are lower throughout this clay section than in the underlying muddy peat, with the exception of peaks in S and Ca near the core top, which contains scattered gastropod shells. Core 4 (216 cm) was extracted 0.5 km east of core 2 at a water depth of ~50 cm. Water and organic percentages, and Ti, Ca, and Sr concentrations are highly variable throughout the bottom section (216–119 cm), which consists of brown to gray clay interlayered with scattered yellow sand layers (Figs 2 and 3). S concentrations are high in this material. The upper 119 cm consists of brown to gray clay with relatively constant water (30–40%), organic (7–10%), and carbonate values (1–2%). Above 119 cm, concentrations of S decrease significantly as Ti increases slightly. Ca concentrations increase near the core top, while Sr concentrations are highly variable throughout the section. Small gastropods are scattered throughout the sediments near the core top. 10.1371/journal.pone.0161568.g003Fig 3 Elemental concentrations of surface samples. Surface samples were collected from the coastal beach and adjacent of the coring sites. Concentrations from surface samples belonging to the same coring site were averaged and plotted. Titanium concentrations decrease eastward across the transects. Concentrations of sulfur, strontium and calcium, commonly-used marine indicators, are higher in the landward sites (2, 4, 6) than in the paired seaward sites (1, 3, 5). Beach samples contain relatively low concentrations of sulfur, strontium and calcium. Core 6 (129 cm) was extracted ~3 km east of core 4 from a depth of ~1 m. An alternating series of dark gray/black sandy peat and coarse gray sand occurs from 129–88 cm, with higher concentrations of Ti, S, Sr, and Ca in the sand layers (Fig 2). A dark gray sand layer from 88–56 cm is marked by low water, organic, and carbonate values (Fig 2), and higher S concentrations. A dark peat occurs from 56–45 cm. The top 45 cm consists of iterations of reddish, gray, and brown clay. A shell hash section at 45–33 cm is marked by a sharp spike in carbonates (32%) and concentrations of Ca and Sr. Ti concentrations are elevated in the reddish clay near the core top. All surface samples are clay-dominant with the exception of the sample near site 6, which contains some shell hash and shiny, flaky silicates, possibly mica. S concentrations are higher near the landward sites 2, 4, and 6, than at the seaward sites 1, 3, and 5 (Fig 3). Sr concentrations are nearly an order of magnitude higher at site 6 than at any of the other five sites: site 1, 2, 3, 4, and 5. Ca concentrations are also higher near site 6. Ti concentrations are higher near the western sites, sites 1, 2, 3, and 4, and lower near the eastern sites 5 and 6. S, Sr, and Ca concentrations are relatively low in the beach samples (Fig 3). Discussion The chronologies and stratigraphies of the six cores along two transects can be cross-correlated and synthesized to reveal the depositional processes and provenances of the source material, a critical necessity for evaluating the occurrence of prehistoric marine intrusions in this coastal zone. Paleoenvironmental reconstruction The multi-proxy data from core 5 are presented in detail elsewhere [32] to provide a 7000-year reconstruction of the coastal paleoenvironmental evolution and Holocene paleoclimatic changes at Laguna Mitla, and are thus not the main objectives of this paper. However, brief descriptions of regional environmental history and site evolution are presented here to provide a context for understanding the depositional patterns and processes within the lagoon, the focus of this paper. Core 5, the longest and best-dated core, provides a sedimentary record spanning the last 7000 years (Fig 2). At the beginning of the record (~ 6900 years BP), the core site was dominated by Rhizophora (red mangrove) [32]. By ~6200 years BP, the rising seas began depositing a mix of offshore clastics and entrained terrestrial sediments, marking the initiation of the beach barrier as documented along Mexico’s Pacific coast [33]. Clastic input was temporally variable due to varying degrees of wave energy from the spatially discontinuous and highly transient beach barrier, perhaps frequently subjected to perturbations from overwash and hydrodynamic processes. This highly dynamic period lasted until the rate of sea level rise slowed at ~5200 years BP [33], at which time the barrier became more stable and consolidated. From ~5200 years BP to present, the beach ridge plain has prograded seaward, and the site has existed as a backbarrier system isolated from the sea by the beach barrier. The build-up and consolidation of the beach barrier increasingly isolated the backbarrier lagoon from the ocean, so that water level at the coring sites has since been essentially controlled by precipitation, rather than sea level. Throughout the last 5000 years, significant environmental changes at the site were registered sedimentologically, geochemically, and palynologically by the episodic alternations between peat and inorganic sediments. Peat, dominated by Rhizophora and Laguncularia (white mangrove) pollen [32] indicates low water levels and a wetland environment, with salinity levels affected by the seepage of marine water through the barrier. On the other hand, the inorganic sediments, characterized by clay rich in Ti and the regional pollen signal, suggest higher backbarrier water levels and the transformation of the site from a wetland to an open-water lagoon, driven by an increase in precipitation and fluvial discharge to the basin. In core 5, seven such clay bands occurred during the past 4500 years, indicating episodes of wet climate and lagoon phases existing at approximately 4430–4270, 4080, 3950, 3680–3490, 3170–3080, 2990–2870, and 1680–0 years BP. The timing of these wet periods corresponds well with paleoclimatological evidence from the nearby coastal Laguna Tetitlan [34] and the Middle American Trench, located offshore of Oaxaca [35]. These abrupt sedimentary transitions are recorded for matching temporal intervals in cores 2, 3, 4, and 6. Distinct peat/clay transitions occurred at approximately 4500, 4300, and 1900 years BP in core 2, at 4600 years BP in core 3, and at 950 years BP (inferred) in core 6. It should be noted that not all six oscillations recorded in core 5 occur in all cores, and some undated, less distinct organic/inorganic transitions are present in other cores (e.g., the three LOI peaks in the lower half of core 6). The lithological differences among the six cores can be explained by the sites’ different sensitivities in recording water level changes, which are a function of water depth, sediment supply, sedimentation rate, and local habitat and vegetation (e.g., mangrove swamp versus mud flat), all of which vary spatially. Nevertheless, the onset of several of these clay layers, especially the ~4500 years BP event, indicating lagoon phases or wet episodes, are broadly synchronous among cores. It is also remarkable that the uppermost sediments in all six cores are characterized by lagoonal clay, which certainly represents the limnic deposits formed in the depositional environment of the modern shallow lagoon. Radiocarbon dates and inferred ages obtained from cores 2, 5, and 6 suggest that the modern lagoon phase has existed continuously for the last 1900 years. While we had considered the possibility that the clay intervals in Laguna Mitla result from land movements attributable to either sudden subsidence or uplift from earthquakes or slow-slip events covering longer time periods, this notion has been rejected due to a lack of sedimentological evidence of sudden land movements (e.g., erosive contacts) in the cores. In addition, geochemical evidence of marine incursions is notably lacking; instead, the clay intervals are characterized by the enrichment of terrigenous chemical elements such as Ti. Along the coast of Guerrero, documented land movements from earthquakes range from ~7–15 cm [11] for coseismic uplift and 5–14 cm for slow slip events [36], insufficient to explain the formation of ~1 m deep lagoons, nor their relatively sudden transformation into mangrove-dominated swamps or mud flats. Sedimentary processes and provenances The distributions of the elemental concentrations display distinct spatial patterns. The concentrations of Ti, which is commonly applied to identify terrestrial sediments [37] are many times higher in surface sediments collected from the western sites (1, 2, 3, and 4) than the eastern sites (5 and 6) (Fig 3). This suggests that slopewash, associated with large precipitation events and delivered by the ephemeral stream located in the northwest edge of the study site, is the primary source of this element. This spatial relationship remains constant throughout our record as Ti concentrations are higher at all times in core 4 than in the eastern cores 5 and 6 (Fig 4). 10.1371/journal.pone.0161568.g004Fig 4 Elemental concentrations of sediment cores. Titanium, sulfur, calcium, and strontium concentrations vs age for seaward core 5 (black), and landward cores 4 (green) and 6 (red). Titanium concentrations (a.) are higher in core 4, nearest the input stream, than cores 5 and 6, farther east. Concentrations of sulfur (b.), calcium (c.), and strontium (d.) are generally higher in landward cores 4 and 6, than the seaward core 5. Spikes in marine elements are largely synonymous with shells (whole, hash) and/or sand, all largely absent in core 5 with the exception of basal sediments. A major marine event was posited from two landward sites, marked by a star and square in Fig 1 [12, 14]. A matching event is absent in the cores from this study, with no sand occurring at any site from 3000–4000 years BP. Due to their high concentrations in seawater and marine sediments, Sr, S, and Ca are commonly used marine indicators. In Laguna Mitla, however, surface sample concentrations of these elements are higher in the landward sites than the seaward sites (Fig 3), with the highest concentrations of all three elements occurring at site 6, ~ 1 km landward of the lagoon’s edge and adjacent to Isla Magueyes. This suggests that these elements do not originate from marine material delivered over the beach barrier, but rather the clastic and carbonate materials eroded from the relict beach ridge in the center of the lagoon. This is further supported by the low Sr, S, and Ca concentrations in the modern beach sand (Fig 3). Many event types can erode and redeposit these materials into the area of the landward transect, including earthquakes, tropical cyclone rain events, increased runoff during climatologically wet periods, and changes in land use. This spatial pattern extends throughout the entirety of our record. Concentration maxima for S, Ca, and Sr occur in the landward cores 4 and 6, especially in intervals dominated by shells (whole, hash) and/or sand (Fig 4). The spikes in elemental concentrations in the uppermost sections of landward cores 2 and 4 (Fig 2) are not matched by corresponding spikes in seaward cores 3 and 5. This indicates that the material was not transported from a seaward location by marine incursions. Possibly these spikes result from the presence of gastropods, or ancient marine materials eroded from Isla Magueyes and other relict beach ridge remnants in or adjacent to the lagoon. Revisiting geological evidence attributed to marine incursion events Large seawater intrusions can be expected to transport coastal sediments (e.g., sand) and marine organisms (shells, diatoms, foraminifera) inland. Marine deposition can vary from isolated overwash lobes to large sand sheets, often tens of centimeters thick [38, 39]. Tsunami waves are highly dynamic and can be extremely powerful, capable of transporting large objects landward [40], and depositing sand and other fine materials many kilometers inland [41]. Common characteristics of tsunami deposits include landward and seaward layering [42], landward fining [5], upward fining [43], rip-up clasts [7], and basal erosion [44]. The occurrence of a large marine incursion at ~3400 years BP for the area has been suggested [12–14], based on evidence found in core ACA 04–06 extracted from wetlands 5 km behind the coast in the northwestern part of the lagoon (red star, Fig 1), and core ACA 03–02 extracted landward of Isla El Conejo and 4 km from the coast (red square, Fig 1). Core ACA 04–06 [12], otherwise dominated by mud and silt, includes a ~60 cm unit of fine to coarse sand that contains rip-up clasts and high concentrations of such marine indicator elements as Na (sodium), Sr, and Ca, as well as a fining-upward stratigraphy. The authors suggest that this layer resulted from a marine incursion associated with an extremely rapid rise in relative sea level. Due to the layer’s sedimentary characteristics, thickness, and distance from the coast, a large tsunami was suggested as the probable delivery mechanism [12]. The downward movement of this event, unlike the uplift normally associated with historical earthquakes along the Guerrero coast, is posited to have been the result of a megathrust event that ruptured the “entire coupled plate interface” [13]. The ~6 m core ACA 03–02 [14] contains a 93 cm thick interval containing marine diatoms, sponge spicules, shells, and high concentrations of marine elements (Na, Sr, S), posited as resulting from marine intrusion with an extrapolated date of ~3500 years BP. This core does not include a sand unit corresponding to that found in ACA 04–06 [12]. The interval posited as resulting from marine intrusion also contains a peat layer [14]. It would be expected that such a marine incursion would have deposited a clear sedimentary signal at our sites, located behind the modern beach ridge plain and ~4 km closer to the coast than the ACA core locations. However, except for the basal sediments, sand layers are absent in cores 2, 3, and 5. Although distinct sand layers do occur in cores 4 and 6, they are much thinner (~1–30 cm) than the posited event deposits in core ACA 04–06, and they fail to exhibit common characteristics of tsunami deposition, including erosional contacts and deposition of marine shells. The elevated concentrations of S, Sr, and/or Ca in the surface samples, and downcore in cores 4 and 6 most likely represent material eroded from nearby Isla Magueyes. Although fine sand occurs from 64–53 cm in core 1, and clay/silt from ~84–79 cm in core 2 (decrease in LOI), core chronologies and stratigraphic correlations indicate that these layers were deposited much more recently than 3400 years BP, as evidenced by a sample dated to 1990±30 14C years BP (1940 years BP) taken from 30 cm below the clastic layer in core 2. Furthermore, these sedimentological signatures in the westernmost cores 1 and 2 are not apparent in the eastward located cores 3, 4, 5, and 6, and may therefore suggest erosion and redeposition of remnant beach ridge materials mixed with sediments deposited from the slopes. Our results therefore do not support the identification of a significant marine incursion at 3400 years BP [12]; nor do our findings show definitive evidence of a large megathrust earthquake resulting in significant coseismic subsidence at this time [13, 14]. The lack of definitive stratigraphic evidence of tectonic movement, and the coherent, progressive ecological succession displayed in the multi-proxy record from Laguna Mitla [32], instead support the scenario of relative tectonic stability during the Late Holocene, previously implied by the presence of alluvial and deltaic plains observed by Ramírez-Herrera and Urrutia-Fucugauchi [45]. The presence of shells may explain the high levels of marine elements, particularly Ca, a major component of both terrestrial and marine shells, in core ACA 03–02, extracted landward of Isla El Conejo [14]. Either Isla Magueyes or Isla El Conejo, or other relict beach ridges further inland, are possible sources of the sediments and marine diatoms, which are capable of attaching to sand grains [46]. The low, episodic spikes in S, Sr, and Ca for the seaward cores 1, 3, and 5 in this study may possibly be explained by the episodic breaching of Laguna Mitla’s outlet channel following large precipitation events. Such breaching, a common occurrence for barred inner-shelf lagoons, has been documented from the nearby Laguna Nuxco [15], and can result in the intrusion of marine waters until the breach is closed. Minoura and Nakaya [47] documented saltwater intrusions which, while not severe enough to deposit sediments, were capable of leaving a geochemical signature in the sediments. Given the height, width, and stability of the beach ridge we believe that the most likely sedimentary signature of seismic activity over the last ~5000 years would be increased slopewash following the loosening of sediments associated with violent earth movement rather than deposition of thick sediment layers several kilometers inland. Conclusions The main objective of this paper is to investigate the sedimentological processes and the provenances of bottom sediments deposited in Laguna Mitla, a barred inner shelf lagoon located on Mexico’s Pacific coast, in order to detect the sedimentary signal of tsunami-generated marine intrusions. Notably, concentrations of elements often associated with marine sources (Sr, S, Ca) are highest at landward sites 4 and 6, >2 km from the coast. We attribute the high concentrations of these elements to the erosion of clastic and carbonate materials from remnant beach ridges located in the lagoon’s interior. The concentration gradient of Ti is related to the proximity of the main feeder stream, located in the northwestern edge of the site. Previous researchers have posited a sudden rise in relative sea level ~3400 years BP, possibly tsunami-generated, based on sedimentary evidence from two cores extracted ~4–5 km inland, each containing thick clastic layers/sections rich in marine elements, one of which included larger-grained material (sand) [12–14]. However, no corresponding sand layers occur in our three seaward cores, located closer to the ocean and directly behind the modern beach ridge plain. Any marine incursion process that deposited a thick layer of sand in the back side of the lagoon can be expected to have deposited an even thicker layer of sand at our core locations in front of that site. The landward cores extracted near the two relict beach ridges in the middle of the lagoon did possess thick sand layers with high concentrations of S, Sr, Ca, suggesting that this material was not introduced into the site by marine incursion, but by alternative means, such as erosion and reworking of sediments from islands El Conejo and Magueyes, possibly from heavy precipitation events or wave action. Our analysis therefore does not support the interpretation of a tsunami or other significant marine incursions at this site ~3400 years ago. We furthermore suggest that the presence of sand layers with elevated concentrations of marine elements are not reliable indicators of extreme event for barred inner shelf lagoons along Mexico’s Pacific coast. This study highlights the complexity of properly interpreting sedimentary data in this tectonically active coastal region, further indicating the necessity of extracting multiple cores with a large spatial coverage to assess sediment provenance and depositional processes to properly determine tsunami risk and long-term tectonic histories. The lack of evidence of a severe marine intrusion event about 3400 years BP suggests that return periods for regional tsunamis may be lower than previously interpreted. This finding has significant implications for hazard risk assessment for Pacific coastal regions of Mexico. Additional regional sedimentary studies must be undertaken to properly assess the long-term risk. Ideally, these studies would successfully integrate sedimentary analyses over relatively large areas to more fully understand the relevant depositional mechanisms, as a means of definitively identifying tsunami deposits. We would like to thank Dr. Blanca Figueroa Rangel and Dr. Miguel Olvera Vargas for their logistical support and Ulises Cruz-Valera for field assistance. We also thank two anonymous reviewers for their valuable comments that improved the manuscript. ==== Refs References 1 National Geophysical Data Center / World Data Service (NGDC/WDS) (2011) Global Historical Tsunami Database. NOAA. 10.7289/V5PN93H7. Retrieved 04-12-2016, from http://www.ngdc.noaa.gov/hazard/tsu_db.shtml. 2 Corona N and Ramírez-Herrera MT (2012 ) Mapping and historical reconstruction of the great Mexican 22 June 1932 tsunami . Natural Hazards and Earth System Sciences 12 : 1337 –1352 . 3 Minoura K , Gusiakov VG , Kurbatov A , Takeuti S , Svendsen JI , Bondevik S , et al (1996 ) Tsunami sedimentation associated with the 1923 Kamchatka earthquake . Sedimentary Geology 106 : 145 –154 . 4 Sawai Y (2002 ) Evidence for 17th-century tsunamis generated on the Kuril-Kamchatka subduction zone, Lake Tokotan, Hokkaido, Japan . Journal of Asian Earth Sciences 20 : 903 –911 . 5 Goff J , McFadgen BG and Chagué-Goff C (2004 ) Sedimentary differences between the 2002 Easter storm and the 15th-century Okoropunga tsunami, southeastern North Island, New Zealand . Marine Geology 204 : 235 –250 . 6 Tuttle MP , Ruffman A , Anderson T and Jeter H (2004 ) Distinguishing tsunami from storm deposits in eastern North America: The 1929 grand banks tsunami versus the 1991 Halloween storm . Seismological Research Letters 75 : 117 –131 . 7 Morton RA , Gelfenbaum G and Jaffe BE (2007 ) Physical criteria for distinguishing sandy tsunami and storm deposits using modem examples . Sedimentary Geology 200 : 184 –207 . 8 Ramírez-Herrera MT , Lagos M , Hutchinson I , Kostoglodov V , Machain ML , Caballero M , et al (2012 ) Extreme wave deposits on the Pacific coast of Mexico: Tsunamis or storms?—A multi-proxy approach . Geomorphology 139 –140 : 360 –371 . 9 Ramírez-Herrera MT , Bógolo MF , Černỳ J , Goguitchaichvili A , Corona N , Machain ML , et al (2016 ) Historic and ancient tsunamis uncovered on the Jalisco-Colima Pacific coast, the Mexican subduction zone . Geomorphology 259 : 90 –104 . 10 Anderson JG , Singh SK , Espindola JM and Yamamoto J (1989 ) Seismic strain release in the Mexican subduction thrust . Physics of the Earth and Planetary Interiors 58 : 307 –322 . 11 Ortiz M , Singh SK , Kostoglodov V and Pacheco J (2000 ) Source areas of the Acapulco-San Marcos, Mexico earthquakes of 1962 (M 7.1;7.0) and 1957 (M 7.7), as constrained by tsunami and uplift records . Geofisica Internacional 39 : 337 –348 . 12 Ramírez-Herrera MT , Cundy A , Kostoglodov V , Carranza-Edwards A , Morales E and Metcalfe S (2007 ) Sedimentary record of late-Holocene relative sea-level change and tectonic deformation from the Guerrero Seismic Gap, Mexican Pacific Coast . The Holocene 17 : 1211 –1220 . 13 Ramírez-Herrera MT , Kostoglodov V and Urrutia-Fucugauchi (2011 ) Overview of recent coastal tectonic deformation in the Mexican Subduction Zone . Pure and Applied Geophysics 168 : 1415 –1433 . 14 Ramírez-Herrera MT , Cundy AB , Kostoglodov V and Ortiz M (2009 ) Late Holocene tectonic land-level changes and tsunamis at Mitla lagoon, Guerrero, Mexico . Geofisica Internacional 48 : 195 –209 . 15 Lankford RR (1977 ) Coastal Lagoons of Mexico: their origin and classification In: Wiley M. , editor. Estuarine Processes . Academic Press Inc pp. 182 –215 . 16 National Meteorológical Service of Mexico (2012) Normales climatológicas 1951–2010. Retrieved 2012, from http://smn.cna.gob.mx/climatologia/normales/5110/NORMAL12142.TXT. 17 Jáuregui E (2003 ) Climatology of landfalling hurricanes and tropical storms in Mexico . Atmósfera 16 : 193 –204 . 18 Cruz G and Wyss M (1983 ) Large earthquakes, mean sea level, and tsunamis along the Pacific Coast of Mexico and Central America . Bulletin of the Seismological Society of America 73 : 553 –570 . 19 Cavalié O , Pathier E , Radiguet M , Vergnolle M , Cotte N , Walpersdorf A , et al (2013 ) Slow slip event in the Mexican subduction zone: Evidence of shallower slip in the Guerrero seismic gap for the 2006 event revealed by the joint inversion of InSAR and GPS data . Earth and Planetary Science Letters 367 : 52 –60 . 20 Yáñez-Arancibia A (1977) Taxonomia, Ecologia Y Estructura de las Comunidades de Peces en Lagunas Costeras con Bocas Efimeras del Pacifico de Mexico. Annual Meeting of the American Society of Limnology and Oceanography. Savannah, Georgia, USA. 21 Contreras-Espinosa F and Warner BG (2004 ) Ecosystem Characteristics and Management Considerations for Coastal Wetlands in Mexico . Hydrobiologia 511 : 233 –245 . 22 Páez-Osuna F and Mandelli EF (1985 ) 210Pb in a tropical coastal lagoon sediment core . Estuarine Coastal and Shelf Science 20 : 367 –374 . 23 Liu KB and Fearn ML (2000 ) Reconstruction of prehistoric landfall frequencies of catastrophic hurricanes in northwestern Florida from lake sediment records . Quaternary Research 54 : 238 –245 . 24 Stuiver M, Reimer PJ and Reimer RW (2014) CALIB 7.0 (program and documentation). http://calib.qub.ac.uk/calib/. 25 Reimer PJ , Bard E , Bayliss A , Beck JW , Blackwell PG , Bronk Ramsey C , et al (2013 ) Intcal13 and Marine13 radiocarbon age calibration curves 0–50,000 years CAL BP . Radiocarbon 55 : 1869 –1887 . 26 Woodruff JD , Donnelly JP and Okusu A (2009 ) Exploring typhoon variability over the mid-to-late Holocene: evidence of extreme coastal flooding from Kamikoshiki, Japan . Quaternary Science Reviews 28 : 1774 –1785 . 27 Liu KB , McCloskey TA , Bianchette TA , Keller G , Lam NSN , Cable JE , et al (2014 ) Hurricane Isaac storm surge deposition in a coastal wetland along Lake Pontchartrain, southern Louisiana . Journal of Coastal Research , SI 70 : 266 –271 . 28 Chen ZY , Chen ZL and Zhang WG (1997 ) Quaternary stratigraphy and trace-element indices of the Yangtze delta, eastern China, with special reference to marine transgressions . Quaternary Research 47 : 181 –191 . 29 Chagué-Goff C , Dawson S , Goff JR , Zachariasen J , Berryman KR , Garnett DL , et al (2002 ) A tsunami (ca. 6300 years BP) and other Holocene environmental changes, northern Hawke's Bay, New Zealand . Sedimentary Geology 150 : 89 –102 . 30 Nichol SL , Goff JR , Devoy RJN , Chagué-Goff C , Hayward B and James I (2007 ) Lagoon subsidence and tsunami on the west coast of New Zealand . Sedimentary Geology 200 : 248 –262 . 31 Woodruff JD , Donnelly JP , Mohrig D and Geyer WR (2008 ) Reconstructing relative flooding intensities responsible for hurricane-induced deposits from Laguna Playa Grande, Vieques, Puerto Rico . Geology 36 : 391 –394 . 32 Bianchette TA, McCloskey TA and Liu KB (in review) A 7000-year history of coastal environmental changes from Mexico’s Pacific Coast: A multi-proxy record from Laguna Mitla, Guerrero. Submitted to The Holocene. 33 Curray F , Emmel J and Crampton PJ (1969 ) Holocene History of a Strand Plain Lagoonal Coast, Nayarit, Mexico In: Ayala-Castanares A. , editors. Coastal Lagoons: A Symposium . Mexico City : Universidad Nacional Autonoma de Mexico , pp. 63 –100 . 34 González-Quintero L (1980 ) Paleoecologia de un sector costero de Guerrero, Mexico (3000 anos) . Memorias 86 : 133 –157 . 35 Habib D , Thurber D , Ross D and Donahue J (1970 ) Holocene palynology of the Middle American Trench near Tehuantepec, Mexico . Memoirs of the Geological Society of America 126 : 233 –261 . 36 Alva VA and Kostoglodov V (2007 ) Aseismic slow slip events in Mexico from tide gauge records . Geos 27 :119 . 37 Haug GH , Hughen KA , Sigman DM , Peterson LC and Rohl U (2001 ) Southward migration of the intertropical convergence zone through the Holocene . Science 293 : 1304 –1308 . 11509727 38 Foster IDL , Albon AJ , Bardell KM , Fletcher JL , Jardine TC , Mothers RJ , et al (1991 ) High energy coastal sedimentary deposits: An evaluation of depositional processes in southwest England . Earth Surface Processes and Landforms 16 : 341 –356 . 39 Switzer AD , Pucillo K , Haredy RA , Jones BG and Bryant EA (2005 ) Sea level, storm, or tsunami: Enigmatic sand sheet deposits in a sheltered coastal embayment from southeastern New South Wales, Australia . Journal of Coastal Research 21 : 655 –663 . 40 Bussert R and Aberhan M (2004 ) Storms and tsunamis: evidence of event sedimentation in the Late Jurassic Tendaguru Beds of southeastern Tanzania . Journal of African Earth Sciences 39 : 549 –555 . 41 Goto K , Chagué-Goff C , Fujino S , Goff J , Jaffe B , Nishimura Y , et al (2011 ) New insights of tsunami hazard from the 2011 Tohoku-oki event . Marine Geology 290 : 46 –50 . 42 Scasso RA , Concheyro A , Kiessling W , Aberhan M , Hecht L , Medina FA , et al (2005 ) A tsunami deposit at the Cretaceous/Paleogene boundary in the Neuquen Basin of Argentina . Cretaceous Research 26 : 283 –297 . 43 McSaveney MJ , Goff JR , Darby DJ , Goldsmith P , Barnett A , Elliott S , et al (2000 ) The 17 July 1998 tsunami, Papua New Guinea: evidence and initial interpretation . Marine Geology 170 : 81 –92 . 44 Fujiwara O , Masuda F , Sakai T , Irizuki T and Fuse K (2000 ) Tsunami deposits in Holocene bay mud in southern Kanto region, Pacific coast of central Japan . Sedimentary Geology 135 : 219 –230 . 45 Ramírez-Herrera MT and Urrutia-Fucugauchi J (1999 ) Morphotectonic zones along the coast of the Pacific continental margin, southern Mexico . Geomorphology 28 : 237 –250 . 46 Meadows PS and Anderson JG (1968 ) Micro-organisms attached to marine sand grains . Journal of the Marine Biological Association of the United Kingdom 48 : 161 –175 . 47 Minoura K and Nakaya S (1991 ) Traces of tsunami preserved in inter-tidal lacustrine and marsh deposits: Some examples from Northeast Japan . Journal of Geology 99 : 265 –287 .

PMC005xxxxxx/PMC5003390.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757127310.1371/journal.pone.0161903PONE-D-16-24627Research ArticleBiology and Life SciencesCell BiologyChromosome BiologyChromosomesChromosome Structure and FunctionTelomeresTelomere LengthMedicine and Health SciencesPublic and Occupational HealthPhysical ActivityMedicine and Health SciencesOncologyCancers and NeoplasmsBreast TumorsBreast CancerBiology and Life SciencesCell BiologyChromosome BiologyChromosomesChromosome Structure and FunctionTelomeresMedicine and Health SciencesOncologyCancers and NeoplasmsInvasive TumorsBiology and Life SciencesCell BiologyCellular TypesAnimal CellsBlood CellsWhite Blood CellsBiology and Life SciencesCell BiologyCellular TypesAnimal CellsImmune CellsWhite Blood CellsBiology and Life SciencesImmunologyImmune CellsWhite Blood CellsMedicine and Health SciencesImmunologyImmune CellsWhite Blood CellsMedicine and Health SciencesSurgical and Invasive Medical ProceduresBiology and Life SciencesNutritionDietAlcohol ConsumptionMedicine and Health SciencesNutritionDietAlcohol ConsumptionAssociation of Telomere Length with Breast Cancer Prognostic Factors Telomere Length and Breast Cancer Prognostic Factorshttp://orcid.org/0000-0002-0962-2020Ennour-Idrissi Kaoutar 123Têtu Bernard 1245Maunsell Elizabeth 1235Poirier Brigitte 1256Montoni Alicia 7Rochette Patrick J. 7Diorio Caroline 1235*1 Axe Oncologie, Centre de Recherche du CHU de Québec-Université Laval, Quebec city (QC), Canada2 Centre de Recherche sur le Cancer, Université Laval, Quebec city (QC), Canada3 Département de médecine sociale et préventive, Faculté de médecine, Université Laval, Quebec city (QC), Canada4 Département de biologie moléculaire, biochimie médicale et pathologie, Faculté de médecine, Université Laval, Quebec city (QC), Canada5 Centre des Maladies du Sein Deschênes-Fabia, Hôpital du Saint-Sacrement, Quebec city (QC), Canada6 Department de chirurgie, Faculté de médecine, Université Laval, Quebec city (QC), Canada7 Axe Médecine Régénératrice, Centre de recherche du CHU de Québec-Université Laval, Quebec city (QC), CanadaAhmad Aamir EditorUniversity of South Alabama Mitchell Cancer Institute, UNITED STATESCompeting Interests: The authors have declared that no competing interests exist. Conceptualization: KE-I BT EM BP CD. Formal analysis: KE-I CD. Funding acquisition: KE-I EM CD. Investigation: KE-I BT BP CD. Methodology: KE-I BT EM BP CD. Project administration: CD. Resources: KE-I BT BP AM PJR CD. Supervision: BT EM AM PJR CD. Validation: KE-I CD. Visualization: KE-I EM CD. Writing – original draft: KE-I. Writing – review & editing: KE-I BT EM BP AM PJR CD. * E-mail: Caroline.Diorio@crchudequebec.ulaval.ca29 8 2016 2016 11 8 e016190319 6 2016 12 8 2016 © 2016 Ennour-Idrissi et al2016Ennour-Idrissi et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Introduction Telomere length, a marker of cell aging, seems to be affected by the same factors thought to be associated with breast cancer prognosis. Objective To examine associations of peripheral blood cell-measured telomere length with traditional and potential prognostic factors in breast cancer patients. Methods We conducted a cross-sectional analysis of data collected before surgery from 162 breast cancer patients recruited consecutively between 01/2011 and 05/2012, at a breast cancer reference center. Data on the main lifestyle factors (smoking, alcohol consumption, physical activity) were collected using standardized questionnaires. Anthropometric factors were measured. Tumor biological characteristics were extracted from pathology reports. Telomere length was measured using a highly reproducible quantitative PCR method in peripheral white blood cells. Spearman partial rank-order correlations and multivariate general linear models were used to evaluate relationships between telomere length and prognostic factors. Results Telomere length was positively associated with total physical activity (rs = 0.17, P = 0.033; Ptrend = 0.069), occupational physical activity (rs = 0.15, P = 0.054; Ptrend = 0.054) and transportation-related physical activity (rs = 0.19, P = 0.019; P = 0.005). Among post-menopausal women, telomere length remained positively associated with total physical activity (rs = 0.27, P = 0.016; Ptrend = 0.054) and occupational physical activity (rs = 0.26, P = 0.021; Ptrend = 0.056) and was only associated with transportation-related physical activity among pre-menopausal women (rs = 0.27, P = 0.015; P = 0.004). No association was observed between telomere length and recreational or household activities, other lifestyle factors or traditional prognostic factors. Conclusions Telomeres are longer in more active breast cancer patients. Since white blood cells are involved in anticancer immune responses, these findings suggest that even regular low-intensity physical activity, such as that related to transportation or occupation, could be recommended to breast cancer patients. Fondation des Hôpitaux Enfant-Jésus Saint-SacrementDiorio Caroline http://dx.doi.org/10.13039/501100000150Canadian Breast Cancer Research AllianceDiorio Caroline Fondation du cancer du sein du QuébecDiorio Caroline Banque de tissus et de données of the Réseau de recherche sur le cancer of the FRQS (affiliated with the Canadian Tumour Repository Network)Diorio Caroline Cancer Research Centre at Laval Universityhttp://orcid.org/0000-0002-0962-2020Ennour-Idrissi Kaoutar Canadian Breast Cancer Foundation-Canadian Cancer Society Capacity Development award703003Diorio Caroline This project was funded by the “Fondation des Hôpitaux Enfant-Jésus – Saint-Sacrement” (http://tmdesign.ca/client/fondation-hopital/) and the Canadian Breast Cancer Research Alliance (http://www.cwhn.ca/fr/node/20460). Biological specimens were provided by the Fondation du cancer du sein du Québec (http://www.rubanrose.org) and the Banque de tissus et de données of the Réseau de recherche sur le cancer of the FRQS (https://www.economie.gouv.qc.ca/objectifs/informer/recherche-et-innovation/page/intervenants-du-milieu-18854/?tx_igaffichagepages_pi1%5BbackPid%5D=18870&tx_igaffichagepages_pi1%5BcurrentCat%5D=&tx_igaffichagepages_pi1%5Bmode%5D=single&tx_igaffichagepages_pi1%5BparentPid%5D=18800&cHash=d57df5997e5aacb60db91a754b089dda), which is affiliated with the Canadian Tumour Repository Network. KEI received a training award from the “Cancer Research Centre at Laval University” (https://www.crc.ulaval.ca/no_cache/en/home/). CD is a recipient of the Canadian Breast Cancer Foundation-Canadian Cancer Society Capacity Development award (award #703003) (http://www.cbcf.org/Pages/default.aspx-http://www.cancer.ca/en/?region=on) and the FRQS Research Scholar (http://www.frqs.gouv.qc.ca). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Data AvailabilityAll relevant data are within the paper.Data Availability All relevant data are within the paper. ==== Body Introduction Breast cancer is the most common cancer in women worldwide, and the second most common cause of cancer death among women [1, 2]. According to the World Health Organization statistics [1, 2], between 2008 and 2012, incidence increased by more than 20% and mortality by 14%. In western countries, 5-year survival is around 89%, due to early detection and molecularly targeted therapies. However, traditional prognostic factors are still imprecise in predicting breast cancer prognosis and therefore new independent prognostic markers are needed. Telomeres are highly specialized structures capping the ends of linear chromosomes [3–6]. They consist of repeated DNA sequences, 5’-TTAGGG-3’ of 5 to 15 kb length in humans, bound by multiple telomeric interacting proteins. Telomeres ensure the stability of chromosomes and genome integrity during replication. The telomerase enzyme complex, a specialized reverse transcriptase, extends the 3’ end of chromosomes by adding TTAGGG repeats [4, 6]. In absence of telomerase, gradual shortening of telomeres occurs with each cell division due to the end-replication problem (the inability of DNA polymerase to fully replicate chromosomes ends). When telomere shortening reaches a critical point, DNA damage responses are elicited, leading to replicative senescence and cell apoptosis (programmed cell death). Inflammation and oxidative stress have been shown to result in accelerated telomere shortening and several studies suggest that some lifestyle factors like smoking, alcohol abuse, sedentary lifestyle and obesity have an impact on telomere length in healthy individuals [7–10]. These modifiable factors also appear to be associated with breast cancer prognosis [11–13]. In cells that express telomerase, such as blood leukocytes, telomere length seems to be a dynamic feature that responds to processes that can shorten or lengthen telomeres [14]. Therefore, peripheral blood cell telomere length could be a surrogate for both the ability of underlying dynamic processes to restore or maintain telomere homeostasis and for assessing the impact of modifiable environmental factors. The systematic review of literature suggest a trend toward a positive association of longer telomeres with better prognosis [15]. However, the exact prognostic significance of telomere length for breast cancer patients is unclear. The objective of the present study is to evaluate the association of telomere length, measured in peripheral blood cells, with traditional and potential prognostic factors in breast cancer patients. Materials and Methods Study design and population We conducted a cross-sectional study on data collected at the time of surgery. Selection of study population was described elsewhere [16]. Briefly, 164 consecutive women who underwent surgery for unilateral breast cancer were prospectively recruited between January 2011 and May 2012, at a breast cancer reference center, the "Centre des maladies du sein Deschênes-Fabia du CHU de Québec" in Quebec City, Canada. Women were eligible (n = 226) if they were not older than 70 years, were not pregnant, had no previous diagnosis of cancer other than non-melanoma skin cancer, never had any breast surgery including breast reduction or implants, never took a selective estrogen receptor modulator such as Tamoxifen or Raloxifen, and did not receive any treatment prior to surgery. Of the consecutively approached women, 226 were eligible and 164 (73%) accepted to participate. Blood samples were provided by 162 participants. All participants provided written informed consent. The study protocol was reviewed and approved by the Research ethics committee of the Centre de Recherche du CHU de Québec. Data collection Before surgery, a qualified research nurse performed anthropometric measures (weight and height measures) and drew blood samples according to standardized protocols. Information about risk factors was collected on average 24 days after surgery, using standardized questionnaires administered by telephone interview. Interviews included questions on gynecological and obstetric history, hormone use, and important lifestyle factors (smoking, alcohol consumption, physical activity). Questions on physical activity were derived from the Past Year Total Physical Activity Questionnaire [17], which measures all types (i.e., occupational, household, transportation-related and recreational) and all parameters (i.e., frequency, duration, and intensity) of physical activity and enables computation of physical activity data as metabolic equivalent (MET) hours of activity per week [17]. Biological characteristics of the tumor were extracted from pathology reports, including tumor size, lymph node involvement, histologic type, tumor grade, hormonal and growth factor receptor status. Disease stage was established following the American Joint Committee on Cancer (AJCC) cancer staging system for breast cancer [18]. Telomere length measurement Blood samples drawn before surgery were collected in EDTA-treated tubes and processed within 2 hours to obtain buffy coats, which were stored at −80°C until analysis. A salting-out method was used to extract DNA from 50 μl of buffy coat with Gentra PureGene Cell Kit (QIAGEN Inc., Canada) according to the manufacturer’s protocol for 3.5 x 106 white cells. DNA quality and quantity were assessed using the NanoDrop® 2000c spectrophotometer (Thermo Scientific, Fisher Scientific Canada). Mean relative telomere length was measured with the quantitative polymerase chain reaction (qPCR) method first described by Cawthon RM [19], with slight modifications. Briefly, the telomeric repeats (T) were amplified using primers that hybridize to telomeres but have mismatches across their length that prevent primer-dimers formation (i.e. hybridization of two primers); amplification was measured quantitatively and compared to that of a single copy gene (S), to adjust for the amount of DNA in the reaction, assuming that both products are amplified with similar efficiency. The result is a relative telomere length estimation, the T/S ratio. The sequences of telomere primers used were: 5’- GGT TTT TGA GGG TGA GGG TGA GGG TGA GGG TGA GGG T-3’ (forward) and 5’- TCC CGA CTA TCC CTA TCC CTA TCC CTA TCC CTA TCC CTA -3’ (reverse). The single copy gene human beta‐globin (Hbg) primers were: 5’- GCT TCT GAC ACA ACT GTG TTC ACT AGC -3’ (forward) and 5’- CAC CAA CTT CAT CCA CGT TCA CC -3’ (reverse). For each sample, 20 μl of reaction solution was prepared using 1 ng of genomic DNA diluted to 0.2 ng/μl, 10 μl of 2× Brilliant III Ultra-Fast SYBR® Green QPCR Master Mix (Agilent Technologies), and either the telomere primer pair or the Hbg primer pair, each primer at a final concentration of 200 nM. For each sample, quadruplicates of telomere and quadruplicates of Hbg reaction solutions were amplified in the same qPCR run, in the Rotor-Gene Q instrument operated with Q-series software version 2.0.2.4 (Qiagen). The qPCR conditions consisted of three steps with melt, beginning with 95°C incubation for 3 minutes, followed by 40 cycles of: 95°C for 20 sec, 56°C for 60 sec and 72°C for 20 sec. After PCR amplification, melting curves were generated to confirm the specificity of PCR products. A negative control (no DNA template) and a reference DNA sample for normalization between experiments were run in duplicates in each batch. This same reference DNA sample was used to generate standard curves for telomere and Hbg amplifications; efficiency was 90% and 92%, respectively. The mean cycle threshold (Ct) values for both telomere and Hbg at a fluorescence signal threshold of 0.3 were calculated from the three closest values of quadruplicate samples with exclusion of the fourth value when it fell outside two standard deviations (SD) from the mean [20, 21]. The intra-assay coefficient of variation (CV) of the Ct was 1.80% and 0.92% and the inter-assay CV was 3.59% and 2.50% for telomeres and Hbg respectively. The comparative Ct method was used for relative quantification of telomere length, using this formula: relative T/S ratio = 2-ΔΔCt where ΔΔCt = (Ct Telomere−Ct Hbg) sample—(Ct Telomere−Ct Hbg) reference DNA [22]. All assays were performed blinded to the study patients’ characteristics and clinical data. Statistical analysis Telomere length, measured as a relative T/S ratio, was treated as a continuous variable, which is typically positively skewed. A Box-Cox transformation method was used to determine the suitable power transformation for the relative T/S ratio to obtain a normal distribution. General linear models (GLM, models fitted by least squares and weighted least squares using SAS Proc GLM) were conducted to evaluate the association between the square root-transformed relative T/S ratio and each of a set of pre-specified prognostic factors: age (years; quintiles), menopausal status, body mass index (BMI) (kg/m2; quintiles), smoking status (never, former, current), alcohol consumption (drinks per week; quintiles), physical activity (MET-hours of activity per week; quintiles), TNM stage (0, I, II, III), histological type (in-situ ductal, invasive ductal, invasive lobular, other), tumor grade (1, 2, 3), estrogen receptor (ER) status (negative, positive), progesterone receptor (PR) status (negative, positive) and human epidermal growth factor receptor 2 (HER2) status (negative, positive). For one variable (transportation-related physical activity), 77% of the values were zeros, which generated empty cells and less than five observations in all quintile categories; consequently, this variable was dichotomized (presence vs absence of transportation-related physical activity). To comply with statistical modeling assumptions, age- and menopausal status-adjusted associations between the square root-transformed relative T/S ratio and each factor were estimated. The same analyses were performed with stratification for menopausal status. The resulting adjusted estimates were back transformed to adjusted means of relative T/S ratio. Contrast statements were computed to generate tests for linear trends, using appropriate orthogonal polynomial coefficients for unequally spaced means in the GLM procedure. Spearman partial correlations of the relative T/S ratio and each of the above factors, while adjusting for age and menopausal status were computed. Inclusion of all the pre-specified factors in multivariable models and Spearman partial correlations did not change the observed associations, hence, age- and menopausal status- adjusted estimates are presented. Given that our sample size was fixed at 162, the present study was powered to detect a significant correlation ≥0.20 and at least a 0.3 standardized difference with 80% power and a two-sided statistical significance of 5% [23]. All statistical analyses were performed with SAS software version 9.4. Results Telomere length was estimated for all of the 162 Caucasian patients (mean ± SD of relative T/S ratio = 1.06 ± 0.63, median = 0.97, range 0.04 to 3.04). Characteristics of study participants are presented in Table 1. Patients were aged between 30 and 69 years (median = 52), and 50% were pre-menopausal. The majority had an invasive ductal carcinoma stage I or II and none had distant metastasis. More post-menopausal women were obese (28.4% with BMI ≥30 kg/m2) and former smokers (53.1%) compared to premenopausal women (16.1% with BMI ≥30 kg/m2; 35.8% former smokers). Pre-menopausal women were more active, with higher total (137.5 ± 47.2 vs 99.7 ± 48.6 MET-hours per week) and occupational (79.9 ± 37.6 vs 47.4 ± 47.3 MET-hours per week) physical activity. Post-menopausal women had more advanced disease with regard to stage (14.8% vs 5% stage III), more invasive lobular tumors (13.6% vs 4.9%), more hormonal negative tumors (12.4% vs 8.6% for ER, 23.5% vs 8.6% for PR), and slightly more HER2 positive tumors (12.3% vs 9.9%) compared to pre-menopausal women. 10.1371/journal.pone.0161903.t001Table 1 Patient characteristics. All women Pre-menopausal Post-menopausal n = 162 n = 81 n = 81 Number (%) Age (mean ± SD, years) 52.6 ± 7.9 46.8 ± 5.8 58.3 ± 4.9 BMI (mean ± SD, kg/m2) 27.0 ± 5.6 26.3 ± 5.6 27.7 ± 5.5     <25 67 (41.4) 39 (48.2) 28 (34.6)     25-<30 59 (36.4) 29 (35.8) 30 (37.0)     ≥30 36 (22.2) 13 (16.1) 23 (28.4) Smoking status     Never 69 (42.6) 41 (50.6) 28 (34.6)     Former 72 (44.4) 29 (35.8) 43 (53.1)     Current 21 (13.0) 11 (13.6) 10 (12.4) Alcohol consumption (mean ± SD, drink per week) 4.3 ± 4.6 4.5 ± 4.2 4.1 ± 5.0 Physical activity (mean ± SD, MET-hours per week)     Total 118.6 ± 51.4 137.5 ± 47.2 99.7 ± 48.6     Occupational 63.6 ± 45.6 79.9 ± 37.6 47.4 ± 47.3     Transportation-related 0.7 ± 2.0 0.8 ± 2.3 0.7 ± 1.7     Household 36.0 ± 23.3 36.2 ± 24.0 35.7 ± 22.6     Recreational 18.3 ± 17.0 20.6 ± 19.1 16.0 ± 14.3 Stage     0 16 (9.9) 9 (11.1) 7 (8.6)     I 64 (39.5) 33 (40.7) 31 (38.3)     II 66 (40.7) 35 (43.2) 31 (38.3)     III 16 (9.9) 4 (5.0) 12 (14.8) Histological type     Ductal, in-situ 16 (9.9) 9 (11.1) 7 (8.6)     Ductal, invasive 121 (74.7) 66 (81.5) 55 (67.9)     Lobular, invasive 15 (9.3) 4 (4.9) 11 (13.6)     Others* 10 (6.1) 2 (2.5) 8 (9.9) Tumor grade     Non-assessable 26 (16.0) 11 (13.6) 15 (18.5)     1 29 (17.9) 15 (18.5) 14 (17.3)     2 68 (42.0) 36 (44.4) 32 (39.5)     3 39 (24.1) 19 (23.5) 20 (24.7) ER status     Positive 145 (89.5) 74 (91.4) 71 (87.6)     Negative 17 (10.5) 7 (8.6) 10 (12.4) PR status     Positive 136 (84.0) 74 (91.4) 62 (76.5)     Negative 26 (16.0) 7 (8.6) 19 (23.5) HER2 status     Not evaluated 28 (17.3) 11 (13.6) 17 (21.0)     Positive 18 (11.1) 8 (9.9) 10 (12.3)     Negative 116 (71.6) 62 (76.5) 54 (66.7) SD: Standard deviation; BMI: Body mass index; MET-hours: metabolic equivalent hours of activity; ER: Estrogen receptor; PR: Progesterone receptor; HER2: Human epidermal growth factor receptor 2 *: includes mucinous, tubular, adenoid cystic and metaplastic carcinomas Spearman correlations coefficients and results from the GLM models of the associations of telomere length and traditional prognostic factors are presented in Table 2. Telomere length was not associated with age or with menopausal status. No association was observed for telomere length with stage, histological type, tumor grade, ER status, PR status and HER2 status, either before or after stratification according to menopausal status (Table 2). 10.1371/journal.pone.0161903.t002Table 2 Associations of telomere length and traditional prognostic factors. All women Pre-menopausal Post-menopausal Mean (95% CI) Adjusted* mean (95% CI) Adjusted* mean (95% CI) Adjusted* mean (95% CI) Age (quintiles, years) 30–47 1.16 (0.96, 1.38) 1.21 (0.95, 1.50) 30–42 1.05 (0.78, 1.36) 43–54 1.21 (0.94, 1.53) 48–51 0.92 (0.75, 1.12) 0.96 (0.75, 1.19) 43–47 1.26 (0.94, 1.63) 55–57 0.94 (0.70, 1.21) 52–55 1.15 (0.91, 1.43) 1.15 (0.91, 1.42) 48–50 0.99 (0.74, 1.28) 58–59 1.19 (0.90, 1.52) 56–59 1.09 (0.89, 1.31) 1.04 (0.81, 1.31) 51 0.78 (0.50, 1.15) 60–62 1.14 (0.83, 1.50) 60–69 0.99 (0.78, 1.22) 0.94 (0.70, 1.23) 52–57 1.15 (0.81, 1.56) 63–69 0.86 (0.62, 1.14) p-trend 0.475 0.365 0.633 0.180 rs (p-value) -0.035 (0.659) -0.079 (0.321) -0.037 (0.746) -0.150 (0.182) Menopausal status Pre-menopausal 1.04 (0.91, 1.18) 0.97 (0.82, 1.14) Post-menopausal 1.07 (0.94, 1.21) 1.14 (0.97, 1.33) p-value 0.777 0.227 rs (p-value) 0.005 (0.947) 0.073 (0.361) Stage 0 0.93 (0.67, 1.24) 0.93 (0.79, 1.24) 0.80 (0.48, 1.21) 1.13 (0.72, 1.64) I 1.09 (0.94, 1.26) 1.09 (0.94, 1.25) 1.04 (0.83, 1.27) 1.15 (0.94, 1.38) II 1.06 (0.91, 1.22) 1.06 (0.92, 1.22) 1.16 (0.94, 1.40) 0.96 (0.77, 1.17) III 1.04 (0.76, 1.37) 1.03 (0.75, 1.37) 0.92 (0.43, 1.63) 1.07 (0.76, 1.44) p-trend 0.660 0.672 0.649 0.682 rs (p-value) -0.001 (0.994) -0.002 (0.976) 0.099 (0.383) -0.102 (0.370) Histological type Ductal, in-situ 0.93 (0.67, 1.24) 0.93 (0.67, 1.28) 0.80 (0.48, 1.21) 1.13 (0.72, 1.64) Ductal, invasive 1.06 (0.95, 1.18) 1.06 (0.95, 1.18) 1.07 (0.92, 1.24) 1.04 (0.89, 1.20) Lobular, invasive 1.03 (0.75, 1.37) 1.06 (0.77, 1.38) 1.48 (0.82, 2.37) 0.95 (0.64, 1.33) Others*, invasive 1.24 (0.86, 1.70) 1.25 (0.86, 1.70) 1.00 (0.34, 2.10) 1.32 (0.90, 1.83) p-value 0.673 0.676 0.374 0.592 rs (p-value) 0.064 (0.416) 0.071 (0.370) 0.173 (0.125) 0.006 (0.958) Tumor grade 1 1.05 (0.84, 1.29) 1.04 (0.83, 1.28) 1.10 (0.78, 1.47) 0.98 (0.71, 1.31) 2 1.08 (0.94, 1.24) 1.09 (0.95, 1.25) 1.10 (0.89, 1.34) 1.09 (0.89, 1.30) 3 1.07 (0.88, 1.27) 1.06 (0.88, 1.27) 1.16 (0.87, 1.50) 0.97 (0.74, 1.24) p-trend 0.909 0.886 0.792 0.947 rs (p-value) 0.043 (0.587) 0.050 (0.528) 0.176 (0.119) -0.080 (0.480) ER status Positive 1.11 (0.82, 1.43) 1.05 (0.96, 1.16) 1.07 (0.92, 1.22) 1.04 (0.91, 1.19) Negative 1.05 (0.95, 1.15) 1.07 (0.79, 1.40) 0.94 (0.54, 1.46) 1.17 (0.81, 1.60) p-value 0.732 0.905 0.611 0.540 rs (p-value) -0.043 (0.590) -0.033 (0.681) 0.047 (0.680) -0.099 (0.382) PR status Positive 1.05 (0.94, 1.15) 1.05 (0.95, 1.16) 1.07 (0.92, 1.22) 1.04 (0.90, 1.19) Negative 1.12 (0.88, 1.38) 1.08 (0.85, 1.35) 0.92 (0.54, 1.46) 1.13 (0.87, 1.43) p-value 0.602 0.811 0.611 0.558 rs (p-value) -0.044 (0.578) -0.029 (0.715) 0.047 (0.680) -0.079 (0.488) HER2 status Positive 1.08 (0.81, 1.38) 1.07 (0.81, 1.38) 1.12 (0.95, 1.29) 1.03 (0.70, 1.17) Negative 1.06 (0.95, 1.18) 1.06 (0.95, 1.18) 1.12 (0.71, 1.65) 1.01 (0.86, 1.43) p-value 0.938 0.947 0.979 0.914 rs (p-value) 0.031 (0.695) 0.043 (0.593) 0.188 (0.094) -0.104 (0.359) Adjusted means and p-trend values from the general linear models (GLM); * Adjusted for: age and menopausal status, when applicable; rs: Spearman correlation coefficient; ER: Estrogen receptor; PR: Progesterone receptor; HER2: Human epidermal growth factor receptor 2; *: includes mucinous, tubular, adenoid cystic and metaplastic carcinomas Spearman correlations coefficients and results from the GLM models of the associations of telomere length and lifestyle factors are presented in Table 3. Telomere length increased linearly with increasing levels of total physical activity (rs = 0.17, P = 0.033; Ptrend = 0.069), occupational physical activity (rs = 0.15, P = 0.054; Ptrend = 0.054) and transportation-related physical activity (rs = 0.19, P = 0.019; P = 0.005) (Table 3 and Fig 1). When stratified by menopausal status (Table 3 and Fig 1), linear trends for increasing telomere length were observed for total physical activity (rs = 0.27, P = 0.016; Ptrend = 0.054) and occupational physical activity (rs = 0.26, P = 0.021; Ptrend = 0.056) in post-menopausal women, while in pre-menopausal women, TL was only associated with transportation-related physical activity (rs = 0.27, P = 0.015; P = 0.004). No associations were observed for telomere length with recreational or household activities, or the other lifestyle factors considered, namely BMI, smoking status and alcohol consumption, either before or after stratification according to menopausal status (Table 3). 10.1371/journal.pone.0161903.g001Fig 1 Association of telomere length with levels of total, occupational and transportation-related physical activity. Adjusted means and 95% confidence intervals of relative telomere length according to: (a) total physical activity (quintiles), (b) occupational physical activity (quintiles) and (c) transportation-related physical activity (dichotomous) for all, pre-menopausal and post-menopausal women, with adjustment for age (continuous) and menopausal status when applicable. a, b, c: MET-hours per week: metabolic equivalent hours of activity per week. 10.1371/journal.pone.0161903.t003Table 3 Association of telomere length and lifestyle factors. All women Pre-menopausal Post-menopausal Mean (95% CI) Adjusted* mean (95% CI) Adjusted* mean (95% CI) Adjusted* mean (95% CI) BMI (quintiles, kg/m2) 17.2–22.2 1.14 (0.93, 1.38) 1.13 (0.92, 1.37) 17.3–21.6 1.27 (0.94, 1.65) 17.2–22.8 0.95 (0.70, 1.25) 22.3–24.8 0.95 (0.76, 1.16) 0.95 (0.76, 1.16) 21.7–24.1 0.99 (0.70, 1.35) 22.9–25.5 1.33 (1.04, 1.67) 24.9–27.0 1.13 (0.91, 1.36) 1.13 (0.91, 1.36) 24.2–26.2 0.83 (0.58, 1.14) 25.6–27.4 1.06 (0.79, 1.38) 27.1–30.6 1.08 (0.88, 1.31) 1.10 (0.89, 1.33) 26.3–29.3 1.11 (0.79, 1.48) 27.5–32.1 0.90 (0.66, 1.19) 30.7–48.6 1.00 (0.80, 1.22) 0.98 (0.78, 1.20) 29.4–45.9 1.11 (0.81, 1.46) 32.2–48.6 1.04 (0.78, 1.34) p-trend 0.592 0.544 0.900 0.710 rs (p-value) -0.040 (0.614) -0.039 (0.628) -0.039 (0.732) -0.035 (0.760) Smoking status Never 1.05 (0.91, 1.21) 1.07 (0.82, 1.36) 1.01 (0.82, 1.22) 1.12 (0.90, 1.36) Former 1.06 (0.92, 1.21) 1.06 (0.92, 1.21) 1.06 (0.83, 1.32) 1.05 (0.88, 1.24) Current 1.06 (0.81, 1.35) 1.05 (0.91, 1.20) 1.22 (0.84, 1.68) 0.92 (0.61, 1.30) p-trend 0.974 0.911 0.361 0.360 rs (p-value) 0.004 (0.964) 0.001 (0.994) -0.074 (0.513) 0.098 (0.389) Alcohol consumption (quintiles, drink/week) 0–0.2 1.15 (0.89, 1.44) 1.13 (0.88, 1.43) 0.0–0.2 1.28 (0.93, 1.69) 0.0–0.1 1.09 (0.79, 1.43) 0.3–2.0 1.09 (0.92, 1.28) 1.09 (0.92, 1.27) 0.3–3.0 0.81 (0.60, 1.05) 0.2–1.0 1.18 (0.92, 1.48) 2.1–5.0 0.96 (0.77, 1.17) 0.94 (0.75, 1.16) 3.1–5.0 1.05 (0.74, 1.42) 1.1–4.0 1.08 (0.81, 1.41) 5.1–7.0 1.04 (0.84, 1.26) 1.05 (0.85, 1.27) 5.1–7.0 1.09 (0.80, 1.43) 4.1–7.0 0.98 (0.74, 1.25) 7.1–28.0 1.07 (0.83, 1.35) 1.11 (0.86, 1.39) 7.1–16.0 1.27 (0.89, 1.71) 7.1–28.0 0.93 (0.63, 1.30) p-trend 0.758 0.988 0.412 0.300 rs (p-value) -0.070 (0.377) -0.055 (0.490) 0.041 (0.719) -0.127 (0.260) Physical activity (quintiles, MET-hours per week) Total 13.0–77.3 0.86 (0.69, 1.06) 0.85 (0.66, 1.07) 23.7–101.6 0.80 (0.56, 1.10) 13.0–56.6 0.87 (0.62, 1.16) 77.4–108.7 0.92 (0.74, 1.12) 0.91 (0.73, 1.11) 101.7–124.3 1.23 (0.92, 1.59) 56.7–87.6 0.96 (0.71, 1.26) 108.8–125.8 1.38 (1.15, 1.63) 1.37 (1.15, 1.62) 124.4–146.5 1.29 (0.96, 1.67) 87.7–110.1 1.01 (0.74, 1.33) 125.9–158.1 1.01 (0.82, 1.22) 1.02 (0.83, 1.24) 146.6–169.4 0.92 (0.65, 1.26) 110.2–135.1 1.25 (0.95, 1.59) 158.2–317.7 1.11 (0.91, 1.34) 1.13 (0.92, 1.36) 169.5–317.7 1.03 (0.75, 1.35) 135.2–256.7 1.23 (0.94, 1.56) p-trend 0.068 0.069 0.645 0.054 rs (p-value) 0.163 (0.038) 0.169 (0.033) 0.056 (0.622) 0.269 (0.016) Occupational 0.0–5.5 0.82 (0.65, 1.01) 0.80 (0.61, 1.01) 0.0–57.7 1.13 (0.82, 1.49) 0.0–2.4 0.88 (0.67, 1.12) 5.6–60.0 1.27 (1.05, 1.51) 1.25 (1.03, 1.50) 57.8–72.4 0.95 (0.66, 1.30) 2.5–12.0 0.78 (0.51, 1.13) 60.1–76.9 1.02 (0.84, 1.23) 1.04 (0.85, 1.25) 72.5–82.4 0.97 (0.70, 1.30) 12.1–63.0 1.38 (1.08, 1.73) 77.0–99.5 0.91 (0.72, 1.12) 0.92 (0.73, 1.14) 82.5–108.5 1.01 (0.73, 1.35) 63.1–82.4 0.98 (0.72, 1.29) 99.6–265.2 1.27 (1.05, 1.51) 1.28 (1.05, 1.53) 108.6–265.2 1.24 (0.91, 1.63) 82.5–227.8 1.25 (0.96, 1.57) p-trend 0.056 0.054 0.628 0.056 rs (p-value) 0.151 (0.056) 0.153 (0.054) 0.051 (0.654) 0.258 (0.021) Physical activity (quintiles, MET-hours per week) Transportation-related† 0 0.98 (0.88, 1.08) 0.98 (0.88, 1.08) 0.93 (0.79, 1.09) 1.03 (0.89, 1.18) >0 1.33 (1.11, 1.56) 1.32 (1.11, 1.55) 1.45 (1.13, 1.80) 1.17 (0.90, 1.49) p-value 0.004 0.005 0.004 0.361 rs (p-value) 0.192 (0.014) 0.186 (0.019) 0.271 (0.015) 0.094 (0.409) Household 7.1–17.5 1.06 (0.86, 1.29) 1.07 (0.87, 1.30) 7.1–17.5 0.96 (0.70, 1.28) 8.7–18.2 1.19 (0.90, 1.51) 17.6–22.6 0.99 (0.79, 1.21) 0.99 (0.79, 1.22) 17.6–21.0 0.94 (0.66, 1.28) 18.3–24.2 1.03 (0.77, 1.34) 22.7–35.8 1.06 (0.86, 1.28) 1.05 (0.85, 1.27) 21.1–36.7 1.32 (0.98, 1.71) 24.3–35.2 0.81 (0.57, 1.09) 35.9–53.6 1.20 (0.98, 1.44) 1.19 (0.97, 1.43) 36.8–52.5 1.15 (0.83, 1.54) 35.3–53.6 1.21 (0.92, 1.55) 53.7–148.6 0.99 (0.79, 1.21) 0.99 (0.79, 1.21) 52.6–113.9 0.94 (0.67, 1.25) 53.7–148.6 1.07 (0.80, 1.38) p-trend 0.913 0.882 0.824 0.890 rs (p-value) 0.040 (0.610) 0.038 (0.638) 0.084 (0.460) -0.009 (0.938) Recreational 0.0–3.8 0.98 (0.79, 1.20) 0.99 (0.79, 1.10) 0.0–5.2 1.17 (0.85, 1.55) 0.0–3.3 0.90 (0.65, 1.20) 3.9–9.4 1.11 (0.90, 1.34) 1.10 (0.89, 1.34) 5.3–11.0 1.06 (0.77, 1.41) 3.4–8.8 1.15 (0.87, 1.47) 9.5–19.0 1.14 (0.93, 1.37) 1.14 (0.94, 1.37) 11.1–19.5 0.94 (0.65, 1.28) 8.9–14.5 1.14 (0.84, 1.48) 19.1–27.9 0.95 (0.75, 1.17) 0.95 (0.76, 1.17) 19.6–34.6 1.03 (0.75, 1.37) 14.6–27.4 1.08 (0.81, 1.39) 28.0–104.8 1.10 (0.89, 1.33) 1.09 (0.89, 1.32) 34.7–104.8 1.09 (0.78, 1.45) 27.5–61.9 1.04 (0.77, 1.36) p-trend 0.810 0.875 0.896 0.830 rs (p-value) 0.023 (0.775) 0.020 (0.806) 0.019 (0.869) 0.042 (0.714) Adjusted means and p-trend values from the general linear models (GLM) * Adjusted for: age and menopausal status, when applicable †: Dichotomous categories (0: n = 125; >0: n = 37); rs: Spearman correlation coefficient; BMI: Body mass index; MET-hours: Metabolic equivalent hours of activity Discussion The present study aimed at assessing the association of telomere length with traditional and potential prognostic factors. The findings suggest that peripheral white blood cell telomeres are longer in more active breast cancer patients, especially for transportation-related physical activity among pre-menopausal patients, and for total and occupational physical activity among post-menopausal patients. Neither age nor menopausal status, nor tumor prognostic factors nor certain modifiable factors were associated with peripheral white blood cell telomere length. Although modest associations of physical activity with telomere length have been reported in healthy individuals (an increase of 0.07-SD of relative telomere length in moderately or highly active women vs least active women) [9] and breast cancer patients (β = −0.22, 95% confidence interval (CI): -0.41 to -0.03, n = 392 post-menopausal women) [24], the present study is the first to reveal associations of different types of physical activity with telomere length in pre- and post-menopausal breast cancer patients. Mean MET-hours per week of total physical activity in our population was higher than total energy expenditure recommended for healthy adults to achieve health benefits while mean MET-hours per week of transportation-related physical activity was relatively very low. In fact, based on a systematic review of 254 studies there is a dose-response relationship between increased physical activity and health benefits [25]. To achieve health benefits, healthy adults should accumulate at least 150 minutes of moderate- to vigorous-intensity aerobic physical activity per week, which corresponds to an energy expenditure comprised between 500 and 1,000 MET-minutes per week (8.33 and 16.67 MET-hours per week) [26, 27]. Our findings suggest a similar dose-response relationship between physical activity and telomere length, but not for all physical activity domains, and depending on the menopausal status. Low intensity, but probably regularly performed, physical activity seems to be associated with longer peripheral blood cell telomere length. Hence, breast cancer patients, for whom the recommendation is to engage in regular physical activity at least 150 minutes per week [28], but among whom moderate to vigorous exercise may be difficult to achieve, may benefit from regular low-intensity physical activity. Even though older age was found to be related to shorter telomere length in healthy women, with statistically significant correlation coefficients varying from −0.09 (p-value <0.001, n = 7813, of whom 80% were post-menopausal women) [9] to –0.23 (p-value <0.04, n = 58 premenopausal women) [8], it is probably not the only determinant of telomere shortening in breast cancer patients. In fact, only one out of five studies reported a statistically significant adjusted association of older age with shorter peripheral white blood cell telomere length [15], when comparing patients less than 55 years old to those older than 65 (β = −0.26, p-value = 0.02, n = 392 post-menopausal women) [24]. The same observation was made for other modifiable factors (BMI, smoking, alcohol consumption) which were found to be associated with shorter telomere length in healthy women [7, 10], but not in breast cancer patients [15]. Only two studies have assessed the associations of tumor prognostic factors with peripheral blood cell relative telomere length, and one reported no association with ER status after adjustment for age [15, 29]. Moreover, longitudinal cohort studies suggest that telomere length might be an independent prognostic factor [15]. The strengths of the present study include the recruitment of a consecutive series of women presenting with breast cancer and the high participation percentage among eligible women (73%), which minimizes the risk of selection bias. Even though pre-menopausal women constituted half of our population, the distributions of study patients by age category–with 70.4% aged older than 50 years–and tumor characteristics—74.7% had ductal invasive carcinomas and 89.5%, ER-positive tumors—were very similar to those of the breast cancer population [30]. Additionally, telomere length was estimated for all patients who provided blood samples and all participants were included in statistical analyses. Data collection at time of surgery using standardized measures and questionnaires, the use of an appropriate DNA extraction method and the assessment of telomere length with an appropriate and highly reproducible method ensured the quality of the data, and prevented selection bias resulting from missing values and measurement bias. All the participants were approached for information about risk factors when they were not yet aware of their disease severity and stage (an average of 24 days after surgery) or their telomere length, which prevents recall bias. All the laboratory assays were performed blinded to study patient characteristics and clinical data, which prevent bias from differential misclassification. Thus, if a measurement error had occurred, it would result in non-differential misclassification, and would have underestimated the true associations between telomere length and the factors studied. The estimates presented were all adjusted for age and menopausal status, and were not different from those adjusted for all the pre-specified known and potential prognostic factors. However, residual confounding from unknown factors, a common concern in observational studies, may still exist. The limitations include the cross-sectional design that precludes causal inferences. However, it seems very likely that the collected information about risk factors refers to exposure before blood collection, especially physical activity, for which questions were derived from the Past Year Total Physical Activity Questionnaire [17]. Even though peripheral white blood cell telomere length seems to be a dynamic feature [14], it is likely that factors affecting telomere length have a long latency period, as observed with chemotherapy, which seems to induce an initial telomere attrition 3–6 months after treatment that takes 1–5 years to recover [31, 32]. The relatively small sample size could also have been a limitation of the present study. However, the strength of the relationships between telomere length and the studied factors, as reflected by the size of correlation coefficients regardless of statistical significance, were as high as those observed in larger studies [8, 9]. Finally, given that white blood cells are involved in anticancer immune responses, they are likely to be linked with breast cancer prognosis. A longitudinal analysis of the association of telomere length at the time of diagnosis, as a surrogate for both innate adaptive abilities and cumulative exposures to modifiable environmental factors, with survival is still needed to demonstrate the significance of telomere length as an independent prognostic marker. Conclusions Longer peripheral blood cell telomeres seem to be associated with higher levels of physical activity in breast cancer patients, especially for physical activity related to occupation and transportation. Telomere length was not associated with any of the other known or potential prognostic factors. These findings suggest that even regular low-intensity physical activity could be effectively recommended to breast cancer patients, and may contribute to the control of cancer along with conventional therapies. The authors wish to thank all the participants for allowing access to their biological samples and data, and for the time they devoted to answering questions. We are also grateful to all research team members who were involved in participants’ recruitment and data acquisition. ==== Refs References 1 Ferlay J, Soerjomataram I, Ervik M, Dikshit R, Eser S, Mathers C, et al. GLOBOCAN 2012 v1.0, Cancer Incidence and Mortality Worldwide: IARC CancerBase No. 11 [Internet] Lyon, France.: International Agency for Research on Cancer; 2013. Available: http://globocan.iarc.fr, accessed on 14/12/2015. 2 Torre LA , Bray F , Siegel RL , Ferlay J , Lortet-Tieulent J , Jemal A . Global cancer statistics, 2012 . CA Cancer J Clin . 2015 ;65 (2 ):87 –108 . 10.3322/caac.21262 25651787 3 Lu W , Zhang Y , Liu D , Songyang Z , Wan M . Telomeres-structure, function, and regulation . Exp Cell Res . 2013 ;319 (2 ):133 –41 . 10.1016/j.yexcr.2012.09.005 23006819 4 Hug N , Lingner J . Telomere length homeostasis . Chromosoma . 2006 ;115 (6 ):413 –25 . 16741708 5 Aubert G , Lansdorp PM . Telomeres and aging . Physiol Rev . 2008 ;88 (2 ):557 –79 . 10.1152/physrev.00026.2007 18391173 6 Baird DM . Telomeres II . Exp Gerontol . 2008 ;43 (1 ):15 –9 . 17981417 7 Valdes AM , Andrew T , Gardner JP , Kimura M , Oelsner E , Cherkas LF , et al Obesity, cigarette smoking, and telomere length in women . Lancet . 2005 ;366 (9486 ):662 –4 . 16112303 8 Epel ES , Blackburn EH , Lin J , Dhabhar FS , Adler NE , Morrow JD , et al Accelerated telomere shortening in response to life stress . Proc Natl Acad Sci U S A . 2004 ;101 (49 ):17312 –5 . 15574496 9 Du M , Prescott J , Kraft P , Han J , Giovannucci E , Hankinson SE , et al Physical activity, sedentary behavior, and leukocyte telomere length in women . Am J Epidemiol . 2012 ;175 (5 ):414 –22 . 10.1093/aje/kwr330 22302075 10 Sun Q , Shi L , Prescott J , Chiuve SE , Hu FB , De Vivo I , et al Healthy lifestyle and leukocyte telomere length in U.S. women . PLoS One . 2012 ;7 (5 ):e38374 10.1371/journal.pone.0038374 22675460 11 Spark LC , Reeves MM , Fjeldsoe BS , Eakin EG . Physical activity and/or dietary interventions in breast cancer survivors: a systematic review of the maintenance of outcomes . J Cancer Surviv . 2013 ;7 (1 ):74 –82 . 10.1007/s11764-012-0246-6 23179496 12 Gou YJ , Xie DX , Yang KH , Liu YL , Zhang JH , Li B , et al Alcohol Consumption and Breast Cancer Survival: A Meta- analysis of Cohort Studies . Asian Pac J Cancer Prev . 2013 ;14 (8 ):4785 –90 . 24083744 13 Braithwaite D , Izano M , Moore DH , Kwan ML , Tammemagi MC , Hiatt RA , et al Smoking and survival after breast cancer diagnosis: a prospective observational study and systematic review . Breast Cancer Res Treat . 2012 ;136 (2 ):521 –33 . 10.1007/s10549-012-2276-1 23053660 14 Teixeira MT , Arneric M , Sperisen P , Lingner J . Telomere length homeostasis is achieved via a switch between telomerase- extendible and -nonextendible states . Cell . 2004 ;117 (3 ):323 –35 . 15109493 15 Ennour-Idrissi K , Maunsell E , Diorio C . Telomere length and breast cancer prognosis: a systematic review . Cancer Epidemiol Biomarkers Prev . 2016 ;In press. 16 Hanna M , Dumas I , Jacob S , Tetu B , Diorio C . Physical activity, mammographic density, and age-related lobular involution among premenopausal and postmenopausal women . Menopause . 2015 ;22 (9 ):964 –75 . 10.1097/GME.0000000000000433 25710783 17 Friedenreich CM , Courneya KS , Neilson HK , Matthews CE , Willis G , Irwin M , et al Reliability and validity of the Past Year Total Physical Activity Questionnaire . Am J Epidemiol . 2006 ;163 (10 ):959 –70 . 16524954 18 American Joint Committee on Cancer . Breast 7th ed. Edge SB , Byrd DR , Compton C , et al, editors. New York, NY : Springer ; 2010 . 19 Cawthon RM . Telomere measurement by quantitative PCR . Nucleic Acids Res . 2002 ;30 (10 ):e47 12000852 20 Yang IV , Chen E , Hasseman JP , Liang W , Frank BC , Wang S , et al Within the fold: assessing differential expression measures and reproducibility in microarray assays . Genome Biol . 2002 ;3 (11 ):research0062 12429061 21 Xiao H , Leung A , Yieh L . Quality Control of Microarray Data In: Carmen A , Hardiman G , editors. Biochips as pathways to drug discovery . Drug Discovery Series Boca Raton, USA : CRC Press ; 2007 . 22 Livak KJ , Schmittgen TD . Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method . Methods . 2001 ;25 (4 ):402 –8 . 11846609 23 Browner W , Newman T , Hulley S . Estimating Sample Size and Power: applications and examples In: Hully S , Cummings S , Browner W , Grady D , Newman T , editors. Designing Clinical Research . 3rd ed. Baltimore : Lippincott Williams & Wilkins ; 2007 p. 65 –94 . 24 Garland SN , Johnson B , Palmer C , Speck RM , Donelson M , Xie SX , et al Physical activity and telomere length in early stage breast cancer survivors . Breast Cancer Res . 2014 ;16 (4 ):413 10.1186/s13058-014-0413-y 25074648 25 Warburton DE , Charlesworth S , Ivey A , Nettlefold L , Bredin SS . A systematic review of the evidence for Canada's Physical Activity Guidelines for Adults . Int J Behav Nutr Phys Act . 2010 ;7 :39 10.1186/1479-5868-7-39 20459783 26 Tremblay MS , Warburton DE , Janssen I , Paterson DH , Latimer AE , Rhodes RE , et al New Canadian physical activity guidelines . Appl Physiol Nutr Metab . 2011 ;36 (1 ):36 –46 ; 7–58. 10.1139/H11-009 21326376 27 Physical Activity Guidelines Advisory Committee. Physical Activity Guidelines Advisory Committee Report Washington, DC: U.S. Department of Health and Human Services. 2008. http://health.gov/paguidelines/guidelines/appendix1.aspx. Accessed February 2016; 2008 [cited 2016 February 29]. 28 Rock CL , Doyle C , Demark-Wahnefried W , Meyerhardt J , Courneya KS , Schwartz AL , et al Nutrition and physical activity guidelines for cancer survivors . CA Cancer J Clin . 2012 ;62 (4 ):243 –74 . 10.3322/caac.21142 22539238 29 Svenson U , Nordfjall K , Stegmayr B , Manjer J , Nilsson P , Tavelin B , et al Breast cancer survival is associated with telomere length in peripheral blood cells . Cancer Res . 2008 ;68 (10 ):3618 –23 . 10.1158/0008-5472.CAN-07-6497 18483243 30 Lakhani S, Ellis I, Schnitt S, Tan P, van de Vijver MJ. WHO Classification of Tumours of the Breast. 4th ed. Lyon2012. 31 Diker-Cohen T , Uziel O , Szyper-Kravitz M , Shapira H , Natur A , Lahav M . The effect of chemotherapy on telomere dynamics: clinical results and possible mechanisms . Leuk Lymphoma . 2013 ;54 (9 ):2023 –9 . 10.3109/10428194.2012.757765 23240911 32 Benitez-Buelga C , Sanchez-Barroso L , Gallardo M , Apellaniz-Ruiz M , Inglada-Perez L , Yanowski K , et al Impact of chemotherapy on telomere length in sporadic and familial breast cancer patients . Breast Cancer Res Treat . 2015 ;149 (2 ):385 –94 . 10.1007/s10549-014-3246-6 25528024

PMC005xxxxxx/PMC5003391.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757097410.1371/journal.pone.0162034PONE-D-15-55899Research ArticleResearch and Analysis MethodsBioassays and Physiological AnalysisElectrophysiological TechniquesMuscle ElectrophysiologyElectromyographyPhysical SciencesPhysicsThermodynamicsEntropyBiology and Life SciencesBiomechanicsBiological LocomotionWalkingBiology and Life SciencesPhysiologyBiological LocomotionWalkingMedicine and Health SciencesPhysiologyBiological LocomotionWalkingEngineering and TechnologySignal ProcessingSignal FilteringPeople and PlacesPopulation GroupingsAge GroupsElderlyBiology and Life SciencesDevelopmental BiologyOrganism DevelopmentAgingBiology and Life SciencesPhysiologyPhysiological ProcessesAgingMedicine and Health SciencesPhysiologyPhysiological ProcessesAgingPeople and PlacesPopulation GroupingsAge GroupsYoung AdultsEngineering and TechnologySignal ProcessingSignal FilteringBandpass FiltersDifferential Changes with Age in Multiscale Entropy of Electromyography Signals from Leg Muscles during Treadmill Walking Multiscale Entropy of EMG during WalkingKang Hyun Gu 1*Dingwell Jonathan B. 231 Kinesiology, California State University San Marcos, San Marcos, California, United States of America2 Kinesiology and Health Education, University of Texas at Austin, Austin, Texas, United States of America3 Biomedical Engineering, University of Texas at Austin, Austin, Texas, United States of AmericaLebedev Mikhail A. EditorDuke University, UNITED STATESCompeting Interests: The authors have declared that no competing interests exist. Conceived and designed the experiments: HGK JBD. Performed the experiments: HGK. Analyzed the data: HGK. Contributed reagents/materials/analysis tools: JBD. Wrote the paper: HGK JBD. * E-mail: hkang@csusm.edu29 8 2016 2016 11 8 e016203425 12 2015 13 7 2016 © 2016 Kang, Dingwell2016Kang, DingwellThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Age-related gait changes may be due to the loss of complexity in the neuromuscular system. This theory is disputed due to inconsistent results from single-scale analyses. Also, behavioral adaptations may confound these changes. We examined whether EMG dynamics during gait is less complex in older adults over a range of timescales using the multiscale entropy method, and whether slower walking attenuates this effect. Surface EMG was measured from the left vastus lateralis (VL), biceps femoris (BF), gastrocnemius (GA), and tibialis anterior (TA) in 17 young and 18 older adults as they walked on a treadmill for 5 minutes at 0.8x-1.2x of preferred speed. Sample entropy (SE) and the complexity index (CI) of the EMG signals were calculated after successive coarse-graining to extract dynamics at timescales of 27 to 270 Hz, with m = 2 and r = 0.15 SD. SE and CI were lower across the timescales in older adults in VL and BF, but higher in GA (all p<0.001); these results held for VL and GA even after accounting for longer EMG burst durations in older adults. CI was higher during slower walking speed in VL and BF (p<0.001). Results were mostly similar for m = 3 and r = 0.01–0.35. Smaller r was more sensitive to age-related differences. The decrease in complexity with aging in the timescales studied was limited to proximal muscles, particularly VL. The increase in GA may be driven by other factors. Walking slower may reflect a behavioral adaptation that allows the nervous system to function with greater complexity. Whitaker FoundationRG-02-0354Dingwell Jonathan B. This work was supported by Whitaker Foundation Biomedical Engineering Research Grant RG-02-0354 to JBD, and American Society of Biomechanics Grant-in-Aid and University of Texas A.D. Hutchinson Fellowship to HGK. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Data AvailabilityAll EMG files are available at the following URL: http://dx.doi.org/10.5061/dryad.p937t.Data Availability All EMG files are available at the following URL: http://dx.doi.org/10.5061/dryad.p937t. ==== Body Introduction Common gait alterations with aging include slower walking speed [1], increased kinematic variability [2], sensitivity to local perturbations [3], and decreased ankle push off [4]. However, the mechanisms for these behavioral alterations are not clear. These behavioral changes in aging have been hypothesized to be a result of a multi-system dysregulation [5], where interactions between the physiological systems that influence behavior break down, and thus limit function. Some alterations, such as walking slower, may instead be adaptive responses to attenuate the effects of such dysregulation. The changes in physiological dynamics due to aging and disease have been observed in cell membranes [6], the cardiovascular system [7], postural control [8] and gait [9] where the observed dynamics decrease in complexity. Here, complexity is defined as the presence of irregular dynamics over a wide range of time or spatial scales of a physiological variable, as quantified using various entropy metrics. Previous studies of aging motor function, however, have been inconclusive with regards to this question. Various single-scale methods were used to quantify dynamics at a particular timescale [10–14], but each method studied behavior at a different timescale, with different results. Such heterogeneity in methodology has made the synthesis across these results difficult. As these studies indicate, physiological systems appear to exhibit different types of behavior at different timescales. Therefore, studying motor behavior over a range of timescales would allow for a more complete picture. Therefore, our purpose was to more completely describe neuromuscular behavioral changes in gait due to aging, by quantifying the effect of age on the complexity of muscle activation patterns during walking as measured using the multiscale entropy method on surface electromyography (EMG) signals. We tested whether: (1) older adults exhibit less complex dynamics as predicted by the multisystem dysregulation hypothesis once we consider a range of timescales; and (2) these effects are less present when walking slower, possibly attenuating such effects in older adults who may exhibit slow gait. Methods Subjects Eighteen healthy older adults (age 65–85 years) and eighteen height-, weight-, and gender-matched young adults (age 18–28 years), participated after providing written informed consent as approved by the University of Texas at Austin institutional review board (Table 1). Data from one young adult was discarded due to poor quality, resulting in data from 17 young adults. We excluded those who reported recent lower extremity injuries, visible gait asymmetries or disabilities, or taking medications that may influence gait. 10.1371/journal.pone.0162034.t001Table 1 Subject Characteristics [15]. Young adults Older Adults p-value Gender (M/F) 12/5* 12/6 0.55** Age (years) 23.3 ± 2.6 72.1 ± 6.0 <0.0001 Height (m) 1.73 ± 0.094 1.70 ± 0.104 0.36 Body Mass (kg) 71.1 ± 9.86 73.2 ± 12.3 0.58 Preferred Walking Speed (PWS) (m/s) 1.30 ± 0.10 1.29 ± 0.15 0.86 PWS Range (m/s) 1.16–1.56 0.93–1.52 * Reflects the number after the data for one young subject were discarded **Fisher’s Exact Test (χ2). Subjects walked on a treadmill (Desmo S model, Woodway USA, Waukesha WI) while wearing a safety harness (Protecta International, Houston TX). Each subject’s preferred walking speed (PWS) was determined, which allowed for treadmill acclimation. Subjects completed two 5-minute walking trials each at 5 different speeds (0.8, 0.9, 1.0, 1.1, 1.2×PWS), and with in-between rests of 2 minutes. The order of presentation was randomized while preventing two consecutive fast speed trials to avoid fatigue. Subjects were instructed to look straight ahead and avoid extraneous movements. Kinematics were measured using Vicon 612 (Oxford Metrics, UK). Other details of data collection and standard EMG analyses are presented elsewhere [15]. Muscle activation patterns during walking were recorded using surface electromyography (EMG) (Delsys, Boston MA; Bagnoli-8, DE 2.1 electrodes) from the left vastus lateralis (VL), biceps femoris (BF), medial gastrocnemius (GA), and tibialis anterior (TA) [16]. The EMG data were sampled at 1080 Hz, bandpass filtered (passband 20–300 Hz in software; built-in 20–450 Hz hardware filter), notch-filtered at 60 Hz using zero-lag Butterworth filters, and demeaned using MATLAB 7.04 (Mathworks, Natick MA). Data from some trials were discarded from VL (15 of 349 collected), BF (17/349), GS (8/349), and TA (7/349) muscles due to signal quality issues. Also, noisy sections of the signals due to sweating, etc., were identified visually and were not used in subsequent analyses. Multiscale Entropy Calculations for Surface EMG The complexity of each EMG signal was quantified using the multiscale entropy (MSE) method described in detail elsewhere [17]. Briefly, MSE quantifies the degree of irregularity of time series data over multiple time scales. Time series data that are more irregular or entropic over a broad range of time scales are considered more complex than those that show irregular behavior at only a single time scale. Single-scale methods, such as approximate entropy and sample entropy (SampEn), yield higher entropy values for uncorrelated random signals than for signals with long-range correlations, especially with signals of high sampling frequency (i.e., small time scales). However, the reverse occurs at larger time scales, as long range patterns do not exist in uncorrelated noise and yield low entropy values, yet signals with long-range correlations still contain dynamics at these larger timescales [17]. EMG signals exhibit fairly uniform power spectrum throughout the passband. This power spectrum is similar to that of filtered uncorrelated white noise. Both signals display decreasing sample entropy with increasing timescales. Despite their “noise”-like spectral properties, EMG signals contain important information about muscle activation patterns [18]. Physiologically, surface EMG during gait is comprised of the filtered sum of action potentials from the muscle fibers, describing an overall “ensemble” effect of the activation of different motor units. Each motor unit contributes to the overall signal, and therefore its activation can add to the overall complexity of the surface EMG signal. During muscle activation, motor units cycle on and off, and their firing rates can change. We expect that physiological changes associated with aging, such as loss of motor units, to reduce the number of elements that contribute to the overall EMG signal, and thus produce a less complex output. This result is predicted by the multisystem dysregulation hypothesis [5]. However, both motor unit firings and firing patterns can become more variable, which could therefore cause the overall signal to be less predictable and therefore exhibit higher entropy. These different physiological mechanisms may explain the discrepancies observed between different previous studies. In addition, multiscale relationships between EMG signal dynamics and time scales on the results has not been adequately considered in many studies of aging motor function, yielding seemingly conflicting results as they tend to compare different timescales [10–14]. In contrast, the MSE method considers SampEn over a range of timescales, and provides a more complete picture of the signal properties, and thus could clarify this issue. The MSE analysis consists of three steps: 1) coarse-graining the original time series to derive multiple signals, each capturing the system dynamics at different scales; 2) calculating a measure of entropy suitable for finite time series, SampEn in this case, for each coarse-grained time series; and 3) integrating the entropy values over a pre-defined range of scales to obtain an index of complexity (CI). For a time series [x1, …, xN], each element j of the coarse-grained time series y for scale n was calculated according to the Eq 1. yj(n)=1n∑i=(j−1)n+1jnxi(1) MSE, as noted, uses sample entropy to quantify the degree of irregularity of a time series. Sample entropy is a conditional probability measure that quantifies the likelihood that if a vector with m data points matches a template with the same length, then the vector and template will still match when their length increases from m to m+1 data points, within a given tolerance r. SampEn can theoretically vary from 0 to infinity, but generally ranges between 0 and 3 in the literature [17]. Plotting sample entropy for each coarse-grained time series as a function of time scale yields the MSE curve (Fig 1). The complexity index (CI) is the area under the MSE curve [17] that indicates the amount of information, or entropy, in a signal over a certain range of time scales. Consistently high entropy values over a wide range of time scales, and thus high CI, indicate high complexity, and vice versa [17]. The length of the original time series determines the largest scale that can be analyzed [17]. Here, CI was calculated using the trapezoid rule. 10.1371/journal.pone.0162034.g001Fig 1 Schematic of Multiscale Entropy Calculation. (A) After the original EMG signal is bandpass filtered at 20–300 Hz, it is coarse grained to extract dynamics at different time scales. (B) Sample entropy (SampEn) is calculated from each coarse grained signal. For each pattern of m points in the signal (template example: ×-■), places in other parts of the signal where the template is seen are identified within tolerance r. SampEn is calculated as the negative natural log (-ln) of the conditional probability that the pattern of m+1 points (×-■-○) will match if that the pattern of m points (×-■) did match. In other words, after the signal matched the first two parts of the pattern ×-■, this is the probability that pattern match will complete, ×-■-○. The number of ×-■ matches are compared to the number of complete pattern (×-■-○) matches. Higher SampEn indicates that the signal is less predictable, and thus more irregular. (C) Multiscale view of the signal is derived by examining sample entropy of the EMG at each of the coarse-grained time scales. Complexity index CI is defined as the area under the curve. Higher CI indicates that the signal has unpredictable dynamics over a wide range of time scales, and thus more complex. This use of multiple entropy method uses the same amount of tolerance r for all of the timescales. For signals that do not exhibit dynamics at larger timescales, coarse graining will produce signals with low amplitudes relative to r, and that are thus more “regular” or of less entropy. Depending on the timescales in which complex dynamics can be observed, two different signals can produce the same overall complexity index. Thus, scale-by-scale comparisons, as performed here, are also useful. Sample Entropy Parameter Choices Parameter choices in entropy calculations can affect results, particularly in short data sets [19]. We used scales n = 4–40, template length m = 2 and tolerance r = 0.15 of the standard deviation of the processed EMG signal, using previous recommendations [17, 19–21]. Scales n = 4 through 40, equivalent to 1080Hz/4 = 270Hz through 1080Hz/40 = 27 Hz, were chosen to match the 20–300 Hz passband, although larger scales could be analyzed given the length of the time series (324,000), EMG dynamics at lower or higher frequencies did not exist in the signal after the bandpass filtering. Due to the high-pass hardware filtering at 20 Hz, analyses were not extended to larger timescales beyond n = 40 or 50 ms (or lower than 20 Hz) as larger time scale dynamics were removed by filtering. Likewise, dynamics at scale 1 (1080 Hz), scale 2 (540 Hz), and scale 3 (360 Hz) were not used as they would not contain information about the EMG signal after it was filtered [7], and to avoid any potential issues with oversampling. Template length m = 2 was chosen for computational expediency, in line with previous recommendations [18]. Tolerance r = 0.15 SD was chosen a priori based on literature. Although entropy measures are known to be sensitive to the choice of r for short data sets [19], we used long data sets to minimize the effect of this particular parameter choice. We also repeated the analysis for r = 0.01, 0.05, 0.25, and 0.35 SD, as well as m = 3 (and the same range of r). For computational expediency, these analyses were completed for scales n = 4, 8, 12, 16, 20, 24, 28, 32, 36, and 40. Trapezoid rule was used to calculate CI to make the comparisons similar to studying scales n = 4–40. Signal processing was performed using MATLAB 2016a (Mathworks, Natick MA) on Amazon EC2 cloud cluster. Phasic Bursts and Non-stationarity in EMG Because EMG signals during walking have strong phasic bursts during each stride, the EMG signal is locally non-stationary. This posed unique challenges in adapting the multiscale entropy method to the gait EMG signal. This non-stationarity can pose a significant problem in estimating entropy using the template-matching approach explained above. As the signal mean or the amplitude deviate from another part of the signal, the signal will no longer match the template. If only a few template matches occur, our estimate of the conditional probability of template matches would be inaccurate. For an accurate estimate of sample entropy, enough of the non-matches and matches need to occur. We addressed the non-stationarity of the signal mean with a high-pass filter in the post-processing and through using large amount of data. Detrending is a common method to remove the non-stationarity of the signal mean, particularly in short data sets [17]. Detrending was unavoidable in our study due to the built-in high-pass hardware filter and limited our ability to study longer timescales beyond 20 Hz or 50 ms. This high pass process minimized the fluctuation of the signal mean from one burst to another, and made the signal mean stationary across multiple strides. Another approach to create stationary amplitudes that has been used in the literature is to control the experimental protocol to very simple movements [18]. As this was not appropriate for studying gait, we instead collected a large quantity of continuous data (5 minutes @ 1080 Hz = 324,000 data points in a time series). After coarse graining, the time series lengths were between 8100 (= 324,000/40 for scale 40) to 81,000 (= 324,000/4 for scale 4), much longer than the recommended minimum of 150–200. Using long time series data, we ensured that each EMG signal contained enough bursts associated with enough strides, so that each signal was stationary in the large scale, and thus enough template matches would occur to accurately estimate sample entropy. EMG Duty Factor and the Non-stationarity of Amplitude Aside from non-stationarity of the signal mean, the non-stationarity of the amplitude (or variance) is another issue. Due to the phasic bursts, signal amplitudes are not stationary in EMG signals during gait due to gaps of minimal muscle activity between bursts between gait cycles. Likewise, any large spikes in the data posed the same problem, as the overall signal amplitude, as measured using the standard deviation is inflated by these spikes. Both of these issues make the matching tolerance large particularly in the low-amplitude quiescent portions of the EMG signal. Because the local signal amplitude is quite small compared to the large tolerance, many more template matches can occur and thus the signal will exhibit lower entropy. As this is the nature of the EMG signals during gait, this issue is unavoidable. Therefore, in preliminary work, we quantified the effect of burst duration vs. quiescence duration on the complexity index (CI) in a pilot simulation to better understand the potential impact of this effect on CI calculations. We called the proportion of the EMG “on” during the gait cycle “EMG duty factor”. Since aging is associated with longer burst durations, this could potentially confound our interpretation of the final results. To assess the possible effects of EMG duty factor on the complexity index, simulated EMG signals were created as bandpass filtered white noise (20–300 Hz, same settings used for recorded signals). Of note, although filtered white noise does not model all aspects of EMG signals during gait, it is used here because (1) it contained similar power spectra and multiscale entropy profiles within the passband, and (2) it allowed us to test the effect of EMG duty factor with other things being constant. Simulated EMG time series of same lengths (324,000) were created. Within each “gait cycle” of 1.4 seconds, regular gaps were inserted to create signals of 10, 40, 60, and 80% EMG duty factor, where the muscle was “active” for 10–80% of the gait cycle, throughout the simulated 5 minutes of walking. Simulation results indicated that decreasing EMG duty factor (shorter EMG bursts within a gait cycle) led to lowering of sample entropy values during short timescales (Fig 2). Therefore, the EMG duty factors in the recorded EMG signals were calculated as a possible confounding variable. 10.1371/journal.pone.0162034.g002Fig 2 Effect of EMG duty factor on the Multiscale Entropy of Simulated EMG. Increasing EMG duty factor of EMG signals increased the SampEn of the EMG signal at the smaller timescales. This is because the signal is smoother on average in the short timescales, since the quiescent portions become very predictable as templates will match very often. In real EMG signals, EMG duty factor could confound the observed complexity differences between two signals. This positive correlation of CI with EMG duty factor was as expected based on the method of multiscale entropy. With decreasing burst durations and EMG duty factor, there are more “off,” “flat,” “regular,” and thus predictable parts to the signal. More numbers of pattern matches and completions will occur during these quiet periods, since the signal looks very predictable during these “off” parts. With more pattern completions, the sample entropy of the signal would be less, thus would lead to overall lower CI. EMG Duty Factor Calculation To calculate the EMG duty factor in the recorded EMG signals, the signal was divided into 25-ms epochs, and then RMS amplitude of each epoch was calculated. We made the assumption that the lowest 1% of RMS amplitudes would definitely occur when the muscle is “off.” The initial cutoff amplitude that divides the epochs with bottom 1% of the amplitudes from those in the top 99% was identified. Then, the muscle was defined to be “on” if the RMS amplitude of the epoch greater than three times of the initial cutoff (Fig 3). 10.1371/journal.pone.0162034.g003Fig 3 EMG duty factor Calculation. (A) RMS amplitude histogram. The EMG signal was divided into 25-ms portions and the root-mean-square (RMS) amplitude was calculated. Amplitude histogram is shown in (A). The 1st percentile was used to assume that this EMG activity is when the muscle is quiescent or “Off.” “On” was defined as 3× the RMS amplitude. (B) Sample EMG signal denoted with On/Off times. “On” portion is marked in dark blue; “Off” portion is marked in grey. This method identifies phasic bursts as “On” relative to the quiescent “Off” periods. Statistics First, to test whether older adults exhibited lower complexity and whether this effect was attenuated by slower walking speed, we compared the complexity index CI between age groups and across walking speeds using a general linear model analysis of variance (ANOVA) for each of the four muscles (proc glm, subjects nested within age groups as a random factor). Analyses were repeated across m = 2–3 and r = 0.01–0.35 SD. EMG duty factors were also compared likewise. Second, CI was compared between age groups and across walking speeds with EMG duty factor as a continuous covariate, in an analysis of covariance (ANCOVA; using proc glm). Third, for a comprehensive look at the multiscale nature of EMG signals, we compared sample entropy between the age groups and speed at each timescale with α = 0.001 (≈0.05 / 37 scales) to account for multiple comparisons. SAS 9.3 was used (SAS Institute, Cary NC). Results Unadjusted Complexity Indices The complexity indices CI in vastus lateralis (VL) (p < 0.0001) and biceps femoris (BF) (p<0.0001) were lower in older adults (Fig 4). However, in gastrocnemius (GA), older adults exhibited greater complexity CI (p<0.0001). In tibialis anterior (TA), the effect of age was not significant (p = 0.14). A small decrease in the CI was observed with increasing walking speed in VL (p<0.0001) and BF (p = 0.0003; Fig 4), where the two slow speeds were significantly different from the two fast speeds (p<0.05), per pairwise Tukey-Kramer post-hoc tests. Effects of walking speed were not observed in GA or TA. 10.1371/journal.pone.0162034.g004Fig 4 Age and Walking Speed Effects on the Complexity Index. The (unadjusted) complexity index CI are shown for vastus lateralis, biceps femoris, gastrocnemius, and tibialis anterior muscles as a function of the age group and walking speed. Error bars indicate standard deviation. CI decreased slightly with increasing walking speed in the proximal muscles. P-values are shown in the plot. Significant (p<0.05) pairwise Tukey-Kramer post-hoc tests are indicated with an asterisk (*). In vastus lateralis and biceps femoris, the two slow speeds were significantly different from the two fast speeds. After covariate adjustment with EMG duty factor, age-group differences were no longer present in biceps femoris (p = 0.06), but became noticeable in tibialis anterior (p<0.0001). EMG Duty Factors EMG duty factors were higher in older adults in VL, BF (p<0.001) and GA (p = 0.004), but not in TA (p = 0.33). Group differences were more noticeable at slower walking speeds in BF (interaction p = 0.039). A slight quadratic relationship was observed in GA with walking speed, where the EMG duty factor was the lowest at the preferred speed (p = 0.014). TA EMG duty factor increased slightly with speed (p<0.001; Fig 5). 10.1371/journal.pone.0162034.g005Fig 5 Age and Walking Speed Effects on EMG duty factor. The fraction of the time that vastus lateralis, biceps femoris, gastrocnemius, and tibialis anterior muscles as “on” within a gait cycle is plotted as a function of the age group and walking speed. Error bars indicate standard deviation. Older adults exhibited higher EMG duty factor than young adults except in tibialis anterior. Statistically significant P-values are shown in the plot. EMG duty factors were negatively correlated to CI in VL (Pearson r = -0.407, p<0.0001) and BF (r = -.272, p<0.0001), and positively correlated in GA (r = 0.184, p = 0.0007) and TA (r = 0.34, p<0.0001). Correlations strengths were “very weak” (|r|<0.2), “weak” (0.2≤|r|<0.4) to “medium” (0.4≤|r|<0.6) [22]. These were in contrast to the simulated results, where EMG duty factor and CI were positively correlated. Covariate Adjusted Complexity Indices Since EMG duty factors were associated with age, speed, and CI, CI was compared after covariate adjustment. After adjusting for EMG duty factor, the CI for VL were higher in young adults as before (p = 0.0005), but no longer statistically significant in BF (p = 0.06), although the trends were similar. CI in GA was higher in older adults as before (p<0.0001). After adjustment, CI in TA was higher in young adults (p<0.0001). The effect of walking speed remained similar: CI decreased with speed in VL (p<0.0001) and BF (p = 0.0004), but not in GA (p = 0.89) or TA (p = 0.95). The results indicate that although EMG duty factor is associated with CI, age, and speed, it only affects the results in BF and TA. Parameter Choices These unadjusted results were generally consistent across multiple m and r values, with the exception of m = 2 and r = 0.05 SD for the GA, where the young adults exhibited higher CI (Table 2). Smaller r values were more sensitive to age and speed differences, whereas these differences were not observable with larger r values, particularly in BF and TA. Covariate adjusted results were in general similar to the unadjusted results, except at m = 2 and r = 0.15 SD as discussed above. 10.1371/journal.pone.0162034.t002Table 2 Complexity Index Comparisons across m = 2–3 and r = 0.01–0.35 SD with and without Covariate Adjustment. m = 2 r (×SD) Vastus lateralis Biceps femoris Gastrocnemius Tibialis anterior 0.01 Young>Older (Y>O), p<0.0001 Y>O, p<0.0001 O>Y, p<0.0001 Y>O, p<0.0001 Slow>Fast (S>F), p<0.0001 S>F, p<0.0001 S<>F, p = 0.23 S<>F, p = 0.17 After covariate adjustment Y>O, p<0.0001 Y>O, p = 0.002 O>Y, p<0.0001 Y>O, p<0.0001 S>F, p<0.0001 S<>F, p = 0.24 S<>F, p = 0.9 S<>F, p = 0.09 0.05 Y>O, p<0.0001 Y>O, p<0.0001 Y>O, p<0.0001** Y>O, p<0.0001 S>F, p<0.0001 S>F, p<0.0001 S<>F, p = 0.41 S<>F, p = 0.40 Ix p = 0.04* After covariate adjustment Y>O, p<0.0001 Y>O, p<0.0001 Y>O, p<0.0001** Y>O, p<0.0001 S>F, p<0.0001 S>F, p<0.0001 S<>F, p = 0.43 S<>F, p = 0.37 Ix p = 0.02 0.15 Y>O, p<0.0001 Y>O, p<0.0001 O>Y p<0.0001 Y<>O p = 0.14 a priori recommendation S>F, p<0.0001 S>F, p = 0.0003 S<>F, p = 0.91 S<>F, p = 0.22 (see text) After covariate adjustment Y>O, p = 0.0005 Y>O, p = 0.0624 O>Y, p<0.0001 Y>O, p<0.0001 S>F, p<0.0001 S>F, p = 0.0004 S<>F, p = 0.89 S<>F, p = 0.95 0.25 Y>O, P<0.0001 Y>O, p<0.0001 O>Y, p <0.0001 O<>Y, p = 0.93 S>F, p = 0.0003 S>F, p = 0.02 S<>F, p = 0.93 S<>F, p = 0.43 After covariate adjustment Y>O, P = 0.0001 Y>O, p<0.0001 O>Y, p < 0.0001 O<>Y, p = 0.86 S>F p<0.0001 S>F, p = 0.02 S<>F, p = 0.93 S<>F, p = 0.45 0.35 Y>O, p<0.0001 Y>O, p = 0.0007 O>Y, p<0.0001 O<>Y, p = 0.47 S>F, p = 0.007 S<>F = 0.27 S<>F, p = .88 S<>F, p = 0.54 After covariate adjustment Y>O p<0.0001 Y>O, p = 0.0007 O>Y, p<0.0001 O<>Y, p = 0.66 S>F, p = 0.002 S<>F = 0.29 S<>F, p = 0.90 S<>F, p = 0.55 m = 3 r Vastus lateralis Biceps femoris Gastrocnemius Tibialis anterior 0.01 Y>O, p<0.0001 Y>O, p = 0.0001 O>Y, p<0.0001 Y>O, p<0.0001 S>F, p<0.0001 S>F, p = 0.011 S<>F, p = 0.24 F>S, p = 0.04*** After covariate adjustment Y>O, p<0.0001 Y>O, p<0.0001 O>Y, p<0.0001 Y>O, p<0.0001 S>F, p<0.0001 S>F, p<0.0001 S<>F, p = 0.4244 F<>S, p = 0.18 0.05 Y>O, p<0.0001 Y>O, p<0.0001 O>Y, p<0.0001 Y>O, p<0.0001 S>F, p<0.0001 S>F, p = 0.0005 S<>F, p = 0.55 F>S, p = 0.04*** After covariate adjustment Y>O, p<0.0001 Y>O, p<0.0001 O>Y, p<0.0001 Y>O, p<0.0001 S>F, p<0.0001 S>F, p<0.0001 S<>F, p = 0.551 F<>S, p = 0.10 0.15 Y>O, p<0.0001 Y>O, p<0.0001 O>Y, p<0.0001 O<>Y, p = 0.13 S>F, p<0.0001 S>F, p = 0.005 S<>F p = 0.83 S<>F p = 0.10 After covariate adjustment Y>O, p = 0.0023 Y>O, p<0.0001 O>Y, p<0.0001 Y>O, p = 0.04 S>F, p<0.0001 S>F, p<0.0001 S<>F, p = 0.84 F<>S, p = 0.19 0.25 Y>O, p<0.0001 Y>O, p<0.0001 O>Y, p<0.0001 O<>Y, p = 0.83 S>F, p = 0.0005 S<>F, p = 0.12 S<>F p = 0.89 S<>F, p = 0.20 After covariate adjustment Y>O, p = 0.004 Y>O, p<0.0001 O>Y, p<0.0001 O<>Y, p = 0.68 S>F, p<0.0001 S>F, p = 0.0072 S<>F, p = 0.87 S<>F, p = 0.35 0.35 Y>O, p<0.0001 Y>O, p<0.0005 O>Y, p<0.0001 O<>Y, p = 0.36 S>F, p = 0.007 S<>F, p = 0.64 S<>F, p = 0.92 S<>F, p = 0.33 After covariate adjustment Y>O, p = 0.0007 Y>O, p = <0.0001 O>Y, p<0.0001 O<>Y, p = 0.74 S>F, p = 0.0004 S<>F, p = 0.1607 S<>F, p = 0.87 S<>F, p = 0.48 Summary Young>Older and Slow>Fast for all parameters Young>Older and Slow>Fast for r<0.25SD Older>Young for all, except m = 2 and r = 0.05SD Young>Older for r<0.15SD No speed effect for all parameters No speed effect for all parameters, except m = 3, r≤0.05 *Ix: Age-speed interaction was statistically present, where the speed effect was stronger in young adults. ** The only exception to the trend in the gastrocnemius *** Although statistically significant at α = 0.05, these findings are marginal considering the large number of comparisons made. A Bonferroni adjustment would negate these effects. Y>O: CI in young adults higher than in older adults O<Y: CI in older adults higher than in young adults S>F: CI in slow speeds higher than in fast speeds F>S: CI in fast speeds higher than in slow speeds Y<>O, S<>F: age effect or speed effect not statistically significant at α = 0.05 Covariate adjustment did not affect most parameter sets, except for biceps femoris and tibialis anterior at m = 2, r = 0.15SD (see text). Entropy and Timescales Young adults exhibited greater sample entropy (m = 2, r = 0.15 SD) at most timescales in VL and BF, and at small timescales in TA (p<0.001). The opposite was observed in the GA. Age-related differences were more noticeable at small timescales (Figs 6 and 7). Sample entropy overall decreased with increasing timescales. However, the GA in young adults exhibited a nearly flat profile. The effect of walking speed was similar to that in Table 2. Age-group differences were similar across walking speeds (Fig 7). 10.1371/journal.pone.0162034.g006Fig 6 Age and Timescale Effects on the Sample Entropy. Sample entropy of the EMG signals for vastus lateralis, biceps femoris, gastrocnemius, and tibialis anterior were calculated for timescales 4–40 (27–270 Hz) after successive coarse-graining. The timescales and corresponding frequencies are shown on the abscissa of the plots. Error bars indicate standard deviations. SampEn values in general become smaller at larger timescales. Age-related differences are more noticeable at small timescales. Older adults exhibited lower SampEn values in the proximal muscles, and higher values in the gastrocnemius. Significant age-related differences (main effect of age) at α = 0.001 (≈ 0.05/37 scales tested) are shown with the dot (∙) above each timescale. Plotted mean and standard deviation values were pooled across walking speeds. 10.1371/journal.pone.0162034.g007Fig 7 Age and Timescale Effects on Sample Entropy stratified by Speed. Sample entropy values are shown stratified by walking speed. Although the main effect of walking speed was present as shown in Fig 4, the trends between walking speeds are similar. The asterisk (*) denotes significant age-related differences as determined using Welch’s t-test (unequal variances, 2-tail). Compared to Fig 6, the age differences are not statistically significant except with gastrocnemius due to the less powerful t-test and the smaller sample size. Discussion Our goals were to determine whether older adults exhibit lower overall complexity of muscle activation patterns by studying the dynamics of EMG patterns over multiple time scales, and whether slower walking would attenuate this effect. We observed a lower complexity index of EMG activations during gait in older adults, but only in vastus lateralis, biceps femoris and tibialis anterior with smaller r. The opposite was observed in the gastrocnemius, and this effect was more pronounced at the shorter timescales and smaller r. These results were generally robust across different sample entropy parameters, and after adjusting for EMG duty factor. Slower walking speed produced higher complexity in the proximal muscles, but not in the distal muscles. We conclude that the decrease in complexity in EMG signals with aging may be limited to proximal leg muscles, particularly the quadriceps, and that slower walking attenuates this decrease in complexity in the timescales (27–270 Hz) studied. This finding was observed despite longer EMG burst durations within the gait cycle as observed in older adults, which would have increased rather than decreased CI. Therefore, contrary to the multisystem dysregulation hypothesis of aging, greater complexity may also indeed be a sign of deterioration, at least within our analysis range of 27-270Hz in some systems. This may not apply to behaviors occurring in other timescales. Older adults exhibited lower complexity particularly in short timescales as well as longer burst durations. Our pilot simulation indicated that longer bursts would produce higher complexity values, but this relationship was opposite in the proximal muscles. Therefore, we conclude that in the quadriceps, the decreases in complexity are not due to simple changes in burst durations, but despite them. Biceps femoris also exhibited lower complexity in older adults, but this may be due to changes in activation burst patterns rather than changes in the EMG signal complexity, since changes in EMG duty factor could explain the age group differences. However, in the gastrocnemius the longer burst may explain the increased complexity, and in tibialis anterior the age-related decrease in complexity may be masked by the longer burst durations. Aging and Complexity The lower complexity in older adults observed in the proximal muscles was predicted by the multisystem dysregulation hypothesis. In the motor system, this result may be explained by sarcopenia and the associated loss of motor units [23, 24]. As the surface EMG signal is created from a smaller number of motor units, and therefore the resulting signal would be less complex [23]. Age-related differences in SampEn values were more notable in the small timescales and SampEn values of older adults became lower than that of young adults at very large timescales, although the differences were minimal (Fig 3). However, sarcopenia by itself does not explain this phenomenon, as the opposite was found in the gastrocnemius. Sarcopenic deterioration with aging is well known in plantarflexor muscles, as well as proximal muscles [24]. Also, the fiber composition is likewise similar in the vastus lateralis of young and older adults [25]. Therefore, sarcopenia may explain the results of proximal muscles, but not others. One possible explanation for this difference between the muscles is the substantial change in their activation pattern with aging. The activation of the gastrocnemius may change to somehow compensate for the loss of complexity in the proximal muscles. During gait, older adults exhibit increased hip moments and decreased ankle moments [26] and power [27], and activate proximal muscles more [4, 28]. Yet distal muscles activate more during balance challenged walking [29] in older adults, and this may be the case in our study, where the treadmill walking task without handrails was somewhat novel to the older adults. We observed that burst durations were longer in older adults, and this was associated with increased CI. Another possible explanation for our result is the increase in neuromuscular noise with aging. As older adults activate proximal leg muscles more [4, 28], there may be less motor drive to distal muscles. With increased neuromuscular noise and less drive ‘signal’ in the gastrocnemius, the noise portion may be dominating the entropy characteristics of the EMG output. Sample entropy of white noise are higher at smaller timescales [17], similar to the EMG signals we observed in the gastrocnemius (Fig 3). This may explain the unique results in the gastrocnemius of older adults. Also, the positive correlation between EMG duty factor and CI were similar to that of white noise. Therefore, signal complexity profile may represent neuromuscular noise rather than the changes in drive signal itself, as multiscale entropy was found to increase only minimally with increasing isometric contraction intensity [18]. A third possibility is the longer EMG burst durations during walking. Although our analysis does not consider longer timescales that would capture the stride-to-stride fluctuations of muscle burst times or lengths, we did separately calculate the burst durations by way of calculating the EMG duty factor. As our pilot simulation indicated that higher EMG duty factor would produce higher complexity, and this effect was seen in the gastrocnemius. The higher EMG duty factor in the gastrocnemius may partially explain the increased in complexity in older adults. This result is congruent with the reported increased co-contraction [4] and burst duration. In consideration of these three possible explanations, we speculate that the muscle fiber activation in the gastrocnemius becomes more like that of white noise. Given the known progression of sarcopenia, proximal muscles may also show this tendency as aging continues. Gait Speed and Complexity Slow gait in older adults may reflect adaptations to manage the walking task. We examined whether slower walking speed would minimize age-related differences in complexity. Although the age-related differences were not different across walking speeds, faster walking speed led to lowering of the complexity of EMG signals in the proximal muscles. This result supports our hypothesis that older adults may be benefitting by walking slower, and that slower walking may be an adaptive behavior. Certainly, there are physical limitations associated with aging and frailty that can limit walking speed. Nonetheless, we speculate that walking slower may allow the nervous system to function with higher complexity, which may allow for better function. Walking speed is known to produce changes in stepping control and timing, which likely occur in larger timescales (~0.1–1 Hz) not reflected in our analysis, limited to 27–270 Hz. EMG and other gait dynamics at these timescales may need to be examined to better address this question of adaption. EMG Signals and Complexity Ours may be the first work to examine the entropy of muscle activation patterns during gait in a comprehensive multiscale manner. To accomplish this, we measured EMG over 5 minutes of walking, which provided sufficient data to study dynamics that are present over the frequency range of EMG. Previous studies have used a wide range of methods that may have led to the apparent conflicts in the results [10–14], or are limited to isometric contractions [18, 30]. Methodologically, we recommend reporting entropy values over the range of timescales. It is particularly important to study only the timescales that fall within of the passband of the filtering methods used, consider the effects of low frequency non-stationarity on entropy estimates, and confounding effects of burst durations. EMG signals during gait exhibited larger sample entropy at small timescales, and lower entropy at larger timescales. This behavior was similar to that of bandpass filtered uncorrelated white noise. Both EMG and filtered noise have similar power spectrum over the frequencies considered, which may explain this similarity. Practically, it may not be necessary to consider the larger timescales considered in our study. Timescales from 4 to 30 ms may be adequate for future studies, but studying larger scales may be useful for understanding stride-to-stride fluctuations. Limitations and Future Work Our work is descriptive of neuromuscular behavior, and does not explain the sources of the complexity of the signal or their mechanistic changes due to aging. Our work is also limited in studying the passband of the surface EMG signal. Only one decade of timescales (27–270 Hz) was considered, due to the nature of the surface EMG signals. Although the timescales and the dynamics studied are not strictly defined by linear behavior of the signal at each frequency, analyses of timescales beyond the passband will likely result in spurious results as there is no remaining meaningful information after filtering. We started with a priori multiscale entropy parameters based on previous recommendations, although others could have been used. For example, optimization could be used have been used to determine the parameter set that (1) provides most consistent entropy values by providing sufficient pattern matches at m and m+1 [18, 21] or (2) best distinguishes two population groups [31]. In the present examination of the presence of a consistent difference in age groups, using such approaches to find the parameter set that would provide the best results would be circular logic. Therefore we chose to start from a particular recommendation rather than find those supporting our hypothesis. A comprehensive exploration of the parameter space also would provide a more complete answer [19], but the computational costs are very prohibitive. Our use of m = 2–3 and r = 0.01–0.35 identified generally consistent patterns across these parameters. In summary, older adults exhibited lower complexity in proximal leg muscle surface EMG compared to young adults, but higher complexity in the gastrocnemius in the 27–270 Hz range. Slower walking corresponded to higher complexity, and thus may allow better neuromuscular function in older adults who walk slower. These changes may be due to neuromuscular noise or adaptations to the requirements of the motor task. Although lower complexity has been observed in many systems with aging, this phenomenon may be obvious in some physiological systems at certain timescales but not others. Studies of motor behavior over a wider range of timescales may be needed to better understand the effects of aging over the motor system. The authors thank Ann Newstead, Cooper Philips, and Philip Hwang for their assistance in protocol design, pilot work, and data collection. The authors thank Mark Dela for the work in computational and statistical analyses. ==== Refs References 1 Alexander NB . Gait disorders in older adults . J Am Geriatr Soc . 1996 ;44 (4 ):434 –51 . 8636592 2 Kang HG , Dingwell JB . Separating the effects of age and walking speed on gait variability . Gait Posture . 2008 ;27 (4 ):572 –7 . .17768055 3 Kang HG , Dingwell JB . Effects of walking speed, strength and range of motion on gait stability in healthy older adults . J Biomech . 2008 ;41 (14 ):2899 –905 . 10.1016/j.jbiomech.2008.08.002 18790480 4 Schmitz A , Silder A , Heiderscheit B , Mahoney J , Thelen DG . Differences in lower-extremity muscular activation during walking between healthy older and young adults . J Electromyogr Kinesiol . 2009 ;19 (6 ):1085 –91 . 10.1016/j.jelekin.2008.10.008 19081734 5 Lipsitz LA , Goldberger AL . Loss of 'Complexity' and Aging: Potential Applications of Fractals and Chaos Theory to Senescence . J Am Med Assoc . 1992 ;267 (13 ):1806 –9 . 6 Costa M , Ghiran I , Peng CK , Nicholson-Weller A , Goldberger AL . Complex dynamics of human red blood cell flickering: alterations with in vivo aging . Physical Review E, Statistical, Nonlinear, and Soft Matter Physics . 2008 ;78 (2 Pt 1 ):1 . 7 Costa MD , Peng CK , Goldberger AL . Multiscale Analysis of Heart Rate Dynamics: Entropy and Time Irreversibility Measures . Cardiovasc Eng . 2008 ;8 (2 ):88 –93 . Epub 2008/01/04. 10.1007/s10558-007-9049-1 .18172763 8 Kang HG , Costa MD , Priplata AA , Starobinets OV , Goldberger AL , Peng CK , et al Frailty and the degradation of complex balance dynamics during a dual-task protocol . Journals of Gerontology Series A, Biological Sciences and Medical Sciences . 2009 ;64 (12 ):1304 –11 . Epub 2009/08/15. glp113 [pii] 10.1093/gerona/glp113 19679739 9 Costa M , Peng CK , Goldberger AL , Hausdorff JM . Multiscale entropy analysis of human gait dynamics . Physica A . 2003 ;330 (1–2 ):53 –60 . 10 Vaillancourt DE , Newell KM . Changing complexity in human behavior and physiology through aging and disease . Neurobiol Aging . 2002 ;23 (1 ):1 –11 . 11755010 11 Kurz MJ , Stergiou N . The aging humans neuromuscular system expresses less certainty for selecting joint kinematics during gait . Neurosci Lett . 2003 ;348 (3 ):155 –8 . 12932817 12 Vaillancourt DE , Newell KM . Aging and the time and frequency structure of force output variability . J Appl Physiol . 2003 ;94 (3 ):903 –12 . 12571125 13 Heffernan KS , Sosnoff JJ , Ofori E , Jae SY , Baynard T , Collier SR , et al Complexity of force output during static exercise in individuals with Down syndrome . J Appl Physiol . 1985 ;106 (4 ):1227 –33 . 14 Sosnoff JJ , Voudrie SJ . Practice and age-related loss of adaptability in sensorimotor performance . Journal of Motor Behavior . 2009 ;41 (2 ):137 –46 . 10.3200/JMBR.41.2.137-146 19201684 15 Kang HG , Dingwell JB . Dynamics and stability of muscle activations during walking in healthy young and older adults . J Biomech . 2009 ;42 (14 ):2231 –7 . 10.1016/j.jbiomech.2009.06.038 19664776 16 Hermens HJ , Freriks B , Disselhorst-Klug C , Rau G . Development of recommendations for SEMG sensors and sensor placement procedures . J Electromyogr Kinesiol . 2000 ;10 (5 ):361 –74 . .11018445 17 Costa M , Goldberger AL , Peng CK . Multiscale entropy analysis of biological signals . Phys Rev E Stat Nonlin Soft Matter Phys . 2005 ;71 (2 Pt 1 ):021906 Epub 2005/03/24. .15783351 18 Cashaback JG , Cluff T , Potvin JR . Muscle fatigue and contraction intensity modulates the complexity of surface electromyography . J Electromyogr Kinesiol . 2013 ;23 (1 ):78 –83 . 10.1016/j.jelekin.2012.08.004 .22959820 19 Yentes JM , Hunt N , Schmid KK , Kaipust JP , McGrath D , Stergiou N . The appropriate use of approximate entropy and sample entropy with short data sets . Annals of biomedical engineering . 2013 ;41 (2 ):349 –65 . 10.1007/s10439-012-0668-3 .23064819 20 Pincus SM . Approximate entropy as a measure of system complexity . Proc Natl Acad Sci . 1991 ;88 (6 ):2297 –301 . 11607165 21 Lake DE , Richman JS , Griffin MP , Moorman JR . Sample entropy analysis of neonatal heart rate variability . Am J Physiol Regul Integr Comp Physiol . 2002 ;283 (3 ):R789 –97 . .12185014 22 Evans JD . Straightforward statistics for the behavioral sciences . Pacific Grove, CA : Brooks/Cole Publishing ; 1996 . 23 Watanabe K , Kouzaki M , Merletti R , Fujibayashi M , Moritani T . Spatial EMG potential distribution pattern of vastus lateralis muscle during isometric knee extension in young and elderly men . Journal of Electromyography and Kinesiology . 2012 ;22 (1 ):74 –9 . 10.1016/j.jelekin.2011.09.010 21996320 24 Rowan SL , Purves-Smith FM , Solbak NM , Hepple RT . Accumulation of severely atrophic myofibers marks the acceleration of sarcopenia in slow and fast twitch muscles . Experimental gerontology . 2011 ;46 (8 ):660 –9 . 10.1016/j.exger.2011.03.005 .21513786 25 Andersen JL . Muscle fibre type adaptation in the elderly human muscle . Scandinavian journal of medicine & science in sports . 2003 ;13 (1 ):40 –7 . .12535316 26 Franz JR , Kram R . Advanced age and the mechanics of uphill walking: A joint-level, inverse dynamic analysis . Gait & Posture . 2013 ;11 (13 ):00293 –2 . 27 DeVita P , Hortobagyi T . Age causes a redistribution of joint torques and powers during gait . J Appl Physiol . 2000 ;88 (5 ):1804 –11 . 10797145 28 Franz JR , Kram R . How does age affect leg muscle activity/coactivity during uphill and downhill walking? Gait & Posture . 2013 ;37 (3 ):378 –84 .22940542 29 Brown LA , Gage WH , Polych MA , Sleik RJ , Winder TR . Central set influences on gait: Age-dependent effects of postural threat . Exp Brain Res . 2002 ;145 (3 ):286 –96 . 12136378 30 Sung PS , Zurcher U , Kaufman M . Comparison of spectral and entropic measures for surface electromyography time series: a pilot study . Journal of Rehabilitation Research and Development . 2007 ;44 (4 ):599 –609 . 18247257 31 Ghassemi M , Lehman L , Snoek J , Nemati S . Global Optimization Approaches for Parameter Tuning in Biomedical Signal Processing: A Focus of Multi-scale Entropy Computers in Cardiology ; 7–10 9 2014 ; Cambridge, MA : IEEE ; 2014. p. 993 –6 .

PMC005xxxxxx/PMC5003392.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757097710.1371/journal.pone.0161886PONE-D-16-15127Research ArticlePhysical sciencesChemistryChemical compoundsOrganic compoundsVitaminsB vitaminsVitamin KPhysical sciencesChemistryOrganic chemistryOrganic compoundsVitaminsB vitaminsVitamin KBiology and Life SciencesCell BiologyCell ProcessesCell DeathApoptosisMedicine and Health SciencesOncologyCancers and NeoplasmsGenitourinary Tract TumorsBladder CancerMedicine and Health SciencesUrologyBladder CancerBiology and Life SciencesBiochemistryBioenergeticsEnergy-Producing OrganellesMitochondriaBiology and Life SciencesCell BiologyCellular Structures and OrganellesEnergy-Producing OrganellesMitochondriaPhysical SciencesChemistryChemical CompoundsOrganic CompoundsVitaminsPhysical SciencesChemistryOrganic ChemistryOrganic CompoundsVitaminsMedicine and Health SciencesOncologyCancer TreatmentBiology and Life SciencesPhysiologyElectrophysiologyMembrane PotentialMedicine and Health SciencesPhysiologyElectrophysiologyMembrane PotentialResearch and Analysis MethodsSpectrum Analysis TechniquesSpectrophotometryCytophotometryFlow CytometryVitamin K2 Induces Mitochondria-Related Apoptosis in Human Bladder Cancer Cells via ROS and JNK/p38 MAPK Signal Pathways Vitamin K2 Induces Apoptosis in Human Bladder Cancer CellsDuan Fengsen 1Yu Yuejin 2Guan Rijian 2Xu Zhiliang 1Liang Huageng 2*Hong Ling 1*1 Department of Genetics and Developmental Biology, College of Life Science and Technology, Huazhong University of Science and Technology, Wuhan, Hubei, P.R. China2 Department of Urology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei, P.R. ChinaHsieh Yi-Hsien EditorInstitute of Biochemistry and Biotechnology, TAIWANCompeting Interests: The authors have declared that no competing interests exist. Conceived and designed the experiments: FSD LH. Performed the experiments: FSD YJY. Analyzed the data: FSD ZLX. Contributed reagents/materials/analysis tools: FSD RJG ZLX HGL LH. Wrote the paper: FSD. * E-mail: lhong@mail.hust.edu.cn (LH); leonard19800318@hust.edu.cn (HGL)29 8 2016 2016 11 8 e016188614 4 2016 12 8 2016 © 2016 Duan et al2016Duan et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.The effects of vitamin K2 on apoptosis in a variety of cancer cells have been well established in previous studies. However, the apoptotic effect of vitamin K2 on bladder cancer cells has not been evaluated. The aim of this study is to examine the apoptotic activity of Vitamin K2 in bladder cancer cells and investigate the underlying mechanism. In this study, Vitamin K2 induced apoptosis in bladder cancer cells through mitochondria pathway including loss of mitochondria membrane potential, cytochrome C release and caspase-3 cascade. Furthermore, the phosphorylation of c-Jun N-terminal kinase (JNK) and p38 MAPK was detected in Vitamin K2-treated cells and both SP600125 (an inhibitor of JNK) and SB203580 (an inhibitor of p38 MAPK) completely abolished the Vitamin K2-induced apoptosis and loss of mitochondria membrane potential. Moreover, the generation of reactive oxygen species (ROS) was detected in bladder cancer cells, upon treatment of vitamin K2 and the anti-oxidant N-acetyl cysteine (NAC) almost blocked the Vitamin K2-triggered apoptosis, loss of mitochondria membrane potential and activation of JNK and p38 MAPK. Taken together, these findings revealed that Vitamin K2 induces apoptosis in bladder cancer cells via ROS-mediated JNK/p38 MAPK and Mitochondrial pathways. NSFC (P.R. China)30971608Hong Ling NSF of the Hubei Province2009CDB074Hong Ling This work was supported by contract grant sponsor: NSFC (P.R. China), contract grant number 30971608; and contract grant sponsor NSF of the Hubei Province, contract grant number 2009CDB074. Data AvailabilityAll relevant data are within the paper and its Supporting Information files.Data Availability All relevant data are within the paper and its Supporting Information files. ==== Body Introduction Bladder cancer is one of the most common carcinoma and ranks the ninth in worldwide cancer incidence. More than 12 million new cases arise each year globally. In particular, bladder cancer accounts for approximately 180,000 new cancer diagnosis and more than 50,000 deaths annually in the United States and European countries[1,2]. To cure human bladder cancer, traditional and current methods, such as radical cystectomy, chemotherapy, radiotherapy, concurrent chemotherapy and radotherapy, combination of radical cystectomy and chemotherapy and immunotherapy, are widely used[1,3–5]. However, these therapies usually encounter a variety of adverse effect such as distant metastasis, local recurrence, toxicity to health, low survival of patients and cost-effectiveness. Base on the above side effect and poor life quality of patients[4,6,7], new drugs are urgently required to treat bladder carcinoma. Vitamin K is one of the fat-soluble vitamins which are indispensible to human health and rich in a variety of food. Usually, vitamin K exists in three forms including phylloquinone (VK1), menaquinone (VK2) and menadione (VK3). Predominant research on vitamin K has devoted to its role as a critical factor in blood coagulation, a cofactor in bone metabolism and prevention of cardiovascular calcification[8–10]. Recent years, a growing number of studies have revealed that vitamin K exhibited remarkable anti-proliferative effects on cancer cells. Vitamin K2 (Menaquinone) is a series of vitamin K with multi-isoprene units at the 3-position of the naphthoquinone, which are named as MK-n by the number of the prenyl units[9,11]. For instance, MK-4, utilized in this study, is endowed with four isoprene units in its side chain. Original studies have discovered that vitamin K2 was produced by a vast array of bacteria and originally isolated from putrefied fishmeal as a product of microbial synthesis[9]. Recent studies have suggested vitamin K2 can actually be produced by animals and humans via conversion of other forms of vitamin K [12]. Furthermore, as the latest studies indicated, Menaquinone 4 (MK-4, one of vitamin K2 forms) was synthesized by UBIAD1, a geranylgeranyltransferase, in humans from the conversion of phylloquinone (VK1) and menadione (VK3) [12]. To date, abundant studies have shown that vitamin K2 can exhibit anticancer activity in various cancer cell lines, including leukemia, lung cancer, ovarian cancer, prostate cancer and heptocellular cancer [13–17]. Although some studies have revealed vitamin K2 exerted anticancer effect mainly by blocking the cell cycle at the G1 phase and inducing the caspase-3-mediated apoptosis, the detailed mechanism of anticancer effect of vitamin K2 remains unclear[17–19]. In this study, we demonstrated vitamin K2 induced apoptosis in human bladder cancer cells via generation of reactive oxygen species (ROS) which subsequently mediated MAPK and Mitochondrial pathways. Moreover, because vitamin K2 is ubiquitously produced in human and without adverse effects for clinical treatments, we adopted vitamin K2 treatment to nude mice bearing human bladder cancer cells and showed vitamin K2 sufficiently induced apoptosis of bladder cancer cells in vivo. This study was the first time to utilize vitamin K2 to treat human bladder cancer cells and demonstrated the detailed mechanism of anticancer activity of vitamin K2, which provide the basic theories for curing human bladder cancer. Materials and Methods Cell culture The human bladder cancer cell lines (T24, J82 and EJ) and human normal cell lines (L02 and HEK293) were obtained from the American Type Culture Collection (Manassas, VA, USA). The T24, J82 and EJ cells were cultured in Minimum Essential Medium Eagle (MEM) supplemented with 10% Fetal Bovine Serum (FBS). While, the L02 and HEK293 cells were culture in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% Fetal Bovine Serum (FBS). All the cultures were maintained at 37°C in a humidified 5% CO2 incubator. Animal study Twenty female BALB/c nude mice, 4- or 5-week old, were provided by experimental animal center (Tongji Medical college of Huazhong University of Science and Technology). Procedures and handing were strictly conducted in compliance with guidelines approved by the Science and Technology Department of Hubei province. All animal studies were approved by the Animal Experimentation Ethics Committee of Huazhong University of Science and Technology. All the efforts were made to minimize the animals’ suffering and to reduce the number of animals used. Drugs and reagents Vitamin K2 was purchased from Sigma (USA), with carbon 82.6–84.9%, EmM 17.4–18.9 and completely soluted in ethonal. Vitamin K2 was dissolved in 99.9% ethanol at a stock concentration of 50 mM; it was then diluted to working concentration with MEM or DMEM. Ethanol was added to cultures at 0.1% (V/V) as a solvent control. 3-(4,5-dimethyl-2-thiazyl)-2,5-diphenyl-2H-tetrazolium bromide (MTT), N-acetyl cysteine (NAC) and Rhodamine 123 were purchased from Sigma Chemical Co. (St. Louis, MO). Annexin V-FITC (fluorescein isothiocyanate)/PI (propidium iodide) kit was purchased from BD Biosciences (San Jose, CA). The caspase-3 specific inhibitor Z-DEVD-FMK, JNK inhibitor SP600125 and p38 inhibitor SB203580 were purchase from Calbiochem (San Diego, CA). The antibodies against caspase-3, p-JNK, p38 and p-p38 were purchase from Cell signaling Technology. The antibodies against PARP, cytochrome C, COX IV, Bax, Puma, Bcl-2 and JNK were purchase from Proteintech. The antibodies against Actin was purchase from Santa Cruz. Cell viability assay Cells were plated in 96-well plates at a density of approximately 1×105 cells per well. Twenty four hours after plating, the cells were treated with vitamin K2. Cell viability was then evaluated using the MTT assay according to the manufacturer’s protocol. The number of viable cells was evaluated by uptake of MTT, assayed at 590nm. Assays were performed in triplicate on three independent experiments. Cell apoptosis assay To analyze cellular apoptosis, cells were harvested, washed with PBS and resuspended in 500 μl of 1× binding buffer. The resuspended cells were then stained with Annexin V-FITC/PI and incubated in the dark for 15 minutes. The number of apoptotic cells was analyzed by flow cytometry (Beckman coulter FC500) TUNEL assay DNA breaks were evaluated with an in situ cell death detecting kit (Roche Molecular Biochemicals, Basel, Switzerland), according to the manufacture’s instructions. Briefly, Cells were treated with the indicated concentration of vitamin K2 for 24 hours at 37°C in a 5% CO2 incubator, After incubation, cells were washed with PBS and fixed with 4% paraformaldehyde, then the cells were rinsed and subjected to TUNEL staining (terminal deoxynucleotidyl transferase dUTP nick end labeling). The apoptotic cells were observed under the fluorescence microscope (Olympus, Japan). Subcellular fraction The protein in bladder cancer cells was separated into cytosolic and mitochondrial fraction using a special cytosolic and mitochondrial fraction kits (Beyotime, China), according to manufacture’s instructions. Briefly, Cells were harvested and mitochondrial isolation reagents were added and incubated on ice followed by centrifugation at 600g for 10 min. Supernatant was further centrifuged at 11,000g for 15 min. The mitochondrial fraction was contained in the pellets, while supernatant containing the cytosolic fraction. Western blot analysis Cells were lysed with radioimmunoprecipitation (RIPA) buffer (Beyotime, Shanghai) supplemented with a protease inhibitor mixture tablet (Google Biology, Wuhan) for 30 minutes on ice. Total protein samples (40 μg) were then separated by SDS-PAGE (sodium lauryl sulfate (SDS)-polyacrylamide gel (PAGE)) and transferred to PVDF (polyvinylidenedifluoride) membranes (Millipore, USA). The membranes were subsequently blocked with 5% fat-free milk dissolved in Tris-Buffered Saline containing Tween-20 (TBST buffer) for 2 hours at room temperature and then probed with primary antibodies and incubated for overnight at 4°C. After incubation with horseradish peroxidase-conjugated secondary antibodies, The protein signals were detected using a chemiluminescence solution (ECL, Advansta, USA). Band intensity was quantified by Quantity one software (BioRad, USA). Cell mitochondria membrane potential assay To evaluate the changes of mitochondria membrane potential, Rhodamine 123, a mitochondria specific dye, was used. Briefly, Cells were harvested and washed with PBS twice, then stained with 1.5 μM Rhodamine 123 and incubated at 37°C for 30 minutes. The cells were subsequently washed twice with cold PBS to remove the unbound dye. The mitochondria membrane potential was evaluated by the fluorescence of Rhodamine 123 under the flow cytometry with excitation and emission wavelengths of 488 and 525 nm (Beckman, FC500). Intracellular ROS detecting To measure ROS generation, 2',7'-dichlorofluorescein-diacetate (DCFH-DA) was utilized. DCFH-DA, a cell membrane permeable dye, is converted to DCFH (a non-fluorescent cell membrane impermeable compound) by intracellular esterases and highly fluorescent DCF was produced by the oxidation of DCFH by intracellular ROS. Therefore, fluorescent DCF intensity is proportional to the amounts of intracellular ROS. Briefly, cells were harvested and stained with 10 μM DCFH-DA (Beyotime, China) for 30 minutes at 37°C, washed twice with PBS and then immediately analyzed by flow cytometry (Beckman, FC500). To observe intracellular ROS, cells were seeded on coverslips, treated with the indicated concentration of vitamin K2 for 24 hours, then stained with 10 μM DCFH-DA. Before DAPI staining, Cells were fixed in 4% paraformaldehyde, washed with PBS. Observing intracellular ROS was performed by confocal microscope (FV1000, Olympus). In vivo study Human bladder cancer EJ cells (1×107) suspended in PBS were injected subcutaneously into the lower right flank of each mouse. After 2 weeks, when tumors reached approximately 50 mm in diameter, the mice were randomly divided into two groups. 10 mice were used in each group. Treatment was 30mg/kg of vitamin K2 by directly injection at tumor each day as the experiment group, while treatment was the equivalent volume of PBS by directly injection at tumor per day as the control group. Tumor size was measured using a sliding caliper two times per week and the volume (mm3) was calculated by the formula (W2× L) /2. After 21 days, mice were sacrificed and tumors were excised and sectioned for caspas-3, HE staining and TUNEL assays. Statistical analyses All the experiments were performed at least three times. The data were analyzed using GraphPad Prism software. The results are displayed as the mean ± standard deviation, and the differences were measured using Student’s t-test. Statistical significance was set at p<0.05. Results Vitamin K2 reduces bladder cancer cell viability To investigate the cell viability changes in human bladder cancer cells after treatment with vitamin K2, MTT assays were performed. Vitamin K2 significantly decreased the viability of human bladder cancer T24, J82 and EJ cells in a dose- and time-dependent manner. As shown in Fig 1A, T24, J82 and EJ cell viability was remarkably reduced following treatment with increasing concentrations of vitamin K2 (p<0.001). Similarly, viability of T24, J82 and EJ cells was significantly diminished with prolonged treatment with 100 μM vitamin K2 (p<0.001) (Fig 1B). On the other hand, viability of human normal cells (L02 and HEK293) was minimally affected after exposed to high concentration (100 μM) of Vitamin K2 (S1A Fig). These results suggest that vitamin K2 has anticancer activity in human bladder cancer cells, with low cytotoxic effect on human normal cells. 10.1371/journal.pone.0161886.g001Fig 1 Effect of vitamin K2 on the viability of three human bladder cancer cells. (A). Vitamin K2 dose-dependently reduced the viability of human bladder cancer T24, J82 and EJ cells. Cells were treated different concentration of vitamin K2 for 24 hours, respectively and cell viability was measured by MTT assays. (B). Vitamin K2 time-dependently decreased the viability of T24, J82 and EJ cells. Cells were treated with 100 μM vitamin K2 for 0, 6, 12, 18 and 24 hours, respectively, and cell viability was evaluated by MTT assays. Data represent the mean ± SEM of three different experiments with triplicate sets in each assay. * P<0.05, ** P<0.01 and *** P<0.001 vs vitamin K2-untreated group. Vitamin K2 induces significant apoptosis in human bladder cancer cells To evaluate the apoptotic effect of vitamin K2 on bladder cancer cells, T24, J82 and EJ cells were respectively exposed to indicated-concentration of vitamin K2 for 24 hours. As shown in Fig 2A, vitamin K2 remarkably triggered apoptosis in human bladder cancer T24 cells in a dose-dependent manner and approximately 50% of the cells occurred apoptosis after exposed to 100 μM vitamin K2 for 24 hours, compared with less than 10% in control group (0 μM vitamin K2) (Fig 2B). Similarly, J82 and EJ cells also underwent significant apoptosis after treatment with increasing concentration of vitamin K2, with approximately 30% of cells occurred apoptosis in 100 μM vitamin K2-treated group, compared with about 7.0% of cells in control group (Fig 2C and S2 Fig). In addition, TUNEL assays showed that DNA strands was dramatically broken in the T24 cells treated with 100μM vitamin K2 for 24 hours, compared with intact DNA strands in control group (Fig 2D). To further ascertain the apoptotic effect of vitamin K2 on human bladder cancer cells, caspase-3 and PARP, typical apoptotic markers, were measured by western blots. As shown in Fig 2E, cleaved caspase-3 and PARP were induced in vitamin K2 dose dependent-treated T24 cells, which indicated vitamin K2 indeed triggered apoptosis in human bladder cancer T24 cells. Next, to further confirm whether vitamin K2-induce T24 cell apoptosis was caspase-3 dependent, Z-DEVD-FMK, an inhibitor of caspase-3, was used. As indicated in MTT and apoptotic assay, the Z-DEVD-FMK significantly blocked the decreased viability of vitamin K2-treated T24 cells (Fig 2F) and abolished the vitamin K2-induced apoptosis in T24 cells (Fig 2G), which revealed that caspase-3 was involved in vitamin K2-induced apoptosis in T24 cells. To evaluate the apoptotic effect of vitamin K2 on human normal cells, HEK239T cells were utilized. As shown in S1B Fig, no significant apoptosis occurred in HEK239T cells after exposed to the indicated concentration of vitamin K2 for 24 hours. These results suggest that vitamin K2 undoubtedly triggers apoptosis in human bladder cancer cells, but not in human normal cells. 10.1371/journal.pone.0161886.g002Fig 2 Vitamin K2 induced apoptotic cell death in human bladder cancer cells. (A). T24 cells were treated with the indicated concentration of vitamin K2 and the apoptosis was evaluated with Annexin V-FITC/PI dyes and measured by Flow cytometry. (B). The quantification of apoptotic death in vitamin K2-treated T24 cells. (C). Flow cytometry showed that vitamin K2 induced the apoptotic death in another two human bladder cancer J82 and EJ cells. (D). The effect of vitamin K2 on apoptosis in T24 cells was determined by TUNEL method using a detecting kit. Scale bar: 100μm (E). Western blots indicated that vitamin K2 induced activation of caspase-3 and cleavage of PARP in T24 cells. (F) Vitamin K2 inhibited the caspase-3-dependent viability of T24 cells by MTT assays. 10μM Z-DEVD-FMK, a caspase-3 inhibitor, was pretreated for 1 hours before exposure of 100 μM vitamin K2 to T24 cells for 24 hours. (G). Z-DEVD-FMK, a caspase-3 inhibitor, remarkably attenuated the apoptosis in vitamin K2-treated T24 cells. Cell apoptosis was evaluated with Annexin V-FITC/PI dyes and measured by Flow cytometry. * P<0.05, ** P<0.01 and *** P<0.001. Vitamin K2 induces mitochondria-related apoptosis in human bladder cancer cells To explore the underlying mechanism of vitamin K2-induced apoptosis in bladder cancer cells, we investigated whether mitochondria is associated with vitamin K2-induced apoptosis in human bladder cancer cells. As shown in Fig 3A, vitamin K2 remarkably disrupted the mitochondria membrane potential (MMP) of human bladder cancer T24 cells in a dose-dependent manner. As treatments with increasing concentration of vitamin K2 for 24 hours, a large number of T24 cells lost their MMP and approximately 90% of T24 cells had low MMP after exposed to vitamin K2 (100 μM) for 24 hours (Fig 3B). Similarly, vitamin K2 also caused significant collapse of MMP in J82 and EJ cells, another two human bladder cancer cells, in dose-dependent manners (Fig 3C). Moreover, the amount of cytochrome c in the mitochondrial fraction was reduced and conversely elevated in the cytosolic fraction after T24 cells were treated with vitamin K2 (50 μM and 100 μM) for 24 hours (Fig 3D). Next, we investigated whether Bcl-2 family proteins, such as Bax, Puma and Bcl-2, were implicated in the disruption of mitochondria membrane potential, upon vitamin K2 treatment. As shown in Fig 3E, vitamin K2 elevated the expression of Bax and Puma in T24 cells in a time-dependent manner. In contrast, the expression of Bcl-2, an anti-apoptotic protein, was diminished after prolonged treatment of vitamin K2. These results indicate that dysfunction of mitochondria is implicated in vitamin K2-induced apoptosis in human bladder cancer cells. 10.1371/journal.pone.0161886.g003Fig 3 Vitamin K2 triggered mitochondria-related apoptosis in human bladder cancer cells. (A). T24 cells were treated with the indicated concentration of vitamin K2 for 24 hours and the disruption of mitochondria membrane potential was measured using a specific mitochondria dye Rhodamine 123 by flow cytometry. M1 stands for the percentage of cells with low mitochondria membrane potential. (B). Quantification of T24 cells with low mitochondria membrane potential. (C). J82 and EJ cells were treated the indicated concentration of vitamin K2 for 24 hours and cells with low mitochondria membrane potential was determined using the Rhodamine 123 dye by flow cytometry. (D). T24 cells were treated by the indicated concentration of vitamin K2 for 24 hours, then cells were harvested and separated into cytosolic and mitochondrial fractions using a commercial kit. The expression of cytochrome C in cytosol and mitochondria was evaluated by western blots. (E) T24 cells were treated with 100μM vitamin K2 for 0, 12, 18, 24 hours respectively, then the total proteins were isolated from the cells and the expression of Bax and Puma were analyzed by western blots. * P<0.05, ** P<0.01 and *** P<0.001. Activation of JNK and p38 are required for vitamin K2-induced apoptosis in human bladder cancer cells We next investigated whether MAPKs were involved in vitamin K2-induced apoptosis in bladder cancer cells. As shown in Fig 4A and 4B, vitamin K2 significantly induced phosphorylation of JNK and p38 in human bladder cancer T24 cells in a dose and time-dependent manner. To further confirm whether JNK and p38 activation contributed to vitamin K2-triggered apoptosis in human bladder cancer cells, SP600125 (a pharmacological inhibitor of JNK) and SB203580 (a pharmacological inhibitor of p38) were used. As shown in Fig 4C and 4D and S3A Fig, pretreatment of 40 μM SP600125 remarkably attenuated the decrease of cell viability and abrogated the apoptosis in T24 cells after exposed to 100 μM vitamin K2 for 24 hours. Moreover, phosphorylation of JNK, cleaved caspase-3 and PARP induced by vitamin K2 were significantly abolished by SP600125 (Fig 4G). These results indicate that JNK activation is involved in vitamin K2-induced apoptosis in human bladder cancer T24 cells. In addition, addition of 10 μM SB203580 significantly inhibited vitamin K2-induced the decrease of cell viability and blocked vitamin K2-triggered apoptosis in T24 cells (Fig 4E and 4F and S3B Fig). Furthermore, as shown in Fig 4H, SB203580 remarkably attenuated phosphorylation of p38 and inhibited cleaved caspase-3 and PARP in vitamin K2-treated T24 cells. Thus, active p38 is also associated with the apoptosis in vitamin K2-treated T24 cells. 10.1371/journal.pone.0161886.g004Fig 4 Activation of JNK/p38 is required for vitamin K2-triggered apoptosis in human bladder cancer T24 cells. (A). T24 cells were treated with the indicated concentration of vitamin K2 for 24 hours, then the total proteins were isolated from the cells and the phosphorylation of JNK/p38 was analyzed by western blots. (B). Western blots indicated that vitamin K2 at the concentration of 100 μM induced sustained phosphorylation of JNK/p38 in T24 cells. (C). T24 cells were treated with 40 μM SP600125(SP), a pharmacological inhibitor of JNK activation, for 1 hour before treatment with 100 μM vitamin K2 for 24 hours and cell viability was evaluated by MTT assays. (D). T24 cells were treated 40 μM SP600125(SP) for 1 hour before treatment with 100 μM vitamin K2 for 12 hours and apoptotic death was assessed by flow cytometry. (E). T24 cells were treated with 10 μM SB203580(SB), a pharmacological inhibitor of p38 activation, for 1 hour before treatment with 100 μM vitamin K2 for 24 hours, then cell viability was assessed by MTT assays. (F). T24 cells were pre-treated with 10 μM SB203580(SB) for 1 hour, then treated with 100 μM vitamin K2 for 12 hours and apoptotic death was determined by flow cytometry. (G). T24 cells were treated with 40μM SP600125(SP) for 1 hour prior to treatment with 100 μM vitamin K2 for 24 hours. The total proteins extracted from the cells were assessed by western blots. (H). T24 cells were treated with 10 μM SB203580(SB) for 1 hours before treatment with 100 μM vitamin K2 for 24 hours, the total protein was evaluated by western blots. * P<0.05, ** P<0.01 and *** P<0.001. ROS generation is required for vitamin K2-triggered apoptosis in human bladder cancer cells Given that Reactive Oxygen Species (ROS) are able to initiate various stimuli-induced apoptosis, Next, we assessed whether ROS was involved in vitamin K2-triggered apoptosis in human bladder cancer cells. As shown in Fig 5A and S4A Fig, after treatment with 100μM vitamin K2 for 24 hours, intracellular ROS was significantly generated in human bladder cancer T24 cells, compared with control group (0 μM vitamin K2). Moreover, vitamin K2 induced ROS over-production in T24 cells in a dose-dependent manner. As shown in Fig 5B, vitamin K2 at concentration of 50 μM and 100 μM, respectively, elevated ROS level almost by 7.0 and 10.0 fold of the vehicle-treated group in T24 cells. In addition, ROS levels were also remarkably enhanced in J82 and EJ cells, upon treatments with vitamin K2 in dose-dependent manner (Fig 5C). Since exposure of human bladder cancer cells to vitamin K2 triggered ROS generation, we next evaluated the role of ROS generation in vitamin K2-triggered apoptosis in human bladder cancer cells. As shown in Fig 5F and S4B Fig, N-acetyl cysteine (NAC), a ROS scavenger, completely blocked ROS generation in vitamin K2-treated T24 cells. In addition, NAC almost reversed the cell viability decrease (S4C Fig) and abolished the apoptosis in vitamin K2-treated T24 cells (Fig 5D and 5E). Furthermore, as shown in Fig 5G, cleavage of caspase-3 and PARP that induced by vitamin K2 were almost blocked by pre-treatment of NAC, suggesting NAC completely abolished the vitamin K2-triggered apoptosis. Collectively, ROS generation is required and plays an essential role in vitamin K2-triggered apoptosis in human bladder cancer cells. 10.1371/journal.pone.0161886.g005Fig 5 Vitamin K2 induced ROS-mediated apoptosis in human bladder cancer cells. (A). T24 cells were treated with the indicated concentration of vitamin K2 for 24 hours and the intracellular ROS generation was evaluated using the DCFH-DA probe by flow cytometry. M reflects the positive DCF fluorescence (B). Quantification of the intracellular ROS generation in vitamin K2-treated T24 cells. (C). J82 and EJ cells were treated with the indicated concentration of vitamin K2 for 24 hours and intracellular ROS generation was assessed by flow cytometry. (D and E). T24 cells were treated with 5mM antioxidant N-acetyl cysteine (NAC) for 1 hour before the treatment with or without 100 μM vitamin K2 for 24 hours and the apoptotic death was determined by flow cytometry. (F). T24 cells were treated with 5mM antioxidant NAC for 1 hour before the treatment with or without 100 μM vitamin K2 for 12 hours and intracellular ROS generation was evaluated using the DCFH-DA probe by flow cytometry. (G). Activation of caspase-3 and cleavage of PARP were analyzed by western blots after T24 cells were treated with 5mM NAC for 1 hour before the treatment with or without 100 μM vitamin K2 for 24 hours. * P<0.05, ** P<0.01 and *** P<0.001. ROS mediates mitochondria dysfunction and regulates the activation of JNK and p38 in vitamin K2-treated human bladder cancer cells We next investigated the relationship between ROS generation and mitochondria dysfunction. As shown in Fig 6A, antioxidant NAC significantly attenuated the vitamin K2-induced disruption of mitochondria membrane potential in T24 cells, indicating that ROS generation was responsible for vitamin K2-induced mitochondria dysfunction. In addition, NAC remarkable inhibited vitamin K2-induced up-regulation of Bax and Puma (Fig 6B), suggesting that ROS regulated mitochondria dysfunction through ROS-mediated expression of Bax and Puma. Moreover, we continue to investigate whether activation of JNK and p38 were involved in Mitochondria dysfunction. Both SP600125 and SB203580 significantly inhibited the collapse of Mitochondria membrane potential, which reveals that activation of JNK and p38 contribute to Mitochondria dysfunction (Fig 6C and 6D). To verify the relationship between ROS and JNK/p38 in vitamin K2-induced apoptosis in bladder cancer cells, antioxidant NAC was used. As shown in Fig 6E, antioxidant NAC remarkably alleviated the vitamin K2-induced phosphorylation of JNK and p38, suggesting that ROS generation mediated activation of JNK and p38 in vitamin K2-treated T24 cells. These results indicate that vitamin K2 induces T24 cell apoptosis via ROS-JNK/p38-mediated mitochondria dysfunction. 10.1371/journal.pone.0161886.g006Fig 6 ROS mediated the mitochondria dysfunction and regulated activation of JNK/p38 in vitamin K2-triggered apoptosis of human bladder cancer T24 cells. (A). T24 cells were treated with 5mM antioxidant NAC for 1 hour prior to the treatment with or without 100 μM vitamin K2 for 24 hours, then the mitochondria membrane potential was assessed using the Rhodamine 123 dye by flow cytometry. (B). The expression of Bax, Puma and Bcl-2 were changed after treatment with 100 μM vitamin K2 for 24 hours in the present or absent of 5mM antioxidant N-acetyl cysteine (NAC) to human bladder cancer T24 cells. (C). T24 cells were treated 40 μM SP600125(SP) for 1 hour before treatment of 100 μM vitamin K2 for 24 hours, mitochondria membrane potential was evaluated using Rhodamine 123 dye by flow cytometry. (D). T24 cells were treated 10 μM SB203580(SB) for 1 hour before treatment of 100 μM vitamin K2 for 24 hours, mitochondria membrane potential was evaluated using Rhodamine 123 dye by flow cytometry. (E). T24 cells were treated with 5mM NAC for 1 hour before exposure to 100 μM vitamin K2 for 24 hours, then the total proteins were isolated from the cells and activation of JNK/p38 were determined by western blots. ** P<0.01 and *** P<0.001. Vitamin K2 exerts the activity of inhibitory growth in xenografted nude mice model by causing apoptosis To evaluate the effect of vitamin K2 on inhibitory growth in human bladder cancer cells in vivo, human bladder cancer EJ cells were injected subcutaneously into nude mice. When transplanted tumors reached a mean group size of approximately 50 mm3, mice were treated every day for 21 days by directly injection of 30 mg/kg vitamin K2 at tumors and directly injection of the equivalent volume of PBS as controls. As shown in Fig 7A and 7B, in nude mice, vitamin K2 remarkably inhibited the tumor growth and the tumor volume was gradually reduced after the 11th day, compared with the sustained growth of control group. To determine whether the reduced tumor growth was due to the apoptotic effect of vitamin K2, we excised the tumors from the mice and sectioned for caspase-3 activity, TUNEL and HE staining assay. Compared with the control group, vitamin K2 induced activation of caspase-3 in tumor sections. Moreover, the TUNEL and HE staining assay showed the robust apoptosis in tumor sections from vitamin K2-treated mice, compared with the control group (Fig 7C). Taken together, vitamin K2 indeed inhibits the EJ cell growth in vivo by causing apoptotic death. 10.1371/journal.pone.0161886.g007Fig 7 Vitamin K2 inhibited the tumor growth in mouse bearing human bladder cancer cells. Nude mice with EJ transplant tumors were directly injected with 30 mg/kg vitamin K2 at tumors each day for 21 days. (A). Tumor volume changed after administration with 30 mg/kg vitamin K2 everyday. (B). Measurement of tumor volume in mice after treatment with or without 30 mg/kg vitamin K2 each day for 21 days before sacrificed the nude mice. (C). After 21 days treatments, mice were sacrificed and tumors were excised to sections. Activation of caspase-3 in the sections was measured with the immuno-histo-chemistry method using antibody against caspase-3. Staining TUNEL and HE in the sections was measured by the commercial kits, respectively. Scale bar: 50μm. Discussion In this study, the anticancer effect of vitamin K2 on human bladder cancer cells was the first to demonstrate and the related mechanism was elucidated. As shown in Fig 8, vitamin K2 induces mitochondria-related apoptosis in human bladder cancer cells via ROS-JNK/p38 pathways, which explains the reason why vitamin K2 exerts anticancer activity in human bladder cancer cells. 10.1371/journal.pone.0161886.g008Fig 8 Schematic diagram of pathway involved in vitamin K2-induced apoptosis in human bladder cancer cells. As suggested in many recent studies, multiple clinical scenarios were employed to cure bladder cancer, including radical cystectomy, radiotherapy, chemotherapy, immunotherapy and so forth. In particular, radical cystectomy and chemotherapy are considered as effective therapeutic regimen to treat bladder cancer, however, they have many severe adverse effects, such as distant metastasis, local reccurence, toxicity to other normal organs and cost-effectiveness, which dramatically affect the life quality of patients[3,6,7]. Therefore, new therapeutics with less side effects to cure bladder cancer are greatly required. It is widely recognized that vitamin K2 is closely associated with the improvement of human health, including functions as cofactor for blood coagulation, bone metabolism and reduces the arterial calcium deposition avoiding vascular calcification[8,11]. Apart from these functions, interestingly, vitamin K2 also exerts potent anticancer activity in various cancer cells. Accumulating recent studies have documented that vitamin K2 induces growth suppression and apoptosis in a variety of cancer cells, including lung carcinomas, acute myeloid leukemia, ovarian cancer cells, prostate cancer cells and HCC cells[15,17–19]. Consistent with the previous studies, it was indicated in our results that vitamin K2 exerts anticancer activity in bladder cancer cells, including inhibits cell growth by reducing the cell viability and triggers cell apoptosis by DNA breaks, activation of caspase-3 and cleavage of PARP. Moreover, Z-DEVD-FMK, a pharmacological caspase-3 inhibitor, significantly reverses the vitamin K2-induced apoptosis in T24 cells, suggesting caspase-3 mediates vitamin K2-induced apoptosis in bladder cancer cells. A growing number of studies have revealed that Mitochondria plays a pivotal role in regulating apoptotic signal pathways [20–22]. In this regard, targeting the mitochondria might be a novel strategy for cancer therapy. As indicated from the previous studies, the mechanism of mitochondria-mediated apoptosis mainly depends on the dysfunction of mitochondria including loss of mitochondria membrane potential and apoptotic factors, such as cytochrome C, AIF, and smac, release into cytosol, which subsequently, activates caspase cascade. Interestingly, in this study, the remarkable collapse of mitochondria membrane potential was displayed in vitamin K2-treated T24 cells and cytochrome C, in turn, released from the mitochondria to cytosol. These results suggest that vitamin K2 induces mitochondria-related apoptosis in human bladder cancer T24 cells. Recent studies have suggested that Bcl-2 family proteins are greatly responsible for mitochondria dysfunction[23–25]. Up-regulation of Bax, Bak and Puma (proapoptotic proteins of Bcl-2 family) due to cellular stress can directly or indirectly cause collapse of mitochondria membrane potential. Satoki et al. had recently reported that vitamin K2 induces up-regulation of Bax and Bak, which lead to loss of mitochondria membrane potential in Hela cells[25]. In accordance with the former studies, our results showed that up-regulation of Bax and puma was also induced in vitamin K2-treated T24 cells, which may be one of reasons that vitamin K2 caused loss of mitochondria membrane potential in T24 cells. MAPKs are one of the sensors in response to extra-cellular stimuli and mediate the cellular signals[26–28]. ERK is usually associated with cell proliferation and growth. In contrast, JNK and p38 are induced by cellular stress and closely associated with cell death [29–31]. To elucidate the exact mechanism involving in vitamin K2-induced apoptosis in human bladder cancer cells, the effect of vitamin K2 on activation of MAPKs was examined. Our results showed vitamin K2 induced activation of JNK and p38 in human bladder cancer T24 cells. In addition, either the SP600125 (a JNK inhibitor), or SB203580 (a p38 inhibitor) completely blocked the vitamin K2-induced apoptosis in human bladder cancer T24 cells, suggesting activation of JNK and p38 are required and involved in vitamin K2-induced apoptosis in T24 cells. Furthermore, it is interesting that SP600125 as well as SB203580 remarkably alleviated the disruption of mitochondria membrane potential, which indicates JNK, as well as p38, contributes to vitamin K2-induced mitochondria dysfunction in human bladder cancer T24 cells. There are increasing evidences indicating that reactive oxygen species (ROS) mediates the intracellular signal cascades and excessive ROS production leads to intracellular stress, mitochondria dysfunction and ultimately cell apoptosis or necrosis [32–35]. In this study, vitamin K2 induced ROS generation in human bladder cancer cells in a dose-dependent manner. Moreover, antioxidant NAC significantly abolished the apoptosis and collapse of mitochondria membrane potential in vitamin K2-treated T24 cells. These results suggest that ROS remarkably mediates the mitochondria-related apoptosis in vitamin K2-treated T24 cells. Recent studies have elucidated that ROS mediated MAPKs activation in various stimuli-triggered cell apoptosis[26,31,36]. Concordantly, in our hands, antioxidant NAC significantly inhibited phosphorylation of JNK and p38, suggesting that ROS generation activates the JNK and p38, which is supposed to the upstream of vitamin K2-induced apoptotic pathway in human bladder cancer T24 cells. Growing evidence has indicated that many chemotherapeutic agents exhibit anticancer activity in numerious cancer cells by inducing ROS generation, suggesting that vitamin K2, in some extent like chemotherapeutic drugs, exerts anticancer activity in bladder cancer cells by providing oxidative stress. In vivo study, we investigated the effect of vitamin K2 on the growth of human bladder cancer cells. As the results shown, vitamin K2 indeed inhibits the tumor growth in xenografte nude mice. In addition, we further determined that the inhibition of tumor growth was mainly due to vitamin K2-induced apoptotic cell death. These results indicate vitamin K2 is able to induce apoptosis in human bladder cancer cells in vivo. Furthermore, it is indicated from the latest studies that vitamin K2 is not applied in clinical therapy for cancer because of its insufficient strong activity to cancer. Considering the current clinical concerns, the method of directly injection of vitamin K2 at tumor was employed in our studies. Interestingly, treatment by directly injection of vitamin K2 into the tumors is a highly efficient method to kill the human bladder cancer cells in vivo, which maybe provide a new sight for clinical research. In conclusion, our results demonstrated that vitamin K2 was able to induce mitochondria-related apoptosis in human bladder cancer cells via ROS-JNK/p38 MAPK signal pathways, which indicated a detailed mechanism of the anticancer activity of vitamin K2 in human bladder cancer cells. In addition, vitamin K2 also suppresses the growth of human bladder cancer cells in nude mice, which was further confirmed by vitamin K2-induced apoptosis. Considering the potent apoptotic effect on human bladder cancer cells but not on human normal cells, vitamin K2 will become a promising anticancer agent to cure human bladder cancer in future. Supporting Information S1 Fig Vitamin K2 had no siginificant effect on cell viability or apoptosis in human normal cells. (TIF) Click here for additional data file. S2 Fig Vitamin K2 induced apoptosis in human bladder cancer J82 and EJ cells (TIF) Click here for additional data file. S3 Fig Vitamin K2 induced JNK/p38-mediated apoptosis in human bladder cancer T24 cells. (TIF) Click here for additional data file. S4 Fig Intracellular ROS is generated and responsible for the viability decrease in vitamin K2-treated T24 cells. (TIF) Click here for additional data file. We thank the instruments shared platform for technology assistances. ==== Refs References 1 Witjes JA . Bladder cancer in 2015: Improving indication, technique and outcome of radical cystectomy . Nat Rev Urol . 2016 ; 13 : 74 –76 . 10.1038/nrurol.2015.272 26597614 2 Hermans TJ , Mertens LS , van Rhijn BW . Re: Trends in the Use of Perioperative Chemotherapy for Localized and Locally Advanced Muscle-invasive Bladder Cancer: A Sign of Changing Tides . Eur Urol . 2016 ; 69 : 1156 –1157 . 10.1016/j.eururo.2016.02.023 27302138 3 Stevenson SM , Danzig MR , Ghandour RA , Deibert CM , Decastro GJ , Benson MC , et al Cost-effectiveness of neoadjuvant chemotherapy before radical cystectomy for muscle-invasive bladder cancer . Urol Oncol . 2014 ; 32 : 1172 –1177 . 10.1016/j.urolonc.2014.05.001 24998787 4 Feuerstein MA , Goenka A . Quality of Life Outcomes for Bladder Cancer Patients Undergoing Bladder Preservation with Radiotherapy . Curr Urol Rep . 2015 ; 16 :75 10.1007/s11934-015-0547-1 26343030 5 Schepisi G , Santoni M , Massari F , Gurioli G , Salvi S , Conteduca V , et al Urothelial Cancer: Inflammatory Mediators and Implications for Immunotherapy . BioDrugs . 2016 : PMID: 27177757 6 Juffs HG , Moore MJ , Tannock IF . The role of systemic chemotherapy in the management of muscle-invasive bladder cancer . Lancet Oncol . 2002 ; 3 : 738 –747 . 10.1016/S1470-2045(02)00930-0 12473515 7 Hafeez S , Huddart R . Selective organ preservation for the treatment of muscle-invasive transitional cell carcinoma of the bladder: a review of current and future perspectives . Expert Rev Anticancer Ther . 2014 ; 14 : 1429 –1443 . 10.1586/14737140.2014.953938 25263197 8 O'Keefe JH , Bergman N , Carrera-Bastos P , Fontes-Villalba M , DiNicolantonio JJ , Cordain L . Nutritional strategies for skeletal and cardiovascular health: hard bones, soft arteries, rather than vice versa . Open Heart . 2016 ; 3 : e000325 10.1136/openhrt-2015-000325 27042317 9 Lamson DW , Plaza SM . The anticancer effects of vitamin K . Altern Med Rev . 2003 ; 8 : 303 –318 12946240 10 Ebina K , Noguchi T , Hirao M , Kaneshiro S , Tsukamoto Y , Yoshikawa H . Comparison of the effects of 12 months of monthly minodronate monotherapy and monthly minodronate combination therapy with vitamin K2 or eldecalcitol in patients with primary osteoporosis . J Bone Miner Metab . 2016 ; 34 : 243 –250 . 10.1007/s00774-015-0710-2 26303222 11 Shearer MJ , Newman P . Recent trends in the metabolism and cell biology of vitamin K with special reference to vitamin K cycling and MK-4 biosynthesis . J Lipid Res . 2014 ; 55 : 345 –362 . 10.1194/jlr.R045559 24489112 12 Nakagawa K , Hirota Y , Sawada N , Yuge N , Watanabe M , Uchino Y , et al Identification of UBIAD1 as a novel human menaquinone-4 biosynthetic enzyme . Nature . 2010 ; 468 : 117 –121 . 10.1038/nature09464 20953171 13 Yoshida T , Miyazawa K , Kasuga I , Yokoyama T , Minemura K , Ustumi K , et al Apoptosis induction of vitamin K2 in lung carcinoma cell lines: the possibility of vitamin K2 therapy for lung cancer . Int J Oncol . 2003 ; 23 : 627 –632 . 10.3892/ijo.23.3.627 12888897 14 Yaguchi M , Miyazawa K , Katagiri T , Nishimaki J , Kizaki M , Tohyama K , et al Vitamin K2 and its derivatives induce apoptosis in leukemia cells and enhance the effect of all-trans retinoic acid . Leukemia . 1997 ; 11 : 779 –787 .9177427 15 Wei G , Wang M , Hyslop T , Wang Z , Carr BI . Vitamin K enhancement of sorafenib-mediated HCC cell growth inhibition in vitro and in vivo . Int J Cancer . 2010 ; 127 : 2949 –2958 . 10.1002/ijc.25498 21351273 16 Miyazawa K , Yaguchi M , Funato K , Gotoh A , Kawanishi Y , Nishizawa Y , et al Apoptosis/differentiation-inducing effects of vitamin K2 on HL-60 cells: dichotomous nature of vitamin K2 in leukemia cells . Leukemia . 2001 ; 15 : 1111 –1117 . 11455981 17 Samykutty A , Shetty AV , Dakshinamoorthy G , Kalyanasundaram R , Zheng G , Chen A , et al Vitamin k2, a naturally occurring menaquinone, exerts therapeutic effects on both hormone-dependent and hormone-independent prostate cancer cells . Evid Based Complement Alternat Med . 2013 ; 2013 : 287358 10.1155/2013/287358 24062781 18 Tokita H , Tsuchida A , Miyazawa K , Ohyashiki K , Katayanagi S , Sudo H , et al Vitamin K2-induced antitumor effects via cell-cycle arrest and apoptosis in gastric cancer cell lines . Int J Mol Med . 2006 ; 17 : 235 –243 . 10.3892/ijmm.17.2.235 16391821 19 Matsumoto K , Okano J , Nagahara T , Murawaki Y . Apoptosis of liver cancer cells by vitamin K2 and enhancement by MEK inhibition . Int J Oncol . 2006 ; 29 : 1501 –1508 . 10.3892/ijo.29.6.1501 17088989 20 Park GB , Kim YS , Lee HK , Song H , Kim S , Cho DH , et al Reactive oxygen species and p38 MAPK regulate Bax translocation and calcium redistribution in salubrinal-induced apoptosis of EBV-transformed B cells . Cancer Lett . 2011 ; 313 : 235 –248 . 10.1016/j.canlet.2011.09.011 22056078 21 Yang CR , Liao WS , Wu YH , Murugan K , Chen C , Chao JI . CR108, a novel vitamin K3 derivative induces apoptosis and breast tumor inhibition by reactive oxygen species and mitochondrial dysfunction . Toxicol Appl Pharmacol . 2013 ; 273 : 611 –622 . 10.1016/j.taap.2013.10.007 24128853 22 Shibayama-Imazu T , Sonoda I , Sakairi S , Aiuchi T , Ann WW , Nakajo S , et al Production of superoxide and dissipation of mitochondrial transmembrane potential by vitamin K2 trigger apoptosis in human ovarian cancer TYK-nu cells . Apoptosis . 2006 ; 11 : 1535 –1543 . 10.1007/s10495-006-7979-5 16763728 23 Yu J , Zhang L . PUMA, a potent killer with or without p53 . Oncogene . 2008 ; 27 Suppl 1 : S71 –83 . doi: 10.1038/onc.2009.45 PMID: 86043219641508 24 Zhang D , Armstrong JS . Bax and the mitochondrial permeability transition cooperate in the release of cytochrome c during endoplasmic reticulum-stress-induced apoptosis . Cell Death Differ . 2007 ; 14 : 703 –715 . 10.1038/sj.cdd.4402072 17170750 25 Karasawa S , Azuma M , Kasama T , Sakamoto S , Kabe Y , Imai T , et al Vitamin K2 covalently binds to Bak and induces Bak-mediated apoptosis . Mol Pharmacol . 2013 ; 83 : 613 –620 . 10.1124/mol.112.082602 23229512 26 Shi Y , Nikulenkov F , Zawacka-Pankau J , Li H , Gabdoulline R , Xu J , et al ROS-dependent activation of JNK converts p53 into an efficient inhibitor of oncogenes leading to robust apoptosis . Cell Death Differ . 2014 ; 21 : 612 –623 . 10.1038/cdd.2013.186 24413150 27 Song IS , Jun SY , Na HJ , Kim HT , Jung SY , Ha GH , et al Inhibition of MKK7-JNK by the TOR signaling pathway regulator-like protein contributes to resistance of HCC cells to TRAIL-induced apoptosis . Gastroenterology . 2012 ; 143 : 1341 –1351 . 10.1053/j.gastro.2012.07.103 22841785 28 Wang H , Jiang D , Liu J , Ye S , Xiao S , Wang W , et al Compound K induces apoptosis of bladder cancer T24 cells via reactive oxygen species-mediated p38 MAPK pathway . Cancer Biother Radiopharm . 2013 ; 28 : 607 –614 . 10.1089/cbr.2012.1468 23895116 29 Javadov S , Jang S , Agostini B . Crosstalk between mitogen-activated protein kinases and mitochondria in cardiac diseases: therapeutic perspectives . Pharmacol Ther . 2014 ; 144 : 202 –225 . 10.1016/j.pharmthera.2014.05.013 24924700 30 Krilleke D , Ucur E , Pulte D , Schulze-Osthoff K , Debatin KM , Herr I . Inhibition of JNK signaling diminishes early but not late cellular stress-induced apoptosis . Int J Cancer . 2003 ; 107 : 520 –527 . 10.1002/ijc.11331 14520687 31 Xiong XX , Liu JM , Qiu XY , Pan F , Yu SB , Chen XQ . Piperlongumine induces apoptotic and autophagic death of the primary myeloid leukemia cells from patients via activation of ROS-p38/JNK pathways . Acta Pharmacol Sin . 2015 ; 36 : 362 –374 . 10.1038/aps.2014.141 25619389 32 Santabarbara-Ruiz P , Lopez-Santillan M , Martinez-Rodriguez I , Binagui-Casas A , Perez L , Milan M , et al ROS-Induced JNK and p38 Signaling Is Required for Unpaired Cytokine Activation during Drosophila Regeneration . PLoS Genet . 2015 ; 11 : e1005595 10.1371/journal.pgen.1005595 26496642 33 Ghavami S , Kerkhoff C , Los M , Hashemi M , Sorg C , Karami-Tehrani F . Mechanism of apoptosis induced by S100A8/A9 in colon cancer cell lines: the role of ROS and the effect of metal ions . J Leukoc Biol . 2004 ; 76 : 169 –175 . 10.1189/jlb.0903435 15075348 34 Vaseva AV , Marchenko ND , Ji K , Tsirka SE , Holzmann S , Moll UM . p53 opens the mitochondrial permeability transition pore to trigger necrosis . Cell . 2012 ; 149 : 1536 –1548 . 10.1016/j.cell.2012.05.014 22726440 35 Latimer HR , Veal EA . Peroxiredoxins in Regulation of MAPK Signalling Pathways; Sensors and Barriers to Signal Transduction . Mol Cells . 2016 ; 39 : 40 –45 . 10.14348/molcells.2016.2327 26813660 36 Li S , Dong P , Wang J , Zhang J , Gu J , Wu X , et al Icariin, a natural flavonol glycoside, induces apoptosis in human hepatoma SMMC-7721 cells via a ROS/JNK-dependent mitochondrial pathway . Cancer Lett . 2010 ; 298 : 222 –230 . 10.1016/j.canlet.2010.07.009 20674153

PMC005xxxxxx/PMC5003393.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757135610.1371/journal.pone.0160731PONE-D-16-09697Research ArticleBiology and Life SciencesAnatomyCardiovascular AnatomyBlood VesselsAortaMedicine and Health SciencesAnatomyCardiovascular AnatomyBlood VesselsAortaBiology and Life SciencesCell BiologyCellular TypesAnimal CellsBlood CellsWhite Blood CellsNeutrophilsBiology and Life SciencesCell BiologyCellular TypesAnimal CellsImmune CellsWhite Blood CellsNeutrophilsBiology and Life SciencesImmunologyImmune CellsWhite Blood CellsNeutrophilsMedicine and Health SciencesImmunologyImmune CellsWhite Blood CellsNeutrophilsBiology and Life SciencesBiochemistryPeptidesGlutathioneBiology and Life SciencesImmunologyImmune ResponseInflammationMedicine and Health SciencesImmunologyImmune ResponseInflammationMedicine and Health SciencesDiagnostic MedicineSigns and SymptomsInflammationMedicine and Health SciencesPathology and Laboratory MedicineSigns and SymptomsInflammationBiology and Life SciencesAnatomyBody FluidsBloodBlood PlasmaMedicine and Health SciencesAnatomyBody FluidsBloodBlood PlasmaBiology and Life SciencesPhysiologyBody FluidsBloodBlood PlasmaMedicine and Health SciencesPhysiologyBody FluidsBloodBlood PlasmaMedicine and Health SciencesHematologyBloodBlood PlasmaMedicine and Health SciencesVascular MedicineVasoconstrictionMedicine and Health SciencesDiagnostic MedicineSigns and SymptomsLesionsMedicine and Health SciencesPathology and Laboratory MedicineSigns and SymptomsLesionsBiology and Life SciencesBiochemistryNeurochemistryNeurotransmittersBiogenic AminesSerotoninBiology and Life SciencesNeuroscienceNeurochemistryNeurotransmittersBiogenic AminesSerotoninInflammation and Vascular Effects after Repeated Intratracheal Instillations of Carbon Black and Lipopolysaccharide Inflammation and Vascular Effects after Exposure to Carbon Black and LPSChristophersen Daniel Vest 1Jacobsen Nicklas Raun 2Jensen Ditte Marie 1Kermanizadeh Ali 1Sheykhzade Majid 3Loft Steffen 1Vogel Ulla 24Wallin Håkan 12http://orcid.org/0000-0002-2021-1249Møller Peter 1*1 Department of Public Health, Section of Environmental Health, University of Copenhagen, Copenhagen K, Denmark2 The National Research Centre for the Working Environment, Copenhagen, Denmark3 Department of Drug Design and Pharmacology, Section of Molecular and Cellular Pharmacology, Faculty of Health Sciences, University of Copenhagen, Copenhagen, Denmark4 Department of Micro- and Nanotechnology, Technical University of Denmark, Kgs. Lyngby, DenmarkStoeger Tobias EditorGERMANYCompeting Interests: The authors have declared that no competing interests exist. Conceived and designed the experiments: DVC HW NRJ PM SL UV. Performed the experiments: DVC NRJ AK DMJ. Analyzed the data: DVC HW NRJ PM SL UV. Contributed reagents/materials/analysis tools: HW NRJ PM SL UV MS. Wrote the paper: DVC NRJ PM. * E-mail: Pemo@sund.ku.dk29 8 2016 2016 11 8 e01607317 3 2016 25 7 2016 © 2016 Christophersen et al2016Christophersen et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Inflammation and oxidative stress are considered the main drivers of vasomotor dysfunction and progression of atherosclerosis after inhalation of particulate matter. In addition, new studies have shown that particle exposure can induce the level of bioactive mediators in serum, driving vascular- and systemic toxicity. We aimed to investigate if pulmonary inflammation would accelerate nanoparticle-induced atherosclerotic plaque progression in Apolipoprotein E knockout (ApoE-/-) mice. ApoE -/- mice were exposed to vehicle, 8.53 or 25.6 μg nanosized carbon black (CB) alone or spiked with LPS (0.2 μg/mouse/exposure; once a week for 10 weeks). Inflammation was determined by counting cells in bronchoalveolar lavage fluid. Serum Amyloid A3 (Saa3) expression and glutathione status were determined in lung tissue. Plaque progression was assessed in the aorta and the brachiocephalic artery. The effect of vasoactive mediators in plasma of exposed ApoE-/- mice was assessed in aorta rings isolated from naïve C57BL/6 mice. Pulmonary exposure to CB and/or LPS resulted in pulmonary inflammation with a robust influx of neutrophils. The CB exposure did not promote plaque progression in aorta or BCA. Incubation with 0.5% plasma extracted from CB-exposed ApoE-/- mice caused vasoconstriction in aorta rings isolated from naïve mice; this effect was abolished by the treatment with the serotonin receptor antagonist Ketanserin. In conclusion, repeated pulmonary exposure to nanosized CB and LPS caused lung inflammation without progression of atherosclerosis in ApoE-/- mice. Nevertheless, plasma extracted from mice exposed to nanosized CB induced vasoconstriction in aortas of naïve wild-type mice, an effect possibly related to increased plasma serotonin. Danish Working Environment Research Foundation20110092173-3The work was supported by the Danish Nanosafety Centre (grant no. 20110092173-3) from the Danish Working Environment Research Foundation. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Data AvailabilityAll relevant data are within the paper and its Supporting Information files.Data Availability All relevant data are within the paper and its Supporting Information files. ==== Body Introduction Cardiovascular disease is the leading cause of mortality worldwide, with exposure to inhaled airborne particulate matter being a major contributor to the burden of disease [1]. Exposure to particulate matter from ambient air or nanomaterials (NMs) is associated with vascular dysfunction and progression of atherosclerosis, which is believed to be promoted by inflammation, acute phase response and oxidative stress in the lungs and target tissue [2–5]. The role of pulmonary inflammation in particle-induced cardiovascular outcomes has been debated with experimental evidence ranging from the spill-over effect of cytokines from the lung to the circulation [6] to vascular effects occurring independently of pulmonary inflammation [7]. A number of studies have addressed cardiovascular effects of carbon black (CB) exposure in animal models. CB is a widely used carbon-based material in both bulk form and as an NM. The latter has a high particle number to mass ratio and unique surface properties, which make them attractive component in many products, although the small size and large surface area potentially make them more toxic than their larger counterparts [8]. It has been shown that pulmonary exposure to CB by intratracheal instillation (i.t.) once a week for ten weeks in LDL receptor knockout mice on a cholesterol-rich diet promoted progression of atherosclerotic plaque in the aorta [9]. There has also been reported a decrease in the vasorelaxation response in aortic segments from ApoE-/- mice after two i.t. instillations of nanosized CB, whereas the same dose did not cause plaque progression in the aorta or the brachiocephalic artery (BCA) [10]. Nevertheless, 4 weeks inhalation by nose-only exposure to CB did not affect the vasoconstriction and vasorelaxation response in the aorta of rats [11]. Particle-generated vasomotor dysfunction may be related to relatively acute exposures because it has been shown that exposure to nanosized CB did not alter the vasorelaxation response in second order branch interlobar pulmonary arteries of male Wistar rats at day 21 after an i.t. instillation [12]. The same pattern was observed in dyslipidemic Zucker rats following oral administration of CB for 10 weeks; vasomotor dysfunction in the aorta was observed 24 h after the last exposure, whereas this particle-effect had subsided at 13 weeks post-exposure [13]. The mechanistic link between pulmonary/intestinal exposure to CB and cardiovascular outcomes has not been elucidated. The direct application of nanosized CB to artery segments caused dysfunction of the vasorelaxation response and increased vasoconstriction[14, 15]. However, these ex vivo observations are based on concentrations that are much higher than the concentrations that may be relevant in physiological scenarios after pulmonary exposure. The translocation of poorly soluble particles is typically much less than 1% of the deposited dose in the lungs, whereas higher translocation can ben observed for nanomaterials with high solubility such as zincoxide or nanosilver [16]. In addition, systemic low-grade inflammation has been hypothetized to be a mechanism of vascular effects following pulmonary exposure to particulate matter, but a recent systematic review and meta-analysis of the literature do not demonstrate any consistency between systemic inflammation and vascular endpoints [17]. Interestingly, recent studies have shown that pulmonary exposure to particles or ozone can induce the level of bioactive mediators in serum that acts as drivers for systemic toxicity, including loss of vascular integrity and vasomotor dysfunction [18–20]. As such, it is difficult to disentangle the contribution of pulmonary inflammation from other mediators on vascular effects in CB exposed animals because the material dose-dependently causes a strong pulmonary inflammation response [21]. The aim of this study was to investigate if pulmonary inflammation exacerbated nanosized CB induced progression of atherosclerosis. LPS was used as a non-particulate agent because it induces inflammation and accelerates plaque progression in ApoE-/- mice [22]. We carried out a range-finding experiment to find a dose of LPS that would produce a similar extent of pulmonary inflammation as the CB material of choice (i.e. Printex 90). Subsequently, we assessed pulmonary inflammation and atherosclerotic plaque progression after CB and/or LPS exposure in ApoE-/- mice. Plaque progression was evaluated in the aorta and BCA; both arteries are prone to develop atherosclerotic lesions in ApoE-/- mice. Pulmonary inflammation was assessed as a neutrophilic influx in bronchoalveolar lavage fluid (BALF) and the acute phase response as gene expression of Serum Amyloid A3 (Saa3) in lung tissue. The level of oxidative stress in the lungs was assessed by measuring the glutathione balance. Finally, we investigated vasoactive effects of mediators in plasma from nanosized CB, and/or LPS exposed ApoE-/- mice in aortas of naïve wild-type C57BL/6 mice (the ApoE-/- background strain). Material and Methods Particles CB was chosen as a high volume industrial NM. Specifically, we used Printex 90 because it is well-characterized physicochemically and it has been used in several studies as benchmark type on NM that causes both pulmonary inflammation and oxidative stress [23–30]. Printex 90 was a kind gift from Degussa-Hüls, Frankfurt, Germany. We have previously analyzed and reported the physicochemical characteristics of the Printex 90 material [31]. Printex 90 was suspended in nanopure water (< 45 μm pore size) to a concentration of 2.5 mg/ml and sonicated in an ice bath for 16 min with 10% amplitude using a Branson Sonifier S-450D (Branson Ultrasonics Corp., Danbury, CT, USA) equipped with a disruptor horn (Model number: 101-147-037). The instillation vehicle was prepared as described above without particles. For LPS spiked suspensions, 4 μg/ml LPS (Escherichia coli 055: B5, Sigma-Aldrich) was added to the sonicated suspension with or without particles and vortexed for 2 min before i.t. instillation. A fresh suspension was prepared immediately before i.t. instillation. Dynamic light scattering (DLS) Zetasizer Nano ZS (Malvern Instruments Ltd., UK) was used to measure the hydrodynamic size distribution of CB in nanopure water as described previously [29]. Calculations were done by the DTS software using the viscosity of water. S1 Fig depicts particle size in the suspensions. In brief, the particle size of CB was 44 and 38 nm in the low and high dose groups, respectively. Addition of LPS increased the particle size of the suspension to 1281 and 1718 nm, respectively. Likewise, the polydispersity index of the suspensions with CB were lower as compared to the suspensions with CB and LPS. Animal and caging conditions For the LPS pilot study, female C57BL/6-Ntac mice were purchased from Taconic (Ejby, Denmark) at nine weeks of age and allowed one week of acclimatization before the first exposure. For the main studies, female C57BL/6-Apoetm1 ApoE-/- mice were purchased from Taconic (Ejby, Denmark) at eight weeks of age and allowed two weeks of acclimation before the experimental procedure. All mice were randomly assigned to groups of 10 animals and housed in polypropylene cages (Jeluxyl HW 300/500) with sawdust bedding and enrichment, such as pinewood sticks and rodent tunnels. The mice were maintained on a 12:12 h light-dark cycle and with controlled humidity and temperature. Mice had ad libitum access to regular mouse chow (Altromin no. 1324 Christian Petersen, Denmark) and tap water. All animal procedures followed the local institutional and governmental guidelines on animal ethics and welfare issued by the Danish government and the Animal Experimental Inspectorate under the Ministry of Justice approved the study (permission 2010/561-1779). Study design LPS pilot study The LPS dose-range study was designed to find a dose of LPS that in C57BL/6 mice when repeatedly administered, would give a neutrophil influx in BALF similar to that of CB. Furthermore, the purpose of the experiment was to select doses of CB and LPS that would not cause a saturation of the inflammatory response. In the assessment of dose-response relationship, C57BL/6 mice were exposed by i.t. instillation to vehicle, 0.047 μg, 0.094 μg, 0.187 μg, 0.375 μg, 0.750 μg, 1.5 μg, 3.0 μg or 6.0 μg of LPS and euthanized at 24 h post-exposure (n = 3 mice/group). This was followed by a repeated exposure study in which C57BL/6 mice were exposed by i.t. instillation to vehicle, 0.2 μg (low-dose) or 1.0 μg (high-dose) of LPS once a week for four weeks (n = 3 mice/group). The total dose in the repeated exposure study was 0.8 μg (low-dose) and 4 μg (high-dose). The mice were euthanized at 24 h after the last exposure, and the BALF was collected. The vehicle utilized in these experiments was 2% serum obtained from sibling mice diluted with nanopure water under sterile conditions. Main study 1 This study was conducted as a 2x3 factorial design with six exposure groups (n = 10/group) of ApoE-/- mice as follows: 1) vehicle (nanopure water), 2) vehicle spiked with LPS, 3) low-dose CB, 4) low-dose CB spiked with LPS, 5) high-dose CB, and 6) high-dose CB spiked with LPS. All mice were exposed by i.t. instillation once a week for 10 weeks. The total dose given in the 10 administrations was 85.3 μg/mouse for the low-dose and 256 μg/mouse for high-dose nanosized CB and 2 μg/mouse for LPS spiked groups. Main study 2 To increase the statistical power of plaque progression results for the second and third exposure groups, a second study was performed. In this study only the vehicle (nanopure water), vehicle spiked with LPS, and low-dose of CB groups were included. Ten ApoE-/- mice in each group were exposed to vehicle, CB (total dose = 85.3 μg/mouse) or LPS (total dose = 2 μg/mouse) once a week for ten weeks by the same procedure as described in study 1. The exposure doses of 8.53 and 25.6 μg/mouse of CB were selected from earlier observations on i.t. instillations in C57BL/6 mice that showed neutrophilic influx in BALF at the high dose [21, 25, 29, 32, 33]. S2 Fig depicts the dose-response relationship for total cells and neutrophils in earlier studies. Four repeated i.t. instillations of CB to pregnant C57BL/6 mice were not associated with increased influx of neutrophils at a weekly dose of 14 μg/mouse (total dose = 54 μg/mouse) [34]. Moreover, earlier observations in which CB exposure resulted in a stronger influx of neutrophils in BALF in ApoE-/- mice as compared to wild-type counterparts were taken into account [21]. Based on these observations, it was anticipated that CB exposure would yield approximately 50.000 neutrophils in BALF. Hence, an LPS dose of 0.2 μg/mouse (total dose = 2 μg/mouse) was selected for the repeated exposure study in C57BL/6 mice. The lowest of exposure to CB (8.53 μg/mouse, corresponding to approximately 0.30 mg/kg bodyweight per week) is approximately equal to the accumulated dose that humans can encounter during a one-week stay in a working environment with the same aerosol concentration of respirable CB as the Danish threshold limit (i.e. aerosol concentration of 3.5 μg/m3, inhalation of 8 m3/day, 15% deposition in lower respiratory tract and 70 kg bodyweight). Intratracheal instillation All mice were anesthetized using 3–4% isoflurane until fully relaxed and thereafter placed at a 50° angle on a board for the i.t. instillation. A diode light source was positioned on the larynx to visualize the trachea and vocal cords. The tongue was gently pressed towards the lower jaw, and the trachea was intubated using a 24 gauge BD Insyte plastic catheter (#381212, Becton Dickinson, Denmark) with a shortened needle. To ensure that the tube was positioned correctly in the trachea an extremely sensitive pressure transducer was used to measure the respiration frequency as previously described [21]. Each mouse was instilled with 50 μl of suspension, immediately followed by 200 μl of air using a 250 μl SGE glass syringe (250F-LT-GT, Micro- Lab, Aarhus, Denmark). Bronchoalveolar lavage and isolation of organs At 24 h post-exposure, the mice were euthanized using a cocktail of Zoletil® (Tiletamine/Zolazepam), Fentanyl, and Rompun®. Their weight was recorded and blood collected by cardiac puncture in an eppendorf tube containing 36 μl K2EDTA. The plasma was collected by centrifugation. Immediately after blood collection, a lung lavage was performed by cannulation of the trachea using a 22 gauge needle equipped with polyurethane catheter. The lungs were flushed twice using sterile saline (2 x 0.8 ml/mouse). The total BALF volume recovery was estimated to be around 75%. The BALF was kept on ice until it was centrifuged (400 x g, 4°C for 10 min) and the supernatant stored at -80° until use. The pellet was resuspended in 100 μl of media (HAMS F12 (GIBCO #21765) with 10% fetal bovine serum (FBS)). The total BALF cell count of each mouse was determined using 20 μl of cell suspension in an NC-100 Nucleocounter (ChemoMetec A/S, DK). To prepare Cytospin slides 40 μl of cell suspension was centrifugated at 1000 rpm for 4 min. The BALF cells deposited on objective slides were stained using May-Grunwald and Giemsa co-staining. Differential cell count was performed counting 200 cells per slide by a person blinded to the exposure groups. After the BAL procedure, the lungs and liver tissue were dissected, snap frozen in liquid nitrogen and stored at -80°C. The heart and the whole aorta (from the arch to the iliac bifurcation) were dissected and transferred to a petri dish containing ice-cold PBS. The aortic arch and BCA was gently trimmed of fat- and connective tissue under an Olympus SZX7 stereomicroscope. The BCA was embedded in Tissue-Tek® O.C.T™ Compound (Sakura Finetek, Værløse, DK), frozen on dry ice and stored at -80°C. The heart and aorta were kept in phosphate buffer saline (PBS) at 4°C for a few hours before they were trimmed free of fat- and connective tissue. Glutathione quantification Total glutathione was measured by reducing oxidized glutathione dimers (GSSG) with the addition of 7 μl of 10 mM sodium dithionite to all samples and incubating at room temperature for 1 h. Reduced glutathione was quantified in lung tissue from ApoE-/- mice using an o-phthalaldehyde probe that reacts with the reduced form of glutathione, generating a fluorescence signal. Briefly, approximately 100 mg of tissue from the right lung was homogenized on ice utilizing an IKA ULTRA TURRAX® T25 (disperser S25N-10G) in a redox quenching lysis buffer containing 5% trichloroacetic acid. Homogenates were vortex briefly and processed according to a slightly modified version of the protocol adapted from Senft and colleagues [35]. Lysates were diluted 1:10 for analysis and reduced glutathione normalized to the tissue mass. RNA purification and gene expression Saber and colleagues have shown a robust increase in the gene expression of the acute phase reactant Saa3 in the lungs after i.t. instillation of nanosized CB [36]. In addition, Bourdon et al have shown that the expression of Saa3 in lungs is much higher and prolonged as compared to the expression of other acute phase proteins after a single i.t. instillation of CB [23]. Based on their findings we investigated the lung expression of Saa3 in the present study because it is the most differentially expressed acute phase protein after i.t. instillation of CB in mice. Mouse lung RNA was purified in an RNAse free environment using Maxwell® 16 LEV simplyRNA Tissue Kit (Promega, Madison, WI, USA). In brief, 16–22 mg of lung tissue in homogenization solution containing thioglycerol and stainless steel beads was homogenized on a Tissuelyser II (Qiagen, Hilden, Germany) 30 times/second for 60 seconds and stored on ice for a few min. The homogenates were transferred to Maxwell® 16 LEV cartridges (MCE) and purified on AS2000 Maxwell® 16 Instruments (Promega, Madison, WI, USA) according to manufacturer’s instructions. The RNA concentration and purity was measured on a Nanodrop 2000 UV-Vis Spectrophotometer (Thermo Scientific). All RNA samples with A260/280 ratio between 2.0–2.15 were used for cDNA synthesis. cDNA was synthesized from DNAse treated RNA using Taq-Man® Reverse transcriptase reagents (Applied Biosystems) according to manufacturer’s protocol. Saa3 gene expression was measured using real-time qPCR with 18S as reference gene as described previously [37]. In brief, we quantified the relative expression levels of Saa3 and 18S genes using commercial Taq-Man 2xPCR master mix (Applied Biosystems) on a Viia7 sequence detector (Applied Biosystems). cDNA from each lung sample was run in triplicates with target and reference gene in separate wells. Negative controls (i.e. no RNA had been converted to cDNA), a sample without RNA and cDNA, and a plate control were included in each run, the latter to control for day to day variation. The sequences of the Saa3 primers and probe were: Saa3 forward: 59 GCC TGG GCT GCT AAA GTC AT 39, Saa3 reverse: 59 TGC TCC ATG TCC CGT GAA C 39 and Saa3 probe: 59 FAM–TCT GAA CAG CCT CTC TGG CAT CGC T–TAMRA 39. The relative expression of the target gene was quantified using the comparative method 2-ΔCT [38]. Atherosclerotic plaque progression in the aorta The trimmed aorta was cut longitudinally from the arch to the iliac bifurcation, flattened and mounted between an objective glass and a cover slide, avoiding any overlapping tissue. Digital images of the intimal surface were obtained using an Olympus SZX7 stereomicroscope and Olympus Color View I camera. The level of atherosclerotic lesions (plaque percentage of the total surface area of the aorta) was quantified by a person blinded to the exposure groups using image processing software ImageJ. Atherosclerotic plaque progression in brachiocephalic arteries Triplicates of 5 μm thick frozen sections were obtained at 100 and 200 μm distal to the aortic arch on a CM3050 S-Cryostat (Leica Microsystems Nussloch GmbH, DE). The sections were mounted on Super Frost® Plus slides (Thermo Scientific) and stored at -80°C. The mounted sections were fixed in Bouin’s solution (Sigma) overnight, followed by staining with Masson's trichrome stain (Sigma). In brief, sections were stained in Weigert’s Iron Hematoxylin Solution (Ampliqon, DK) for 2 min; Biebrich Scarlet-Acid Fuchsin (HT15-1, Sigma) for 5 min; Phosphotungstic Acid/Phosphomolybdic Acid workings solution (HT152/HT153, Sigma) for 5 min, and Aniline Blue for 5 min. The sections were destained in 1% acetic acid for 2 min, dehydrated through graded ethanol to xylene, and mounted with Pertex for examination by light microscopy. Digital images were captured using an Olympus BX41 microscope and Olympus Color View I camera or using the Hamamatsu NDP slide scanner (Hamamatsu Nanozoomer 2.0HT). Image analysis was carried out using ImageJ or by analyzing the virtual slides by using the Hamamatsu NDP viewer. All analyses were performed by a person blinded to the exposure groups. Each data point was calculated as the mean of triplicates from 100 and 200 μm distal to the aortic arch, respectively (the combination of the mean of the triplicates from the 100 and 200 μm cross-sections was due to no difference being observed in the plaque area between the locations). Plaque progression in the BCA was calculated as the percentage of plaques covering the lumen cross-sectional area. The relative plaque area was evaluated by calculating the intima-media ratio. Classification of atherosclerotic plaques in the BCA was carried out using the American Heart Association guideline [39]. In brief, stages I-III is regarded as clinically silent lesions that are precursors to advanced lesions. Stage IV is advanced lesions called atheroma and has a core of accumulated extracellular lipid. Stage V lesions are advanced fibroatheroma lesions that have multiple lipid cores, fibrotic layers, and calcifications. Stage VI lesions are complicated lesion with surface rupture and hematoma-hemorrhage thrombus, these are not observed in ApoE-/- mice and, therefore, the score is not used in this study [39]. S3 Fig depicts examples of the stages of lesions in the present study. Effects of vasomotor active mediators in plasma The aorta of naïve female C57BL/6 Ntac mice (Taconic, Ejby, Denmark) aged 10–12 weeks was dissected and trimmed free of fat and connective tissue in oxygenated cold physiological saline solution (PSS) buffer after cervical dislocation, as described previously [40]. The thoracic section of the aorta was cut into ring-shaped segments of 2 mm. Two stainless steel wires, 40 μm in diameter, were gently led through the lumen of each ring segment and mounted in the organ bath of a Multi-Wire Myograph 620M (Danish Myo Technology, Aarhus, DK) interfaced to a PowerLab 4/35 recorder (ADInstruments). Each organ bath contained 5 ml of cold oxygenated PSS and was continuously perfused with a 95% O2 and 5% CO2 gas mixture. After mounting the segments, the heat was turned on, fresh 37°C warm PSS added and the segments were allowed to equilibrate. Next the passive length-tension relationships of each segment were determined using the DMT LabChart normalization procedure. After a successful normalization, the vessel viability was confirmed via non-receptor mediated contraction. In brief, 5 ml of K-PSS was added to the organ bath while the contraction was allowed to stabilize (reached a plateau). Thereafter, the vessels were stimulated three times with 37°C K-PSS. Only viable vessels were used and stimulated with plasma from ApoE-/- mice that had been exposed to vehicle, LPS or low-dose CB in order to investigate the vasomotor response. The aorta ring segments were stimulated with 0.5% plasma, and the presence or absence of vessel contraction recorded. At this juncture, an additional 0.5% of plasma was added to the vessel and contraction evaluated. As a final phase, 10 μM of acetylcholine (Ach) was added to assess the endothelium-dependent vasorelaxation (endothelium function). This concentration of Ach induces maximal relaxation of prostaglandin F2α pre-contracted aorta rings. In addition, we have previously shown that exposure to nanomaterials affects the maximal endothelium-dependent vasorelaxation rather than the EC50 value [10, 41–43]. To investigate the origin of the pharmacological modulators in the plasma we antagonized the serotonin receptor with ketanserin. This drug is a S2-serotoninergic antagonist, which also inhibits α1-adrenergic receptors [44]. 1) Aorta ring segments were stimulated with 1% plasma (from CB-exposed ApoE-/- mice), at the plateau of contraction the serotonin receptor antagonist Ketanserin (0.1 μM) was added and the force was recorded. 2) Aorta ring segments were pre-incubated 17 min with Ketanserin (0.1 μM), and 1% plasma (plasma from CB-exposed ApoE-/- mice) was added, and the force was recorded. In addition, the aorta ring segments were stimulated with Phenylephrine (10 μM) to demonstrate whether the artery had preserved the ability to receptor-mediated contraction via a receptor different from serotonin receptors. The serotonin investigations were reproduced in 3 different mice (n = 3). For data analysis, the response was ranked based upon contraction above 0.5 mN in the aorta ring segments. This difference was used for practical reasons because certain plasma samples produced little vasoconstriction, although the basal tonus increased slightly over the incubation period, whereas other samples produced a swift vasoconstriction. An increase in tonus of 0.5 mN was used as threshold for vasoconstriction because this difference is larger than the variation observed in the basal tonus of aorta rings. The threshold for a significant vasorelaxation response to 10 μM Ach was calculated as a decrease of more than 50% from the contraction following incubation with 0.5% plasma. The results of study 1 and 2 have been pooled in the statistical analysis to increase the power of the qualitative assessment. Endotoxin assay To assess whether the CB contained LPS or inhibited the LPS-mediated response, a Limulus amebocyte lysate (LAL) gel clot assay (N194-06, Lonza, Walkersville, MD) was utilized. In brief, a suspension of 5 ml (2.5 mg/ml) CB was sonicated in endotoxin-free water for 16 min and used at the same concentration as the i.t. instillations. LPS (Escherichia coli 055: B5, Sigma-Aldrich) was added to a final concentration of 4 μg/ml, equivalent to 10,000 Endotoxin Unit/ml. The CB working solutions (8.53 μg/ml and 25.6 μg/ml) were serially diluted (two-fold dilution) for analysis and a standard curve was generated according to the manufacturer's protocol. A solid gel clot formation was interpreted as LPS being present in the sample or as no interference from the CB in the sample. The lack of clot formation as was interpreted as no LPS being present in the sample or that CB inhibited the effect of LPS. Statistics The results of the LPS pilot study were analyzed by regression analysis fitted to either a four-parameter sigmoid dose-response (single-dose i.t. instillation) or linear (repeated i.t. instilations) curve. The BALF data of study 1 were analyzed by two-factor ANOVA (interaction analysis) followed by a Fisher’s least statistical difference (LSD) post-hoc test. BALF data for eosinophils were log-transformed to achieve homogeneity of variance (assessed by Bartlett’s test). For lymphocytes, neutrophils and epithelial cells, log-transformation did not produce homogeneity of variance and parametric test on ranks was used with Fisher’s LSD post-hoc test (fold-differences and 95% confidence interval (CI) have been calculated from data on the nominal scale). The BALF data in study 2 were analyzed by one-way ANOVA with Tukey’s post-hoc test. In study 1, data on plaque progression in the aorta and BCA and the intima-media ratio were analyzed by two-factor ANOVA (interaction analysis) with Fisher’s LSD post-hoc test. The data on plaque progression in the aorta and BCA and the intima/media ratio of study 2 were log-transformed to achieve homogeneity of variance and analyzed using one-way ANOVA with Tukey’s post-hoc test. The data for plasma effects on the aorta were analyzed using χ2-test (data from study 1 and 2 were pooled within their respective groups). The data on glutathione in study 1 was analyzed by two-factor ANOVA (interaction analysis) with Fisher’s LSD post hoc test and the data of study 2 using one-way ANOVA with Tukey’s post-hoc test. Data on Saa3 gene expression were log-transformation to achieve homogeneity of variance and parametric test on ranks was used with Fisher’s LSD post-hoc test. The statistical analyzes were carried out in STATA 13 program package (StataCorp LP, College Station, TX, USA), Statistica version 5.5 (StatSoft Inc., Tusla, OK, USA) and GraphPad Prism version 5.00 for Windows, (GraphPad Software, San Diego, CA USA,). All results are reported as the mean and standard error of the mean (SEM). Statistical significance was accepted at 5% level, and all P-values refer to post-hoc tests. Results Pulmonary inflammation after exposure to LPS in wild-type C57BL/6 mice Pulmonary exposure to a single dose of LPS significantly increased the number of neutrophils in BALF (Fig 1). There were unaltered numbers of macrophages, eosinophils, lymphocytes or epithelial cells at any dose (S1 Table). The data fitted reasonably well to a sigmoid dose-response curve (R2 = 0.72), with the statistically significant bottom (1.6x105 cells, 95% confidence interval: 0.9x105 to 2.2x105 cells) and top values (5.0x105 cells, 95% confidence interval: 4.1x105 to 5.9x105 cells). 10.1371/journal.pone.0160731.g001Fig 1 The influx of neutrophils in BALF after a single (A) or repeated (B) i.t. instillation of lipopolysaccharide (LPS) in wild-type C57BL/6 mice. Symbols represent the number of neutrophils in each mouse (mean values are reported in S1 Table). The regression lines represent sigmoid (A) and linear with 95% confidence interval (B) curve fit. The repeated exposure to LPS during four weeks was associated with increased influx of neutrophils in BALF at the low (0.8 μg/mouse: 0.9x105 cells, 95% confidence interval: 0.5x105 to 1.3x105 cells) and high dose (4.0 μg/mouse: 4.5x105 cells, 95% confidence interval: 2.7x105 to 6.3x105 cells). The exposure did not significantly affect the numbers of other leukocytes or epithelial cells at any dose (S1 Table). Pulmonary inflammation in ApoE-/- mice after repeated exposure to CB and/or LPS Study 1 The statistical analysis showed an interaction between the exposure to CB and LPS for total cells (P<0.01), lymphocytes (P<0.01) and neutrophils (P<0.001). The exposure to LPS was associated with increased number of total cells (1.9x105 cells, 95% CI: 0.9x105 to 3.0x105 cells), whereas there were slightly lower numbers of total cells in the LPS-exposed mice that also received high-dose CB (-0.5x105 cells, 95% CI: -1.5x105 to 0.6x105 cells) relative to the LPS-exposed mice (Table 1). There was a strong and significant increase in the number of neutrophils in the low-dose CB (0.8x105 cells, 95% CI: 0.1x105 to 1.4x105 cells) and LPS (2.5x105 cells, 95% CI: 1.9x105 to 3.1x105 cells) exposed mice compared to the vehicle group. Additionally, there was a lower number of neutrophils in the low-dose CB+LPS (-1.0x105 cells, 95% CI: -1.6x105 to -0.3x105 cells) and high-dose group CB+LPS (-0.6x105 cells, 95% CI: -1.3x105 to -0.02x105 cells) groups as compared to the LPS group. The number of lymphocytes was increased in the low-dose (0.5x105 cells, 95% CI: 0.3x105 to 0.7x105 cells) and high dose (0.2x105 cells, 95% CI: 0.01x105 to 0.3x105 cells) CB exposed animals as compared to the controls. Moreover, the group of LPS exposed mice had decreased the number of epithelial cells as compared to the groups not exposed to LPS (P<0.01). 10.1371/journal.pone.0160731.t001Table 1 BALF cell number and distribution at 24 h post-exposure after repeated i.t. instillations of carbon black (CB) and/or lipopolysaccharide (LPS) in ApoE-/- mice. Group Vehicle Low-dose CB High-dose CB LPS Low-dose CB+LPS High-dose CB+LPS Study 1 Total cells (x 103) 187.2 ± 32.1 378.9 ± 39.7*** 243.5 ± 37.7 405.0 ± 44.5*## 358.5 ± 39.0## 297.1 ± 30.2*## Neutrophils (x 103) 8.5 ± 5.2 85.2 ± 16.5*** 52.1 ± 11.5*** 261.0 ± 25.8## 164.5 ± 32.9*##$ 196.5 ± 26.1*##$ Macrophage (x 103) 98.1 ± 15.3 138.4 ± 10.2* 134.9 ± 13.4* 108.0 ± 13.2## 130.7 ± 14.7## 86.7 ± 12.9 Eosinophils (x 103) 60.1 ± 20.7 87.2 ± 24.1 26.3 ± 15.7 20.0 ± 8.9 32.7 ± 7.8 5.4 ± 3.0## Lymphocytes (x 103) 3.2 ± 1.1 53.5 ± 9.3*** 20.7 ± 7.3*** 10.0 ± 2.4 20.3 ± 7.2 4.0 ± 0.9 Epithelial (x 103) 17.1 ± 4.1 14.6 ± 3.8 9.4 ± 1.7 6.1 ± 2.7## 10.3 ± 2.1 4.5 ± 1.4## Study 2 Total cells (x 103) 103.3 ± 17.4 104.8 ± 16.7 ND 294.4 ± 27.3*** ND ND Neutrophils (x 103) 1.6 ± 0.6 11.6 ± 4.4*** ND 176.7 ± 23.3*** ND ND Macrophage (x 103) 72.3 ± 12.1 66.2 ± 7.1 ND 98.1 ± 8.8 ND ND Eosinophils (x 103) 14.7 ± 5.2 14.2 ± 7.9 ND 9.8 ± 3.3 ND ND Lymphocytes (x 103) 1.8 ± 0.9 2.6 ± 0.8 ND 2.1 ± 0.8 ND ND Epithelial (x 103) 13.0 ± 2.4 10.1 ± 2.1 ND 7.8 ± 2.1 ND ND * P<0.05 *** P<0.001 in CB exposed group compared to vehicle group. ## P<0.01 in LPS group compared to vehicle control group. $ P<0.05 in CB+LPS exposed group compared to LPS group. Data is presented as mean ± SEM. The BALF cells were not determined (ND) in all groups in study 2. Study 2 There was a significant increase in total cell number and neutrophil influx in BALF in the LPS exposed (P<0.0001) and low-dose CB exposed mice (P<0.001) (Table 1). There was no significant difference between the groups for macrophages, lymphocytes, eosinophils and epithelial cell. Increased expression of Saa3 in lung tissue of ApoE-/- mice after pulmonary exposure to CB and/or LPS Saa3 expression was measured as a marker of acute-phase response in lung tissue. It has previously been shown that nanoparticle-mediated neutrophil influx in BALF correlates with Saa3 expression in mouse lung [5, 36]. In the present study, we only measured Saa3 expression in study 1. The LPS exposure (P<0.01) and high-dose CB+LPS (P<0.001) caused a significant increase in the expression of Saa3 in lung tissue 24 h post-exposure (Fig 2). 10.1371/journal.pone.0160731.g002Fig 2 Normalized Saa3 mRNA levels in lung tissue from ApoE-/- mice after repeated i.t. instillations of carbon black (CB) and/or lipopolysaccharide (LPS). Fold increase in mRNA levels of Saa3 compared to the vehicle group are shown. Open circles and squares represent the individual mice. Minus (-) denotes no exposure, plus denotes low (+) or high-dose (++) exposure. Lines represent the mean in each group (N = 9–10 per group). Asterisk denotes ***P<0.001 compared to the LPS exposed group, and #P<0.05, ##P<0.01, ###P<0.001 compared to the vehicle control, two-factor ANOVA with Fisher’s LSD post-hoc test. Unaltered levels of glutathione status in lung tissue of ApoE-/- mice after pulmonary exposure to CB and/or LPS As a measure of oxidative stress, total and reduced glutathione were measured in lung homogenate 24 h after last exposure (Fig 3). There were no differences in total and reduced glutathione between the exposure groups and controls in both studies. In study 1 there was a trend of higher total glutathione levels in the low-dose CB exposure group, which was not observed in study 2. 10.1371/journal.pone.0160731.g003Fig 3 Glutathione status in lung tissue of ApoE-/- mice after repeated i.t. instillations of carbon black (CB) or lipopolysaccharide (LPS). Total and reduced glutathione was measured at 24 h post-exposure in lung homogenate from ApoE-/- mice of study 1 and 2. Data is presented as mean and SEM (n = 10 mice per group). Unaltered progression of atherosclerosis in the aorta of ApoE-/- mice after repeated exposure to CB and/or LPS We used en face without staining of lipids to evaluate atherosclerotic plaque progression in aortas of ApoE-/- mice (Fig 4A). In study 1 there was a statistically significant interaction between the high-dose CB+LPS group, and the LPS group (P<0.05), although there was no difference between the low-dose CB+LPS and LPS only group. However, there was 1.6-fold (95% CI: 0.9–2.7 fold) higher plaque percentage in mice only exposed to LPS compared to vehicle treated (P = 0.09, Post-hoc Fisher LSD test). For the particle only exposed mice, there were no significant differences in plaque progression between the vehicle, low-dose CB, and high-dose CB exposed mice. In study 2, there was no significant difference in plaque progression between vehicle, LPS and low-dose CB exposed mice. Pooled analysis of the results of study 1 and 2 showed no statistical significance (S2 Table). 10.1371/journal.pone.0160731.g004Fig 4 Progression of atherosclerotic plaques in the aorta and brachiocephalic artery from ApoE-/- mice exposed to carbon black (CB) and/or lipopolysaccharide (LPS) by i.t. instillation. A) Atherosclerotic plaque area is expressed as the percentage of the luminal surface of the aorta covered with plaques. Calculations were made on whole aorta preparations from ascending aorta to the iliac bifurcation. B) Atherosclerotic plaque area expressed as the percentage of the lumen occupied by plaques in BCA. Six sections of BCA per animal were analyzed; three sections at 100 μm and three sections at 200 μm after the branch from the aortic arch (n = 6–10 mice per group). C) The intima-media ratio in BCA is calculated by dividing the area of the intima with the area of the media layer (n = 6–10 mice per group). Black bars represent the groups that did not receive LPS and white bars the groups that did receive LPS. Minus (-) denotes no exposure, plus denotes low (+) or high-dose (++) exposure. Data are presented as mean and SEM. Asterisks denote statistical significance *P<0.05, using one-way ANOVA with Tukey’s post-hoc test, and **P<0.01, two-factor ANOVA with Fisher’s LSD post-hoc test. Unaltered progression of atherosclerosis in the BCA aorta of ApoE-/- mice after repeated exposure to CB and/or LPS There was no significant difference between the groups exposed to vehicle, low-dose CB, and high-dose CB or the groups exposed to LPS, low-dose CB+LPS, and high-dose CB+LPS (Fig 4B). In study 2 there was a statistically significant decrease in the plaque area in the BCA lumen of the LPS exposed (P<0.01), and low-dose CB exposed mice (P<0.01) as compared to the vehicle exposed mice (Fig 4B). Pooled analysis of the results of study 1 and 2 showed no statistical significance, whereas the plaque percentage was 2.8-fold (95% CI: 1.0–5.1) higher in study 2 as compared to study 1 (S2 Table). There was no difference in intima-media ratio between the groups exposed to vehicle, low-dose CB, and high-dose CB or the groups exposed to LPS, low-dose CB + LPS, and high-dose CB+LPS in study 1 (Fig 4C). In Study 2 there was a significant decrease in the intima-media ratio in the mice exposed to LPS (P<0.05) and low-dose CB (P<0.01) compared to the vehicle exposed mice (Fig 4C). Pooled analysis of the results of study 1 and 2 showed no statistical significance, whereas there was a 3.0-fold (95% CI: 1.4–6.2 fold) higher level in study 2 as compared to study 1 (S2 Table). The atherosclerotic lesion stage score was assessed by visual classification. S2 Fig shows plaques of present study which are considered representative of the AHA classification score [39]. Overall, there was no significant difference in the plaque morphology between the exposed groups and their respective controls (Table 2). 10.1371/journal.pone.0160731.t002Table 2 Plaque classification score in brachiocephalic arteries from ApoE-/- after repeated i.t. instillations of carbon black (CB) and/or lipopolysaccharide (LPS). Group Vehicle Low-dose CB High-dose CB LPS Low-dose CB+LPS High-dose CB+LPS Study 1 2.6± 0.6 1.8 ± 0.5 2.3 ± 0.4 2.5 ± 0.5 2.7 ± 0.5 2.2 ± 0.45 Study 2 2.2 ± 0.5 1.6 ± 0.3 ND 1.8 ± 0.3 ND ND The table presents the classification of atherosclerotic plaque in study 1 and 2 based on morphological characteristics according to the American Heart Association classification guideline [39]. Data in numeric numbers are presented as mean ± SEM calculated from a randomized and blinded scoring of same sections 3 times. The plaque score was not determined (ND) in all groups in study 2. Serum of CB-exposed ApoE-/-mice contains factors that cause vasoconstriction in aorta rings of unexposed wild-type mice We investigated the response induced by plasma from ApoE -/- exposed mice (vehicle, LPS, or CB) on vasomotor activity in aorta ring segments from naïve wild-type mice using wire myography. The addition of 0.5% plasma from low-dose CB exposed mice caused contraction in aorta ring segments (6 out of 19, P<0.05, χχ2-test), while no contraction was observed using 0.5% plasma from the vehicle (0 out of 19) or LPS (2 out of 17) exposed mice (Table 3). When the plasma concentration was increased to 1%, contraction was observed in all three groups. The ability of endothelial-dependent vasorelaxation was preserved in all three groups after vasoconstriction in response to plasma from exposed animals. 10.1371/journal.pone.0160731.t003Table 3 Vasoconstriction in response to vasoactive mediators in plasma from ApoE-/- mice. Group 0.5% plasma 1% plasma Vasorelaxation (ACh) Vehicle 0/19 9/19 8/9 LPS 2/17 4/17 4/4 Low-dose CB 6/19 * 10/19 8/10 The data represent the distribution of vasoconstriction and vasorelaxation in naïve aorta ring segments after adding plasma 0.5% and 1% from the vehicle, LPS, and low-dosed exposed ApoE-/- to the vessel ex vivo. A contractility force above 0.5 mN was regarded as a significant response and a force below 0.5 mN an insignificant response. A vasorelaxation response above 50% of the maximal plasma-mediated contraction after adding 10 μM ACh was considered a significant response. * P<0.05 using χ2- test. To further investigate the origin of the pharmacological modulators in the plasma, we investigated if the above-observed responses were serotonin-driven (Fig 5). The plasma mediated contraction was decreased after the addition of the serotonin receptor antagonist Ketanserin (0.1 μM). Additionally, a 15 min incubation of Ketanserin (0.1 μM) completely suppressed the plasma mediated contraction in aorta ring segments. 10.1371/journal.pone.0160731.g005Fig 5 Serotonin receptor antagonist inhibition of plasma-mediated vasomotor contraction in aorta rings. The figure shows two graphical illustrations of the vasomotor function in real-time recorded in Lab Chart 8 myograph module (ADInstruments & DMT). A) 1% plasma from CB-exposed ApoE-/- mice causes contraction of the naïve aorta ring segment. The addition the serotonin receptor antagonist Ketanserin (0.1 μM) on the plateau of the constriction response releases the plasma-mediated contraction. B) 17 min pre-incubation with Ketanserin (0.1 μM) inhibits the 1% plasma-mediated vessel contraction. Stimulation of the adrenergic receptors in the naïve aorta ring segments with Phenylephrine (10 μM), demonstrates that the artery has preserved the ability for receptor-mediated contraction via a receptor not specific for serotonin. The findings presented in the figure was reproduced in 3 different mice (n = 3) with the same result. The activity of LPS in the Limulus amebocyte lysate assay is inhibited by addition of CB The observation of a statistically significant interaction between the high-dose CB and LPS on pulmonary inflammation suggested that CB may inhibit the biological activity of LPS through a physical interaction (Table 4). Firstly, the gel clot formation in the LAL assay at 4.26 μg/ml of CB indicated the presence of LPS in the sample. This was also tested with the suspensions of CB used for the in vivo experiments. Three 2-fold dilutions of the low dose CB+LPS suspension promoted clotting in the LAL assay (corresponding to 1.07 μg/ml of CB and 125 EU/ml of LPS). However, clotting did not occur in the high dose CB+LPS, despite the substantially higher concentration of both CB and LPS. Therefore, this suggests a CB concentration-dependent inhibition of the effects of LPS. 10.1371/journal.pone.0160731.t004Table 4 Interaction between carbon black (CB) and lipopolysaccharide (LPS). CB dilution (μg/ml) Replicate 25.6 12.8 8.3 6.4 4.3 3.2 2.1 1.1 0.5 LPS-free water 1 + + + + + - - - - - 2 + + + + + + - - - - Low-dose CB+LPS dilution (μg/ml) / (EU/ml) Replicate 8.5 / 1000 4.3 / 500 2.1 / 250 1.1 / 125 0.5 / 62.5 LPS-free water 1 + + + + - - 2 + + + + - - High-dose CB+LPS dilution (μg/mL) / (EU/ml) Replicate 25.6 / 1000 12.8 / 500 6.4 / 250 3.2 / 125 1.6 / 62.5 LPS-free water 1 - - - - - - 2 - - - - - - Nanosized CB was sonicated in LPS free water and diluted in LPS free water to the concentration used for in vivo study exposure (the highest concentration of low-dose (8.5 μg/mouse), and high-dose (25.6 μg/mouse)). Each assay was performed in duplicates. “+” indicates a clot equals LPS in the sample, and “–”indicates no-clot that is equal to no LPS in the sample. Discussion In the present study, we investigated the in vivo pulmonary and cardiovascular effects following repeated exposure to nanosized CB and/or LPS. Our findings show that repeated pulmonary exposure to CB increases the total cell number in BALF with a modest yet significant influx of neutrophils. The data are in concordance with results from previous studies on repeated i.t. instillation of Printex 90 to Balb/c mice [45]. It has also been shown that multiple i.t. instillations of LPS increased the influx of neutrophils in BALF [46]. We observed a similar pattern following 10 weeks of pulmonary exposure to LPS, regardless of co-exposure to CB. The total BALF cell number in the LPS exposed groups was significantly increased and mainly driven by the strong neutrophil influx contributing to approximately 50% of the total cell number. Moreover, the macrophage numbers were significantly increased in the LPS- and low-dose CB+LPS group, but not the high-dose CB+LPS group following repeated exposures. We only found increased expression of Saa3 in the lungs of mice exposed to high-dose CB+LPS; this is different from observations in studies with a single i.t. instillation of nanosized CB, which increased Saa3 expression levels in the lungs of C57BL/6 mice at day 1, 3 and 28 post-exposure [36]. The differences in Saa3 expression levels witnessed here and previous studies might be due to variances in the acute phase response between single and repeated exposures (i.e. high levels of Saa3 after a single exposure). If inflammation persists, it may lead to a redox imbalance favoring a pro-oxidant milieu causing the depletion of antioxidant enzymes [47]. It is recognized that altered glutathione metabolism occurs in inflammatory lung diseases [48]. Nevertheless, the levels and balance of total and reduced glutathione in lung tissue of ApoE-/- mice at 24 h after last exposure were not affected, indicating that the pulmonary exposure did not cause oxidative stress in the lungs (or it was too late to investigate this end-point). It is our experience that it requires relatively high bolus doses of NMs to cause glutathione depletion in the lungs; e.g. i.t. instillation of 15 μg ZnO caused both cytotoxicity and depletion of glutathione in mouse lung [49]. However, it should also be noted that the lung tissue samples had a relatively high content of oxidized glutathione (approximately 30%), which may indicate spontaneous oxidation of the samples during storage at -80°C. The repeated exposure of CB and/or LPS may have been associated with subtle differences in the glutathione status, but not an adaptation in terms of increased glutathione levels in the lungs. The exposure to CB did not accelerate plaque progression in the aorta or the BCA. An earlier study on repeated exposure to CB showed that a much higher dose administered by i.t. instillation (1000 μg/mouse per week for ten weeks) was associated with increased plaque progression in LDL receptor knockout mice on a cholesterol-rich diet [9]. It is well documented that cholesterol-enriched diet per se accelerates plaque progression in both ApoE-/- and LDL receptor knockout mice [50]. However, it is not possible to disentangle the effect of high CB dose from diet-induced susceptibility to plaque progression. In an earlier study, we exposed 48–49 weeks old ApoE-/- mice to Printex 90 by i.t. instillation (0.5 mg/kg once a week for 2 weeks), but the study was terminated prematurely because of a high mortality rate after each round of exposure [10]. Pulmonary exposure to LPS in study 1 indicated a tendency towards accelerated plaque progression in the aorta; therefore, study 2 was carried out to increase the statistical power. However, the results of study 2 did not support the initial findings of study 1. Previous studies have indicated that LPS exposure by intraperitoneal injection was associated with increased plaque progression in the aorta [51, 52]. Surprisingly, we found that mice exposed to high-dose CB+LPS had lower plaque progression and less pronounced pulmonary inflammation as compared to the LPS exposed group. This led us to investigate if CB could inhibit the effect of LPS using the LAL assay. As shown in Table 4 the high dose CB was able to inhibit gel clot formation i.e. the LPS-mediated effect, whereas low-dose CB+LPS did not. Indeed, the LAL assay showed that 1 μg CB was enough to eliminate the effect of at least 39 endotoxin units. However, clotting indicated that LPS was present in CB in concentrations down to 4.3 μg/ml. Thus, it suggests that CB affects the action of LPS and its downstream signaling cascade. One possible explanation for this inhibition could be the adsorption of LPS by CB [53]. Endothelial dysfunction, including impaired vasomotor function, is a hallmark of early progression of atherosclerosis [54]. Vasomotor dysfunction also occurs by pulmonary exposure to particles in wild-type and atherosclerosis-prone animals as well as in humans with clinical manifestation of atherosclerosis [4]. Although this may be related to structural or molecular changes in the endothelium or intima, recent evidence also indicates that pulmonary toxicity can be conveyed systemically with circulating vasoactive molecules capable of modulating vasomotor function in arteries from naïve animals [18–20]. Thus, we hypothesized that vasomotor active molecules in the plasma from CB or LPS exposed ApoE-/- mice would alter the vasomotor function in aortas from naïve C57BL/6 mice. There was an increased vasoconstriction response in aorta rings that were incubated with 0.5% plasma from the low-dose CB exposed mice. Increasing the concentration to 1% caused a contraction in all three groups. This effect was attributed to increased level of serotonin in plasma of the exposed mice because incubation with ketanserin abolished the vasoconstriction response. The antihypertensive action of ketanserin in humans has been ascribed to its action on both the S2-serotoninergic and α1-adrenergic receptors [44]. However, it has been shown that the ex vivo contractile response of arteries to serum was abolished in the presence of ketanserin, whereas the α1-adrenergic receptor prazosin had no effect [55]. Extended ex vivo cultures of arteries in the presence of serum for 4 days also produced a progressive constriction and arterial wall remodeling, which was not released by treatment with ketanserin, indicating a different mechanism of constriction than soluble vasoconstrictors in long-term cultures [56]. Unfortunately, we did not have sufficient quantities of plasma to measure the serotonin concentration. The concentration of serotonin in plasma is lower than serum; e.g. studies that have assessed serotonin concentration in samples from the same individuals have shown 3–15 versus 72–137 ng/ml in plasma and serum, respectively [57, 58]. The higher serotonin concentration in serum could be related to release during the blood clot formation, which may cause degranulation of platelets. Indeed, it does seem that incubation with serum had a strong constriction response in other studies; e.g. 1% serum produced a constriction response in aorta rings that was comparable to that induced by KPSS [14]. Other observations from the same group showed that aorta rings from unexposed mice had a basal tonus of 8.8 mN and addition of 2.5% serum increased it with 5.3 mN [20]. In comparison, the basal tonus was approximately 5 mN in aorta segments and addition of 0.5% plasma from CB-exposed mice produced approximately 40% increased tonus (i.e. a net increase of 2 mN). It is possible that the elevated serotonin concentrations in the plasma originate from platelets as either a cause or consequence of prothrombotic propensity (direct systemic administration of CB in mice has demonstrated to increase prothrombotic activity) [59]. In addition, i.t. instillation of CB in rats has resulted in platelet hyperactivity [60], although other studies showed no effect on plasma levels of coagulation factors or infarct size after cardiac ischemic/reperfusion injury [61–63]. An increased plasma concentration of serotonin might be a contributing factor to hypertension in humans due to increased vasoconstriction. Exposure to particulate matter in air pollution is associated with hypertension [64]. Interestingly, a panel study with personal black carbon measurements (i.e. a proxy-measure of air pollution) showed association between exposure and rapid changes in carotid arterial stiffening [65]. Increased arterial stiffening is a predictor of cardiovascular disease mortality in patients with essential hypertension [66]. Moreover, it has also been shown that inhalation exposure to concentrated ambient air particulate matter in ApoE-/- mice augmented the vasoconstrictor response to serotonin in aorta rings [67]. Conclusion This study shows that 10 weeks of i.t. instillation with nanosized CB, LPS or a combination in ApoE-/- mice induced pulmonary inflammation that was mainly characterized by an influx of neutrophils. The lung antioxidant defense (glutathione) and pulmonary expression of Saa3 were unaffected by repeated exposures. We did not find evidence of accelerated progression of atherosclerosis in the aorta or BCA after exposure to CB or LPS. Nevertheless, plasma from ApoE-/- mice exposed to CB caused vasoconstriction when added to aorta rings from naïve wild-type mice, which appeared to be related to increased plasma levels of serotonin. Supporting Information S1 Fig Dynamic Light Scattering measurements number distribution of nanosized CB suspended in nanopure water. A) Low dose CB (170 μg/ml), average PDI = 0.18, average size = 44 nm. B) Low dose CB (170 μg/ml) spiked with LPS (2 μg/ml), average PDI = 0.35, average size = 1281 nm. C) High dose CB (512 μg/ml), average PDI = 0.25, average size = 38 nm. High dose CB (512 μg/ml) spiked with LPS (2 μg/ml), average PDI = 0.57, average size = 1718 nm. (DOCX) Click here for additional data file. S2 Fig Standardized mean difference (SMD) in total cells (top) or neutrophils (bottom) in BALF as a function of the dose of Printex 90 administered by intratracheal instillation in C57BL/6 mice (0.67, 2.6, 18, 54 or 162 μg/mouse). The animals were sacrificed at 24 h post-exposure. The SMD has been calculated using Review Manager (RevMan) version 5.0 (The Nordic Cochrane Centre. The Cochrane Collaboration. 2008). The SMD is the difference between the two groups divided by the pooled standard deviation. The SMD in the top and bottom graphs cannot be compared with nominal values (i.e. number of cells) because they represent different scales. The SMD for total cells is close to zero (i.e. no effect), whereas there is a slightly increased influx of neutrophils at the low doses. The dose of 18 μg/mouse shows increased total cells as compared to low doses and a similar level of neutrophils. The responses at doses 54 and 162 μg/mouse suggest a plateau for both total cells and neutrophils in BALF. (DOCX) Click here for additional data file. S3 Fig Representative sections of BCA from the ApoE-/- mice. The sections were stained with Masson’s trichrome stain. Classification of atherosclerotic lesion was based on guidelines from American Heart Association. Stages I-III are clinically silent lesions and precursors to advanced lesions. Stage IV is an advanced lesions called atheroma and have a core of accumulated extracellular lipid. Stage V represents advanced lesions called fibroatheroma lesions and has multiple lipid cores, fibrotic layers and calcifications (Stary et al. 1995). (DOCX) Click here for additional data file. S1 Table BALF cell number and distribution 24 h post-exposure to LPS. Upper table shows the BALF cell number and distribution 24 h after a single exposure to LPS. Lower table shows the BALF cell distribution post 24.h after last exposure (one i.t. instillations once a week for 4 weeks). Asterisk denote ***P<0.001. **P<0.01 and *P<0.05 cells influx in exposed group compared to vehicle group. Data are presented as mean ± SEM. Statistical analyses were performed using one-way ANOVA with Tukey’s post-hoc test. (DOCX) Click here for additional data file. S2 Table Atherosclerosis in the BCA of ApoE-/- mice at 24 h post-exposure. (DOCX) Click here for additional data file. The authors would like to thank Lise Kristine Vesterdal, Michael Guldbrandsen, Eva Terrida, Lourdes Pedersen, Elzbieta Christiansen, and Anne-Karin Asp and Camilla Skånstrøm Dall for their technical assistance. We would also like to thank senior scientist Rikke Kaae Kirk at Novo Nordisk A/S for helping us with the Masson’s trichrome stain protocol, and letting us using the Hamamatsu NDP slide scanner. Abbreviations AchAcetylcholine AHAAmerican Heart Association ApoEApolipoprotein E BALBronchoalveolar lavage BCABrachiocephalic artery CBcarbon black EDTAEthylenediaminetetraacetic acid GSSGGlutathione disulphide I.t.Intratracheal instillation K-PSSPotassium physiological saline solution LALLimulus amebocyte lysate LDLLow density lipoprotein LPSLipopolysaccharide LSDLeast statistical differences NMNanomaterial PSSPhysiological saline solution qPCRquantitative-PCR Saa3Serum Amyloid A 3 ==== Refs References 1 Lim SS , Vos T , Flaxman AD , Danaei G , Shibuya K , Adair-Rohani H , et al A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990–2010: a systematic analysis for the Global Burden of Disease Study 2010 . Lancet 2012 , 380 :2224 –2260 . 10.1016/S0140-6736(12)61766-8 23245609 2 Araujo JA , Nel AE . Particulate matter and atherosclerosis: role of particle size, composition and oxidative stress . Part Fibre Toxicol 2009 , 6 :24 10.1186/1743-8977-6-24 19761620 3 Brook RD , Rajagopalan S , Pope CA III, Brook JR , Bhatnagar A , Diez-Roux AV , et al Particulate matter air pollution and cardiovascular disease: An update to the scientific statement from the American Heart Association . Circulation 2010 , 121 :2331 –2378 . 10.1161/CIR.0b013e3181dbece1 20458016 4 Møller P , Mikkelsen L , Vesterdal LK , Folkmann JK , Forchhammer L , Roursgaard M , et al Hazard identification of particulate matter on vasomotor dysfunction and progression of atherosclerosis . Crit Rev Toxicol 2011 , 41 :339 –368 . 10.3109/10408444.2010.533152 21345153 5 Saber AT , Jacobsen NR , Jackson P , Poulsen SS , Kyjovska ZO , Halappanavar S , et al Particle-induced pulmonary acute phase response may be the causal link between particle inhalation and cardiovascular disease . Wiley Interdiscip Rev Nanomed Nanobiotechnol 2014 , 6 :517 –531 . 10.1002/wnan.1279 24920450 6 Kido T , Tamagawa E , Bai N , Suda K , Yang HH , Li Y , et al Particulate matter induces translocation of IL-6 from the lung to the systemic circulation . Am J Respir Cell Mol Biol 2011 , 44 :197 –204 . 10.1165/rcmb.2009-0427OC 20378751 7 Nurkiewicz TR , Porter DW , Barger M , Millecchia L , Rao KM , Marvar PJ , et al Systemic microvascular dysfunction and inflammation after pulmonary particulate matter exposure . Environ Health Perspect 2006 , 114 :412 –419 . 16507465 8 Brown DM , Wilson MR , MacNee W , Stone V , Donaldson K . Size-dependent proinflammatory effects of ultrafine polystyrene particles: a role for surface area and oxidative stress in the enhanced activity of ultrafines . Toxicol Appl Pharmacol 2001 , 175 :191 –199 . 11559017 9 Niwa Y , Hiura Y , Murayama T , Yokode M , Iwai N . Nano-sized carbon black exposure exacerbates atherosclerosis in LDL-receptor knockout mice . Circ J 2007 , 71 (7 ):1157 –1161 . 17587728 10 Vesterdal LK , Folkmann JK , Jacobsen NR , Sheykhzade M , Wallin H , Loft S , et al Pulmonary exposure to carbon black nanoparticles and vascular effects . Part Fibre Toxicol 2010 , 7 :33 10.1186/1743-8977-7-33 21054825 11 Kim JK , Kang MG , Cho HW , Han JH , Chung YH , Rim KT , et al Effect of Nano-sized Carbon Black Particles on Lung and Circulatory System by Inhalation Exposure in Rats . Saf Health Work 2011 , 2 :282 –289 . 10.5491/SHAW.2011.2.3.282 22953212 12 Courtois A , Andujar P , Ladeiro Y , Baudrimont I , Delannoy E , Leblais V , et al Impairment of NO-dependent relaxation in intralobar pulmonary arteries: comparison of urban particulate matter and manufactured nanoparticles . Environ Health Perspect 2008 , 116 :1294 –1299 . 10.1289/ehp.11021 18941568 13 Folkmann JK , Vesterdal LK , Sheykhzade M , Loft S , Møller P . Endothelial dysfunction in normal and prediabetic rats with metabolic syndrome exposed by oral gavage to carbon black nanoparticles . Toxicol Sci 2012 , 129 :98 –107 . 10.1093/toxsci/kfs180 22610611 14 Aragon MJ , Chrobak I , Brower J , Roldan L , Fredenburgh LE , McDonald JD , et al Inflammatory and Vasoactive Effects of Serum Following Inhalation of Varied Complex Mixtures . Cardiovasc Toxicol 2016 , 16 :163 –171 . 10.1007/s12012-015-9325-z 25900702 15 Vesterdal LK , Mikkelsen L , Folkmann JK , Sheykhzade M , Cao Y , Roursgaard M , et al Carbon black nanoparticles and vascular dysfunction in cultured endothelial cells and artery segments . Toxicol Lett 2012 , 214 :19 –26 . 10.1016/j.toxlet.2012.07.022 22885096 16 Kermanizadeh A , Balharry D , Wallin H , Loft S , Møller P . Nanomaterial translocation—the biokinetics, tissue accumulation, toxicity and fate of materials in secondary organs—a review . Crit Rev Toxicol 2015 , 45 :837 –872 . 10.3109/10408444.2015.1058747 26140391 17 Møller P , Christophersen DV , Jacobsen NR , Skovmand A , Gouveia AC , Andersen MH , et al Atherosclerosis and vasomotor dysfunction in arteries of animals after exposure to combustion-derived particulate matter or nanomaterials . Crit Rev Toxicol 2016 , 46 :437 –476 . 10.3109/10408444.2016.1149451 27028752 18 Aragon MJ , Chrobak I , Brower J , Roldan L , Fredenburgh LE , McDonald JD , et al Inflammatory and Vasoactive Effects of Serum Following Inhalation of Varied Complex Mixtures . Cardiovasc Toxicol 2016 , 163 –171 . 10.1007/s12012-015-9325-z 25900702 19 Paffett ML , Zychowski KE , Sheppard L , Robertson S , Weaver JM , Lucas SN , et al Ozone Inhalation Impairs Coronary Artery Dilation via Intracellular Oxidative Stress: Evidence for Serum-Borne Factors as Drivers of Systemic Toxicity . Toxicol Sci 2015 , 146 :244 –253 . 10.1093/toxsci/kfv093 25962394 20 Robertson S , Colombo ES , Lucas SN , Hall PR , Febbraio M , Paffett ML , et al CD36 mediates endothelial dysfunction downstream of circulating factors induced by O3 exposure . Toxicol Sci 2013 , 134 :304 –311 . 10.1093/toxsci/kft107 23650127 21 Jacobsen NR , Møller P , Jensen KA , Vogel U , Ladefoged O , Loft S , et al Lung inflammation and genotoxicity following pulmonary exposure to nanoparticles in ApoE-/- mice . Part Fibre Toxicol 2009 , 6 :2 10.1186/1743-8977-6-2 19138394 22 Gitlin JM , Loftin CD . Cyclooxygenase-2 inhibition increases lipopolysaccharide-induced atherosclerosis in mice . Cardiovasc Res 2009 , 81 :400 –407 . 10.1093/cvr/cvn286 18948273 23 Bourdon JA , Halappanavar S , Saber AT , Jacobsen NR , Williams A , Wallin H , et al Hepatic and pulmonary toxicogenomic profiles in mice intratracheally instilled with carbon black nanoparticles reveal pulmonary inflammation, acute phase response, and alterations in lipid homeostasis . Toxicol Sci 2012 , 127 :474 –484 . 10.1093/toxsci/kfs119 22461453 24 Bourdon JA , Saber AT , Halappanavar S , Jackson PA , Wu D , Hougaard KS , et al Carbon black nanoparticle intratracheal installation results in large and sustained changes in the expression of miR-135b in mouse lung . Environ Mol Mutagen 2012 , 53 :462 –468 . 10.1002/em.21706 22753103 25 Bourdon JA , Saber AT , Jacobsen NR , Jensen KA , Madsen AM , Lamson JS , et al Carbon black nanoparticle instillation induces sustained inflammation and genotoxicity in mouse lung and liver . Part Fibre Toxicol 2012 , 9 :5 10.1186/1743-8977-9-5 22300514 26 Hogsberg T , Jacobsen NR , Clausen PA , Serup J . Black tattoo inks induce reactive oxygen species production correlating with aggregation of pigment nanoparticles and product brand but not with the polycyclic aromatic hydrocarbon content . Exp Dermatol 2013 , 22 :464 –469 . 10.1111/exd.12178 23800057 27 Jacobsen NR , Pojana G , White P , Møller P , Cohn CA , Korsholm KS , et al Genotoxicity, cytotoxicity, and reactive oxygen species induced by single-walled carbon nanotubes and C(60) fullerenes in the FE1-Mutatrade markMouse lung epithelial cells . Environ Mol Mutagen 2008 , 49 :476 –487 . 10.1002/em.20406 18618583 28 Jacobsen NR , White PA , Gingerich J , Møller P , Saber AT , Douglas GR , et al Mutation spectrum in FE1-MUTA(TM) Mouse lung epithelial cells exposed to nanoparticulate carbon black . Environ Mol Mutagen 2011 , 52 :331 –337 . 10.1002/em.20629 20963790 29 Kyjovska ZO , Jacobsen NR , Saber AT , Bengtson S , Jackson P , Wallin H , et al U: DNA strand breaks, acute phase response and inflammation following pulmonary exposure by instillation to the diesel exhaust particle NIST1650b in mice . Mutagenesis 2015 , 30 :499 –507 . 10.1093/mutage/gev009 25771385 30 Møller P , Jacobsen NR , Folkmann JK , Danielsen PH , Mikkelsen L , Hemmingsen JG , et al Role of oxidative damage in toxicity of particulates . Free Radic Res 2010 , 44 :1 –46 . 10.3109/10715760903300691 19886744 31 Jacobsen NR , Saber AT , White P , Møller P , Pojana G , Vogel U , et al Increased mutant frequency by carbon black, but not quartz, in the lacZ and cII transgenes of muta mouse lung epithelial cells . Environ Mol Mutagen 2007 , 48 :451 –461 . 17584883 32 Saber AT , Jensen KA , Jacobsen NR , Birkedal R , Mikkelsen L , Møller P , et al Inflammatory and genotoxic effects of nanoparticles designed for inclusion in paints and lacquers . Nanotoxicology 2012 , 6 :453 –471 . 10.3109/17435390.2011.587900 21649461 33 Saber AT , Koponen IK , Jensen KA , Jacobsen NR , Mikkelsen L , Møller P , et al Inflammatory and genotoxic effects of sanding dust generated from nanoparticle-containing paints and lacquers . Nanotoxicology 2012 , 6 :776 –788 . 10.3109/17435390.2011.620745 21995293 34 Jackson P , Hougaard KS , Boisen AM , Jacobsen NR , Jensen KA , Møller P , et al Pulmonary exposure to carbon black by inhalation or instillation in pregnant mice: effects on liver DNA strand breaks in dams and offspring . Nanotoxicology 2012 , 6 :486 –500 . 10.3109/17435390.2011.587902 21649560 35 Senft AP , Dalton TP , Shertzer HG . Determining glutathione and glutathione disulfide using the fluorescence probe o-phthalaldehyde . Anal Biochem 2000 , 280 :80 –86 . 10805524 36 Saber AT , Lamson JS , Jacobsen NR , Ravn-Haren G , Hougaard KS , et al Particle-induced pulmonary acute phase response correlates with neutrophil influx linking inhaled particles and cardiovascular risk . PLoS One 2013 , 8 :e69020 10.1371/journal.pone.0069020 23894396 37 Saber AT , Halappanavar S , Folkmann JK , Bornholdt J , Boisen AM , Møller P , et al Lack of acute phase response in the livers of mice exposed to diesel exhaust particles or carbon black by inhalation . Part Fibre Toxicol 2009 , 6 :12 10.1186/1743-8977-6-12 19374780 38 Livak KJ , Schmittgen TD . Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method . Methods 2001 , 25 :402 –408 . 11846609 39 Stary HC , Chandler AB , Dinsmore RE , Fuster V , Glagov S , Insull W Jr, et al A definition of advanced types of atherosclerotic lesions and a histological classification of atherosclerosis. A report from the Committee on Vascular Lesions of the Council on Arteriosclerosis, American Heart Association . Circulation 1995 , 92 :1355 –1374 . 7648691 40 Hansen CS , Sheykhzade M , Møller P , Folkmann JK , Amtorp O , Jonassen T , et al Diesel exhaust particles induce endothelial dysfunction in apoE-/- mice . Toxicol Appl Pharmacol 2007 , 219 :24 –32 . 17234226 41 Mikkelsen L , Sheykhzade M , Jensen KA , Saber AT , Jacobsen NR , Vogel U , et al Modest effect on plaque progression and vasodilatory function in atherosclerosis-prone mice exposed to nanosized TiO(2) . Part Fibre Toxicol 2011 , 8 :32 10.1186/1743-8977-8-32 22074227 42 Vesterdal LK , Folkmann JK , Jacobsen NR , Sheykhzade M , Wallin H , Loft S , et al Modest vasomotor dysfunction induced by low doses of C60 fullerenes in apolipoprotein E knockout mice with different degree of atherosclerosis . Part Fibre Toxicol 2009 , 6 :5 10.1186/1743-8977-6-5 19243580 43 Vesterdal LK , Jantzen K , Sheykhzade M , Roursgaard M , Folkmann JK , Loft S , et al Pulmonary exposure to particles from diesel exhaust, urban dust or single-walled carbon nanotubes and oxidatively damaged DNA and vascular function in apoE(-/-) mice . Nanotoxicology 2014 , 8 :61 –71 . 10.3109/17435390.2012.750385 23148895 44 Vanhoutte P , Amery A , Birkenhager W , Breckenridge A , Buhler F , Distler A , et al Serotoninergic mechanisms in hypertension. Focus on the effects of ketanserin . Hypertension 1988 , 11 :111 –133 . 3277910 45 Shwe TT , Yamamoto S , Kakeyama M , Kobayashi T , Fujimaki H . Effect of intratracheal instillation of ultrafine carbon black on proinflammatory cytokine and chemokine release and mRNA expression in lung and lymph nodes of mice . Toxicol Appl Pharmacol 2005 , 209 :51 –61 . 16331831 46 Corbel M , Theret N , Caulet-Maugendre S , Germain N , Lagente V , Clement B , et al Repeated endotoxin exposure induces interstitial fibrosis associated with enhanced gelatinase (MMP-2 and MMP-9) activity . Inflamm Res 2001 , 50 :129 –135 . 11339500 47 Møller P , Christophersen DV , Jensen DM , Kermanizadeh A , Roursgaard M , Jacobsen NR , et al Role of oxidative stress in carbon nanotube-generated health effects . Arch Toxicol 2014 , 88 :1939 –1964 . 10.1007/s00204-014-1356-x 25212906 48 Lee KS , Kim SR , Park HS , Park SJ , Min KH , Lee KY , et al A novel thiol compound, N-acetylcysteine amide, attenuates allergic airway disease by regulating activation of NF-kappaB and hypoxia-inducible factor-1alpha . Exp Mol Med 2007 , 39 :756 –768 . 18160846 49 Jacobsen NR , Stoeger T , van den Brule S , Saber AT , Beyerle A , Vietti G , et al Acute and subacute pulmonary toxicity and mortality in mice after intratracheal instillation of ZnO nanoparticles in three laboratories . Food Chem Toxicol 2015 , 85 :84 –95 . 10.1016/j.fct.2015.08.008 26260750 50 Kennedy AJ , Ellacott KL , King VL , Hasty AH . Mouse models of the metabolic syndrome . Dis Model Mech 2010 , 3 :156 –166 . 10.1242/dmm.003467 20212084 51 Andoh Y , Ogura H , Satoh M , Shimano K , Okuno H , Fujii S , et al Natural killer T cells are required for lipopolysaccharide-mediated enhancement of atherosclerosis in apolipoprotein E-deficient mice . Immunobiology 2013 , 218 :561 –569 . 10.1016/j.imbio.2012.07.022 22954709 52 Ostos MA , Recalde D , Zakin MM , Scott-Algara D . Implication of natural killer T cells in atherosclerosis development during a LPS-induced chronic inflammation . FEBS letters 2002 , 519 :23 –29 . 12023012 53 Cai X , Ramalingam R , Wong HS , Cheng J , Ajuh P , Cheng SH , et al Characterization of carbon nanotube protein corona by using quantitative proteomics . Nanomedicine 2013 , 9 :583 –593 . 10.1016/j.nano.2012.09.004 23117048 54 Libby P , Ridker PM , Hansson GK . Progress and challenges in translating the biology of atherosclerosis . Nature 2011 , 473 :317 –325 . 10.1038/nature10146 21593864 55 De Mey JG , Uitendaal MP , Boonen HC , Vrijdag MJ , Daemen MJ , Struyker-Boudier HA . Acute and long-term effects of tissue culture on contractile reactivity in renal arteries of the rat . Circ Res 1989 , 65 :1125 –1135 . 2791222 56 Bakker EN , van Der Meulen ET , Spaan JA , VanBavel E . Organoid culture of cannulated rat resistance arteries: effect of serum factors on vasoactivity and remodeling . American Journal of Physiology Heart and Circulatory Physiology 2000 , 278 :H1233 –1240 . 10749719 57 Koch DD , Kissinger PT . Determination of serotonin in serum and plasma by liquid chromatography with precolumn sample enrichment and electrochemical detection . Analytical Chemistry 1980 , 52 :27 –29 . 7356172 58 Lee GS , Simpson C , Sun BH , Yao C , Foer D , Sullivan B , et al Measurement of plasma, serum, and platelet serotonin in individuals with high bone mass and mutations in LRP5 . Journal of Bone and Mineral Research: the official journal of the American Society for Bone and Mineral Research 2014 , 29 :976 –981 . 59 Holzer M , Bihari P , Praetner M , Uhl B , Reichel C , Fent J , et al Carbon-based nanomaterials accelerate arteriolar thrombus formation in the murine microcirculation independently of their shape . Journal of Applied Toxicology 2014 , 34 :1167 –1176 . 10.1002/jat.2996 24531921 60 Kim H , Oh SJ , Kwak HC , Kim JK , Lim CH , Yang JS , et al The impact of intratracheally instilled carbon black on the cardiovascular system of rats: elevation of blood homocysteine and hyperactivity of platelets . J Toxicol Environ Health A 2012 , 75 :1471 –1483 . 10.1080/15287394.2012.722519 23116452 61 Gilmour PS , Ziesenis A , Morrison ER , Vickers MA , Drost EM , Ford I ,. et al Pulmonary and systemic effects of short-term inhalation exposure to ultrafine carbon black particles . Toxicol Appl Pharmacol 2004 , 195 :35 –44 . 14962503 62 Harder V , Gilmour P , Lentner B , Karg E , Takenaka S , Ziesenis A , et al Cardiovascular responses in unrestrained WKY rats to inhaled ultrafine carbon particles . Inhal Toxicol 2005 , 17 :29 –42 . 15764481 63 Tong H , McGee JK , Saxena RK , Kodavanti UP , Devlin RB , Gilmour MI . Influence of acid functionalization on the cardiopulmonary toxicity of carbon nanotubes and carbon black particles in mice . Toxicol Appl Pharmacol 2009 , 239 :224 –232 . 10.1016/j.taap.2009.05.019 19481103 64 Fuks KB , Weinmayr G , Foraster M , Dratva J , Hampel R , Houthuijs D , et al Arterial blood pressure and long-term exposure to traffic-related air pollution: an analysis in the European Study of Cohorts for Air Pollution Effects (ESCAPE) . Environ Health Perspect 2014 , 122 :896 –905 . 10.1289/ehp.1307725 24835507 65 Provost EB , Louwies T , Cox B , Op't Roodt J , Solmi F , Dons E , et al Short-term fluctuations in personal black carbon exposure are associated with rapid changes in carotid arterial stiffening . Environment International 2016 , 88 :228 –234 . 10.1016/j.envint.2015.12.023 26773393 66 Laurent S , Boutouyrie P , Asmar R , Gautier I , Laloux B , Guize L , et al Aortic stiffness is an independent predictor of all-cause and cardiovascular mortality in hypertensive patients . Hypertension 2001 , 37 :1236 –1241 . 11358934 67 Sun Q , Wang A , Jin X , Natanzon A , Duquaine D , Brook RD , et al Long-term air pollution exposure and acceleration of atherosclerosis and vascular inflammation in an animal model . JAMA 2005 , 294 :3003 –3010 . 16414948

PMC005xxxxxx/PMC5003394.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 27571421PONE-D-16-0888210.1371/journal.pone.0161440Research ArticleComputer and Information SciencesCryptographyPhysical SciencesMathematicsCryptographyComputer and Information SciencesComputing MethodsCloud ComputingPhysical SciencesMathematicsAlgebraPolynomialsEngineering and TechnologyManagement EngineeringOutsourcingPhysical SciencesMathematicsApplied MathematicsAlgorithmsResearch and Analysis MethodsSimulation and ModelingAlgorithmsResearch and Analysis MethodsDatabase and Informatics MethodsDatabase SearchingComputer and Information SciencesComputer NetworksInternetSocial SciencesLinguisticsSemanticsComputational SemanticsPrivacy-Aware Relevant Data Access with Semantically Enriched Search Queries for Untrusted Cloud Storage Services Privacy-Aware Search for Cloud Storage ServicesPervez Zeeshan 1Ahmad Mahmood 2Khattak Asad Masood 3Lee Sungyoung 2Chung Tae Choong 4* 1 School of Engineering and Computing, University of the West of Scotland, Paisley, PA1 2BE, United Kingdom 2 Ubiquitous Computing Lab, Department of Computer Engineering, Kyung Hee University, Global Campus, 1 Seocheon-dong, Giheung-gu, Yongin-si, Gyeonggi-do 446-701, South Korea 3 College of Technological Innovation, Zayed University, Abu Dhabi, United Arab Emirates 4 Artificial Intelligent Lab, Department of Computer Engineering, Kyung Hee University, Global Campus, 1 Seocheon-dong, Giheung-gu, Yongin-si, Gyeonggi-do 446-701, South Korea Wang Yeng-Tseng Editor Kaohsiung Medical University, TAIWAN Competing Interests: The authors have declared that no competing interests exist. Conceived and designed the experiments: ZP AMK SL TC. Analyzed the data: MA AMK. Wrote the paper: ZP MA AMK. Conceptualized the research: ZP MA. Designed and implemented the privacy-aware relevant data access: ZP MA AMK. Contributed in reviewing and revising the manuscript: ZP SL TC. * E-mail: tcchung@khu.ac.kr2016 29 8 2016 11 8 e016144011 3 2016 7 8 2016 © 2016 Pervez et al2016Pervez et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Privacy-aware search of outsourced data ensures relevant data access in the untrusted domain of a public cloud service provider. Subscriber of a public cloud storage service can determine the presence or absence of a particular keyword by submitting search query in the form of a trapdoor. However, these trapdoor-based search queries are limited in functionality and cannot be used to identify secure outsourced data which contains semantically equivalent information. In addition, trapdoor-based methodologies are confined to pre-defined trapdoors and prevent subscribers from searching outsourced data with arbitrarily defined search criteria. To solve the problem of relevant data access, we have proposed an index-based privacy-aware search methodology that ensures semantic retrieval of data from an untrusted domain. This method ensures oblivious execution of a search query and leverages authorized subscribers to model conjunctive search queries without relying on predefined trapdoors. A security analysis of our proposed methodology shows that, in a conspired attack, unauthorized subscribers and untrusted cloud service providers cannot deduce any information that can lead to the potential loss of data privacy. A computational time analysis on commodity hardware demonstrates that our proposed methodology requires moderate computational resources to model a privacy-aware search query and for its oblivious evaluation on a cloud service provider. http://dx.doi.org/10.13039/501100002597Kyung Hee UniversityKHU-20130439Chung Tae Choong This work was supported by a grant from the Kyung Hee University in 2013 (KHU-20130439). Data AvailabilityData are available from https://github.com/EDSReseach/SemanticallyEnrichedSearch.Data Availability Data are available from https://github.com/EDSReseach/SemanticallyEnrichedSearch. ==== Body Introduction We are living through a post-PC era in which computing facilities are regarded as the fifth utility [1]. These facilities, which are primarily related to computational and storage services, are provisioned to subscribers on a pay-as-you-go basis. This new service provisioning model is known as cloud computing [2]. Advances in virtualization technologies and the availability of high-speed Internet have fostered this on-demand computing paradigm. It provides an abstraction of unlimited computational and storage facilities to its subscribers, enabling them to dynamically scale services or applications according to their specific requirements [3]. These on-demand and virtualized services are provisioned by a cloud service provider (CSP). The underlying cloud infrastructure (processing power, storage capacity, and networking facility) is owned, managed, and operated by a CSP. Subscribers do not need to take care of the cloud infrastructure, the assurance related to uninterrupted service provisioning is delineated in a service contract that is signed between the CSP and its subscribers. Cloud-based storage service is a generalization of cloud-enabled data sharing, archiving, collaboration, and synchronization services [4]. These services leverage their subscribers to store their data for much a longer duration without the concerns of data availability and accessibility from varied devices, i.e., desktop computers, laptops, and smartphones. As lucrative as it sounds, there are data privacy concerns when confidential and personal data are outsourced to cloud-based storage services owned and managed by a CSP [5], [6], [7], [8]. Since these services are provisioned beyond the federated domain of subscribers over which they do not have any control, the CSP is considered to be an untrusted entity [9]. The most obvious solution to ensure data confidentiality in untrusted domain is to encrypt personal and confidential data before it can be outsourced to a cloud-based storage service. Since these services are provisioned on a pay-as-you-go basis, each data access request is charged according to the amount of data transferred between the subscriber and the CSP. Thus, the capability of a subscriber to access relevant encrypted data is very important. It ensures data privacy and can also increase the utility of the cloud-based storage services. To access relevant data within the untrusted domain of a CSP, two main methodologies are employed, namely a search over encrypted data [10], [11], [12] and an index-based data search [13], [14], [15]. A search over encrypted data exploits the mathematical properties (trapdoors) of the cryptographic protocol to identify encrypted data that contain a particular keyword. These methodologies ensure the privacy of the outsourced data and the search query, preventing the CSP from deducing any information about the outsourced data that can lead to a potential loss of privacy. An index-based data search employs a different methodology than an encrypted data search. Instead of executing a search query over encrypted data, the search query is evaluated for the index (inverted index) associated with the outsourced data. Trusted entities can be employed to persist the index and evaluating the search query. In contrast to that, index can be stored in cloud storage in encrypted form along with the outsourced data and concealed search queries can be used to search the cloud. The aforementioned methodologies provide accessibility to relevant data and also ensure data privacy. However, these methodologies are fairly limited in their functionality and greatly affect the utility of cloud-based storage services. A search over encrypted data can only search for predefined keywords for which trapdoors are defined, and in the case of data sharing and collaborative services, these trapdoors are shared among subscribers. In an index-based data search, where the index is used by a trusted entity, the cloud storage is underutilized for only the outsourced data, whereas the search queries that are handled by a trusted entity can only retrieve the data that have an exact match between the search criterion and index entries. Similarly, when the index is outsourced to cloud storage, the CSP can learn the access patterns of subscribers and can deduce confidential information about the outsourced data and subscribers. For instance, if the outsourced data of a patient is searched by a medical doctor specializing in diabetes mellitus, this leads to a possibility that either the outsourced data contains information regarding diabetes mellitus or that the patient is suffering from diabetes mellitus. Considering the limitations of conventional methodologies to efficiently retrieve outsourced data taking data privacy into consideration, there is a need for a searching methodology that can achieve semantic data retrieval and for an oblivious data search. Semantic data retrieval will ensure that relevant data can be discovered even if there is no exact match between the outsourced data and the search criteria defined by a subscriber. This will greatly increase the efficacy of the searching methodology in data sharing and collaboration services where the exact contents of the outsourced data are not known to participating subscribers, and only abstract ideas/concepts are communicated between them. For example, the employees of an insurance company who are collaborating on a task to define premium rates for next year’s insurance policy, do not know the actual contents of the survey reports shared by their colleagues. However, they want to determine if there are any surveys on viral diseases in a certain vicinity. An oblivious search will lead to maximized utilization of the cloud infrastructure without relying on a trusted third party and will enable the CSP to evaluate the encrypted search queries. In this research, we propose an privacy-aware content discovery methodology that enables subscribers of a cloud storage service to locate relevant data contents without using actual keywords from the outsourced data. It is an index-based privacy-aware data searching methodology that does not rely on a trusted third party to evaluate the search query. It realizes privacy-aware content discovery, which ensures that only authorized subscribers are able to search the outsourced data. It also prevents the CSP and unauthorized subscribers from learning the presence or absence of any keywords and deducing information that can lead to a potential loss of privacy, encompassing the outsourced data and the subscribers’ personal information. With the proposed methodology of privacy-aware content discovery, we make the following contributions in the area of cloud-based storage services: Privacy-aware search for encrypted data by utilizing semantic information to identify similarities between search criteria and outsourced data. The search criteria defined by a subscriber need not be exactly the same as in the outsourced data. If there exists a semantic relation between the search criteria and outsourced data, the relevant data contents can be retrieved; Privacy-aware data search without the need to share trapdoor information, and authorized subscribers can define their own search criteria. Their ability to access relevant data is not restricted to the information communicated by the data owner who outsources the data to the cloud storage; Maximized utilization of cloud storage services by persisting encrypted index and evaluated encrypted search queries within the domain of untrusted CSP; and Index and search query expansion by using semantic technologies to realize an encrypted data search similar to data contents for searching over the Internet. The rest of this paper is organized as follows: Section 2 presents the related work. Section 3 defines the system design goals and the architectural and security model along with the assumptions. Section 4 is dedicated to the descriptive details of the proposed methodology. Section 5 discusses the implementation details, followed by evaluation of the results in Section 6. Section 7 presents the discussion on security, and Section 8 concludes the paper and discusses future directions. Related Work In this section, we present methodologies to search encrypted data within an untrusted domain. Throughout this section, we mainly focus on cryptosystems, which exploit the mathematical properties of underlying cryptographic primitives to search encrypted data (i.e., trapdoor functions), and enterprise products, which define their protocols to match encrypted search queries and data. We mainly discuss the effects of these conventional methodologies on the efficacy and utility of cloud-based storage services within the context of data sharing and collaboration services. Symmetric key cryptography (SKC) enables a search over encrypted data [10] by utilizing a trapdoor defined for a particular key only to identify a match between a search query (trapdoor function) and encrypted data. SKC has been used in various schemes for searching over encrypted data, in which trapdoors are used to identify a match between the index of encrypted keywords instead of encrypted data [13], [14], [15]. However, the basic principles of the trapdoor’s definition and the matching remain the same. A trapdoor-based search for public key cryptography (PKC) was proposed by Boneh et al. [11]. It leverages an untrusted server to search encrypted data using a public key, without the need to decipher concealed data. Schemes to search encrypted data that are based on SKC and PKC are limited in functionality because encrypted data can only be searched for keywords having corresponding trapdoors that are defined by the data owner who encrypts the data. Also, these trapdoors must be transmitted to authorized users, enabling them to access relevant data using search queries. Thus, methodologies relying on trapdoor-based cryptography assume guaranteed availability of the data owner or a trusted third party (TTP) to transmit a trapdoor to authorized users according to their access privileges. To search confidential personal healthcare records, Li et al. proposed the Authorized Private Keyword Search (APKS) [12]. APKS utilizes Hierarchical Predicate Encryption (HPE) to realize a search over encrypted data [16] and employs a TTP to distribute capabilities (trapdoors) to authorized users according to their access privileges. These capabilities are then submitted to the CSP to evaluate the search query. Wang et al. proposed a methodology to rank search results according to their relevance with the selection criteria (trapdoor) [17]. However, it only supports a single trapdoor-based search query, greatly reducing its efficacy in defining the complex selection criterion and lacking the realism to search a large amount of data. A Searchable Cryptographic Cloud Storage System (CS2) focusing on dynamic data updates also provides a search over encrypted data [18]. Instead of searching the entire encrypted data repository, CS2 utilized the inverted index. However, CS2 is limited to cloud-based storage services and is not applied for cloud-based data sharing and collaboration services. Recently, Wenhai Sun et al. presented a privacy-preserving multi-keyword text search (MTS) with similarity-based ranking [19]. MTS utilizes tree-based indexing with adaption methods for a multi-dimensional algorithm. It ensures the confidentiality of the search query and the index data structure. However, it assumes that the user searching the cloud storage always behaves honestly, whereas the cloud server is honest but curious. This assumption can greatly affect the practicality of MTS for cloud-based storage services, focusing on data sharing and collaboration, in which users can behave maliciously to determine the presence or absence of a particular keyword(s). Oblivious Term Matching (OTM) realizes an encrypted index search, where the index is computed over encrypted outsourced data. OTM obliviously evaluates encrypted search queries, where it does not consider relevant data access with consideration of semantic enrichment of the encrypted index or search queries [20]. The proposed methodology of semantic data search uses OTM to identify similarities between search criteria and outsourced data. To achieve efficient data retrieval over large data contents, enterprises rely on search products that are customized to their specific needs and requirements. The Google search appliance [21] and Windows enterprise search products [22] offer such search solutions. These products create a searchable centralized enterprise-wide index that is used within the enterprise’s data center or can be configured to use cloud repositories. The search queries are evaluated and the results are filtered according to the access privileges of a user. Since these products evaluate access privileges after the execution of a search query, they require search services to be hosted within the federated domain of an enterprise. Thus, these search services retrain the migration of an enterprise-to-cloud ecosystem as it has to engage its own dedicated computation and storage resources for customizable search services. The authors in [23] have shown that, by carefully modeling search queries, malicious users can deduce confidential information from the centralized index, even if their access privileges do not allow them to access encrypted data. The aforementioned methodologies for searching encrypted data focus on the confidentiality of the search query and the outsourced data. However, these methodologies do not consider the privacy of the query evaluation process that is employed to identify the relevant data contents. It can be exploited by a malicious CSP to deduce information that can lead to a potential loss of privacy. In cloud-based data sharing services, if multiple users are searching for a similar keyword, the CSP can effortlessly identify the importance of the outsourced data and can concentrate its malicious intents to deduce confidential information from the data. For instance, if the employees from the accounts and planning departments of an organization are searching data that has been outsourced to a folder called projected income statements, the CSP can determine the irregularity in access patterns and consequently affect the highly sensitive stock market, thereby disrupting the stock prices. Thus, a methodology that can obliviously search cloud-based repositories is of great importance, as it restrains the capability of a CSP to deduce or infer confidential information. Fig 1 highlights the important features of existing methodologies for encrypted data search i.e., availability requirement for involved entities, entity responsible for evaluating the search query, and capability of a user to define arbitrary search queries. Although these methodologies realize encrypted data search, however their functionality is limited to exact matching between the search query and encrypted data i.e., trapdoors and encrypted index. Also these methodologies restrain authorized users to define their own search queries. In the subsequent sections, relevant data access with semantically enriched search queries is presented. The proposed methodology realizes semantic search enabling authorized users to define their own conjunctive search queries without compromising privacy of the outsourced data and search queries as well. 10.1371/journal.pone.0161440.g001Fig 1 Features of conventional encrypted data search methodologies. Design Goals, System and Security Model, Main Idea, Assumptions and Notations System Design Goals A data search within a cloud storage service allows subscribers to locate the required data contents. However, when data is outsourced to an untrusted domain of a public cloud service provider in encrypted form, standard search queries do not work, as the search criteria cannot be mapped to encrypted data. These search queries can also reveal confidential information about the outsourced data and the data owner. The design goal of our proposed system is to allow the subscribers of a public cloud storage service to search encrypted data in a similar way as contents are discovered over the Internet. However, search queries should not reveal any information to the cloud service provider, which can lead to the potential loss of privacy, affecting the outsourced data and personal information. System Model To search encrypted data similar to content discovery works over the Internet, the public cloud storage service provider, repository owner, and content contributor are considered as the involved entities. For the sake of simplicity in the subsequent descriptive details, we refer to these entities as the cloud server, owner, and subscriber, respectively. The cloud server owns the cloud infrastructure (i.e., storage, computation, and network) and provisions its access on a subscription basis. The owner is a cloud storage subscriber who creates a shared repository that is accessible to other authorized subscribers. Subscribers contribute to the shared repository by outsourcing data contents. The owner and authorized subscribers search the cloud storage (shared repository) by submitting search queries to the cloud server. Search queries are obliviously evaluated by the cloud server, and the search results are provided to the respective entity according to its access privileges. Security Model We consider the cloud server to be an untrusted entity that can collude with unauthorized subscribers to compromise the privacy of the outsourced data. It can assist unauthorized subscribers to search the outsourced data. Since the search query is evaluated by the cloud server, its result can be exploited to deduce confidential information about the outsourced data. To ensure the confidentiality of the outsourced data, only encrypted data is outsourced to the cloud server. In addition, to prevent the cloud server from inferring confidential information about the outsourced data, the encrypted search query is obliviously evaluated. This restrains the cloud server and unauthorized subscribers from learning of the presence or absence of a particular word or concept in the outsourced data. Main Idea Suppose the Daily News is a nationwide newspaper that provides coverage of national and international events. Alice is a subeditor working for the Daily News. She oversees the department that focuses on financial corruption. At any particular point in time, she is working on multiple cases. She has assigned evidence collection and report compilation tasks to her subordinate journalists. To deal with the problem of content accessibility on her office and mobile devices, she has subscribed to a public cloud storage service that is provisioned by Eve. Since her subordinate journalists share confidential information with her, she does not want Eve to learn or deduce any information about the outsourced data. To ensure the privacy of the data, each journalist outsources encrypted data to the repository shared by Alice. Bob and Mallory are Daily News journalists who work with Alice. Bob is an expert at retrieving information from online resources. Mallory’s expertise is in finding the ground truth by contacting the concerned authorities. Both are directed to submit their findings on a financial scam that was recently exposed by the Fraud and Financial Crime Division of the State. Alice has provisioned access to both Bob and Mallory to a cloud-based shared repository. Bob retrieves all of the related information from online resources, whereas Mallory compiles her report using the information she has collected from the appropriate authorities. Before outsourcing their findings to a shared repository, they index the information. The index is then further enhanced by augmenting it with missing relevant information. After that, the findings and the enhanced index are encrypted and outsourced to the shared repository. Whenever Alice needs to search for a file containing particular information, she defines a search criterion. The search criterion is then enriched by adding missing relevant information. The augmented search criterion is then encrypted with the secret key, and after that, Alice models an oblivious search query using the encrypted search criterion. The oblivious query is then submitted to the cloud server, which replies with the response. Alice processes the cloud server response and determines the presence / absence of keywords that were defined in the search criterion. From the processes of query formation, evaluation, and post-processing of the result, the cloud server learns nothing about the outsourced data or the search query; the search evaluation is oblivious to the cloud server. If an unauthorized subscriber tries to search the repository, the proposed system generates a randomized response. Fig 2 illustrates the conceptual model of our proposed system for searching encrypted data in an untrusted domain. 10.1371/journal.pone.0161440.g002Fig 2 Semantic search over encrypted data—conceptual model. Assumptions and Notations The proposed system focuses on semantic search for encrypted data. We assume that the owner has shared a symmetric encryption key with the authorized subscribers. The data that is outsourced to a shared repository is always encrypted with that key. In the subsequent descriptive details of the proposed system, the specifics of sharing data within an untrusted domain are intentionally neglected for the sake of simplicity. Readers can refer to [24] and [25] for descriptions of efficient and secure data sharing within public cloud storage services. The proposed system address the problem of privacy-aware relevant data access in untrusted cloud storage services. Ensuring data integrity and correctness is beyond the scope of research undertaken in this work; interested reader can refer to [26] for more details on public auditing. Table 1 presents the notations used in the descriptive details of our proposed system to semantically search the encrypted data. F represents a file that is outsourced to cloud storage. I stands for an index computed over F that contains the keywords and their respective frequencies i.e., I={〈kw0,f0〉…〈kwn,fn〉}, where n is the size of the index. Is represents a semantic index that is generated by identifying synonyms and the root word for each kw0…n:kwi∈I i.e., Is={〈kw0,syn00…ν,rw0,f0〉…〈kwn,synn0…ν,rwn,fn〉}, where syni0…ν is the list of synonyms of kwi, and rwi is its root word. H is an encoding function that is publicly known and encodes variable sized keywords into integer values of fixed length. EH and DH are homomorphic encryption algorithms. σpk and σsk are public and secret keys respectively, that are used by homomorphic encryption algorithms. These algorithms enable the processing of encrypted values (search query and encrypted index) without the need to decrypt them. ES and DS are symmetric encryption algorithms with a secret key k. F and Is are encrypted with symmetric encryption algorithms before they can be outsourced to a cloud server. EA and DA are asymmetric encryption algorithms associated with kpub and kpri public and private keys, respectively. α0…n represents a list of polynomial coefficients that are used to formulate an oblivious search query. Δy0…n is a list of oblivious values that are obtained as a result of the oblivious search query execution by the cloud server. 10.1371/journal.pone.0161440.t001Table 1 Notations used in the descriptive detail of semantically enriched encrypted data search. Notation Description F File outsourced to a shared repository. I={〈kw0,f0〉…〈kwn,fn〉} Index file that contains n keywords. Is= {〈kw0, syn00…ν, rw0, f0〉…〈kwn, synn0…ν, rwn, fn〉} Semantic index—an enriched form of I. kw is a keyword from I, syn0…ν is a list of its synonyms and rw is its root/parent word. H Publicly known encoding function that transforms an arbitrary-sized string to an integer value of q modulo, where q is a large prime. EH, DH Homomorphic encryption and decryption algorithms. σpk, σsk Public and secret key pair for homomorphic encryption algorithms. EA, DA Asymmetric encryption and decryption algorithms. kpub, kpri Public and private key pair for asymmetric encryption algorithms. ES, DS Symmetric encryption and decryption algorithms. k Secret key of symmetric encryption algorithms. It is shared with authorized users only. α0…n List of coefficients of a polynomial P which defines a search query. Δy0…n List of oblivious values generated as a result of query execution by the cloud server. Proposed System The proposed methodology of encrypted data search based on semantically enriched index and search queries is presented in this section. It is divided into five cohesive steps: indexing, data outsourcing, query formulation, query execution, and post-processing of results. Indexing A semantic search over encrypted data is achieved by evaluating search queries for an enriched inverted index (Is) associated with the outsourced data (F). Since, we want subscribers to search outsourced data using search queries that are semantically equivalent, the inverted index (I) is augmented with extra information. This extra information enables us to identify outsourced data that contains relevant information instead of finding an exact match between the search query and the keywords extracted from the outsourced data. To achieve this, the indexing is further divided into two phases. In the first phase, for each F that needs to be stored in cloud storage, I is generated. It contains all of the keywords (kw0…kwn) that appear in F, along with their respective frequencies i.e, I={〈kw0,f0〉,…〈kwn,fn〉}. After that I is further processed to augment it with semantic information. For that, kw0…n:kwi∈I are searched in a lexical database. This enables us to identify synonyms (syn0…ν) of kwi that do not exist in I but where syni0…ν and kwi semantically equivalent. Further root word (rwi) of each kwi is also extracted from the lexical database. rw assists us in finding the keywords that share the same root word, consequently identifying the relevancy between the search query and the encrypted outsourced data. Once syn0…ν and rw are identified, I is augmented with this extra information and is transformed into semantic index i.e., ⊎(I,syn0…ν,rw)→Is; where ⊎(⋅) is a function that appends syn0…ν and rw to I removing any duplicate values, where Is={〈kw0,syn00…ν,rw0,f0〉…〈kwn,synn0…ν,rwn,fn〉}. Data Outsourcing To ensure that the cloud server cannot exploit Is by deducing confidential information about the outsourced data. Is is concealed using a symmetric encryption algorithm before it can be outsourced to a cloud server. The secret key (k) for the symmetric encryption algorithm is shared among all of the authorized subscribers by the owner who having ownership rights over the shared cloud based repository. The scope of this paper is limited to encrypted data search, readers may refer to [27] for more details secret key sharing and user revocation in untrusted domain. In order to ensure that the search query can be obliviously evaluated by the cloud server, the owner encodes each keyword (kwi∈Is) using a publiclly known encoding function H(Is)→I^s. I^s is then encrypted using a symmetric encryption algorithm ES(I^s,k)→I^sk. Once the confidentiality of Is is ensured, I^sk along with Fk are outsourced to the cloud server. Since, k is only shared among the authorized subscribers and the owner, unauthorized subscribers cannot deduce any information about the outsourced data, even if they conspire with the cloud server. Fig 3 illustrates the entire process of securing inverted index with symmetric encryption. 10.1371/journal.pone.0161440.g003Fig 3 Encoding semantically enriched index and securing its confidentiality through symmetric encryption. Query Formulation To search encrypted data, a subscriber defines a search criterion (Ckw0…j), which consists of a set of keywords (kw0…j) that are used to search the relevant encrypted outsourced data. Since, we want to realize a semantic search over encrypted data, the search criteria defined by the subscriber is further enriched by identifying synonyms of kw0…j:kwi∈Ckw0…j. Once the relevant keywords (syn0…ν) are identified Ckw0…j is enriched by adding syn0…ν to Ckw0…j i.e., ⊎(Ckw0…j,syn0…ν)→Ckw0…l, where j < l, and ⊎(⋅) is a function that appends syn0…ν to Ckw0…j. Since, search queries are evaluated by the cloud server, there is a need to conceal Ckw0…l using an appropriate symmetric encryption algorithm. To prevent the cloud server from deducing any information about the encrypted outsourced data, the owner encodes Ckw0…l using a publicly known encoding function. For example, H(Ckw0…l)→C^kw0…j—H(·) must be the same encoding function as that used in the data outsourcing; otherwise, an oblivious search query cannot be successfully evaluated. After that C^kw0…l is encrypted with the symmetric encryption i.e., ES(C^kw0…l,k)→C^kw0…lk, where k is a shared symmetric encryption key, which is the same as that used in the data outsourcing to conceal I^kw0…n. To this stage, C^kw0…l has been concealed, however in order to realize an oblivious query evaluation there is a need to further process C^kw0…lk. A polynomial (P(x)) is defined such that the concealed kw0…n:kwi∈C^kw0…lk are the root of P(x) i.e., P(x∈C^kw0…lk)=∑i=0lαixi=0, where α0…l are the coefficients of P(x). Once the polynomial P(x) has been defined, a homomorphic encryption key pair (σpk, σsk) is initialized. Homomorphic encryption enables the cloud server to process the encrypted search query and also restrains its ability to learn the result of the query evaluation. After that, α0…l are encrypted i.e., EH(α0…l,σsk)→α0…lσsk, α0…lσsk along with σpk are transfered to the cloud server. α0…lσsk are used as the encrypted search query whereas σpk enables the evaluation of an encrypted search query without the need to decipher I^sk and α0…lσsk. Fig 4 describes the entire process of query formation. 10.1371/journal.pone.0161440.g004Fig 4 Encoding semantically enriched search criteria and modeling search query for oblivious computation. Query Execution To semantically identify the encrypted data, the search query is obliviously executed by the cloud server. α0…lσsk, which is submitted by a subscriber, is evaluated for I^sk. By using σpk, for kw0…n:kwi∈I^sk, the cloud sever computes the oblivious value i.e., Δy0…n=r.P(yi∈I^sk), where yi=kwi∈I^sk and r is a random number. The computation of the oblivious value ensures that the owner can identify the match between α0…lσsk and I^sk. Fig 4 describes the oblivious query execution process. Since, we are employing homomorphic encryption, the cloud server cannot learn whether kwi∈I^sk is a root of α0…lσsk. Thus, it cannot identify a match between the encrypted search criterion and the encrypted index. Once Δy0…n = r.P(y0…n) are computed, the cloud server transfers the result of the search query evaluation to the subscriber. Post-processing of results The oblivious values that the subscriber receives from the cloud server can only be deciphered using the valid homormophic key, which is the secret key, σsk. This ensures that the cloud server cannot collude with malicious subscribers to exploit the oblivious query evaluation process. On receiving the cloud server’s response, the subscriber deciphers Δy0…n i.e., DH(Δ0…n,σsk)=ψ0…n, where ψi can be a zero or non-zero randomized value. Since, the search query α0…lσsk submitted by the subscriber is constituted of root values from C^kw0…lk the decryption of ψi turns out to be zero for all those yi=kwi∈I^sk that are equal to the root value of P(yi), i.e., kwi∈C^kw0…lk∧kwj∈I^kw0…nk where kwj = kwi. For all other values where kwj ≠ kwi, ψi would turn out to be a random value (see Eq 1). P(y)=∑i=0lαiyi{=0if y is root of P(y).≠0a random value r for all other index entries.(1) Thus, only by deciphering Δy0…n with valid σsk owner can learn the result of encrypted search query. However, for the cloud server the evaluation of the search query will remain oblivious. Implementation The proposed methodology for a semantically enabled search of encrypted data is realized using jdk 1.7. We implemented a Java based desktop application and web service. The desktop application performs keyword extraction, indexing and search query augmentation, and post-processing of the result, whereas the web service is solely responsible for the oblivious evaluation of the encrypted search queries. Fig 5 illustrates the core functionalities of desktop application (data owner and authorized users) and web service. 10.1371/journal.pone.0161440.g005Fig 5 Core functionalities—desktop application and web service. In desktop application for data owner we generates an inverted index from the plain text, i.e., the data that needs to be outsourced to cloud storage. For this, we employ Apache Lucene API [28], which is a fully-featured text search engine that is focused on high performance. Apache Lucene enables us to extract all keywords, avoiding indexing of the stop word and repeated keywords. Once the keywords are extracted from the plain text in the form of the inverted index, we augment them with semantic information, i.e., synonyms and root words, using WordNet [29]. To use the augmented keywords to search the encrypted data, a hash of the individual keywords is computed using the SHA-512 hashing algorithm. The hashed keywords are then encoded into BigInteger values of arbitrary size. Once encoded, the inverted index entries are encrypted with the symmetric encryption algorithm and are outsourced to the web service. User desktop application authorized users model their search query in the form of a polynomial and learn the semantic map between their search criteria and the outsourced encrypted index. The search criteria defined by a user is augmented and encoded in a similar way, as discussed for the inverted index. To evaluate the encrypted search query, we utilize the Pascal Paillier cryptosystem [30]. The secret key of Pascal Paillier is used to conceal the search criteria, whereas a public key is used by the cloud server to evaluate the search query. For each encrypted keyword in the encrypted index, the search is evaluated and the result is transmitted back to the user. Evaluation The proposed methodology for the encrypted data search was evaluated on a 2.60 GHz Windows 7 PC with 2.0 GB of main memory. We opted for a relatively low-end machine to demonstrate that the proposed methodology can be realized for any public cloud-based storage service, since it does not have any special computational requirements. In the subsequent section, we first present the computational complexity of the semantic search for encrypted data. We then discuss the computational analysis of augmenting the inverted index and the search criteria with the semantic information. In the last section, we show the computational load of the oblivious query evaluation, where the search query is composed of multiple search criteria. Complexity Analysis The computational complexity and the amount of data transmitted between the entities are analyzed in order to illustrate the efficacy of our proposed encrypted data search. Both of these parameters are directly proportional to the size of the encrypted index outsourced to the cloud storage and the size of the encrypted search query. Table 2 shows the set of operations performed in each step of our proposed methodology, along with the input size and the amount of transmitted data. For the sake of simplicity, we regarded the cryptographic and hashing operations as constant time operations, because the proposed methodology is not confined to any particular encryption or hashing algorithm. 10.1371/journal.pone.0161440.t002Table 2 Complexity analysis of semantic search for encrypted data. Steps Operations Input Size Computational Complexity Transmitted Values Indexing Public encoding & Symmetric encryption N O(N) – Data outsourcing – N O(N) N Query formulation Asymmetric encryption & Polynomial modeling n O(n3) n + 2 Query execution Polynomial evaluation (n + 1)N O(n2.N) N Post-processing of results Asymmetric decryption N depends on n depends on n Indexing: To extract keywords and to identify the semantically equivalent words for each extracted keyword, we utilize freely available libraries, such as Apache Lucene and Wordnet. Since these libraries complement our proposed system, we consider their execution at a constant time. Thus, the computational complexity of the index is O(N), where N is the size of the augmented index containing both the synonym and the root word. Data outsourcing: We regard the computational complexity of the data outsourcing to be O(N), where N is the size of the augmented index. In total, N values are transmitted to the cloud server. Query formulation: The query formation is comprised of two steps. In the first step, the user defines the search criteria, which is then expanded with semantic information and finally encoded into a fixed length integer value using a hashing algorithm. In the second step, encoded values are used to model a polynomial, which is then concealed using the Pascal Paillier homomorphic encryption algorithm. Since the retrieval of the synonyms and root word and the encoding of the expanded search criteria are regarded as constant time operations, the computational complexity of the first step is O(n), where n is the size of the expanded search criteria. For the second step, the first individual encoded search criterion is modeled as a polynomial (where the search criterion is a root of the polynomial), the individual polynomials are multiplied together, and the coefficients of the resultant polynomial are concealed with the private key of the homomorphic encryption algorithm. Since the encrypted search query is modeled in three steps, its computational complexity is O(n3), where n is the size of the encoded search criteria. In total, n + 2 values are transmitted to the cloud server, where n + 1 is the number of coefficients, and there is one public key of the homomorphic encryption algorithm. Query execution: The encrypted search query is evaluated for each encrypted entry in the index outsourced to a cloud server. The computation complexity of the query execution depends on two factors: the size of the index, N, and the size of the polynomial that models the search query, n + 1. Thus, the computational complexity of an oblivious search query evaluation in terms of the Big-O notation can be expressed as O(n2 N). The size of search query results is also directly proportional to the size of the index. In total, N values are transmitted to the user as a result of the search query execution. Post-processing of results: Post-processing of the results is relatively a simple process, and it only deciphers the number of oblivious values, N, sent by the cloud server. Since we consider the computational load of the cryptographic operations as a constant, the computational complexity of the post-processing of the result can be regarded as O(N). The computational time required to post-process an individual oblivious record depends on size of search query i.e., number of keywords used to model conjunctive search query. Computational Analysis The computational time required to enrich the inverted index and the search query with semantic information is presented along with the amount of time required to model an encrypted search query. We studied the computational time of conjunctive search queries and presented the time required to evaluate those encrypted search queries over the enriched inverted index. Table 3 shows the average computational time of the aforementioned steps computed over 100 iterations. To measure computational time we used Java time logging mechanism. System time to the precision of nanoseconds was logged at the beginning and end of the process, difference between logged timestamps was regarded as the time required to completely execute the process. 10.1371/journal.pone.0161440.t003Table 3 Computational time analysis of semantic search for encrypted data. Query Size (No. of keywords) Query formulation (ms) Query execution (ms) 2 238 245 4 411 791 6 590 1169 8 778 2811 10 982 4230 12 1187 6018 14 1405 8796 Synonym identification: The index and search query expansion are two important steps which enable a semantic search over encrypted data. For the computational analysis, we evaluated Wordnet API over a batch of 50 words. These words can be regarded as keyword entries in the inverted index and search criteria defined by authorized subscribers. For a batch, the total number of synonyms and the execution time are noted first, and we extracted 872 synonyms in 408 ms. This total number of synonyms is then divided by 50 to calculate the average number of synonyms per word, which is approximately 18 synonyms per word. Finally, the time required to extract these average synonyms per word is calculated as the total execution time divided by the total time multiplied by the average number of synonyms per word, i.e., (408/872) * 18 = 8.42 ms, which represents the average execution time per word. This exercise is repeated over 10 batches of different word sets for a more realistic time calculation. For the evaluation, we selected the standard implementation of WordNet and did not consider an optimization strategy. Query formulation: The query formulation is comprised of the polynomial modeling and the asymmetric encryption of polynomial coefficients. As discussed in the complexity analysis, the computational cost of the query formulation depends upon the size of the search criteria that constitutes the encrypted search query. Unlike the conventional methodology, the proposed method supports a conjunctive search query, allowing authorized subscribers to set multiple search filters instead of relying on a single search criterion. Table 3 shows the average computational cost to model an encrypted search query with multiple search criteria. Query execution: For each index entry (encrypted keyword), the cloud server evaluates the search query. The query evaluation is merely a process of polynomial evaluation at a certain value, and that value happens to be an individual keyword in the outsourced enhanced index. The computational time of the query execution depends on the size of the enhanced index and the encrypted search query. The entire process of query execution utilizes the homomorphic property of the Pascal Paillier cryptosystem. The result of the query execution is oblivious to the cloud server. The computational time of the encrypted search query comprises a range of two to ten search criteria, as shown in Table 3, which shows how the increase in the number of search criteria affects the computational time required to obliviously execute a search query. Security Analysis In this section, we present the security analysis of our proposed methodology. Particularly, we focus on the capabilities of malicious entities to learn the encrypted search query and to deduce confidential information about the encrypted outsourced data. We examine the advantage of an untrusted cloud service provider to learn the result of the search query evaluation and to deduce information that can lead to a potential loss of privacy. We then discuss the scenario in which an unauthorized subscriber attempts to search encrypted data to which it does not have access. The proposed methodology utilizes a number of cryptographic primitives to ensure execution of the encrypted search queries and to restrain malicious entities from deducing information that assists them in compromising the privacy of the outsourced data. As illustrated in the descriptive details of our proposed methodology, the inverted index is encrypted with symmetric encryption. To ensure oblivious evaluation of the search queries, homomorphic encryption is utilized along with a private matching protocol [31]. For the security analysis of these cryptographic primitives, readers can refer to [30] and [32]. In the subsequent sections, we examine the capabilities of malicious entities to deduce confidential information within the context of a semantic search over encrypted data. Malicious Cloud Server The proposed methodology for encrypted data search utilizes the computational power and storage facility of a cloud server to execute search queries, instead of relying on a trusted third party. The cloud server uses an encrypted index I^sk, that is comprised of encrypted keywords. To compromise the privacy of the outsourced data, the cloud server either has to decipher the inverted index or deduce information from the evaluation of the encrypted search queries. In addition, the search queries are submitted in an encrypted format (α0…lσsk), and are evaluated by using a private matching protocol i.e., (P(y0…n∈I^kw0…nωu)=Δy0…n). Since, search queries are encrypted and the result of query evaluation is oblivious to the cloud server, the cloud server cannot learn any information about the keywords concealed in the search query. In order to compromise the privacy of the outsourced data, the cloud server needs access to the secret key, k, which is shared by the repository owner. Once the cloud server has access to the secret key it can effortlessly decipher the keywords that comprises the inverted index. However, only authorized subscribers have access to the secret key as it is encrypted with their respective public key (ωuikpub). Thus, for a cloud server the computational complexity to compromise the privacy of the outsourced is equivalent to that of asymmetric encryption. However, even if the cloud server manages to gain access to the secret key it can only decipher the encrypted keywords that are associated with the outsourced data—so that confidentialitly of the outsourced data is preserved as it is encrypted with a symmetric encryption key, which is only disseminated to authorized subscribers. Since our proposed methodology deals with the encrypted data search, the topic of authorized data access is beyond its scope. Malicious Subscriber The proposed methodology of encrypted data search not only realizes encrypted search in untrusted domain it also tackles the problem of unauthorized data search by malicious users. It ensures that unauthorized subscribers are not able to deduce any information about the encrypted outsourced data by simply learning the presence or absence of keywords. It does provide protection against conspired attacks by unauthorized subscribers and untrusted cloud server. Since, encrypted index is concealed with secret key that is only shared amongst authorized collaborating subscribers, malicious subscriber can not successfully evaluate their search query. To search the encrypted data, the search criteria (Ckw0…l) is concealed with a secret key i.e., ES(C^kw0…l,k)=C^kw0…lk. Once concealed it is then used to the model search query, which is comprised of the encrypted coefficients α0…lσsk, of polynomial P(x∈C^kw0…lk). Since, only authorized subscribers have secret keys, search queries from unauthorized subscribers cannot be evaluated successfully. Also, unauthorized subscribers cannot intercept valid search query to modify the search criteria. This is because unauthorized subscriber does have valid secret key to model new or a part of valid search request. The concealed search criteria are only comparable with the encrypted index if the search criteria are also encrypted with the same key. Even if unauthorized subscribers collude with the cloud server, the execution of unauthorized search queries cannot assist them in learning any useful information. The search criterion encrypted with the arbitrary secret key is not compatible with the concealed inverted index, i.e., C^kw0…lk?∉I^kw0…nk. Thus, for unauthorized subscribers, it is computationally infeasible to deduce any information that can lead to the potential loss of data privacy. Conclusion and Future Directions This paper addresses the problem of privacy-aware data search within the untrusted domain of a cloud service provider. It proposes an index-based privacy-aware data search methodology which can identify a semantic match between encrypted data and search criteria. Unlike the conventional methodology, the proposed privacy-aware data search leverages authorized subscribers to access relevant data by defining conjunctive search queries without relying on any trapdoors defined by the data owner. It realizes an oblivious data search, which ensures that the cloud service provider can only assist in the execution of encrypted search queries; however, the CSP can not learn or deduce confidential information from the execution of the search query, which can lead to the potential loss of data privacy. The security analysis demonstrated that, for malicious subscribers and untrusted cloud service providers, the proposed methodology always generates a randomized response that restrains them from learning of the presence or absence of a particular keyword in the outsourced encrypted data. Since the proposed methodology is an index-based data search, it does not have a requirement to encrypt the outsourced data with a particular encryption algorithm. The encryption of outsourced data with an arbitrary encryption algorithm does not affect the operation of the proposed methodology. The computational analysis shows that the proposed methodology exerts a reasonable computational load on authorized subscribers to model their encrypted search queries. So far, the proposed methodology can only identify exact matches between the extended search criteria and the inverted index. In the future, we plan to include wildcard-enabled search queries, which can be used to match a substring while ensuring oblivious execution and privacy-awareness of search queries. This work was supported by a grant from the Kyung Hee University in 2013[KHU-20130439]. ==== Refs References 1 Buyya R , Yeo CS , Venugopal S , Broberg J , Brandic I . Cloud computing and emerging IT platforms: Vision, hype, and reality for delivering computing as the 5th utility. vol. 25 Amsterdam, The Netherlands, The Netherlands : Elsevier Science Publishers B. V. ; 2009 p. 599 –616 . Available from: http://portal.acm.org/citation.cfm?id=1528937.1529211. 2 Armbrust M , Fox A , Griffith R , Joseph AD , Katz R , Konwinski A , et al A view of cloud computing . Commun ACM . 2010 4 ;53 :50 –58 . Available from: http://doi.acm.org/10.1145/1721654.1721672. 10.1145/1721654.1721672 3 Buyya R , Yeo CS , Venugopal S . Market-oriented cloud computing: Vision, hype, and reality for delivering IT services as computing utilities, in . In: Department of Computer Science and Software Engineering (CSSE), The University of Melbourne , Australia He; 2008 p. 10 –1016 . 4 Velte T , Velte A , Elsenpeter R . Cloud Computing, A Practical Approach . 1st ed New York, NY, USA : McGraw-Hill, Inc. ; 2010 . 5 Archer J, Cullinane D, Puhlmann N, Boehme A, Kurtz P, Reavis J. Security Guidance for Critical Areas of Focus in Cloud Computing V3.0; 2011. https://cloudsecurityalliance.org/guidance/csaguide.v3.0.pdf. Available from: http://dl.acm.org/citation.cfm?id=1833515.1833621. 6 Ristenpart T, Tromer E, Shacham H, Savage S. Hey, you, get off of my cloud: exploring information leakage in third-party compute clouds. In: Proceedings of the 16th ACM conference on Computer and communications security. CCS’09. New York, NY, USA: ACM; 2009. p. 199–212. Available from: http://doi.acm.org/10.1145/1653662.1653687. 7 Kaufman LM . Data Security in the World of Cloud Computing . IEEE Security and Privacy . 2009 7 ;7 :61 –64 . Available from: http://dl.acm.org/citation.cfm?id=1591890.1592198. 10.1109/MSP.2009.87 8 Hacigümüş H, Iyer B, Li C, Mehrotra S. Executing SQL over encrypted data in the database-service-provider model. In: Proceedings of the 2002 ACM SIGMOD international conference on Management of data. SIGMOD’02. New York, NY, USA: ACM; 2002. p. 216–227. Available from: http://doi.acm.org/10.1145/564691.564717. 9 Zhou M , Mu Y , Susilo W , Yan J , Dong L . Privacy enhanced data outsourcing in the cloud . Journal of Network and Computer Applications . 2012 ;35 (4 ):1367 –1373 . Available from: http://www.sciencedirect.com/science/article/pii/S1084804512000367. 10.1016/j.jnca.2012.01.022 10 Song DX, Wagner D, Perrig A. Practical techniques for searches on encrypted data. In: Security and Privacy, 2000. S P 2000. Proceedings. 2000 IEEE Symposium on; 2000. p. 44 –55. 11 Boneh D , Crescenzo GD , Ostrovsky R , Persiano G . Public Key Encryption with Keyword Search In: EUROCRYPT ; 2004 p. 506 –522 . 12 Li M, Yu S, Cao N, Lou W. Authorized Private Keyword Search over Encrypted Data in Cloud Computing. In: Distributed Computing Systems (ICDCS), 2011 31st International Conference on; 2011. p. 383 –392. 13 cheng Chang Y, Mitzenmacher M. Privacy Preserving Keyword Searches on Remote Encrypted Data. In: In Proc. of 3rd Applied Cryptography and Network Security Conference (ACNS; 2005. p. 442–455. 14 Curtmola R , Garay J , Kamara S , Ostrovsky R . Searchable Symmetric Encryption: Improved Definitions and Efficient Constructions ; 2006 . 15 Yang Z, Zhong S, Wright RN. Privacy-Preserving Queries on Encrypted Data. In: In Proc. of 11th European Symposium On Research In Computer Security (Esorics); 2006. p. 479–495. 16 Okamoto T , Takashima K . Hierarchical Predicate Encryption for Inner-Products In: Matsui M , editor. Advances in Cryptology– ASIACRYPT 2009. vol. 5912 of Lecture Notes in Computer Science . Springer Berlin Heidelberg ; 2009 p. 214 –231 . Available from: 10.1007/978-3-642-10366-7_13 . 17 Wang C, Cao N, Li J, Ren K, Lou W. Secure Ranked Keyword Search over Encrypted Cloud Data. In: Distributed Computing Systems (ICDCS), 2010 IEEE 30th International Conference on; 2010. p. 253 –262. 18 Kamara S, Papamanthou C, Roeder T. CS2: A Searchable Cryptographic Cloud Storage System. Microsoft Research; 2011. MSR-TR-2011-58. Available from: http://research.microsoft.com/apps/pubs/default.aspx?id=148632. 19 Sun W, Wang B, Cao N, Li M, Lou W, Hou YT, et al. Privacy-preserving multi-keyword text search in the cloud supporting similarity-based ranking. In: Proceedings of the 8th ACM SIGSAC symposium on Information, computer and communications security. ASIA CCS’13. New York, NY, USA: ACM; 2013. p. 71–82. Available from: http://doi.acm.org/10.1145/2484313.2484322. 20 Pervez Z , Awan A , Khattak A , Lee S , Huh EN . Privacy-aware searching with oblivious term matching for cloud storage . vol. 63 Springer US ; 2013 p. 538 –560 . Available from: 10.1007/s11227-012-0829-z . 21 Google Search Appliance.; 2013. Available from: http://www.google.co.uk/enterprise/search/gsa.html. 22 Enterprise Search Server Solutions.; 2013. http://sharepoint.microsoft.com/en-us/product/capabilities/search/Pages/Search-Server.aspx. 23 Singh A , Srivatsa M , Liu L . Search-as-a-service: Outsourced search over outsourced storage . ACM Trans Web . 2009 9 ;3 :13:1 –13:33 . 24 Goyal V, Pandey O, Sahai A, Waters B. Attribute-based encryption for fine-grained access control of encrypted data. In: Proceedings of the 13th ACM conference on Computer and communications security. CCS’06. New York, NY, USA: ACM; 2006. p. 89–98. Available from: http://doi.acm.org/10.1145/1180405.1180418. 25 Kamara S , Lauter K . Cryptographic Cloud Storage In: Financial Cryptography and Data Security. vol. 6054 of Lecture Notes in Computer Science ; 2010 p. 136 –149 . 26 Wang C, Wang Q, Ren K, Lou W. Privacy-Preserving Public Auditing for Data Storage Security in Cloud Computing. In: INFOCOM, 2010 Proceedings IEEE; 2010. p. 1–9. 27 Pervez Z , Khattak AM , Lee S , Lee YK . SAPDS: self-healing attribute-based privacy aware data sharing in cloud . The Journal of Supercomputing . 2012 ;62 (1 ):431 –460 . 10.1007/s11227-011-0727-9 28 Lucene. Apache Lucene Core; 2013. http://lucene.apache.org/core/. 29 WordNet. About WordNet—a large lexical database of English; 2013. http://wordnet.princeton.edu/wordnet/. 30 Paillier P. Public key cryptosystems based on composite degree residuosity classes. In: Proceedings of the 17th international conference on Theory and application of cryptographic techniques. EUROCRYPT’99. Berlin, Heidelberg: Springer-Verlag; 1999. p. 223–238. 31 Freedman M , Nissim K , Pinkas B . Efficient Private Matching and Set Intersection . Springer-Verlag ; 2004 p. 1 –19 . 32 Goldreich O , Israel R , Dana T . Foundations of Cryptography ; 1995 .

PMC005xxxxxx/PMC5003395.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757136810.1371/journal.pone.0161904PONE-D-16-22330Research ArticleBiology and Life SciencesEcologyPlant EcologyPlant-Animal InteractionsPlant-Herbivore InteractionsEcology and Environmental SciencesEcologyPlant EcologyPlant-Animal InteractionsPlant-Herbivore InteractionsBiology and Life SciencesPlant SciencePlant EcologyPlant-Animal InteractionsPlant-Herbivore InteractionsBiology and Life SciencesEcologyPlant EcologyPlant-Animal InteractionsHerbivoryEcology and Environmental SciencesEcologyPlant EcologyPlant-Animal InteractionsHerbivoryBiology and Life SciencesPlant SciencePlant EcologyPlant-Animal InteractionsHerbivoryBiology and Life SciencesEcologyCommunity EcologyTrophic InteractionsHerbivoryEcology and Environmental SciencesEcologyCommunity EcologyTrophic InteractionsHerbivoryBiology and Life SciencesDevelopmental BiologyMetamorphosisLarvaeBiology and Life SciencesAgriculturePest ControlBiology and Life SciencesPlant SciencePlant PhysiologyPlant DefensesPhysical SciencesChemistryChemical CompoundsOrganic CompoundsVolatile Organic CompoundsPhysical SciencesChemistryOrganic ChemistryOrganic CompoundsVolatile Organic CompoundsBiology and Life SciencesAgricultureCrop ScienceCropsFruitsBiology and Life SciencesOrganismsPlantsFruitsBiology and Life SciencesOrganismsPlantsDoes the Slow-Growth, High-Mortality Hypothesis Apply Below Ground? Plant Traits and Belowground Interactionshttp://orcid.org/0000-0002-4183-8673Hourston James E. 1*Bennett Alison E. 2Johnson Scott N. 23Gange Alan C. 11 School of Biological Sciences, Royal Holloway, University of London, Egham Hill, Egham, TW20 0EX, England2 The James Hutton Institute, Invergowrie, Dundee, DD2 5DA, Scotland3 Hawkesbury Institute for the Environment, Western Sydney University, Sydney, AustraliaHe Zhili EditorUniversity of Oklahoma, UNITED STATESCompeting Interests: The authors have declared that no competing interests exist. Conceptualization: JH AB SJ AG. Formal analysis: JH. Funding acquisition: AB SJ AG. Investigation: JH. Methodology: JH AB SJ AG. Project administration: JH. Resources: JH AB SJ AG. Supervision: AB SJ AG. Validation: JH. Visualization: JH. Writing – original draft: JH AB SJ AG. Writing – review & editing: JH AB SJ AG. * E-mail: james.hourston@rhul.ac.uk29 8 2016 2016 11 8 e01619043 6 2016 12 8 2016 © 2016 Hourston et al2016Hourston et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Belowground tri-trophic study systems present a challenging environment in which to study plant-herbivore-natural enemy interactions. For this reason, belowground examples are rarely available for testing general ecological theories. To redress this imbalance, we present, for the first time, data on a belowground tri-trophic system to test the slow growth, high mortality hypothesis. We investigated whether the differing performance of entomopathogenic nematodes (EPNs) in controlling the common pest black vine weevil Otiorhynchus sulcatus could be linked to differently resistant cultivars of the red raspberry Rubus idaeus. The O. sulcatus larvae recovered from R. idaeus plants showed significantly slower growth and higher mortality on the Glen Rosa cultivar, relative to the more commercially favored Glen Ample cultivar creating a convenient system for testing this hypothesis. Heterorhabditis megidis was found to be less effective at controlling O. sulcatus than Steinernema kraussei, but conformed to the hypothesis. However, S. kraussei maintained high levels of O. sulcatus mortality regardless of how larval growth was influenced by R. idaeus cultivar. We link this to direct effects that S. kraussei had on reducing O. sulcatus larval mass, indicating potential sub-lethal effects of S. kraussei, which the slow-growth, high-mortality hypothesis does not account for. Possible origins of these sub-lethal effects of EPN infection and how they may impact on a hypothesis designed and tested with aboveground predator and parasitoid systems are discussed. http://dx.doi.org/10.13039/501100000270Natural Environment Research CouncilNE/I018921/1Gange Alan C This research was funded by the Natural Environment Research Council grant number NE/I018921/1, http://www.nerc.ac.uk/, and The James Hutton Institute, http://www.hutton.ac.uk/. Data AvailabilityThe underlying research data are openly available from Figshare at the DOI: 10.6084/m9.figshare.3490640.Data Availability The underlying research data are openly available from Figshare at the DOI: 10.6084/m9.figshare.3490640. ==== Body Introduction The black vine weevil, Otiorhynchus sulcatus Fabricius (Curculionidae) causes significant damage to a range of silvicultural and horticultural crops throughout the world’s temperate regions [1]. Adult O. sulcatus feed on the foliage of a huge range of plants, inflicting relatively minor damage when compared to the root feeding larvae, which can result in reduced plant growth and if an infestation is severe, the death of a host plant [2]. Conventional control of O. sulcatus is achieved using soil drench treatments of chemical pesticides. Until very recently the most commonly used treatment for an O. sulcatus infestation was the neonicotinoid; Imidacloprid, which has been temporarily withdrawn from use in the EU since 2014 due to non-target effects on bees. Future strategies to control O. sulcatus would be wise to therefore consider pesticide free alternatives as part of an integrated approach to pest management [3]. An alternative method of control popular for the treatment of plants that may be at risk of, or already under O. sulcatus attack, is entomopathogenic nematodes (EPNs). These have been shown in many studies to be effective in reducing both the performance and increasing mortality of O. sulcatus [4–6]. One of the primary plant hosts of O. sulcatus which is of major economic importance is the red raspberry, Rubus idaeus L.(Rosaceae), with over 13.8 thousand tons produced in 2013, worth $135.7 million (GPB/USD 1.52 on 23/11/2015) to the UK economy [7] and in the USA, in 2014 a production of 97 thousand tons worth $499.3 million, by Washington state, California and Oregon alone [8]. The production of R. idaeus is, in the UK, almost entirely under the protection of plastic tunnels which can raise temperatures by around 4°C compared to the surrounding field conditions and results in greatly increased growth [9]. However these conditions are also very favorable for O. sulcatus performance with the insects consuming more R. idaeus biomass, completing their life cycles faster and with adults being more fecund [9]. Two cultivars that have been studied previously with respect to their tolerance to O. sulcatus attack are Glen Ample and Glen Rosa [10]. Despite being closely related autumn mid-season fruiters these cultivars differ in their usage, with Glen Ample being a major commercial variety, due to fruit size and quality, and Glen Rosa being more popular on the amateur market due to its better tolerance to pests and diseases [11,12]. This study utilized these two cultivars of R. idaeus, demonstrated to be more (Glen Rosa) and less (Glen Ample) resistant to O. sulcatus [10]. Comparing two differently resistant cultivars is likely to result in different growth rates in O. sulcatus and which enables the testing of Feeny’s (1976) [13] slow growth, high mortality hypothesis when these two O. sulcatus populations are exposed to natural enemies. This hypothesis predicts that when plant traits impose a cost to the fitness of larvae, increasing developmental time, either via sub lethal effects via antixenosis, antibiosis or the reallocation of resources away from sites of herbivory, that herbivores will be vulnerable for longer to predation. Thus counteracting the effects of reduced host plant nutritional quality that might result in herbivores compensating by increasing consumption of plant tissues. The inclusion of natural enemies in order to truly understand plant-herbivore interactions has long been championed in the field of tri-tropic interactions [14] and continues to be further explored in light of new evidence [15]. Otiorhynchus sulcatus is a good model for testing the hypothesis that resistant cultivars and EPNs may be combined as a technique to suppress root herbivore populations as O. sulcatus has a history, particularly in the horticultural sector, of being treated with a range of EPNs [16]. This alongside evidence from field studies reporting that heavy infestations of O. sulcatus reduced yield by 39% and 66% in Glen Rosa and Glen Ample [10], respectively, provides a framework for a convenient multi-trophic system in which to study potential interactions. The two EPN species incorporated into the experiment are both widely recommended and commercially available specifically for use against O. sulcatus [4]. Steinernema kraussei Steiner is cold tolerant, active at <10°C, whereas Heterorhabditis megidis Poinar, Jackson & Klein is active at >10°C, both are known to alter their dispersal and taxis depending on the substrate they are in and possess different bacterial endosymbiont communities [5,17,18]. To assess how the different EPN treatments influenced O. sulcatus mortality and performance we proposed two hypotheses. First that EPN treatments would decrease O. sulcatus abundance, as one would expect from a tried and tested biological control agent. Following on from this we predicted that the presence of the more resistant host, Glen Rosa, would, in combination with EPNs, result in lower larval mass and higher levels of O. sulcatus control. This hypothesis was based on Feeny's, (1976) hypothesis that slow growth leads to high mortality from natural enemies. So, we would expect that on a more resistant host, O. sulcatus would have a slower growth and consequently be more susceptible to EPN infection. Plant biomass was calculated at the end of the experiment to quantify the plant response to EPN treatments. We hypothesized that EPN treatments on plants infested with O. sulcatus would promote an increase in plant biomass as the plants should be suffering less herbivore damage. The proportion of root biomass to shoot biomass was used as an indirect measure of changes in carbon allocation in response to herbivory. We predicted that the EPN treatments and subsequent abundance of O. sulcatus would influence the proportion of root and shoot biomass in R. idaeus cultivars, potentially allowing a recovery in root biomass following successful O. sulcatus control. Materials and Methods Rootstock from existing R. idaeus plants of two cultivars; Glen Ample and Glen Rosa was sourced from The James Hutton Institute’s (Dundee, UK) breeding stock. The rootstock was washed with a 4% sodium hypochlorite solution, rinsed with water then planted in twice autoclaved (with a 12hr gap between autoclave runs) compost (Keith Singleton sterilized loam, Nethertown, UK) and grown on under-heated benches in a controlled greenhouse environment (16:8 days at 18˚C). Four weeks after this, plants were transplanted into 1.8L pots containing 1.6L of a homogenized, twice autoclaved 1:1 soil (Keith Singleton sterilized loam, Nethertown, UK) and sand mix. Compost was autoclaved in batches to control for the possibility of there being any nematodes living in the compost which could interact with added treatments [19]. Plants from each cultivar were equally and randomly distributed between the 4 treatments giving a replication of 26 plants for each treatment. To account for any possible environmental heterogeneity within the glasshouse the plants were then incorporated into blocks each equally representing individuals from each treatment in a randomized order to adhere to a randomized block design. Two weeks after the plants were transplanted, and before any herbivore or EPN treatments were added, plant height was recorded in order to be used later as a random effect in statistical models to account for the initial variation in height between plants. Five weeks after R. idaeus were transplanted into individual pots, 40 O. sulcatus eggs were added into a 10mm indent in the soil surface, 20mm away from the stem of each plant. This egg density was selected to simulate arrival of a gravid adult feeding on plants for several weeks [10]. Four weeks after plants were infested with O. sulcatus, EPNs were added to plants, with control plants remaining untreated. Three weeks after nematodes were added, the plants were harvested and O. sulcatus larvae were retrieved, counted and fresh mass taken. A subset of half the plants were then freeze dried to ascertain dry mass (Fig 1). Otiorhynchus sulcatus eggs were taken from a culture of adults maintained at 18˚C with a 16:8 light: dark cycle at The James Hutton Institute, Dundee. The EPNs used in the experiment were purchased from commercial suppliers and were advertised as being a specific line to control for O. sulcatus. S. kraussei (Becker and Underwood®, Littlehampton, UK) and H. megidis (Biobest®, Milton Bridge, UK). They were both added to plants as separate treatments at their recommended dosages. When formulated to the commercially recommended dosages this results in 7735 ± 531 S. kraussei added per pot and approximately 12275 ± 780 H. megidis added to each pot according to the product formulation guidelines. 10.1371/journal.pone.0161904.g001Fig 1 The timing of key stages to the experimental setup and execution. This created a 2 × 4 factorial experiment which was conducted under controlled conditions (16:8 days at 18˚C). Two R. idaeus cultivars (Glen Ample and Glen Rosa) were subdivided into one of four treatments. A control treatment, where neither O. sulcatus or EPNs were added, an O. sulcatus ‘only’ treatment where the herbivore was observed in the absence of EPNs and two treatments in which, in addition to O. sulcatus, either S. kraussei or H. megidis were added to plants. Statistical analysis The mean mass and abundance of O. sulcatus larvae on each plant was analyzed using generalized linear mixed models (GLMMs) incorporating Gaussian and Poisson errors respectively. These response variables were tested against the cultivar and EPN treatment and the interactions between the two. Experimental block and autoclave batch were included as random effects. The biomass data taken from the dry mass of R. idaeus plants was also analyzed using a GLMM incorporating Gamma errors with a log link, using nematode treatment, O. sulcatus abundance and mean O. sulcatus mass as explanatory variables with experimental block included as a random effect. All analyses were carried out using R3.2.3 "Wooden Christmas-Tree" [20] using the lmer4 [21] and car [22] packages for GLMMs. Models were simplified where appropriate using AIC, with the best fitting minimal models reported. Pairwise comparisons were conducted using the R package phia [23]. Results Insect herbivore performance The abundance of O. sulcatus recovered at the end of the experiment was affected by both the R. idaeus cultivar and the EPN treatment added to plants (Table 1). In both cultivars the addition of S. kraussei resulted in lower abundance when compared to control treatments and treatments where H. megidis were added (Fig 2A). In addition to the observed differences between cultivars on O. sulcatus abundance, the mass of recovered larvae was found to be lower (χ2 = 74.11, d.f. 1, P<0.001) in Glen Rosa when compared to Glen Ample (Fig 2B). The mass of larvae was also lower on S. kraussei treated plants when compared to O. sulcatus only treatments (χ2 = 7.39, d.f. 1, P<0.05). 10.1371/journal.pone.0161904.g002Fig 2 A Mean O. sulcatus abundance per plant across different R. idaeus cultivar and EPN treatments. B The mean mass of O. sulcatus larvae per plant recovered from different R. idaeus cultivars and EPN treatments. Statistical differences between treatments are indicated by different letters above the bars. 10.1371/journal.pone.0161904.t001Table 1 Summary table of post hoc contrasts carried out on O. sulcatus abundance to describe the interactions between R. idaeus cultivar and mean O. sulcatus mass and R. idaeus cultivar and EPN treatment. Contrast: Category df χ2 value P-value Cultivar: Mean O. sulcatus mass 1 4.46 <0.05 O. sulcatus only -H. megidis: Glen Ample 1 1.54 0.21 O. sulcatus only -S. kraussei: Glen Ample 1 60.37 <0.001 H. megidis-S. kraussei: Glen Ample 1 45.89 <0.001 O. sulcatus only -H. megidis: Glen Rosa 1 5.83 <0.05 O. sulcatus only -S. kraussei: Glen Rosa 1 26.07 <0.001 H. megidis-S. kraussei: Glen Rosa 1 9.06 <0.001 Glen Ample-Glen Rosa: O. sulcatus only 1 14.66 <0.001 Glen Ample-Glen Rosa: H. megidis 1 20.62 <0.001 Glen Ample-Glen Rosa: S. kraussei 1 0.65 0.41 Differences observable between the abundance of O. sulcatus on the two cultivars were entirely driven by an interaction (Table 1) between the mean mass of O. sulcatus and the cultivar of R. idaeus. Larvae on Glen Ample had a greater mass and suffered lower mortality, while those on Glen Rosa had a low masses and suffered greater levels of mortality (Fig 3A). There was an impact of EPNs on this relationship within Glen Ample with S. kraussei exhibiting a different relationship as observed in the H. megidis and O. sulcatus only treatments (χ2 = 4.54, d.f. 1, P<0.05). While the H. megidis and O. sulcatus only treatments followed the general trend observed for Glen Ample in Fig 3A, S. kraussei caused a reduction (χ2 = 7.28, d.f. 1, P<0.05) in O. sulcatus mass as well abundance (Fig 3B). Within Glen Rosa, there was not found to be a significant difference between the O. sulcatus only, and EPN treatments, so far as their impact on the relationship between O. sulcatus abundance and mean body mass was concerned. 10.1371/journal.pone.0161904.g003Fig 3 A The relationship between O. sulcatus abundance and mean mass across two R. idaeus cultivars. B The relationship between O. sulcatus abundance and mean mass within Glen Ample across EPN and O. sulcatus only treatments. Plant biomass The total plant biomass calculated at the end of the experiment showed that the two R. idaeus cultivars performed differently when herbivores were present (Fig 4A). A decrease in biomass was particularly clear in Glen Ample, where average biomass fell by >50% when herbivores were added (χ2 = 18.48, d.f. 1, P<0.001) a pattern that was not reflected in Glen Rosa’s total biomass once variation in initial plant height was taken into account. In the herbivore free control treatment, the two cultivars had a different mean total biomass with Glen Ample having a significantly higher biomass than Glen Rosa (χ2 = 5.07, d.f. 1, P<0.05). 10.1371/journal.pone.0161904.g004Fig 4 A The total biomass of R. idaeus of different cultivars across EPN treatments. B The proportion of root to shoot biomass measured at the end of the experiment separated by R. idaeus cultivar and EPN treatment. Statistical differences between treatments are indicated by different letters above the bars. The distribution of R. idaeus biomass between the above and belowground portions of the plant also appeared to have been disrupted by the presence of O. sulcatus (Fig 4B). The proportion of root to shoot biomass showed a general trend for higher root biomass relative to shoot biomass in the control treatment. In the O. sulcatus only, and H. megidis treatments both cultivars showed very similar trends in how biomass was distributed. The only exception was in plants treated with S. kraussei, where a higher proportion of root biomass relative to shoot biomass was recorded in Glen Rosa than in Glen Ample (χ2 = 15.37, d.f. 1, P<0.001). Discussion The comparison of two commercially available EPN species showed that S. kraussei was more effective at controlling O. sulcatus than H. megidis in both cultivars of R. idaeus, with the abundance and growth of O. sulcatus being substantially reduced. Both these EPN species are considered to be capable of cruise foraging, meaning they actively seek out hosts in the soil [17,24]. The experiment was held at a constant 18°C meaning both species were operating within their optimal temperature range. Their contrasting performance could hence be due to other differences in behavior and biology. There have been several studies that show soil media or substrate can have a significant effect on the dispersal behavior of EPNs, with different species showing greater taxis towards hosts in different media [5,17]. This could explain some of the variation between these species and consequently results may not be the same in the field. This said, S. kraussei has a lower cold tolerance (4°C) when compared to H. megidis (10°C) making it a better choice when treating plants at the beginning or end of a growing season [4]. This is ideal for the protection of both Glen Ample and Glen Rosa, as these are both mid-season fruiting varieties, and the beginning of the season represents a period of critical growth, prior to flowering [12]. The poorer performance of H. megidis could be due to a better O. sulcatus immune response. Possibly resulting in successful encapsulation of EPNs, or resistance to the associated symbiotic bacteria, Photorhabdus spp., which normally causes death by septicemia, different to the Xenorhabdus spp. associated with S. kraussei [25]. Otiorhynchus sulcatus performed significantly better on Glen Ample plants than on Glen Rosa as shown in their larval mass, and this is supported by previous studies which found Glen Ample to be a cultivar less resistant to O. sulcatus when compared to other R. idaeus cultivars [10,26]. This is likely due to the different traits bred into these two cultivars. Glen Ample is a more popular variety as it produces a higher yield of larger, sweeter fruit and is favored commercially. Glen Rosa however is more tolerant to pests and diseases. It has been bred to have an A10 resistance gene which confers resistance to the large raspberry aphid, Amphorophora idaei Börner, but has smaller fruit and typically produces smaller yields when compared to Glen Ample [12]. The testing of Feeny’s (1976) slow growth, high mortality hypothesis using the comparison between O. sulcatus performance on Glen Ample and Gen Rosa would appear to be highly appropriate as exactly this relationship of O. sulcatus growth and mortality was observed between the two cultivars. It is apparent that H. megidis fits to this model, as when O. sulcatus mass was low on Glen Rosa, O. sulcatus abundance fell accordingly. However, the same degree of conformity to Feeny’s hypothesis was not observed in S. kraussei, where abundance of O. sulcatus was found to be low regardless of changes in O. sulcatus mass. There may be several reasons for this nonconformity, firstly the mass of O. sulcatus larvae recovered in the experiment was not just affected by the difference in cultivar. The addition of S. kraussei also appeared to directly decrease larval mass in O. sulcatus. This might suggest that there are potential sub-lethal effects being observed in the surviving larvae. O. sulcatus may be infected by EPNs and this stress may impact on feeding rates and larval development [27] but not result in death. The level of tolerance to EPNs is known to vary greatly, with some insect immune systems able to encapsulate and withstand up to 20 EPNs before the insect’s death [28]. Secondly as Feeny’s hypothesis was based around more classical aboveground predator/parasitoid systems it may be that the more complex communal life strategy employed by EPNs may be less appropriate. There is even evidence that EPNs can interact directly with plant defense chemistry, inducing systemic resistance in plants [29]. Perhaps greater than the interactions that occur directly between the plant and EPNs is the influence of the staggeringly complex soil microbial community. The soil microbial community, when studied in model plants, has been found to be extremely large, taxonomically diverse and specific to certain soil types [30,31]. Indeed the differences in soil structure and composition that have been identified as being a determining factor in EPN efficacy [5,17] also have a strong effect in determining the soil microbial community composition [32]. There are many examples of soil microbes interacting with plants to bolster plant defenses against pests and pathogens [33] but also, soil based entomopathogenic microbes can provide competition for EPNs which can lead to their competitive exclusion from insect hosts [34]. It is therefore likely that in such a complicated system as soil, the slow-growth, high-mortality hypothesis is unlikely to prove a comprehensive explanation for the myriad interacting organisms and their impacts in plant/herbivore/natural enemy interactions. The difference in total plant biomass observed between the R. idaeus cultivars in the control treatment followed what would be expected from the traits bred into these lines. The more vigorous growth more typically associated with Glen Ample [12] would be expected to lead to greater average biomass than in Glen Rosa. The lower biomass observed, particularly in Glen Ample, when O. sulcatus was present fitted well with field observations [10]. Host plants have an arsenal of different defences that they can deploy against herbivores such as antixenosis, antibiosis and tolerance [35]. Even differing plant traits such as variation in nutritional value of plant tissues between two cultivars or species can reduce slower growth in herbivores resulting in higher natural enemy related mortality [36]. Differences in the nutritional chemistry of the roots of the two cultivars may play a role in differing O. sulcatus performance. O. sulcatus larval abundance has been shown to be positively correlated with levels of nitrogen and magnesium and negatively so with respects to iron, but this was not found to be significantly different between Glen Ample and Glen Rosa cultivars [26]. The A10 resistance gene bred into Glen Rosa after being isolated from R. occidentalis L. is thought to be effective against aphids through both antixenosis and antibiosis [37]. It has been suggested in other studies that the presence of this gene in Glen Rosa may be conferring resistance against O. sulcatus [11,26]. This differing plant trait bred into the two cultivars may well explain the observed reduction in larval performance on Glen Rosa. In this experiment, herbivores are not given a choice of host plants and O. sulcatus are known to readily consume both Glen Ample and Glen Rosa in the absence of choice [11] and so the effects observed are not driven by antixenosis. There are many different ways by which antibiosis can be affected, either through abiotic, for example via increased plant nutrition [38], or biotic, via association with beneficial microbes [33], means. Conducting controlled glasshouse experiments minimises the impact of many of these factors and so effects that are observed are most likely as a consequence of plant traits. When a plant is damaged by a feeding herbivore, constitutive and inducible defences are activated which usually involve the increased production of secondary metabolites. Compounds such as alkaloids, glucosinolates, terpenoids, and phenolics can have a variety of different, lethal and sub lethal effects on plant herbivores [39]. Phenolics for example have been identified as likely to act as an antifeedant in Ribes nigrum L. decreasing O. sulcatus performance [40,41]. Higher concentrations of phenolic compounds or similar secondary metabolites in Glen Rosa, relative to Glen Ample, could explain the reduction in the performance of O. sulcatus larvae when EPNs were not present. Such a plant trait would be a classic example of how plant defence chemistry can extend the most vulnerable period of the insect life cycle, exposing herbivores to predation. Induced defences often include an increase in the concentration of volatile organic compounds (VOCs) present in plant tissues which can act in many different ways, either as a direct toxin or a feeding deterrent [42]. These VOCs are exuded by the plant both above and belowground and in some cases this is used by additional herbivores to locate and identify an already damaged plant, but this can also act as an attractant for natural enemies that can come to the aid of the attacked plant. The EPN, H. megidis has been shown to be attracted to volatile emissions from plants that have been attacked by herbivores [43]. It may be that the decreased abundance of O. sulcatus on Glen Rosa could be attributed to greater concentrations of VOCs released from root tissues at sites of tissue damage which could be attracting EPNs towards to their host’s location. Examining the interplay between plants, herbivores, natural enemies and how VOC emissions tie them all together is a growing field that may in future provide new territory for breeding in new types of pest resistant traits [44]. There are various definitions and means of measuring plant tolerance to herbivory but one that can be assessed in this experiment is the difference in fitness between damaged and undamaged plants compared between cultivars [45]. If the biomass collected at the end of the experiment could be considered an indication of R. idaeus fitness then it is clear that compared to Glen Ample, Glen Rosa is exhibiting tolerance to O. sulcatus herbivory. With Glen Ample suffering a large decrease in biomass as a consequence of the presence of O. sulcatus but Glen Rosa maintaining a similar biomass. There is evidence of compensation for lost biomass in Glen Rosa, a classic tolerance mechanism. It is however hard to determine if this may come at a cost to fitness as the plants were not grown for long enough to assess seed production. The ability of a plant to shift carbon stores from roots to shoots, thus changing the distribution of biomass, is another recognised indication that a plant tolerance mechanism is occurring [45]. There were no clear indications that root to shoot biomass demonstrated this tolerance mechanism as there was only a trend of decreased root biomass relative to shoot biomass when O. sulcatus were present. This general trend was reversed however in the S. kraussei treated plants, where the successful reduction in O. sulcatus abundance appears to restore a more vigorous root growth in Glen Rosa. The general pattern of decreased root biomass relative to shoot biomass under root herbivory is not unsurprising. A pattern of resource reallocation away from insect herbivory has been observed in previous studies, suggesting this may be an established plant defense strategy [46,47]. Conclusions S. kraussei performed best out of the two EPN species tested, possibly due to better suitability to the soil substrate, a key factor influencing EPN efficacy. Another key factor that commonly affecting EPN efficacy [30,31], temperature, was discounted as having an effect as both species tested were within their optimum temperature range. Differences between the two raspberry cultivars tested were likely due to herbivore resistance bred into the Glen Rosa cultivar. The presence of high concentrations of phenolic compounds have been known to affect O. sulcatus in other soft fruit crops [40,41]. Despite its relative susceptibility to O. sulcatus, Glen Ample remains the commercial favourite due to its high yield of large fruits. However, as pesticides that are effective for controlling O. sulcatus are withdrawn from the market, over safety and environmental concerns, more resistant plants may increasingly become attractive as part of an integrated crop management approach. Lack of conformity by the EPN S. kraussei to the slow growth, high mortality hypothesis could be explained by lower O. sulcatus larval masses in S. kraussei treatments which indicates sub-lethal effects of exposure to this EPN. This hypothesis was not originally devised with EPNs in mind and has been primarily been tested with data from predator and parasitoid natural enemies [48,49] and may not, for this reason, sufficiently explain such interactions especially in complex soil ecosystems. The authors would like to Sean Hackett and Anna Macrae for practical assistance. Dr Tina Steinbrecher for assistance with graphics. This research was funded by the Natural Environment Research Council and The James Hutton Institute, grant number NE/I018921/1. ==== Refs References 1 Alford D V . Pests of Fruit Crops—A Colour Handbook . Manson Publishing, UK ; 2007 . 2 Penman DR , Scott RR . Adult emergence and egg production of the black vine weevil in Canterbury . New Zeal J Exp Agric . 1976 ;4 : 385 –389 . 3 Gill S , Lutz J , Shrewsbury P , Raupp M . Evaluation of biological and chemical control methods for black vine weevil, Otiorhynchus sulcatus (Fabricius)(Coleoptera: Curculionidae), in container grown perennials . J Environ Hortic. Horticultural Research Institute ; 2001 ;19 : 166 –170 . 4 Haukeland S , Lola-Luz T . Efficacy of the entomopathogenic nematodes Steinernema kraussei and Heterorhabditis megidis against the black vine weevil Otiorhynchus sulcatus in open field-grown strawberry plants . Agric For Entomol . 2010 ;12 : 363 –369 . 10.1111/j.1461-9563.2010.00497.x 5 Ansari MA , Butt TM . Effect of potting media on the efficacy and dispersal of entomopathogenic nematodes for the control of black vine weevil, Otiorhynchus sulcatus (Coleoptera: Curculionidae) . Biol Control . Elsevier Inc.; 2011 ;58 : 310 –318 . 10.1016/j.biocontrol.2011.05.016 6 Bruck DJ , Shapiro-Ilan DI , Lewis EE . Evaluation of application technologies of entomopathogenic nematodes for control of the black vine weevil . J Econ Entomol . 2005 ;98 : 1884 –9 . 10.1603/0022-0493-98.6.1884 16539109 7 DEFRA. Basic Horticultural Statistics 2013. 2013. 8 USDA. USDA Economic Research Service. In: Fruit Yearbook for Berries [Internet]. 2015 p. Table d5. Available: http://www.ers.usda.gov/data-products/fruit-and-tree-nut-data/yearbook-tables.aspx 9 Johnson SN , Petitjean S , Clark KE , Mitchell C . Protected raspberry production accelerates onset of oviposition by vine weevils (Otiorhynchus sulcatus) . Agric For Entomol . 2010 ; 277 –283 . 10.1111/j.1461-9563.2010.00473.x 10 Clark KE , Hartley SE , Brennan RM , Jennings NN , McMenemy LS , McNicol JW , et al Effects of cultivar and egg density on a colonizing vine weevil (Otiorhynchus sulcatus) population and its impacts on red raspberry growth and yield . Crop Prot . Elsevier; 2012 ;32 : 76 –82 . 10.1016/j.cropro.2011.10.008 11 Clark KE , Hartley SE , Brennan RM , MacKenzie K , Johnson SN . Oviposition and feeding behaviour by the vine weevil Otiorhynchus sulcatus on red raspberry: effects of cultivars and plant nutritional status . Agric For Entomol . 2011 ;14 : 157 –163 . 10.1111/j.1461-9563.2011.00554.x 12 Hall H , Hummer K , Jamieson A , Jennings S ., Weber CA. Raspberry breeding and genetics . Plant Breed Rev . 2008 ;32 : 39 –382 . 13 Feeny P . Plant Apparency and Chemical Defense In: Wallace JW , Mansell RL , editors. Biochemical Interaction Between Plants and Insects . volume 10 Springer US ; 1976 pp. 1 –40 . 10.1007/978-1-4684-2646-5_1 14 Price PW , Bouton CE , Gross P , McPheron B a , Thompson JN , Weis AE . Interactions among three trophic levels: Influence of plants on interactions between insect herbivores and natural enemies . Annu Rev Ecol Syst . 1980 ;11 : 41 –65 . 10.1146/annurev.es.11.110180.000353 15 Ode PJ . Plant Chemistry and Natural Enemy Fitness: Effects on Herbivore and Natural Enemy Interactions . Annu Rev Entomol . 2006 ;51 : 163 –185 . 10.1146/annurev.ento.51.110104.151110 16332208 16 Georgis R , Koppenhöfer A , Lacey L , Bélair G , Duncan L , Grewal P , et al Successes and failures in the use of parasitic nematodes for pest control . Biol Control . Elsevier; 2006 ;38 : 103 –123 . 10.1016/j.biocontrol.2005.11.005 17 Kruitbos LM , Heritage S , Hapca S , Wilson MJ . The influence of habitat quality on the foraging strategies of the entomopathogenic nematodes Steinernema carpocapsae and Heterorhabditis megidis . Parasitology . 2010 ;137 : 303 –9 . 10.1017/S0031182009991326 19835647 18 Forst S , Dowds B , Boemare N , Stackebrandt E . Xenorhabdus and Photorhabdus spp.: bugs that kill bugs . Annu Rev Microbiol . 1997 ;51 : 47 –72 . 10.1146/annurev.micro.51.1.47 9343343 19 Jagdale GB , Somasekhar N , Grewal PS , Klein MG . Suppression of plant-parasitic nematodes by application of live and dead infective juveniles of an entomopathogenic nematode, Steinernema carpocapsae, on boxwood (Buxus spp.) . Biol Control . 2002 ;24 : 42 –49 . 10.1016/S1049-9644(02)00004-X 20 R Core Team . R: A Language and Environment for Statistical Computing [Internet]. Vienna, Austria : R Foundation for Statistical Computing ; 2013 Available: http://www.r-project.org 21 Bates D , Maechler M , Bolker B , Walker S . Fitting Linear Mixed-Effects Models Using lme4 . J Stat Softw . 2015 ;67 : 1 –48 . 10.18637/jss.v067.i01 22 Fox J , Weisberg S . An {R} Companion to Applied Regression , Second Edition [Internet]. Thousand Oaks CA : Sage ; 2011 Available: http://socserv.socsci.mcmaster.ca/jfox/Books/Companion 23 De Rosario-Martinez H. phia: Post-Hoc Interaction Analysis. R package version 0.2–1 [Internet]. 2015. Available: http://cran.r-project.org/package=phia 24 Campbell JF , Lewis EE , Stock SP , Nadler S , Kaya HK . Evolution of host search strategies in entomopathogenic nematodes . J Nematol . 2003 ;35 : 142 –5 . 19265988 25 Dowds B , Peters A . Virulence Mechanisms In: Gaugler R , editor. Entomopathogenic Nematology . Wallingford UK : CAB International ; 2002 pp. 79 –99 . 26 Clark KE , Hartley SE , Brennan RM , Mackenzie K , Johnson SN . Investigating preference-performance relationships in aboveground-belowground life cycles: a laboratory and field study with the vine weevil (Otiorhynchus sulcatus) . Bull Entomol Res . 2011 ;61 : 1 –8 . 10.1017/S0007485311000368 27 Alchanatis V , Navon A , Glazer I , Leviski S . An Image Analysis System for measuring Insect Feeding Effects caused by Biopesticides . J Agric Eng Res . 2000 ;77 : 289 –296 . 10.1006/jaer.2001.0709 28 Thurston GS , Yule WN , Dunphy GB . Explanations for the Low Susceptibility of Leptinotarsa decemlineata to Steinernema carpocapsae . Biological Control . 1994 pp. 53 –58 . 10.1006/bcon.1994.1010 29 Jagdale GB , Kamoun S , Grewal PS . Entomopathogenic nematodes induce components of systemic resistance in plants: Biochemical and molecular evidence . Biol Control . Elsevier Inc.; 2009 ;51 : 102 –109 . 10.1016/j.biocontrol.2009.06.009 30 Lundberg DS , Lebeis SL , Paredes SH , Yourstone S , Gehring J , Malfatti S , et al Defining the core Arabidopsis thaliana root microbiome . Nature . Nature Publishing Group; 2012 ;488 : 86 –90 . 10.1038/nature11237 22859206 31 Bulgarelli D , Rott M , Schlaeppi K , Ver Loren van Themaat E , Ahmadinejad N , Assenza F , et al Revealing structure and assembly cues for Arabidopsis root-inhabiting bacterial microbiota . Nature . Nature Publishing Group; 2012 ;488 : 91 –5 . 10.1038/nature11336 22859207 32 Young IM , Crawford JW . Interactions and self-organization in the soil-microbe complex . Science (80-). 2004 ;304 : 1634 –1637 . Available: <Go to ISI>://000221934300039 15192219 33 Gadhave KR , Hourston JE , Gange AC . Developing Soil Microbial Inoculants for Pest Management: Can One Have Too Much of a Good Thing? J Chem Ecol . 2016 ; 1 –9 . doi: 10.1007/s10886-016-0689-8 26662358 34 Kaya HK, Koppenhöfer AM. Effects of Microbial and Other Antagonistic Organism and Competition on Entomopathogenic Nematodes. http://dx.doi.org/101080/09583159631334. Taylor & Francis Group; 2010; 35 Dent D. Host plant resistance. In: Dent D, editor. Insect pest management. Wallingford, UK; 2000. pp. 123–179. doi: 10.1079/9780851993409.0123 36 Moreira X , Abdala-Roberts L , Hernandez-Cumplido J , Rasmann S , Kenyon SG , Benrey B . Plant species variation in bottom-up effects across three trophic levels: A test of traits and mechanisms . Ecol Entomol . 2015 ;40 : 676 –686 . 10.1111/een.12238 37 McMenemy LS , Mitchell C , Johnson SN . Biology of the European large raspberry aphid (Amphorophora idaei): Its role in virus transmission and resistance breakdown in red raspberry . Agric For Entomol . 2009 ;11 : 61 –71 . 10.1111/j.1461-9563.2008.00409.x 38 Low PA , McArthur C , Fisher K , Hochuli DF . Elevated volatile concentrations in high-nutrient plants: Do insect herbivores pay a high price for good food? Ecol Entomol . 2014 ;39 : 480 –491 . 10.1111/een.12124 39 Wink M . Plant breeding: importance of plant secondary metabolites for protection against pathogens and herbivores . Theor Appl Genet . 1988 ;75 : 225 –233 . 10.1007/BF00303957 40 Johnson SN , Barton AT , Clark KE , Gregory PJ , McMenemy LS , Hancock RD . Elevated atmospheric carbon dioxide impairs the performance of root-feeding vine weevils by modifying root growth and secondary metabolites . Glob Chang . Wiley Online Library; 2011 ;17 : 688 –695 . 10.1111/j.1365-2486.2010.02264.x 41 Coyle DR , Clark KE , Raffa KF , Johnson SN . Prior host feeding experience influences ovipositional but not feeding preference in a polyphagous insect herbivore . Entomol Exp Appl . 2011 ;138 : 137 –145 . 10.1111/j.1570-7458.2010.01083.x 42 Bezemer TM , van Dam NM . Linking aboveground and belowground interactions via induced plant defenses . Trends Ecol Evol . 2005 ;20 : 617 –24 . 10.1016/j.tree.2005.08.006 16701445 43 Rasmann S , Köllner TG , Degenhardt J , Hiltpold I , Toepfer S , Kuhlmann U , et al Recruitment of entomopathogenic nematodes by insect-damaged maize roots . Nature . Nature Publishing Group; 2005 ;434 : 732 –737 . 10.1038/nature03451 15815622 44 Stenberg JA , Heil M , Ahman I , Bjorkman C . Optimizing Crops for Biocontrol of Pests and Disease . Trends Plant Sci . 2015 ;20 : 698 –712 . 10.1016/j.tplants.2015.08.007 26447042 45 Strauss S , Agrawal A . The ecology and evolution of plant tolerance to herbivory . Trends Ecol Evol . 1999 ;14 : 179 –185 . Available: http://www.ncbi.nlm.nih.gov/pubmed/10322530 10322530 46 Robert CAM , Erb M , Hibbard BE , Wade FB , Zwahlen C , Turlings TCJ . A specialist root herbivore reduces plant resistance and uses an induced plant volatile to aggregate in a density-dependent manner . Thompson K , editor. Funct Ecol . 2012 ;26 : 1429 –1440 . 10.1111/j.1365-2435.2012.02030.x 47 Newingham BA , Callaway RM , BassiriRad H . Allocating nitrogen away from a herbivore: A novel compensatory response to root herbivory . Oecologia . 2007 ;153 : 913 –920 . 10.1007/s00442-007-0791-2 17619205 48 Benrey B , Denno RF . The Slow-Growth–High-Mortality Hypothesis: A Test Using The Cabbage Butterfly . Ecology . 1997 ;78 : 987 –999 . 10.2307/2265852 49 Williams IS . Slow-growth, high-mortality-a general hypothesis, or is it? Ecol Entomol . 2001 ;24 : 490 –495 . 10.1046/j.1365-2311.1999.00217.x

PMC005xxxxxx/PMC5003396.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757120810.1371/journal.pone.0161887PONE-D-16-15144Research ArticleBiology and Life SciencesAgricultureAgrochemicalsFungicidesMedicine and Health SciencesInfectious DiseasesFungal DiseasesBiology and Life SciencesPlant SciencePlant AnatomyLeavesBiology and Life SciencesMicrobiologyMedical MicrobiologyMicrobial PathogensFungal PathogensMedicine and Health SciencesPathology and Laboratory MedicinePathogensMicrobial PathogensFungal PathogensBiology and Life SciencesMycologyFungal PathogensBiology and Life SciencesPlant SciencePlant PathologyPlant PathogensPlant Fungal PathogensBiology and Life SciencesPlant SciencePlant AnatomySeedsMedicine and Health SciencesInfectious DiseasesInfectious Disease ControlPhysical SciencesMaterials ScienceMaterials by AttributeCoatingsEngineering and TechnologyManufacturing ProcessesSurface TreatmentsCoatingsThe Evolution of Fungicide Resistance Resulting from Combinations of Foliar-Acting Systemic Seed Treatments and Foliar-Applied Fungicides: A Modeling Analysis Fungicide Resistance Caused by Seed- and Foliar TreatmentsKitchen James L. 1van den Bosch Frank 1Paveley Neil D. 2Helps Joseph 1van den Berg Femke 1*1 Computational and Systems Biology, Rothamsted Research, Harpenden, Hertfordshire, United Kingdom2 Plant Pathology Department, ADAS, High Mowthorpe, Duggleby, North Yorkshire, United KingdomHeneberg Petr EditorCharles University in Prague, CZECH REPUBLICCompeting Interests: Neil D. Paveley and employed by and leads the crop protection group of ADAS UK ltd, which is a commercial research organisation. The group’s research is funded by government, levy and industry. Neither the group, nor the author, benefit from patents, development of products or marketed products which result from the research. The research reported here was funded by government and agricultural research levy. This commercial affiliation does not alter our adherence to PLOS ONE policies on sharing data and materials. Conceptualization: JK F. van den Bosch NP F. van den Berg. Formal analysis: JK. Funding acquisition: F. van den Bosch NP F. van den Berg. Methodology: JK F. van den Bosch NP F. van den Berg. Project administration: F. van den Bosch. Software: JK JH F. van den Berg. Supervision: F. van den Bosch F. van den Berg. Validation: JK F. van den Bosch F. van den Berg. Visualization: JK JH. Writing – original draft: JK F. van den Bosch NP F. van den Berg. Writing – review & editing: F. van den Berg. * E-mail: femke.vandenberg@rothamsted.ac.uk29 8 2016 2016 11 8 e016188714 4 2016 12 8 2016 © 2016 Kitchen et al2016Kitchen et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.For the treatment of foliar diseases of cereals, fungicides may be applied as foliar sprays or systemic seed treatments which are translocated to leaves. Little research has been done to assess the resistance risks associated with foliar-acting systemic seed treatments when used alone or in combination with foliar sprays, even though both types of treatment may share the same mode of action. It is therefore unknown to what extent adding a systemic seed treatment to a foliar spray programme poses an additional resistance risk and whether in the presence of a seed treatment additional resistance management strategies (such as limiting the total number of treatments) are necessary to limit the evolution of fungicide-resistance. A mathematical model was developed to simulate an epidemic and the resistance evolution of Zymoseptoria tritici on winter wheat, which was used to compare different combinations of seed and foliar treatments by calculating the fungicide effective life, i.e. the number of years before effective disease control is lost to resistance. A range of parameterizations for the seed treatment fungicide and different fungicide uptake models were compared. Despite the different parameterizations, the model consistently predicted the same trends in that i) similar levels of efficacy delivered either by a foliar-acting seed treatment, or a foliar application, resulted in broadly similar resistance selection, ii) adding a foliar-acting seed treatment to a foliar spray programme increased resistance selection and usually decreased effective life, and iii) splitting a given total dose—by adding a seed treatment to foliar treatments, but decreasing dose per treatment—gave effective lives that were the same as, or shorter than those given by the spray programme alone. For our chosen plant-pathogen-fungicide system, the model results suggest that to effectively manage selection for fungicide-resistance, foliar acting systemic seed treatments should be included as one of the maximum number of permitted fungicide applications. http://dx.doi.org/10.13039/501100000268Biotechnology and Biological Sciences Research CouncilKitchen James L. Chemicals Regulation Directoratehttp://dx.doi.org/10.13039/501100000277Department for Environment, Food and Rural AffairsJK and NP received support from the Agriculture and Horticulture Development Board (project RD2012-3801) and the UK Chemicals Regulation Directorate of the Health and Safety Executive and the Department for Environment, Food and Rural Affairs (project PS2728). JH, FvdBerg and FvdBosch received support from the Biotechnology and Biological Sciences Research Council of the United Kingdom. ADAS provided support in the form of salaries for authors [NP], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section. Data AvailabilityAll relevant data are within the paper and its Supporting Information files.Data Availability All relevant data are within the paper and its Supporting Information files. ==== Body Introduction Foliar plant diseases can cause significant yield loss. In wheat, for example, the disease Septoria tritici leaf blotch, caused by the fungus Zymoseptoria tritici, can reduce grain yields by up to 30–40% [1]. To reduce disease-induced yield loss, growers apply fungicides that reduce the rate of epidemic progress. Fungicides can be applied directly to the foliage of the crop, protecting it from infection by fungal spores. Systemic fungicides, which are taken up by the plant and distributed through the leaf and other tissues, reduce pathogen infection, increase the pathogen’s latent period and reduce spore production. This makes them particularly effective at slowing down epidemics. Such fungicides can in some cases also be added to a seed coating. After germination the fungicide is then gradually taken up by the plant and distributed into the growing leaves. Therefore, the fungicide from a seed coating treatment is more active in the lower part of the canopy as compared to the fungicides applied with foliar sprays later in the crop growing season. Seed treatments with foliar-active systemic action can be attractive because they provide additional disease control and allow some flexibility in the subsequent foliar treatments program. For example, a seed treatment can provide insurance if an early foliar treatment is delayed or missed [2]. Fungicides affect one or more biochemical pathways in the pathogen hampering its growth. The QoI fungicides, for example, affect mitochondrial respiration, thereby shutting down the pathogen’s energy source. Fungicides that affect the same enzyme (target site) within a biochemical pathway are said to have the same mode-of-action (MOA). Fungal pathogens can develop resistance/insensitivity to a MOA, rendering the active substances within that group of fungicides less effective. Such resistance develops due to one or more mutation (or other genetic changes) in the pathogen’s genome. Systemic fungicides, which usually act against a single target site, are more prone to resistance than non-systemic fungicides which often act against multiple sites [3]. For example, one single mutation in the mitochondrial genome causes complete resistance to all QoI fungicides [4]. To prevent or at least delay the build-up of resistance in pathogen populations manufacturers in collaboration with regulatory authorities often put resistance management measures in place when a fungicide is introduced onto the market. Resistance management normally is a set of rules for the treatments program. For example, the number of applications per growing season can be restricted or the fungicide is marketed as a mixture with a fungicide with another mode of action [5]. For foliar application programs there is a considerable body of evidence showing that application programs have an effect on selection for fungicide resistance. Both laboratory and field experiments (summarized in [6]) and modelling studies (summarized in van den Bosch and Gilligan [7]) are published. This body of evidence has led to a clear set of recommendations on resistance management [8, 9]. For example, it is now generally accepted that mixing two fungicides with different modes of action reduces the rate of build-up of resistance. Surprisingly little is known about the selection pressure imposed by seed treatments. Only one study has been published on the rate of fungicide resistance development due to seed treatments [10] as compared to over 70 experimental studies on foliar treatments [6]. Moreover, virtually nothing is known about the selection pressure due to seed treatments as compared to foliar treatments. The temporal and spatial differences in the activity of seed compared to foliar treatments, as discussed above, may lead to significantly differences in selection pressure. However, based on results from the only published experiment on the development of resistance to triademinol in powdery mildew of barley [10] concluded that the selection pressures were similar. It is of key importance in the development of effective resistance management programs to understand and be able to predict the relative rate of selection for fungicide resistance by foliar and seed treatments. Manufacturers will continue to develop and register seed treatments when they add to effective disease control and are cost effective. A case in point is the recent development of the SDHI fungicides that can be applied as foliar as well as seed treatments. A decision needed to be made on resistance management for fungicide programs incorporating seed treatments. The Fungicide Resistance Action Committee (FRAC) decided that a seed treatment used against any foliar pathogen should be counted as one of the maximum two SDHI treatments per growing season [11]. This quite restrictive decision was taken in order to be precautionary in protecting future activity of this MOA group, given the lack of evidence. The objective of this project was to build a model and use it to analyze the development of fungicide resistance with both foliar and seed treatments and to address the knowledge gaps described above. We parameterized the model for Septoria leaf blotch controlled by a fungicide representative of the SDHIs. In developing this model we ran into two significant problems. Firstly, there is little quantitative information in the literature about how plants take up fungicides from the seed coating. Secondly, few experiments on the efficacy of seed treatments against foliar pathogens are published, and none have been published on the efficacy of foliar-acting SDHI fungicides. This makes parameterizing the seed treatment elements of the model difficult. We addressed these problems as follows: We developed a model incorporating two very different methods of fungicide uptake from the seed coating. Each can be switched on or off. We developed two widely contrasting parameterizations for the seed treatment dynamics. Each question posed to the model was then analyzed for all four combinations of uptake mechanism and parameterization. Combined structural and numerical sensitivity analyses were conducted to show how critically results depend on the assumptions regarding the uptake mechanism and parameter values. The two main questions we focus on in this paper are: Does adding a seed treatment to the currently allowed two foliar applications affect resistance development more or less than adding a third foliar application? Does using a seed treatment instead of one of the foliar applications affect resistance development differently as compared to using two foliar applications? Materials and Methods Model Overview The model was used to simulate a population of the pathogen Zymoseptoria tritici, comprised of a fungicide-sensitive and a fungicide-resistant strain on winter wheat under selection pressure from a fungicide treatment program comprising a foliar-acting seed treatment and/or foliar applied treatments. The Canopy The wheat crop canopy growth model developed is an extension of the model described by van den Berg, van den Bosch [12] and was used to simulate the growth and senescence of eleven explicit leaf layers (between nine and fourteen layers can be produced depending on sowing date and environment [13]). Following the convention in previous resistance articles the flag leaf as was numbered as leaf 1 and the bottom leaf as leaf 11. The leaf area of each leaf layer was measured in units of area index, defined as the ratio of the total leaf area to ground area [13]. The life cycle of each leaf layer i contained three phases: a growth phase, where leaf area increased according to a monomolecular function to a maximum; a lag phase, where leaf area remained constant, and a senescence phase, where leaf area decreased due to necrosis (Fig 1). Dead leaf tissues were removed from the simulation. 10.1371/journal.pone.0161887.g001Fig 1 Graphical representation of the 11 leaves in the crop canopy growth model in the absence of disease. Leaf lives overlap and were comprised of a monomolecular growth phase, a lag phase and a logistic senescence phase. The Pathogen The pathogen was modelled using compartments of latently infected tissue, comprised of non-sporulating mycelium, and infectious tissue, carrying sporulating fruiting bodies (Fig 2). At the start of the crop growing season the epidemic was initiated by primary inoculum, made up of asco-spores, produced by ascii on dead stuble left in the field after the previous season’s harvest. The primary inoculum gave rise to lesions producing pycnidiospores. During the rest of the crop growing season the pycnidiospores produced a series of asexual generations that formed the secondary inoculum of the epidemic. This secondary inoculum was dispersed by rain splash. Transmission of inoculum between leaf layers was reduced through stem extension, which led to a greater separation between the leaf layers. 10.1371/journal.pone.0161887.g002Fig 2 Flow diagram of the disease sub-model for leaf layer i. Healthy leaf area not lost by senescence (rate σ(t)) was infected at rate ε by primary inoculum (ascospores) and at rate ρ from secondary inoculum (pycnidiospores). The mean infectious period was 1/μ and the mean latent period was 1/ δ. The pathogen model included two strains; a fungicide sensitive strain, S, and a fungicide resistant strain, R. Parameters δ, ρ and ε for the sensitive strain were reduced according to the fungicide dose response functions. The Fungicides Latent and infectious tissues belonged to either a fungicide-sensitive or a fungicide-resistant strain of the pathogen. The latter strain was assumed to be unaffected by the fungicide within the range of doses permitted and infected new leaf tissues at the same rate in the presence or absence of the fungicide. Use of the fungicide caused selection for the fungicide-resistant strain. The fungicide was applied as a T1 and/or a T2 foliar spray, applied at the full emergence of the third eventual leaf or the flag leaf, respectively, or as a foliar-acting seed treatment. The fungicide was systemic and exhibited both protectant (transmission rate limiting) and eradicant (latent period prolonging) activity towards the fungicide-sensitive strain. The systemic seed treatment fungicide was initially present on, or around, the seed and then gradually translocated to the leaves. The fungicide concentration present in each leaf decayed exponentially over time due to breakdown from exposure to light and plant catabolism. Fungicide Effective Life To compare the effect of selection pressure between foliar sprays and systemic seed treatments, we used the fungicide effective life. The effective life is the number of consecutive years the fungicide treatment program was able to maintain effective disease control. Hereby, loss of effective disease control was defined as a reduction in healthy area duration (HAD), as measured on leaves 1–3, greater than 5%. We used this quantification of effective disease control because for wheat HAD is closely correlated with yield [14]. Model Description Plant growth dynamics We denote by subscript i the ith leaf layer. Leaf layers emerged in reverse chronological order, thus leaf layer i = 11 was the first leaf layer to emerge and leaf layer i = 1 was the flag leaf layer. Each leaf layer began growth at t = tinitiation_i, was fully emerged at t = temergence_i, began senescence at t = tsenes_i and was removed from the simulation at t = tdeath_i. The total leaf layer area index, Ai, grew according to a monomolecular function [15] at rate g and reached an asymptote at the maximum area index for leaf layer i, Amax_i: dAidt={0t<tinitiation_ig(Amax_i−Ai)t<tsenes_i0t≥tdeath_i(1) The healthy area index (HAI) of each leaf layer, Hi, was comprised of infection-free photosynthetic leaf tissue and in the absence of disease was equal to Ai up until t = t_senes_i, after which Hi decreased at rate σi(t) until t = tdeath_i: dHidt={0t<tinitiation_idAidtt≤t<tsenes_i−σi(t)Hitsenes_i≤t<tdeath_i(2) The senescence rate of leaf layer i, σi(t), was calculated by: σi(t)=es(t−tdeath_i) tsenes_i≤t<tdeath_i(3) leading to a sigmoidal decline until tdeath_i, at which point the leaf layer was removed from the simulation. Pathogen dynamics The epidemic was initiated through infection by wind-blown ascospores of the fungicide-sensitive and fungicide-resistant strains, which were produced by pseudothecia on plant debris. A flow diagram of the epidemic model is given in Fig 2. The rate of influx of ascospores, X(t), is given by: X(t)=ηt2e−λt(4) where η and λ are parameters (Fig 3). When multiplied with the ascospore deposition rate and the ascospore infection efficiency, γ, our ascospore transmission rate, ε(t), was obtained: ε(t)=γX(t)(5) 10.1371/journal.pone.0161887.g003Fig 3 Seasonal ascospore spore concentrations. The peak of the function was during the winter (0–1200 degree days) and declined to 1% of the peak value at the time of death of leaf 5 (2094 degree days). The transmission rate of fungicide-sensitive ascospores was reduced according to the concentration of fungicide at time t on leaf layer i and was thus denoted by εSi(t). It was assumed that fungicide-resistant ascospores were fully resistant to the fungicide and the transmission rate of fungicide-resistant ascospores was denoted by εRi. Parameter θ0 denotes the initial proportion of ascospores which were resistant to the fungicide. During the simulation, the proportion of fungicide-resistant ascospores released from pseudothecia on plant debris, θ, was calculated at the end of every growing season according to the fraction of fungicide-resistant infectious leaf tissue on the top five leaves: θ=∑i=15IRi∑i=15ISi+∑i=15IRi(6) ISi and IRi are the infectious leaf tissue occupied by fungicide-sensitive and fungicide resistant infectious lesions, respectively. Once initial infection as a result of influx of ascospore comprising primary inoculum had occurred the epidemic was driven by infection from pycnidiospores arising from mycelium on the remaining active leaf layers. Each infectious lesion produced a constant number of pycnidiospores per time unit. The transmission rate parameter for pycnidiospores, ρ, was found by multiplying the spore production rate with the infection efficiency. The transmission rate for fungicide-sensitive pycnidiospores was reduced according to the fungicide concentration on leaf layer i at time t and was denoted by ρSi(t), whereas the transmission rate for fungicide-resistant pycniodiospores on leaf layer i was unaffected by fungicide and was denoted by ρRi. Unlike wind-dispersed ascospores, pycnidiospores were splash dispersed between leaf layers [16], and the number of pycnidiospores that were dispersed between leaf layers was reduced by the distance between the layers. Stem extension only occured between leaf layers 1 to 4. We denoted the probability of splash dispersal between leaf layer j and leaf layer i as Pj,i(Dj,i), which was a function of the distance between leaf layers j and i, Dj,i: Pj,i(Dj,i)={1i=je−σdownDj,ii>je−σupDj,ii<j(7) where σdown and σup were parameters for the ease of downward and upward splash dispersal, respectively. For each leaf layer i, Pj,i(Dj,i) was summed over every active leaf layer j. We denoted the number of active leaf layers for any given t as u. By including both sources of inoculum, Eq (2) was extended to: dHidt={0t<tinitiation_idAidt−(HiAi)(ρSi(t)∑j=1uPj,i(Dj,i)ISj+ρRi∑j=1uPj,i(Dj,i)IRj+(1−θ)εSi(t)+θεRi )tinitiation_i≤t<tsenes_i−(HiAi)(ρSi(t)∑j=1uPj,i(Dj,i)ISj+ρRi∑j=1uPj,i(Dj,i)IRj+(1−θ)εSi(t)+θεRi )−σi(t)Hitsenes_i≤t<tdeath_i(8) Latently infected leaf tissue transitioned to infectious lesions at rate δ and was reduced due to the eradicant properties of the fungicide on leaf layer i at time t. Hence, the rate with which latently infected leaf tissue transitioned to infectious lesions was denoted by δSi(t) and δRi for the fungicide-sensitive and fungicide-resistant pathogen strains, respectively. Following van den Berg, van den Bosch [12] and Cunniffe, Stutt [17] we introduced m latently infected compartments. This resulted in a more realistic gamma distributed latent period compared to the exponentially distributed latent period that would have arisen from one compartment [17]. The rate at which latently infected leaf tissue transitioned between the m latent compartments for the fungicide-sensitive and resistant strains was hence mδSi(t) and mδRi, respectively. Like healthy leaf tissue, latently infected leaf tissue senesced at rate σi(t). Eqs 9 & 10 define our differential equations for the growth of the first latently infected leaf tissue compartment for the fungicide-sensitive and fungicide-resistant strain, respectively: dL1Sidt=(HiAi)(ρSi(t)∑j=1uPj,i(Dj,i)ISj+(1−θ)εSi(t))−L1Si(mδSi(t)+σi(t))(9) dL1Ridt=(HiAi)(ρRi∑j=1uPj,i(Dj,i)IRj+θεRi )−L1Ri(mδRi+σi(t))(10) The differential equations for each remaining latently infected leaf tissue compartment n where n ϵ {2, …, m} are defined for the fungicide-sensitive and fungicide-resistant strain in Eqs 11 and 12. dLnSidt=m(L(n−1)Si−LnSi)δSi(t)−σi(t)LnSi(11) dLnRidt=m(L(n−1)Ri−LnRi)δRi−σi(t)LnRi(12) Infectious leaf tissue of the fungicide-sensitive Eq (11) and fungicide-resistant Eq (12) strains grew due to the influx of latently infected leaf tissue from latent compartment m at rate mδSi(t) and  mδRi and was removed at rate μ: dISidt=mδSi(t)LmSi−μISi(13) dIRidt=mδRiLmRi−μIRi(14) Foliar spray dynamics The dose of the fungicide that was intercepted by leaf layer i, Fi, at t = tspray, was measured in units of mg m-2 of leaf area. In analogy with the transmission of light through a turbid medium (Beer-Lambert law), the dose intercepted by leaf layer i increased with its AI and decreased according to the product of the AI of each leaf layer j above that intercepts the sprayed fungicide: Fi=F0(1−e−τAi(tspray))(∏j=1i−1e−τAj(tspray)) tinitiation_j≤t<tdeath_j(15) where τ was the angle of leaf layer i, measured as a projection onto a horizontal surface, which ranged from zero when fully vertical to 1 at a fully horizontal projection. F0 was the total dose in mg m-2 of ground that was sprayed onto the field at t = tspray. The fungicide concentration within leaf layer i, measured in mg m-3 of leaf volume at time unit t − tspray, equal to ffoliar_i(t), was then calculated by: ffoliar_i(t)=FiqAie−υ(t−tspray)(16) where q represented leaf thickness, and υ was the fungicide breakdown rate. Seed treatment dynamics We modelled a continuous flow of systemic fungicide from the treated seed coating into the leaf layers where it accumulated and subsequently decayed. We did not simulate export of fungicides from leaf layers. The initial seed treatment fungicide dose was a quantity in mg per seed. The flow of seed treatment fungicide from the seed coating into each leaf layer was calculated in units of mg per time unit, and the dose that reached each leaf layer i was converted into a concentration of mg m-3 of leaf volume, which was then summed with the foliar spray concentration to determine the protectant and eradicant effects of the fungicide on the pathogen. The systemic fungicide that was within the seed reservoir, Nseed, was depleted over time at rate β(t): dNseeddt=−β(t)Nseed(17) As discussed in the introduction we used two models of seed treatment uptake. In the first model we assumed a constant rate of uptake, β(t) = β. In the second model, the seed treatment fungicide was drawn through the plant xylem resulting from a transpirational pull at rate β(t), which was a sigmoidal function of time: β(t)=b+yceatceat+(1−c)(18) where b and y are parameters for the baseline and maximum transpiration rate, respectively, and c and a are growth parameters. We assumed that the influx of seed treatment fungicide from the seed coating into each leaf layer i, zi, was proportional to the fraction of the leaf layer’s healthy leaf tissue, Hi, relative to the total healthy leaf tissue across all leaf layers, which was equal to: zi=βNseedHi∑j=1uHj tinitiation_j≤t<tdeath_j(19) The systemic fungicide dose accumulated in leaf i, Nlayer_i, increased at rate zi and decreased from breakdown and leaf senescence at rates b and σ(t), respectively (Fig 4): dNleaf_idt=zi−bNleaf_i−σ(t)Nleaf_i(20) 10.1371/journal.pone.0161887.g004Fig 4 Flow diagram for the seed treatment fungicide sub-model. The fungicide in the seed coating (Nseed) decreased at rate β and moved into each leaf according to the product of β and the proportion of the leaf’s healthy area index relative to the total canopy healthy area. Seed treatment loss from each leaf was due to senescence and catabolism, at rates σ and υ respectively. The concentration of systemic fungicide applied as a seed treatment in mg per sown seed was given by: fseed_treatment_i=Nleaf_iqhi(21) Where hi is the total area in m2 for leaf i per m2 ground area and q is the leaf thickness in m. See Fig 5 for a graphical representation of how the two different seed treatment uptake models affect the seed treatment fungicide depletion. 10.1371/journal.pone.0161887.g005Fig 5 Seed treatment fungicide depletion for the constant versus transpiration-based seed treatment uptake model. Both models were parameterized such that 99.99% of the fungicide has been depleted from the seed at the time that leaf layer 5 is dead. Dose response curves Our dose response equations were of exponential type and not traditional probit-log models [18], as the exponential curve fitted to the observed data [19]. The modelled fungicide had both protectant activity, which reduced the ascospore and pycnidiospore transmission rates εi and ρi, and eradicant activity, which increased the length of the latent period, as given by 1 / δi. However, the eradicant activity of systemic fungicides was only effective during the early latent stages of infection [20, 21], thus following the approach of van den Berg, van den Bosch [12] only lesions that are in the first half of the latent period are affected by the eradicant action of the fungicide (i.e. only latent compartments n that satisfy the condition 1 ≤ n ≤ (m / 2) are affected by the fungicide). The dose response functions had two parameters: α, which we defined as the maximum proportional reduction in the target pathogen parameter ω, where ω ϵ {ρ, δ, ε}; and k, which served as the dose response curve shape parameter. αω(t)=αmax_ω(1−e−kω(ffoliar_i(t)+fseed_treatment))(22) The pathogen parameters of the fungicide-sensitive strain were then deduced according to: ρSi(t)=ρ(1−αρ(t))(23) εSi(t)=ε(1−αε(t))(24) δSi(t)={δ(1−αδ(t))in Ln with n≤m/2δin Ln with n > m/2(25) The target pathogen parameters were left unchanged for the fungicide-resistant strain, which we assumed to exhibit absolute resistance. Example time courses of the sensitive and resistant pathogen strains when fungicide treatments consist of a T1 foliar treatment only or a seed treatment only are given in Figs 6 and 7, respectively. 10.1371/journal.pone.0161887.g006Fig 6 Time course of the sensitive and resistant pathogen strain for a T1 foliar treatment. The simulation was run for a low fungicide breakdown rate and a constant seed treatment uptake model. Blue and red lines indicate fungicide-sensitive and fungicide-resistant area index, respectively. Top panel: leaf layer 11. Middle panel: leaf layer 5. Bottom panel: leaf layer 1. A T1 spray was applied at 20 mg/m2 and the simulation was run for 10 growing seasons. 10.1371/journal.pone.0161887.g007Fig 7 Time course of the sensitive and resistant pathogen strain for a seed treatment. The simulation was run for a low fungicide breakdown rate and a constant seed treatment uptake model. Blue and red lines indicate fungicide-sensitive and fungicide-resistance area index, respectively. Top panel: leaf layer 11. Middle panel: leaf layer 5. Bottom panel: leaf layer 1. A seed treatment was applied at 4.5 mg/m2, which provides a HAD gain that approximates that of a T1 spray, and the simulation was run for 10 growing seasons. Definition of loss of effective control To compare the selection for fungicide resistance within the simulated treatment regimens we calculated the fungicide effective life [7, 22, 23], which is defined as the number of years after first introduction over which the fungicide is able to maintain effective disease control. Hereto, we first calculated the healthy area duration (HAD) of the crop canopy [14] according to the equation: HAD=∫t=2100t=3100∑i=13(Hi+∑n=1m(LnSi+LnRi)) dt(26) The integral in Eq 26 was calculated using the healthy and latently infected leaf tissue of the top three leaf layers, from anthesis at Zadock’s GS61 until the end of the simulation at GS91. HAD was calculated numerically according to the method described in [24], as implemented in the NAG Numerical Library [25]. For wheat the HAD experienced during the yield forming period is closely correlated with final yield [14], and was used to determine whether a fungicide application was still controlling the epidemic. In the absence of disease we calculated a reference value, HAD0, and defined that a HAD loss of 5% or greater, relative to HAD0, indicated a loss of effective disease control. Either an inadequate fungicide programme (insufficient treatments and/or dose per treatment) or selection for fungicide-resistance will have caused effective disease control to fail. The effective life was then the number of consecutive growing seasons before disease control was lost. Parameter Values For a detailed description of the estimation of all model parameters we refer the reader to S1 file. The parameter values are summarized in Table 1. 10.1371/journal.pone.0161887.t001Table 1 Parameter symbols, descriptions, values and units. Parameter name Definition Value Units g Leaf growth rate 0.034 t-1 s Leaf senescence rate 0.05 t-1 τ Measure of leaf projection onto horizontal surface 0.77 Dimensionless tanthesis_early Time of early anthesis 2100 t tharvest Time of harvest 3100 t q Leaf thickness 0.001 m Dmax Maximum stem extension between leaves 1–4 10 cm η Ascospore influx coefficient 1 t-3 λ Ascospore influx decay rate 0.0035 Dimensionless θ0 Initial proportion of fungicide-resistance in population 1.00E-05 Dimensionless σup Rate of change of reduction of inoculum from upward splash dispersal 0.1 Dimensionless σdown Rate of change of reduction of inoculum from downward splash dispersal 0.01 Dimensionless 1/μ Infectious period 456 t 1/δ Latent period 244 t m Number of latent compartments 10 Dimensionless γ Infection efficiency per individual ascospore 4.00E-10 HAI spore-1 ε Ascospore transmission rate Variable HAI t-1 ρ Pycnidiospore transmission rate 0.007 t-1 υlow Low fungicide breakdown rate 0.0046 Dimensionless υhigh High fungicide breakdown rate 0.009 Dimensionless β Seed treatment uptake rate 0.0055 t-1 αδ_high; αε_high; αρ_high; Maximum reduction in pathogen parameters from fungicides with a high breakdown rate 0.5 Dimensionless αδ_low; αε_low; αρ_low; Maximum reduction in pathogen parameters from fungicides with a low breakdown rate 0.45 Dimensionless kδ_high; kε_high; kρ_high; Dose response curve shape parameter for fungicides with a high breakdown rate 0.003 Dimensionless kδ_low; kε_low; kρ_low; Dose response curve shape parameter for fungicides with a low breakdown rate 0.0025 Dimensionless b Intercept of transpiration based seed treatment uptake 0.002 t-1 c Coefficient for transpiration based seed treatment uptake 0.0015 Dimensionless y Asymptote for transpiration based seed treatment uptake 0.027 t-1 a Rate of increase for transpiration based seed treatment uptake 0.003 Dimensionless lshort Number of leaf layers used for calculation of resistance proportion, with a short ascospore longevity 5 Dimensionless llong Number of leaf layers used for calculation of resistance proportion, with a long ascospore longevity 11 Dimensionless Here we only describe the estimation of parameter values used to model the systemic seed treatment efficacy. The initial amount of fungicide in the seed coating, Nseed, in units of mg seed-1 was estimated by searching for values that resulted in a specific level of disease control. In our simulations, we used two different parameterizations: Parameterization 1 The initial seed treatment dose was set to provide the same antifungal effect as a T1 spray. This was achieved by adjusting Nseed until the calculated HAD at the end of the first growing season equaled the HAD value obtained from a T1 spray. This parameterization provided the upper bound for the disease control provided by a systemic seed treatment, as it is unlikely that a systemic seed treatment could provide a stronger reduction in disease severity than a T1 spray. The estimated doses varied according to the fungicide breakdown rate and the seed treatment uptake model in use (S1 Table). Parameterization 2 A dataset was obtained from Parker and Lovell [26], which contained spore-washing data in spores ml-1 on leaves of winter wheat that were infected by Septoria leaf spot, comparing untreated plots with plots treated with a seed treatment (fluquinconazole; product name Jockey). Areas under the disease progress curve (AUDPC) values for the spore washing data were calculated by numerically integrating the spore washing values using the method described in [24], as implemented in [25]. The percentage AUDPC remaining after treatment was observed and recorded for each data point from both sets. Averaging over all data points for both sets yielded average reductions in AUDPC values of 60% after treatment. Values of Nseed were then adjusted in the model to obtain a 60% reduction in AUDPC (S2 Table). Results Selection for Fungicide-Resistance The model was used to calculate fungicide-sensitive and fungicide-resistant infectious leaf tissue over ten growing seasons on leaf layers 1, 5 and 11 (representing the upper, mid- and lower canopy, respectively, after applying either a solo T1 foliar spray or a solo systemic seed treatment (Figs 6 and 7, respectively). The fungicide in these calculations was parameterized with a low breakdown rate, and a constant uptake model for the systemic seed treatment fungicide was used (refer to Table 1 for values). To provide an appropriate comparison between both treatments, the input seed treatment dose was parameterized to approximate the HAD gain of a T1 spray (S1 Table). As expected, applying a seed treatment led to the fungicide-sensitive infectious leaf tissue being reduced most in the lower canopy, particularly on leaf layer 11. However, the fungicide-sensitive infectious leaf tissue on leaf layers 1 and 5 were affected to a similar extent when treated with a systemic seed treatment as compared to a T1 foliar spray. There were no substantial differences in the growth of the fungicide-resistant strain after either treatment, indicating that the selection pressure for fungicide-resistance was similar in both cases. To obtain a quantitative measure of the selection pressure, median selection ratios [23] were calculated over the ten growing seasons plotted in Figs 5 and 6. The selection ratio represents the factor by which the frequency of the resistant strain is multiplied over one growing season. The selection ratios were calculated per growing season over the ten growing seasons and then the median of the resultant distribution was calculated. Median selection ratios were calculated at 4.37 and 4.0 for Figs 6 and 7 respectively (S3 Table), which suggested a slightly larger rate of increase of the fungicide-resistant strain when a solo T1 spray was applied. However, this trend was reversed when the fungicide half-life was increased (llong rather than lshort), leading to median selection ratios for a solo T1 spray and a seed treatment of 3.61 and 3.86, respectively. Effective Fungicide Lives Effective lives were compared between four treatment programmes over a range of doses. The regimens were labelled in results Tables 2–4 as the following: ST + T1: A seed treatment and a foliar spray at T1 were applied each growing season. ST + T2: A seed treatment and a foliar spray at T2 were applied each growing season. T1 + T2: Two foliar sprays at T1 and T2 were applied each growing season. ST + T1 + T2: A seed treatment and two foliar sprays at T1 and T2 were applied every growing season. In this regimen the doses of the foliar treatments at T1 and T2 were equal. 10.1371/journal.pone.0161887.t002Table 2 Effective lives in the presence of low and high fungicide breakdown rates. ST dose foliar dose (per treatment) Low breakdown High breakdown ST+T1 ST+T2 ST+T1+T2 ST+T1 ST+T2 ST+T1+T2 0 0 - - - - - - 0 0.2 - - - - - - 0 0.4 - - 5 - - - 0 0.6 - - 4 - - 5 0 0.8 - - 4 - - 5 0 1 - - 4 - - 5 0.6 0 - - - - - - 0.6 0.2 4 5 4 - - 4 0.6 0.4 4 5 4 4 5 4 0.6 0.6 4 4 3 4 5 4 0.6 0.8 4 4 3 4 5 4 0.6 1 4 4 3 4 5 3 1 0 - - - - - - 1 0.2 4 5 4 4 4 4 1 0.4 4 4 3 4 5 4 1 0.6 4 4 3 4 5 4 1 0.8 4 4 3 4 4 3 1 1 4 4 3 4 4 3 Effective lives (in years) in the presence of high and low fungicide breakdown rates. All effective lives were calculated for the scenario of a constant seed treatment uptake rate model, and the seed treatment being parameterised such that it provides the same level of control as a T1 spray when applied at dose 1. ST refers to a seed treatment and T1 and T2 refer to foliar treatments at the full emergence of eventual leaf 3 and the flag leaf, respectively. The table has been truncated for brevity. Full tables, also containing results for dose 0.2, 0.4 and 0.8, are in S4 and S5 Tables. Dashes indicate simulations for which effective disease control was not achieved during the first growing season. 10.1371/journal.pone.0161887.t003Table 3 Effective lives for two seed treatment efficacy parameterizations. ST_dose foliar_dose (per treatment) ST = T1 ST = 40% of AUDPC ST_T1 ST_T2 ST_T1_T2 ST_T1 ST_T2 ST_T1_T2 0 0 - - - - - - 0 0.2 - - - - - - 0 0.4 - - 5 - - 5 0 0.6 - - 4 - - 4 0 0.8 - - 4 - - 4 0 1 - - 4 - - 4 0.6 0 - - - - - - 0.6 0.2 4 - 4 - - 5 0.6 0.4 4 5 4 - - 4 0.6 0.6 4 4 3 4 5 4 0.6 0.8 4 4 3 4 5 4 0.6 1 4 4 3 4 5 4 1 0 - - - - - - 1 0.2 4 5 4 - - 4 1 0.4 4 4 3 4 5 4 1 0.6 4 4 3 4 5 4 1 0.8 4 4 3 4 4 3 1 1 4 4 3 4 4 3 Effective lives (in years) for two different seed treatment (ST) efficacy parameterizations: 1) ST = T1, whereby the seed treatment provides the same level of control as a T1 spray and 2) ST = 40% of AUDPC, whereby the seed treatment provides a 60% reduction in AUDPC when applied at dose 1. All simulations were run for the scenario of a low fungicide breakdown rate and the transpiration-based seed treatment uptake model. The table has been truncated for brevity. Full tables, also containing results for dose 0.2, 0.4 and 0.8, are in S9 and S11 Tables. Dashes indicate simulations for which effective disease control was not achieved during the first growing season. 10.1371/journal.pone.0161887.t004Table 4 Effective lives for the constant versus transpiration-based seed treatment uptake model. ST_dose foliar_dose Constant uptake Transpiration-based uptake ST_T1 ST_T2 ST_T1_T2 ST_T1 ST_T2 ST_T1_T2 0 0 - - - - - - 0 0.2 - - - - - - 0 0.4 - - - - - - 0 0.6 - - 5 - - 5 0 0.8 - - 5 - - 5 0 1 - - 5 - - 5 0.6 0 - - - - - - 0.6 0.2 - - 4 - - 5 0.6 0.4 4 5 4 4 4 4 0.6 0.6 4 5 4 4 5 4 0.6 0.8 4 5 4 4 5 4 0.6 1 4 5 3 4 5 4 1 0 - - - - - - 1 0.2 4 4 4 4 5 4 1 0.4 4 5 4 4 5 4 1 0.6 4 5 4 4 4 4 1 0.8 4 4 3 4 4 3 1 1 4 4 3 4 4 3 Effective lives (in years) for two different seed treatment uptake models: constant and transpiration-based. All simulations were run for the scenario of a high fungicide breakdown rate and a seed treatment (ST) which at dose 1 provides the same level of control as a T1 spray. The tables have been truncated for brevity. Full tables, also containing results for dose 0.2, 0.4 and 0.8, are in S5 and S8 Tables. Dashes indicate simulations for which effective disease control was not achieved during the first growing season. High and low fungicide breakdown rates Effective lives for both a high and a low fungicide breakdown rate are shown in Table 2. The maximum seed treatment dose was set to provide an equivalent HAD gain to a T1 spray, and the constant uptake model was used. Note that the column ST+T1+T2 is the T1+T2 spray program for entries of the table where the seed treatment dose is zero. The table leads to three key conclusions for this set of simulations. Firstly, the ST+T1+T2 column shows that adding a seed treatment to a two foliar spray program shortens the effective life of the fungicide. This holds for both the low and the high fungicide breakdown rate. Secondly, the effective life of all spray programs that include a seed treatment is equal to or smaller than a spray program with two foliar sprays. Again this holds for low and high fungicide breakdown rates. Thirdly, comparing treatment programs with equal total fungicide dose used, the effective life of the spray program including a seed treatment is equal to or smaller than that of a program without a seed treatment. See also S4 and S5 Tables for this comparison in more detail. Clearly these qualitative conclusions are not affected by the fungicide decay rate showing that the conclusions are not sensitive to the decay rate parameter. In the following paragraphs we will not vary this parameter, but we have checked that the conclusions are not sensitive to the rate of fungicide decay. Maximum seed treatment dose value The effective lives for the two different seed treatment efficacy parameterizations in the presence of a low fungicide breakdown rate and the transpiration-based uptake model are given in Table 3. In the columns marked ‘ST = T1’ the seed treatment provides a HAD gain equal to that of a T1 spray. In the columns marked ‘ST = 40% of AUDPC’ the maximum seed treatment dose reduced the AUDPC by 60%. From Table 3 we can draw the same three qualitative conclusions as we reached from Table 2, (i) adding a seed treatment to a spray program with two foliar sprays reduces the effective life of the fungicide, (ii) the effective life of all spray programs that includes a seed treatment is equal to or smaller than a spray program of only two foliar applications, and (iii) comparing spray programs with equal total dose the effective life of the program including a seed treatment is equal to or smaller than that of the program without a seed treatment. These conclusions hold irrespective of the chosen efficacy of the seed treatment (ST = T1 or ST = 40% of AUDPC). Seed treatment uptake model Finally, in Table 4 we compare the effective life for the constant fungicide uptake model with those for the transpirational uptake model. Again the same set of three key conclusions relating to the effective lives obtained by the different fungicide treatment programs were present. These conclusions were therefore also not affected by the choice of fungicide uptake dynamics from the seed coating. This study showed that irrespective of the choice in the uptake dynamics of the fungicide from the seed coating, the efficacy of the seed treatment and the fungicide breakdown rate used, the key qualitative conclusions regarding the effective lives remained the same. Tables 2, 3 and 4 are only a subset of all possible permutations of uptake model, breakdown rate and seed treatment efficacy. However, we have calculated all possible combinations of these three aspects and present the results in S4–S11 Tables. The reader can verify from this supplementary material that the conclusions hold for any combination of the three factors. Ascospore longevity and length of period for HAD calculation There is little published information describing over what time period, or from what parts of the crop canopy, ascospores, which cause infection of crops in the following season, are produced. To test if the results were sensitive to this, in S12 Table, the effect of increasing the longevity of the simulated ascospores that remained in pseudothecia on crop debris left over from the previous season (and are therefore carried forward as the founder population for the following season) was extended by summing over all 11 leaves to calculate the proportion of fungicide-resistant individuals in the population (see Eq 6), and effective lives were calculated over the dose range. The aforementioned trends in the effective lives remained, therefore the model output did not seem sensitive to these changes. Discussion We developed a model to simulate epidemics of Zymoseptoria tritici on winter wheat controlled by applying a systemic fungicide to seed and foliage. We used the model to compare the selection pressures for fungicide resistance between both fungicide treatment types. Unlike foliar sprays, for which information on dose response curves are readily available, there is little information on the efficacy and the uptake dynamics of systemic seed treatments [27–30]. We therefore did a sensitivity analysis with respect to the parameter values for the systemic seed treatment model and combined them with a structural sensitivity analysis, using two different approaches to model the uptake dynamics of the fungicide from the seed coating. Our results show that the qualitative trends in the model output are insensitive to: (i) the value of the parameter scaling the seed treatment efficacy, (ii) the half life time of the fungicide, and (iii) the model description of the fungicide uptake by the plant from the seed coating. As further mentioned a range of other parameters were explored by sensitivity analysis and also showed that the conclusions were very robust to these parameter changes. The model outputs result in three key conclusions about the effect of seed treatment on fungicide resistance development as compared to foliar sprays: (i) Adding a seed treatment to a spray program with two foliar sprays reduces the effective life of the fungicide. (ii) The effective life of all spray programs that include a seed treatment is equal to or shorter than a spray program of only two foliar applications. (iii) Comparing spray programs with equal total dose the effective life of the program including a seed treatment is equal to or smaller than that of the program without a seed treatment. For Zymoseptoria tritici on wheat, and for SDHI type solo fungicides, the selection for fungicide resistance of a seed treatment is equal or at least comparable to that of a foliar spray, if the seed treatment efficacy is equal or comparable to that of a foliar spray. This finding is consistent with the current FRAC guideline concerning seed treatments of SDHI fungicides [31]. Our model predictions suggest that for epidemics of Z. tritici on winter wheat being treated by SDHI fungicides, there is no gain in the effective life that can be obtained from seed treatments compared to foliar treatments as the maximum effective life was consistently attained by two foliar sprays. However, we recognize that the effective fungicide life is only one aspect of the usefulness of a fungicidal seed treatment, and that systemic seed treatments may still be useful for controlling disease. The current FRAC guidelines recommended at most two SDHI treatments unless the following ‘risk modifiers’ were in place: i) if the epidemic is being propagated by a low-risk foliar pathogen or a seed or soil-borne pathogen, ii) if the SDHI is mixed with a different mode of action that is able to solely provide control, and iii) if the following foliar spray does not contain a SDHI [31]. The first of the modifiers is irrelevant for foliar-based epidemics of Zymoseptoria tritici. However, the second and third modifiers correspond to fungicide mixture and alternation strategies, and as suggested by FRAC may allow SDHI seed treatments to be combined with more than one foliar spray without significantly reducing the effective life. Supporting Information S1 File Model parameter derivation. (DOCX) Click here for additional data file. S2 File Seed treatment program code.–C++ code used within Visual Studio 2015. (DOCX) Click here for additional data file. S1 Table Estimated initial seed treatment dose (ST) resulting in % HAD losses (in year 1) similar to those achieved with a T1 foliar spray for a range of model scenarios. (DOCX) Click here for additional data file. S2 Table Estimated initial seed treatment dose (ST) leading to a 60% reduction in AUDPC for a range of model scenarios. (DOCX) Click here for additional data file. S3 Table Median selection ratios under three treatment regimes. (DOCX) Click here for additional data file. S4 Table Effective lives (in years) for a seed treatment that provides the same level of control as a T1 foliar treatment in the presence of the constant seed treatment uptake model and a low fungicide breakdown rate. (DOCX) Click here for additional data file. S5 Table Effective lives (in years) for a seed treatment that provides the same level of control as a T1 foliar treatmentin the presence of the constant seed treatment uptake model and a high fungicide breakdown rate. (DOCX) Click here for additional data file. S6 Table Effective lives (in years) for a seed treatment leading to a 60% reduction in AUDPC in the presence of a constant seed treatment uptake rate and a high fungicide breakdown rate. (DOCX) Click here for additional data file. S7 Table Effective lives (in years) for a seed treatment leading to a 60% reduction in AUDPC in the presence of a constant seed treatment uptake rate and a low fungicide breakdown rate. (DOCX) Click here for additional data file. S8 Table Effective lives (in years) for a seed treatment that provides the same level of control as a T1 foliar treatment in the presence of the transpiration-based seed treatment uptake model and a high fungicide breakdown rate. (DOCX) Click here for additional data file. S9 Table Effective lives (in years) for a seed treatment that provide the same level of control as a T1 foliar treatment in the presence of the transpiration-based seed treatment uptake model and a low fungicide breakdown rate. (DOCX) Click here for additional data file. S10 Table Effective lives (in years) for a seed treatment leading to a 60% reduction in AUDPC in the presence of the transpiration-based seed treatment uptake model and a high fungicide breakdown rate. (DOCX) Click here for additional data file. S11 Table Effective lives (in years) for a seed treatment leading to a 60% reduction in AUDPC in the presence of the transpiration-based seed treatment uptake model and a low fungicide breakdown rate. (DOCX) Click here for additional data file. S12 Table Effect of different ascospore initiation. (DOCX) Click here for additional data file. JK and NP received support from the Agriculture and Horticulture Development Board (project RD2012-3801) and the UK Chemicals Regulation Directorate of the Health and Safety Executive and the Department for Environment, Food and Rural Affairs (project PS2728). JH, FvdBerg and FvdBosch received support from the Biotechnology and Biological Sciences Research Council of the United Kingdom. ==== Refs References 1 Morais D , Laval V , Sache I , Suffert F . Comparative pathogenicity of sexual and asexual spores of Zymoseptoria tritici (Septoria tritici blotch) on wheat leaves . Plant Pathol . 2015 ;64 (6 ):1429 –39 . 10.1111/ppa.12372 2 Bartlett DW , Clough JM , Godwin JR , Hall AA , Hamer M , Parr-Dobrzanski B . The strobilurin fungicides . Pest Manag Sci . 2002 ;58 (7 ):649 –62 . 10.1002/ps.520 12146165 3 Grimmer MK , van den Bosch F , Powers SJ , Paveley N . Evaluation of a matrix to calculate resistance risk . Pest Manag Sci . 2014 ;70 :1008 –16 . 10.1002/ps.3646 24013934 4 Fernandez-Ortuno D , Tores JA , de Vicente A , Perez-Garcia A . Mechanisms of resistance to QoI fungicides in phytopathogenic fungi . Int Microbiol . 2008 ;11 (1 ):1 –9 . 18683626 5 Brent KJ H D . Fungicide Resistance in Crop Pathogens: How can it be managed? FRAC Monograph No 1 (second, revised edition ). Brussels : GIFAP ; 2007 . 6 van den Bosch F , Oliver R , van den Berg F , Paveley N . Governing Principles Can Guide Fungicide-Resistance Management Tactics . Annu Rev Phytopathol . 2014 ;52 (1 ):175 –95 . 10.1146/annurev-phyto-102313-050158 24848413 7 van den Bosch F , Gilligan CA . Models of Fungicide Resistance Dynamics . Annu Rev Phytopathol . 2008 ;46 (1 ):123 –47 . 10.1146/annurev.phyto.011108.135838 18680425 8 van den Bosch F , Fraaije BA , Oliver R , van den Berg F , Paveley N . The use of mathematical models to guide fungicide resistance management decisions In: Ishii H , Hollomon DW , editors. Fungicide resistance in plant pathogens . Tokyo : Springer ; 2015 p. 49 –62 . 9 van den Bosch F , Paveley N , Fraaije BA , van den Berg F , Oliver R . Evidence-based resistance management: a review of existing evidence In: Ishii H , Hollomon DW , editors. Fungicide resistance in plant pathogens . Tokyo : Springer ; 2015 p. 63 –76 . 10 Brent KJ C G , Hollomon DW , Hunter T , Locke T , Proven M . Factors affecting build-up of fungicide resistance in powdery mildew in spring barley . Neth J Plant Pathol . 1989 ;95 :31 –41 . 10.1007/BF01974282 11 Statement on SDHI fungicides and resistance risk in cereals issued by FRAG-UK 2015. Available: https://cereals.ahdb.org.uk/press/2015/december/16/statement-on-sdhi-fungicides-and-resistance-risk-in-cereals-issued-by-frag-uk.aspx. 12 van den Berg F , van den Bosch F , Paveley ND . Optimal Fungicide Application Timings for Disease Control Are Also an Effective Anti-Resistance Strategy: A Case Study for Zymoseptoria tritici (Mycosphaerella graminicola) on Wheat . Phytopathology . 2013 ;103 (12 ):1209 –19 . 10.1094/phyto-03-13-0061-r 23859011 13 Sylvester Bradley R , Berry P , Blake J , Kindred D , Spink J , Bingham I , et al HGCA Wheat growth guide . Caledonia House , London, UK : 2008 . 14 Waggoner PE B R . Defoliation, disease, and growth . Phytopathology . 1987 ;77 :393 –8 . 15 Thornley JHM , Johnson IR . Plant and crop modelling . Oxford : Clarendon Press ; 1990 . 16 Shaw MW . Assessment of Upward Movement of Rain Splash Using a Fluorescent Tracer Method and Its Application to the Epidemiology of Cereal Pathogens . Plant Pathol . 1987 ;36 (2 ):201 –13 . 10.1111/j.1365-3059.1987.tb02222.x 17 Cunniffe NJ , Stutt ROJH , van den Bosch F , Gilligan CA . Time-Dependent Infectivity and Flexible Latent and Infectious Periods in Compartmental Models of Plant Disease . Phytopathology . 2012 ;102 (4 ):365 –80 . 10.1094/Phyto-12-10-0338 22106830 18 Bliss CI . The Method of Probits . Science . 1934 ;79 (2037 ):38 –9 . 10.1126/science.79.2037.38 17813446 19 Lockley D, Clark WS. Fungicide dose-response trials in wheat: The basis for choosing ‘appropriate dose’. HGCA, 2005 Project report 373. 20 Paveley N, Blake J, Gladders P, Cockerell V. HGCA wheat disease management guide 2012. 21 Paveley ND , Lockley D , Vaughan TB , Thomas J , Schmidt K . Predicting effective fungicide doses through observation of leaf emergence . Plant Pathol . 2000 ;49 (6 ):748 –66 . 10.1046/j.1365-3059.2000.00518.x 22 Hobbelen PH , Paveley ND , Oliver RP , van den Bosch F . The Usefulness of Fungicide Mixtures and Alternation for Delaying the Selection for Resistance in Populations of Mycosphaerella graminicola on Winter Wheat: A Modeling Analysis . Phytopathology . 2013 ;103 (7 ):690 –707 . 10.1094/PHYTO-06-12-0142-R 23384858 23 Hobbelen PHF , Paveley ND , van den Bosch F . Delaying Selection for Fungicide Insensitivity by Mixing Fungicides at a Low and High Risk of Resistance Development: A Modeling Analysis . Phytopathology . 2011 ;101 (10 ):1224 –33 . 10.1094/Phyto-10-10-0290 21679038 24 Gill PE , Miller GF . An Algorithm for the Integration of Unequally Spaced Data . Compu J . 1972 ;15 (1 ):80 –3 . 10.1093/comjnl/15.1.80 25 The NAG Numerical Library. Available: www.nag.co.uk. 26 Parker SR, Lovell DJ. Quantifying the benefits of seed treatment for foliar disease control. In: Biddle AJ, editor. Seed Treatment: Challenges & Opportunities, Proceedings. British Crop Protection Council Symposium Proceedings. Farnham: British Crop Protection Council; 2001. p. 181–8. 27 Egorycheva MT , Burlakova SV . Effectiveness of pre-sowing seed dressing . Zashchita i Karantin Rastenii . 2009 ;(8 ):43 –4 . CABI:20103025991. 28 Lindstrom O . Mechanism of Liquid Seed Treatment—Vapor Action and Adhesion, Radioactive Studies of Initial Liquid Distribution, and Investigations with Radioactive Panogen Formulations . Journal of Agricultural and Food Chemistry . 1958 ;6 (4 ):283 –98 . 10.1021/Jf60086a003 . ISI:A1958WR76600006. 29 Lipps PE , Madden LV . Effect of Triadimenol Seed Treatment and Triadimefon Foliar Treatment on Powdery Mildew Epidemics and Grain-Yield of Winter-Wheat Cultivars . Plant Disease . 1988 ;72 (10 ):887 –92 . 10.1094/Pd-72-0887 . WOS:A1988Q687500017. 30 Poletine JP , Maciel CDD , da Silva TRB , Zanotto MD . Efficiency of seed treatment with fungicides in castor bean crop genotypes . Journal of Food Agriculture & Environment . 2012 ;10 (2 ):512 –6 . ISI:000305165300010. 31 FRAC. SDHI Guidelines—Cereals 2015. Available: http://www.frac.info/working-group/sdhi-fungicides/general-use-recommendations/cereals-and-soybeans.

PMC005xxxxxx/PMC5003397.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757136110.1371/journal.pone.0161151PONE-D-16-11188Research ArticlePhysical SciencesMaterials ScienceMaterials by AttributeCoatingsEngineering and TechnologyManufacturing ProcessesSurface TreatmentsCoatingsPhysical SciencesMaterials ScienceMaterials by StructureThin FilmsResearch and Analysis MethodsSpectrum Analysis TechniquesElectron Beam Spectrum Analysis TechniquesX-Ray Photoelectron SpectroscopyBiology and Life SciencesBiochemistryProteinsCytoskeletal ProteinsVimentinBiology and Life SciencesCell BiologyCellular TypesAnimal CellsConnective Tissue CellsOsteoblastsBiology and Life SciencesAnatomyBiological TissueConnective TissueConnective Tissue CellsOsteoblastsMedicine and Health SciencesAnatomyBiological TissueConnective TissueConnective Tissue CellsOsteoblastsBiology and Life SciencesBiochemistryProteinsContractile ProteinsActinsBiology and Life SciencesBiochemistryProteinsCytoskeletal ProteinsActinsBiology and Life SciencesBiotechnologyBiomaterialsPhysical SciencesMaterials ScienceBiomaterialsResearch and analysis methodsBioassays and physiological analysisBiochemical analysisColorimetric assaysMTT assayResearch and analysis methodsBioassays and physiological analysisBiochemical analysisEnzyme assaysMTT assayIn Vitro Biocompatibility of Si Alloyed Multi-Principal Element Carbide Coatings In Vitro Biocompatibility of Si Alloyed Multi-Principal Element Carbide CoatingsVladescu Alina 1Titorencu Irina 2Dekhtyar Yuri 3Jinga Victor 2Pruna Vasile 2Balaceanu Mihai 1Dinu Mihaela 1Pana Iulian 14Vendina Viktorija 3http://orcid.org/0000-0002-2543-5866Braic Mariana 1*1 National Institute for Optoelectronics, Magurele-Bucharest, Romania2 Institute of Cellular Biology and Pathology "Nicolae Simionescu" of the Romanian Academy, Bucharest, Romania3 Riga Technical University, 1Kalkustr, Rīga, Latvia4 Faculty of Physics, Bucharest University, Magurele-Bucharest, RomaniaMukherjee Amitava EditorVIT University, INDIACompeting Interests: The authors have declared that no competing interests exist. Conceived and designed the experiments: AV YD VJ M. Braic. Performed the experiments: AV IT MD IP VP VV M. Braic. Analyzed the data: AV YD VJ M. Balaceanu M. Braic. Contributed reagents/materials/analysis tools: AV YD VJ VV M. Braic. Wrote the paper: AV IT YD M. Balaceanu M. Braic. Provided expertise and editing: AV YD M. Balaceanu M. Braic. * E-mail: mariana.braic@inoe.ro29 8 2016 2016 11 8 e016115123 3 2016 1 8 2016 © 2016 Vladescu et al2016Vladescu et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.In the current study, we have examined the possibility to improve the biocompatibility of the (TiZrNbTaHf)C through replacement of either Ti or Ta by Si. The coatings were deposited on Si and 316L stainless steel substrates by magnetron sputtering in an Ar+CH4 mixed atmosphere and were examined for elemental composition, chemical bonds, surface topography, surface electrical charge and biocompatible characteristics. The net surface charge was evaluated at nano and macroscopic scale by measuring the electrical potential and work function, respectively. The biocompatible tests comprised determination of cell viability and cell attachment to the coated surface. The deposited coatings had C/(metal+Si) ratios close to unity, while a mixture of metallic carbide, free-carbon and oxidized species formed on the film surface. The coatings’ surfaces were smooth and no influence of surface roughness on electrical charge or biocompatibility was found. The biocompatible characteristics correlated well with the electrical potential/work function, suggesting a significant role of surface charge in improving biocompatibility, particularly cell attachment to coating's surface. Replacement of either Ti or Ta by Si in the (TiZrNbTaHf)C coating led to an enhanced surface electrical charge, as well as to superior biocompatible properties, with best results for the (TiZrNbSiHf)C coating. CNCS-UEFISCDIPN-II-ID-PCE-2011-3-1016http://orcid.org/0000-0002-2543-5866Braic Mariana POSINOVA-OPTIMA project, SMIS code 49164, contract no. 658/2014This work was funded by a grant of the Romanian National Authority for Scientific Research, CNCS-UEFISCDI, project number PN-II-ID-PCE-2011-3-1016. Part of the analysis were carried out by using the equipment acquired by the infrastructure project INOVA-OPTIMA SMIS code 49164, contract no. 658/2014. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Data AvailabilityAll relevant data are within the paper.Data Availability All relevant data are within the paper. ==== Body Introduction The superior biocompatibility of metallic alloys containing biocompatible metals was proved to derive from the high corrosion resistance and biocompatibility of the oxides formed at their surfaces, the oxidation process being intensified in body fluids [1]. However, health problems related to metallosis (release of metallic ions from the alloys in the surrounding tissue, resulting in adverse physiological effects that lead to implant failure) have been reported [2]. The value of bone-fixation devices and implants is over 44% of the overall biomedical devices market. Because of this the use of coatings to enhance the wear and corrosion resistance and biocompatible characteristics of the metallic implants has been the focus of intensive research work [3]. Up to now, various types of hard coatings have been proposed, mostly consisting of the “classical” binary or ternary transition metal compounds such as nitrides (TiN, ZrN, NbN, TiAlN, TiHfN, TiSiN) [4–6], carbides (TiC, ZrC, NbC, TaC) [7–11] or carbonitrides (TiCN, ZrCN, TiAlN) [5,12–15] that exhibit superior biocompatibility as compared to pure metals. Recently, a new class of multicomponent coatings, with already confirmed biocompatible qualities [16], has been developed. These coatings, based on the concept of high entropy alloys (HEA) [17] and commonly known as multi-principal element (MPE) coatings, were produced either as metallic or as MPE compound (nitride or carbide) films. These coatings contain at least five principal elements in almost equiatomic percentage and form either simple crystalline solid solutions or amorphous structures [17–19]. The properties of the coatings can be engineered by the proper choice of the constituents. Different valuable characteristics such as high hardness, stiffness and toughness, high thermal stability, hydrophobicity, super-elasticity, superior wear, corrosion and oxidation resistance, reported for MPE coatings, are determined by their high mixing entropy, reduced diffusion kinetics, severe lattice distortion, and “cocktail effect” [19][20]. Various MPE nitride coatings have been prepared (e.g. (TiHfZrVNb)N [21,22][20,21], (ZrTaNbTiW)N [23], (TiVCrZrHf)N [24,25]), while studies on MPE carbide or carbonitride coatings are limited (e.g. (CuSiTiYZr)C [26], (CrCuNbTiY)C [27] (AlCrTaTiZr)NxCy [28]). In our previous papers, we investigated the mechanical and tribological characteristics, as well as the corrosion resistance and biocompatibility of a MPE carbide coating, namely (TiNbZrTaHf)C, which proved to be a suitable protective coating for biomedical applications [16,29]. The goal of the present study was to examine the possibility to further improve the biocompatibility of this coating through substitution of one metallic constituent (either Ti or Ta) by Si. As reported earlier, Si addition to binary or ternary carbide or nitride coatings improves their mechanical, anticorrosive and tribological properties [30,31]. The biocompatibility of Si and SiC is also well documented in the literature, being demonstrated that the presence of Si in different biomaterials determines the proliferation and differentiation of human osteoblast-like cell and accelerates the osseointegration of metallic implants [32,33]. Since cell integration with the implant surface is influenced by surfaces roughness and electrical charge [34], the possible correlation between coatings biocompatibility and these factors was also explored. Considering the role of electrostatic interactions in many biological events, it should be mentioned that charged surfaces have been proposed as being conductive to tissue integration [35,36]. In orthopaedic and dental applications, the surface-charge of the implant plays an important role in determining a good adhesion of bone cells to implant and also bone mineralization at the bone-implant interface [37–39]. The surface charge was characterized at nano- and macroscopic scale by the electrical potential and the work function, respectively. In addition, elemental compositions, chemical bonds and roughness of the deposited coatings were analysed. A special attention was devoted to the in vitro biological investigation using the osteosarcoma cells, in order to reveal the effect of Ti or Ta replacement by Si on the coatings' biocompatibility. Material and Methods Preparation of coatings The coatings were prepared by magnetron sputtering using an ATC ORION unit (AJA Int.) equipped with 5 cathodes (2" diameter) [40]. The targets were made of pure Ti, Zr, Nb, Hf, Ta or Si (99.99% purity, from Kurt Lesker Comp.). All coatings were deposited simultaneously on two types of substrates: Si (111) square pieces (l = 20 mm), cut from wafers (Si-Mat Silicon Materials Comp.), and 316 L discs (3 mm thick, 20 mm diameter; Grant Metal SA). The 316L discs were progressively polished using different emery papers (up to 4000 grit) and then polished with 0.5 μm diamond suspension to a Ra roughness of 50 nm. All the substrates were ultrasonically cleaned in ethanol alcohol and flushed with dried nitrogen. Each coating was deposited in the same run on two Si and four stainless steel substrates and then on six stainless steel substrates, in order to get the necessary number of replicates for characterisation. The deposition chamber was initially pumped down to about 2×10−5 Pa, while the total gas (CH4 +Ar) pressure was of 0.67 Pa. Prior deposition, the samples were sputter cleaned with Ar+ (1keV) for 15 min. The deposition parameters were as follows: power applied to cathodes ~75 W (Ti), ~70 W (Zr), ~48 W (Nb), ~220 W (Hf), ~54 W (Ta); ~75 W (Si); total gas flow rate = 10 sccm; CH4/(CH4+Ar) flow rate ratio = 0.16; substrate bias voltage = –100 V; substrate temperature during deposition = 300°C. Deposition durations (120–140 min) were chosen to produce films of ~1.5 μm thick. Characterization of coatings The elemental composition was determined by energy-dispersive X-ray spectroscopy (EDS), using a Bruker Quantax 70 EDS system. For each type of coating, EDS measurements were performed on two replicates in five different areas of each one, the results being averaged (arithmetic mean) and the standard deviation (SD) was calculated. The phase composition and the preferred orientation were analysed by X-ray diffraction (XRD) technique using a Rigaku MiniflexII diffractometer with a CuKα radiation; measurements were carried out on one replicate of each coating. The chemical bonds in the coatings were investigated by X-ray photoelectron spectroscopy (XPS), using a VG ESCA 3MK II spectrometer using monochromatic X-rays (Al Kα(1486.61 eV)) and a hemispherical analyzer operated with constant pass energy. Survey spectra (low resolution, 1000 eV scan) were acquired using a pass energy of 50 eV with a step of 1 eV. High resolution spectra of the chemical species core levels were acquired over a smaller range (30 eV) with a resolution of 0.59 eV. The area of analysis was 300×700 μm2. All measurements were carried out in the analysis chamber in ultra-high vacuum conditions (~ 10−7 Pa). The XPS spectra were charge corrected to the binding energy of C1s line (285.0 eV). The deconvolution of XPS lines was performed using the Spectral Data Processor v 2.3 (SDP) software using the Gaussian-Lorentzian product [41]. The surface morphology was examined by scanning electron microscopy (SEM, XL-30–ESEM TMP) using one replicate for each type of coating. The surface roughness was measured at meso-scale on 316 L coated samples, by surface profilometry (Dektak—Bruker), on an area of (150 × 150) μm2. The surface roughness at nano-scale on Si coated samples was determined by an AFM Innova Bruker microscope, on an area of (3 × 3) μm2. The roughness measurements carried out for each type of coating were performed on two replicates in 5 different areas randomly chosen on each one, the results being averaged. The electrical potential was determined using the Scanning Kelvin Probe technique on a Kelvin Probe atomic force instrumentation (Solver–PRO47 microscope). The Kelvin probe force microscopy method was used for measuring the contact potential difference between the investigated sample and the tip of the atomic force probe [42], thus obtaining information on the Fermi level energy. Because the coatings with different compositions were measured with the same tip, the contact potential difference indicated the specific shift of the Fermi energy for each type of coating. The measurements were done for each type of coating on three random locations on each of the three replicates, the results being averaged (arithmetic mean). Measurement uncertainty was taken to be the SD over the entire scanned area. The electron work function (φ) was taken as an index for the surface charge at macroscopic scale. The value of φ is the minimal energy required for an electron to escape from a solid. The film composition, its crystalline structure and the electrical field induced by the surface electrical charge contribute to the value of φ. To measure φ, the pre-threshold photoelectron emission detection was performed in vacuum conditions (10−10–10−11 Pa) using an ultra-violet photon emission spectrometer [43]. The spectrometer, as described in detail in Ref. [44], is composed of a vacuum system, an UHV measurement chamber and a very sensitive electron detector for electron emission measurement, the noise current of the secondary electron multiplier being of about 0.5 electron/s [45]. The system uses a deuterium LSB-210 lamp (Lot quantum design) as UV source. The required photon energy was selected by means of a MDM2 (LOMO) monochromator, with a step <0.015 eV. The specimens were irradiated by soft UV light at 3–6 eV (the range of the expected value of φ) to release the electron and the photoelectron emission current (I) was measured. The value of φ was identified as the energy of the photons when I = 0. In vitro biocompatibility tests were conducted using human osteosarcoma cells (MG63, American Type Culture Collection). For the in vitro tests were used six replicates of each type of coating and the assay was performed duplicates. MG63 cells were plated at a density of 5×105 cells/ml and cultivated in DMEM low glucose medium (Sigma) supplemented with 10% inactivated fetal bovine serum, penicillin (100 U/ml), streptomycin (72,000 U/l) and neomycin (50 U/l). Cell viability was evaluated by MTT assay and cell morphology by actin staining. MTT assay was performed for in vitro evaluation of cell metabolism level. The cultured cells were rinsed with warm PBS and incubated for 3 h with the yellow tetrazolium MTT (3-(4, 5-dimethylthiazolyl-2)-2, 5-diphenyltetrazolium bromide) salt solution prepared in phenol red–free medium without serum. This salt was reduced in metabolically active cells by the cellular mitochondrial dehydrogenase enzymes resulting in an insoluble purple formazan, which was then solubilised with 0.1N HCl in isopropanol. The formazan solution concentration is directly correlated with cellular enzymatic reduction activity and it was determined by measuring the optical density at λ = 570 nm. For fluorescence microscopy, the culture medium was removed and the cells were rinsed with warm PBS and then permeabilized and fixed with a 4% PFA and 0.1% Triton X 100 solution in PBS. Then, the cells were washed and the nonspecific binding sites were blocked with 0.1% albumin for 15 min and then incubated for an hour with FITC (fluorescein isothiocyanate) coupled phalloid, for actin cytoskeleton detection and anti-vimentin primary antibody for vimentin network detection. The cells were washed and incubated with secondary antibody goat anti-mouse IgG Alexa Fluor 568 at a 1/2000 dilution and mounted using Fluoroshield with DAPI (4',6-diamidino-2-phenylindole). The images were recorded using an Axio Observer fluorescence microscope (Carl Zeiss) equipped with MRc5 digital camera. The data resulted from the biocompatibility tests were statistically analyzed by paired Student’s t-test (α = 0.005, as significant level of confidence). The coated 316 L discs were used for EDS, SEM, work function, electrical potential and surface profilometry measurements, as well as for biocompatibility tests. XPS and AFM measurements were carried out on coated Si (111) substrates. Results and Discussion Elemental composition and chemical bonds The mean values of the atomic concentrations measured for each type of coating in five random locations on each of the two replicates, together with C/(metal+Si) concentration ratio, are given in Table 1. The accepted precision in EDS measurement, when no standard sample is available (as in our case), in expressed as the relative SD (RDS) and is expressed as RDS(%) = (σat/Cav)*100, where (σat represents the calculated SD (%) of 10 measurements, 5 on each of the 2 replicates, and Cav represents the arithmetic mean content of the constituent element. We mention that for metallic (heavy) elements, the usual SRD value is around 1%, but if no standards are used, the SRD value is about 3%, which was reduced by using long counting (acquisition) times [46,47]. The calculated RDS values are also presented in Table 1. As seen, all coatings have C/(metal+Si) ratios close to unity (1.06–1.08). 10.1371/journal.pone.0161151.t001Table 1 Elemental composition of the coatings (EDS). Coating Elemental composition (at.%) C/(metal+Si) Ti Zr Nb Ta Hf Si C O (TiZrNbTaHf)C 9.2±0.2 9.9±0.2 9.5±0.2 8.4±0.2 9.4 ±0.2 - 50.2±1.5 3.4±0.1 1.08 (SiZrNbTaHf)C - 10.3±0.2 9.6±0.2 9.2±0.2 9.9 ±0.2 7.5±0.2 50.4±1.5 3.1±0.1 1.08 (TiZrNbSiHf)C 9.4±0.2 10.5±0.2 9.7 ±0.2 - 9.6±0.2 7.8±0.2 49.8±1.5 3.2±0.1 1.06 The XPS analysis was carried out on the surfaces of the as-deposited coatings. Fig 1 shows, as an example, the XPS Zr 3d, Nb 3p, Ta 4d, Hf 4f, Si 2p and C1s peaks of the (SiZrNbTaHf)C coating. The peak assignment was performed according to Ref. [48] as follows. The Zr 3d5/2 peaks at 179.4, 181.6 and 185.2 eV were associated to ZrC, ZrCOx and ZrO2 compounds, respectively; Nb 3p3/2 peaks at 362.4 and 365.8 eV to NbC and Nb2O5, respectively; Ta 4d3/2 peaks at 241.2 and 243.9 eV to TaCOx and TaOx, respectively; Hf 4f peaks at 14.9 and 17.5 eV to HfC and HfOx, respectively; Si 2p3/2 at 100.8, 102.1 and 103.1 eV to SiC, SiCOx and SiO2, respectively; C1s peaks at 282.5, 285.0, 286.9 and 289.0 eV to metal carbide, C graphite and COx, respectively. As resulted from the XPS analysis, the surface composition of the coatings consists of a mixture of metallic carbide, free-carbon (graphite-like) and oxides phases. The formation of surface oxides has been often reported for coatings based on transition metal compounds [49–51], being the result of the oxidation process during sample exposure in free atmosphere. 10.1371/journal.pone.0161151.g001Fig 1 XPS Si 2p, Zr 3d, Nb3p, Ta 4d, Hf 4f and C 1s spectra of the (SiZrNbTaHf)C coating. The X-ray diffraction patterns of the coatings are presented in Fig 2. The measurements were done on one replicate of each type of coatings. The reference coating exhibited a strong (111) preferred orientation. The replacement of Ti or Ta by Si determined the decrease of the intensity of (111) peak, while the (220), (311) and (222) maxima vanished. The crystallite size was estimated from the (111) peak broadening using the Scherrer formula (d = 0.89λ/(βcosθ)). Si addition resulted in the reduction of the crystallite size, from 11.2 nm for the reference coating to 8.7 nm for (SiZrNbHfTa)C and 8.2 nm for TiZrNbSiHf)C. The observed decrease was ascribed to the increase of the amorphous phase (C = C; Si–C–O) content, as also resulted from the XPS analysis (Fig 1, C1s spectrum). 10.1371/journal.pone.0161151.g002Fig 2 XRD diffractograms of the coating deposited on 316L substrate (S): (a) (TiZrNbTaHf)C; (b) (SiZrNbTaHf)C; (c) (TiZrNbSiHf)C; d = crystallite size. Surface morphology Surface morphology of the coatings deposited on 316L steel, as observed by SEM, is illustrated in Fig 3. The surfaces look smooth and dense, without cracks. The morphology of the surface at nano-scale, as measured by AFM, indicates also smooth surfaces, without pronounced hills or valleys (Fig 4). It is to note that the roughness parameters at nano- and meso-scale (RMS and Ra, respectively) exhibited similar dependence on coating type (Fig 5). 10.1371/journal.pone.0161151.g003Fig 3 SEM surface images of the coated 316L samples. Original magnification: × 10,000. 10.1371/journal.pone.0161151.g004Fig 4 AFM surface images of the coated Si samples. Representative AFM images of the coatings: scanned area: 3 × 3 μm2. 10.1371/journal.pone.0161151.g005Fig 5 Ra and RMS roughness parameters of the coatings, at meso- and nano-scale. Data show the mean and SD values obtained for each type of coating on two replicates in 5 different areas. Ra roughness measurements were performed by surface profilometer (scanned area: (150 × 150) μm2), and the RMS roughness measurements were performed by AFM microscopy (scanned area: (3 × 3) μm2). Electrical potential and electron work function The measurements were carried out on three replicates of each coating type, on three random locations. The surface electrical potential (V) of the investigated coatings is shown in Fig 6. As compared with the (TiZrNbTaHf)C reference, the Si containing coatings have lower electrical potentials, indicating a more negative surface charge. One may also observe that the electrical potential is not related to the surface roughness (Fig 5), probably because the roughness parameters are very low and are varying within a narrow range. 10.1371/journal.pone.0161151.g006Fig 6 Electrical potential of the (TiZrNbTaHf)C, (SiZrNbTaHf)C and (TiZrNbSiHf)C coatings. Data show the mean and SD values obtained for each type of coating on two replicates in 5 different areas. Ra roughness measurements were performed by surface profilometer (scanned area: (150 × 150) μm2), and the RMS roughness measurements were performed by AFM microscopy (scanned area: (3 × 3) μm2). Electron work function of the investigated coatings is illustrated in Fig 7. The (TiZrNbSiHf)C coating exhibits the highest work function, followed by (SiZrNbTaHf)C and (TiZrNbTaHf)C. As can be seen, the changes in work function values correlate well with those of the electrical potential. This finding is in line with the results reported in Ref. [52]which showed that φ increases when the surface acquires a negative charge. Considering the measurements of the work function on polycrystalline materials, due to the random orientations of the crystallites, each area corresponding to a certain crystallite facet presents a specific value of the work function, as well documented by surface anisotropy studies [53,54]. In the current study, the work function measured at macroscopic scale represents the averaged value of all work function of the crystalline facets weighted by their area [55]. 10.1371/journal.pone.0161151.g007Fig 7 Work function of the investigated coatings. Data show the mean and SD values. For each type of coating, the measurements were done on three random locations on each of the three replicates, the results being averaged (arithmetic mean). Biocompatibility Cell viability was investigated, for each coating type on six coated 316L replicates, after 3, 5 and 7 days in vitro culture using MTT test and the results are presented in Fig 8. It is clear that the cells are viable on all the coatings. After 5 and 7 days, a significant increase in the number of metabolically active cells was found for all the coatings. However, after 7 days of culture, an enhanced surface functionality of the (TiZrNbSiHf)C coating compared to the other two was observed, indicating the beneficial effect of Ta replacement by Si. Even if the differences obtained between the (SiZrNbHfTa)C and (TiZrNbSiHf)C coatings were not statistically significant, there are significant positive differences in the cell viability for the cells grown on both Si containing coatings, compared to the reference coating. 10.1371/journal.pone.0161151.g008Fig 8 Results of MTT cell viability assay of osteosarcoma cells on the investigated coatings after 3, 5 and 7 days of culture. Data show the mean and SD values. For each type of coating, the measurements were done on 6 replicates, and the assay was performed duplicates. The actin and vimentin filaments were labelled to observe cell morphology at 3 days after seeding the coated substrates surfaces. The actin is a main structural protein which provides information about the cells ability to adhere and spread [56], while the vimentin is responsible for maintaining cell shape and integrity [56,57], including the maintenance of the overall integrity of cytoplasm [58,59]. It is known that vimentin contributes to the construction of cytoskeleton architecture and generate cellular mechanical strength and cell integrity [59]. The immunofluorescent staining of the osteosarcoma cells on the coated substrates after 3 days (early time point) of incubation can be observed in Fig 9. Concomitant labelling of actin and vimentin was used to evaluate the cell attachment and cytoskeleton organization. The cells on all tested samples showed positive staining for actin and vimentin. The MG63 cells adhered to all surfaces and spread well, maintaining their typical spindle morphology, and b-actin was assembled into distinct filaments. Also the vimentin network displayed a predominantly perinuclear localization in cells grown on all tested coatings. Further, no differences in cell morphology were observed and there was no evidence of membrane damage, cytoplasmic vacuolation or cell death. This result shows that the cells attached and proliferated to all the tested coatings. A lower level of actin filaments and vimentin was observed for the (TiZrNbTaHf)C reference coating, indicating a limited capacity of this coating to promote cell adherence. 10.1371/journal.pone.0161151.g009Fig 9 Analysis of cytoskeleton organization in MG63 cells grown after 3 days on: (TiZrNbTaHf)C, (SiZrNbTaHf)C and (TiZrNbSiHf)C. Representative fluorescent images of actin cytoskeleton and vimentin intermediate filaments were presented. Immunostaining of vimentin (purple), actin (green) are shown in separate channels. DNA was stained with DAPI (blue). Scale bar: 10 μm. Cell density was 5×105 cells/ml. Discussions The cell attachment to an implant is a complex process that is controlled by many factors such as implant surfaces’ physico-chemical properties, crystalline structure, surface topography and roughness, characteristics of the surface oxide layer etc. The electrostatic interaction between the cells and biomaterials was the subject of numerous studies [57–62]. The experimental results demonstrated the key role played by the electrical charge of the biomaterials in cell adhesion. Superior adhesion of the osteoblast and fibroblast cells was observed for the coatings with more negatively charged surfaces [57,58,63]. It is well documented that electronegative potentials occur in non-stressed bone in areas of active growth and repair [64]. In human body, the cells, especially osteoblasts, are negatively charged [65], and it is expected to be electrostatically repelled by the negatively charged implant surface. However, Gongadze et al. demonstrated that titanium implants with low values of the surface potential promote osteoblast adhesion and the formation of the new bone [66]. The extensive review of Guo et al. [38] demonstrates that the bone cell adhesion on a biomaterial and the initial stage of bone proliferation are quite sensible to the surface charge and its polarity. Pattanayak et al. [67] reported that the negatively charged surfaces placed in a biological environment are promoting osteoblast cells-implant interaction in titanium dental implants. Krukowski et al. [36] also reported that the negatively charged surfaces promote craniofacial and intramedullary bone formation. Ohgaki et al. found that negatively charged biomaterial surfaces the cells proliferation was more intense, such as manifold layers of cells and broaden colonies of osteoblast-like cells were observed [68]. Even if the osteoblast cells are negatively charged [69], they can be attracted to negatively charged surfaces, the interaction being mediated by proteins [61,69] which are either positively charged, or which have positively charged tips, due to the presence of a quadrupolar internal charge distribution [70]. These proteins provide a substrate for the subsequent attachment of negatively charged osteoblasts. It is to note that vimentin presents a positively charged amino terminal [71]. This finding was also supported by Maroudas’ early report, which presented the dependence of cell adhesion and spreading on the surface charge of the implant [72]. Also, it was reported that many actin binding domains are rich in positive charge [73]. Fig 10 presents a schematic image of the positive electrically charged proteins (P) near a negatively charged surface (S), attracting towards the surface S the negatively charged osteoblast cells (O). 10.1371/journal.pone.0161151.g010Fig 10 Schematic image of positive electrically charged proteins (P) attracted to a negatively charged surface (S), and the negatively charged osteoblast cells (O) attracted towards the surface. However contradictory results have been also reported in the literature, as also positively charge surfaces were shown to have the same effect [34] [74]. The influence of the surface charge on cells-coating interface is certainly more complex, requiring further studies on different types of materials. According to our results, all the investigated coatings can support osteoblast cell proliferation. As compared to the reference coating, both Si containing coatings, exhibiting higher work function values and presented ~30% increase of the cell viability after 7 days of culture. As commonly admitted, the negative surface net charge depends on elemental composition at material’s surface, topography, thickness of the surface oxide (e.g. [59]). To increase the surface charge, several approaches have been tried such as topography modification [34] or thermal and plasma treatments [75]. For the coatings under investigation in this work, it was found that Ti or Ta replacement by Si led to improved cell attachment to coating surface. On the other hand, the Si containing coatings proved to have a higher surface charge compared to the (TiZrNbTaHf)C coating, as resulted from the electrical potential and work function measurements (the correlation between the electrical potential, work function and cell viability is illustrated in Fig 11). Taking into account the major role of the negatively charged surfaces in cell attachment, it is reasonable to presume that the beneficial effects induced by Si incorporation into the (TiZrNbTaHf)C coating on the cell attachment are due, at least partially, to the enhanced negative surface charge as reflected by the electrical potential decrease or work function increase. Consequently, either surface electrical potential or work function could be taken as relative predictors for evaluating a material from a biological point of view: considering a material with recognized biocompatible characteristics as reference, we could appreciate that different modifications in its composition, structure or surface properties would improve or not its biocompatibility by measuring changes in electrical potential or work function. 10.1371/journal.pone.0161151.g011Fig 11 Correlation between the electrical potential, work function and cell viability of the investigated coatings. It is also to be noted that the effect of Si addition on coating biocompatibility is different when either Ta or Ti is removed from the reference coating. The specific mechanisms according to which one metal replacement by Si in multi-principal element carbide coatings led to an increase in the negative surface charge, may be partially ascribed to the increased Pauling electronegativity. Si (1.90) is more electronegative than Ti (1.54) or Ta (1.50), such as it was observed a net increase of the overall electronegativity of the (TiZrNbSiHf)C coating, as compared to (SiZrNbTaHf)C. It should be also mentioned that Si addition to metal carbides results in the formation of a multiphase composition consisting, beside the crystalline carbide phase, of a mixture of amorphous SiC, SiCO and SiOx phases, as also resulted from our XPS measurements. Therefore, it is likely that these new phases, which have a pronounced insulator character, also contribute to the negative charge enhancement. As previously reported [59,60], in the case of Ti6Al4V alloy, the surface oxide, particularly its thickness, was found to control the net charge of the titanium alloy. It should be underlined that the measured electronegative potentials in the studied coatings were well correlated with their in vitro biocompatibility, and may be related to adequate bone growth [64] conditions. Conclusions The present study explored the possibility to improve the biocompatibility of the (TiNbZrTaHf)C coating through replacement of either Ti or Ta by Si. The coatings were deposited by reactive magnetron sputtering in an Ar+CH4 mixed atmosphere. Considering the important role of electrostatic interactions between cells and biomaterial surface in cell attachment, the effects of surface charge as characterized by electrical potential and work function, on coatings’ biocompatibility was examined. A significant correlation between electrical potential, work function and coating biocompatibility, as derived from osteoblasts viability and attachment to coatings surfaces, was found. Consequently, either the electrical potential or the work function are proposed as relative predictors for biocompatible characterization of the investigated coatings. Among the coatings, (TiZrNbSiHf)C, with low electrical potential and the high work function, exhibited the best biocompatible properties. This work was funded by a grant of the Romanian National Authority for Scientific Research, CNCS-UEFISCDI, project number PN-II-ID-PCE-2011-3-1016. EDS analysis was carried out by using the equipment acquired by the infrastructure project INOVA-OPTIMA SMIS code 49164, contract no. 658/2014. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The authors thank dr. C. Logofatu from the National Institute for Materials Physics (Magurele, Romania) for XPS measurements. ==== Refs References 1 Bauer S , Schmuki P , von der Mark K , Park J . Engineering biocompatible implant surfaces. Part I: Materials and surfaces . Prog Mater Sci . 2013 ;58 : 261 –326 . 2 Holzwarth U, Cotogno G. Total Hip Arthroplasty—State of the Art, Challenges and Prospects. JRC Scientific and policy Reports, EU Commission. 2012. 10.2788/31286; Available: http://publications.jrc.ec.europa.eu/repository/handle/JRC72428 3 Nedoma J , Stehlík J , Hlavácǎek I , Daneǎk J , Dostálová T , Prǎecǎková P , editors. Total Replacement of Human Joints Mathematical and Computational Methods in Biomechanics of Human Skeletal Systems . New Jersey : John Wiley & Sons Inc ; 2011 pp. 34 –72 . 4 Balla VK , Bose S , Davies NM , Bandyopadhyay A . Tantalum—A bioactive metal for implants . JOM . 2010 pp. 61 –64 . 5 Serro AP , Completo C , Colaço R , dos Santos F , da Silva CL , Cabral JMS , et al A comparative study of titanium nitrides, TiN, TiNbN and TiCN, as coatings for biomedical applications . Surf Coatings Technol . 2009 ;203 : 3701 –3707 . 6 Probst J , Gbureck U , Thull R . Binary nitride and oxynitride PVD coatings on titanium for biomedical applications . Surf Coatings Technol . 2001 ;148 : 226 –233 . 7 Brama M , Rhodes N , Hunt J , Ricci A , Teghil R , Migliaccio S , et al Effect of titanium carbide coating on the osseointegration response in vitro and in vivo . Biomaterials . 2007 ;28 : 595 –608 . 17049981 8 Chang YY , Huang HL , Chen YC , Hsu JT , Shieh TM , Tsai MT . Biological characteristics of the MG-63 human osteosarcoma cells on composite tantalum carbide/amorphous carbon films . PLoS One . 2014 ;9 : e95590 10.1371/journal.pone.0095590 24760085 9 Zoita C. , Braic L , Kiss A , Braic M . Characterization of NbC Coatings Deposited by Magnetron Sputtering Method . Surf Coatings Technol . 2010 ;204 : 2002 –2005 . 10 Braic M , Braic V , Balaceanu M , Vladescu A , Zoita CN , Titorencu I , et al Preparation and characterization of biocompatible Nb-C coatings . Thin Solid Films . 2011 ;519 : 4064 –4068 . 11 Chu CL , Ji HL , Yin LH , Pu YP , Lin PH , Chu PK . Fabrication, properties, and cytocompatibility of ZrC film on electropolished NiTi shape memory alloy . Mater Sci Eng C . 2011 ;31 : 423 –427 . 10.1016/j.msec.2010.10.023 12 Cotrut CM , Braic V , Balaceanu M , Titorencu I , Braic M , Parau AC . Corrosion resistance, mechanical properties and biocompatibility of Hf-containing ZrCN coatings . Thin Solid Films . 2013 pp. 48 –55 . 13 Braic M , Balaceanu M , Braic V , Vladescu A , Pavelescu G , Albulescu M . Synthesis and characterization of TiN, TiAIN and TiN/TiAIN biocompatible coatings . Surf Coatings Technol . 2005 ;200 : 1014 –1017 . 14 Braic M , Balaceanu M , Vladescu A , Zoita CN , Braic V . Study of (Zr,Ti)CN, (Zr,Hf)CN and (Zr,Nb)CN films prepared by reactive magnetron sputtering . Thin Solid Films . 2011 ;519 : 4092 –4096 . 15 Vladescu A ; Kiss A ; Popescu A ; Braic M ; Balaceanu M ; Braic V ; Tudor I ; Logofatu C ; Negrila CC ; Rapeanu R . Influence of bilayer period on the characteristics of nanornetre-scale ZrN/TiAIN multilayers . J Nanosci Nanotechnol . 2008 ;8 : 717 –721 . 18464396 16 Braic V , Balaceanu M , Braic M , Vladescu A , Panseri S , Russo A . Characterization of multi-principal-element (TiZrNbHfTa)N and (TiZrNbHfTa)C coatings for biomedical applications . J Mech Behav Biomed Mater . 2012 ;10 : 197 –205 . 10.1016/j.jmbbm.2012.02.020 22520431 17 Tsai M , Yeh J . High-Entropy Alloys: A Critical Review . Mater Res Lett . 2014 ;2 : 107 –123 . 10.1080/21663831.2014.912690 18 Yeh JW , Chen SK , Lin SJ , Gan JY , Chin TS , Shun TT , et al Nanostructured high-entropy alloys with multiple principal elements: Novel alloy design concepts and outcomes . Adv Eng Mater . 2004 ;6 : 299 –303 . 10.1002/adem.200300567 19 Yeh JW , Chen YL , Lin SJ , Chen SK . High-Entropy Alloys–A New Era of Exploitation . Mater Sci Forum . 2007 ;560 : 1 –9 . 10.4028/www.scientific.net/MSF.560.1 20 Yeh JW . Recent progress in high-entropy alloys . Ann Chim Sci des Mater . 2006 ;31 : 633 –648 . 21 Pogrebnjak AD , Yakushchenko IV , Bagdasaryan AA , Bondar OV , Krause-Rehberg R , Abadias G , et al Microstructure, physical and chemical properties of nanostructured (Ti–Hf–Zr–V–Nb)N coatings under different deposition conditions . Mater Chem Phys . 2014 ;147 : 1079 –1091 . 22 Bondar OV, Beresnev VM, Yakuschenko IV, Takeda Y, Krause-Rehberg R, Kolesnikov DA. Influence of deposition and annealing parameters on phase-elemental composition of high entropy alloys nitrides (Ti-Zr-Hf-V-Nb)N. CriMiCo 2013–2013 23rd International Crimean Conference Microwave and Telecommunication Technology, Conference Proceedings. 2013. pp. 802–803. 23 Feng X , Tang G , Ma X , Sun M , Wang L . Characteristics of multi-element (ZrTaNbTiW)N films prepared by magnetron sputtering and plasma based ion implantation . Nucl Instruments Methods Phys Res Sect B Beam Interact with Mater Atoms . 2013 ;301 : 29 –35 . 24 Liang SC , Tsai DC , Chang ZC , Sung HS , Lin YC , Yeh YJ , et al Structural and mechanical properties of multi-element (TiVCrZrHf)N coatings by reactive magnetron sputtering . Appl Surf Sci . 2011 ;258 : 399 –403 . 25 Liang SC , Chang ZC , Tsai DC , Lin YC , Sung HS , Deng MJ , et al Effects of substrate temperature on the structure and mechanical properties of (TiVCrZrHf)N coatings . Appl Surf Sci . 2011 ;257 : 7709 –7713 . 26 Braic M , Balaceanu M , Vladescu A , Zoita CN , Braic V . Deposition and characterization of multi-principal-element (CuSiTiYZr)C coatings . Appl Surf Sci . 2013 ;284 : 671 –678 . 27 Braic V , Parau AC , Pana I , Braic M . Effects of substrate temperature and carbon content on the structure and properties of (CrCuNbTiY)C multicomponent coatings . Surf Coatings Technol . 2014 ;258 : 996 –1005 . 28 Chang S-Y , Lin S-Y , Huang Y-C . Microstructures and mechanical properties of multi-component (AlCrTaTiZr)NxCy nanocomposite coatings . Thin Solid Films . 2011 ;519 : 4865 –4869 . 29 Braic V , Vladescu A , Balaceanu M , Luculescu CR , Braic M . Nanostructured multi-element (TiZrNbHfTa)N and (TiZrNbHfTa)C hard coatings . Surf Coatings Technol . 2012 ;211 : 117 –121 . 30 Vepřek S . Conventional and new approaches towards the design of novel superhard materials . Surf Coatings Technol . 1997 ;97 : 15 –22 . 31 Jansson U , Lewin E . Sputter deposition of transition-metal carbide films—A critical review from a chemical perspective . Thin Solid Films . 2013 pp. 1 –24 . 10.1016/j.tsf.2013.02.019 32 Santavirta S , Takagi M , Nordsletten L , Anttila A , Lappalainen R , Konttinen YT . Biocompatibility of silicon carbide in colony formation test in vitro. A promising new ceramic THR implant coating material . Arch Orthop Trauma Surg . 1998 ;118 : 89 –91 . 9833115 33 Shtansky DV , Gloushankova NA , Sheveiko AN , Kiryukhantsev-Korneev P , Bashkova IA , Mavrin BN , et al Si-doped multifunctional bioactive nanostructured films . Surf Coatings Technol . 2010 ;205 : 728 –739 . 34 Khlusov IA , Dekhtyar Y , Khlusova MY , Gostischev EA , Sharkeev YP , Pichugin VF , et al Novel Concepts of “Niche-Relief” and “Niche-Voltage” for Stem Cells as a Base of Bone and Hematopoietic Tissues Biomimetic Engineering . IFMBE . 2013 pp. 99 –102 . 35 Hamamoto N , Hamamoto Y , Nakajima T , Ozawa H . Histological, histocytochemical and ultrastructural study on the effects of surface charge on bone formation in the rabbit mandible . Arch Oral Biol . 1995 ;40 : 97 –106 . 7540834 36 Krukowski M , Shively RA , Osdoby P , Eppley BL . Stimulation of craniofacial and intramedullary bone formation by negatively charged beads . J Oral Maxillofac Surg . 1990 ;48 : 468 –475 . 1691778 37 Ramazanoglu M , Oshida Y . Osseointegration and Bioscience of Implant Surfaces—Current Concepts at Bone-Implant Interface . Dent—A Rapidly Evol Pract . 2011 ; 57 –80 . 10.5772/16936 38 Guo CY , Matinlinna JP , Tang ATH . Effects of surface charges on dental implants: Past, present, and future . Int J Biomater . 2012 ;2012 : 381535 :1 –5 . 10.1155/2012/381535 39 Novaes AB , de Souza SLS , de Barros RRM , Pereira KKY , Iezzi G , Piattelli A . Influence of implant surfaces on osseointegration . Brazilian Dental Journal . 2010 pp. 471 –481 . 10.1590/S0103-64402010000600001 21271036 40 Braic L , Zoita NC . Influence of the deposition time and temperature on the texture of InN thin films grown by RF-magnetron sputtering . Optoelectron Adv Mater Rapid Commun . 2010 ;4 : 2013 –2017 . 41 Negrila CC , Logofatu C , Ghita R V. , Cotirlan C , Ungureanu F , Manea AS , et al Angle-resolved XPS structural investigation of GaAs surfaces . J Cryst Growth . 2008 ;310 : 1576 –1582 . 42 Nonnenmacher M , O’Boyle MP , Wickramasinghe HK . Kelvin probe force microscopy . Appl Phys Lett . 1991 ;58 : 2921 –2923 . 10.1063/1.105227 43 Bystrov VS , Paramonova EV , Dekhtyar Y , Pullar RC , Katashev A , Polyaka N , et al Polarization of poly(vinylidene fluoride) and poly(vinylidene fluoride-trifluoroethylene) thin films revealed by emission spectroscopy with computational simulation during phase transition . J Appl Phys . 2012 ;111 10.1063/1.4721373 44 Akmene RJ , Balodis AJ , Dekhtyar Yu D , Markelova GN , Matvejevs JV , RozenfeldsLB , et al UAI . Exoelectron emission specrometre complete set of surface local investigation . Surf physics, Chem Mech . 1993 ;8 : 125 –128 . 45 Yu Dekhtyar . D . Functionalized Nanoscale Materials, Devices and Systems . NATO Science for Peace and Security Series B: Physics and Biophysics . 2008 pp. 169 –183 . 46 Agarwal BK . Springer Series in Optical Sciences, Vol. 15 : X-Ray Spectroscopy: An Introduction . 2nd ed Berlin Heidelberg : Springer Verlag ; 1991 . 47 Goldstein J , Newbury DE , Joy DC , Lyman CE , Echlin P , Lifshin E , et al Scanning Electron Microscopy and X-ray Microanalysis . 3-rd ed Springer ; 2003 10.1007/978-1-4615-0215-9 48 Crist B . Handbooks of Monochromatic XPS Spectra. Handbooks of Monochromatic XPS Spectra . XPS International LLC : Mountain View, CA, USA ; 2004 . 49 Lopes C , Parreira NMG , Carvalho S , Cavaleiro A , Rivière JP , Le Bourhis E , et al Magnetron sputtered Ti-Si-C thin films prepared at low temperatures . Surf Coatings Technol . 2007 ;201 : 7180 –7186 . 50 Krzanowski JE , Wormwood J . Microstructure and mechanical properties of Mo-Si-C and Zr-Si-C thin films: Compositional routes for film densification and hardness enhancement . Surf Coatings Technol . 2006 ;201 : 2942 –2952 . 51 Eklund P , Joelsson T , Ljungcrantz H , Wilhelmsson O , Czigany Z , Hogberg H , et al Microstructure and electrical properties of Ti-Si-C-Ag nanocomposite thin films . Surface and Coatings Technology . 2007 pp. 6465 –6469 . 52 Dekhtyar Y , Dvornichenko MV , Karlov AV , Khlusov IA , Polyaka N , Sammons R , et al Electrically functionalized hydroxyapatite and calcium phosphate surfaces to enhance immobilization and proliferation of osteoblasts in vitro and modulate osteogenesis in vivo . IFMBE Proceedings . 2009 pp. 245 –248 . 53 Skriver HL , Rosengaard NM . Surface energy and work function of elemental metals . Phys Rev B . 1992 ;46 : 7157 –7168 . 54 Fall J , Binggeli N , Baldereschi A . Theoretical maps of work-function anisotropies . Phys Rev B . 2001 ;65 : 045401 . 55 Orf ND , Baikie ID , Shapira O , Fink Y . Work function engineering in low-temperature metals . Appl Phys Lett . 2009 ;94 : 16 –19 . 10.1063/1.3089677 56 Curtis A , Riehle M . Tissue engineering: the biophysical background . Phys Med Biol . 2001 ;46 : 47 –65 . 57 Pegueroles M , Aparicio C , Bosio M , Engel E , Gil FJ , Planell JA , et al Spatial organization of osteoblast fibronectin matrix on titanium surfaces: Effects of roughness, chemical heterogeneity and surface energy . Acta Biomater . 2010 ;6 : 291 –301 . 10.1016/j.actbio.2009.07.030 19635598 58 Richards RG . The effect of surface roughness on fibroblast adhesion in vitro . Injury . 1996 ;27 : S/C38 –S/C43 . 59 Rapuano BE , MacDonald DE . Surface oxide net charge of a titanium alloy: Modulation of fibronectin-activated attachment and spreading of osteogenic cells . Colloids Surfaces B Biointerfaces . 2011 ;82 : 95 –103 . 10.1016/j.colsurfb.2010.08.023 20884181 60 MacDonald DE , Rapuano BE , Schniepp HC . Surface oxide net charge of a titanium alloy: Comparison between effects of treatment with heat or radiofrequency plasma glow discharge . Colloids Surfaces B Biointerfaces . 2011 ;82 : 173 –181 . 10.1016/j.colsurfb.2010.08.031 20880672 61 Smith IO , Baumann MJ , McCabe LR . Electrostatic interactions as a predictor for osteoblast attachment to biomaterials . J Biomed Mater Res A . 2004 ;70 : 436 –441 . 15293317 62 Ohgaki M , Kizuki T , Katsura M , Yamashita K . Manipulation of selective cell adhesion and growth by surface charges of electrically polarized hydroxyapatite . J Biomed Mater Res . 2001 ;57 : 366 –373 . 11523031 63 Martin JY , Schwartz Z , Hummert TW , Schraub DM , Simpson J , Lankford J , et al Effect of titanium surface roughness on proliferation, differentiation, and protein synthesis of human osteoblast-like cells (MG63) . J Biomed Mater Res . 1995 ;29 : 389 –401 . 7542245 64 Galkowski V , Brad P , Drew B , David D . Bone stimulation for fracture healing: What’s all the fuss? Indian Jourbal Orthop . 2009 ;43 : 117 –120 . 65 Lotfi M , Neijib M , Naceur M . Cell Adhesion to Biomaterials: Concept of Biocompatibility . Advances in Biomaterials Science and Biomedical Applications . InTech; 2013 pp. 1 –34 . 66 Gongadze E , Kabaso D , Bauer S , Slivnik T , Schmuki P , van Rienen U , et al Adhesion of osteoblasts to a nanorough titanium implant surface . Int J Nanomedicine . 2011 ;6 : 1801 –1816 . 10.2147/IJN.S21755 21931478 67 Pattanayak DK , Kawai T , Matsushita T , Takadama H , Nakamura T , Kokubo T . Effect of HCl concentrations on apatite-forming ability of NaOH-HCl- and heat-treated titanium metal . J Mater Sci Mater Med . 2009 ;20 : 2401 –2411 . 10.1007/s10856-009-3815-0 19585225 68 Ohgaki M , Kizuki T , Katsura M , Yamashita K . Manipulation of selective cell adhesion and growth by surface charges of electrically polarized hydroxyapatite . J Biomed Mater Res . 2001 ;57 : 366 –73 . 11523031 69 Smeets R , Kolk A , Gerressen M , Driemel O , Maciejewski O , Hermanns-Sachweh B , et al A new biphasic osteoinductive calcium composite material with a negative Zeta potential for bone augmentation . Head Face Med . 2009 ;5 : 13 10.1186/1746-160X-5-13 19523239 70 Kabaso D , Gongadze E , Perutková S , Matschegewski C , Kralj-Iglic V , Beck U , et al Comput Methods Biomech Biomed Engin . 2011 ;14 : 469 –82 . 10.1080/10255842.2010.534986 21516531 71 Perides G , Harter C , Traub P . Electrostatic and hydrophobic interactions of the intermediate filament protein vimentin and its amino terminus with lipid bilayers . J Biol Chem . 1987 ;262 : 13742 –13749 . 3308882 72 Maroudas NG . Adhesion and spreading of cells on charged surfaces . J Theor Biol . 1975 ;49 : 417 –424 . 1121188 73 Tang JX , Janmey PA . The polyelectrolyte nature of F-actin and the mechanism of actin bundle formation . J Biol Chem . 1996 ;271 : 8556 –8563 . 10.1074/jbc.271.15.8556 8621482 74 Anselme K . Osteoblast adhesion on biomaterials . Biomaterials . 2000 ;21 : 667 –681 . 10711964 75 Khang G . Evolution of gradient concept for the application of regenerative medicine . Biosurface and Biotribology . 2015 ;1 : 202 –213 .

PMC005xxxxxx/PMC5003398.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757120610.1371/journal.pone.0161526PONE-D-16-12079Research ArticlePhysical SciencesPhysicsCondensed Matter PhysicsSolid State PhysicsCrystallographyCrystal StructureBiology and Life SciencesBiochemistryEnzymologyEnzymesProteasesBiology and Life SciencesBiochemistryProteinsEnzymesProteasesPhysical SciencesChemistryComputational ChemistryMolecular DynamicsMedicine and Health SciencesNeurologyNeurodegenerative DiseasesMovement DisordersParkinson DiseaseBiology and Life SciencesBiochemistryBiochemical SimulationsBiology and Life SciencesComputational BiologyBiochemical SimulationsPhysical SciencesMaterials ScienceMaterials by StructureCrystalsBiology and Life SciencesMolecular BiologyMacromolecular Structure AnalysisProtein StructureBiology and Life SciencesBiochemistryProteinsProtein StructurePhysical SciencesMathematicsOptimizationDistinct 3D Architecture and Dynamics of the Human HtrA2(Omi) Protease and Its Mutated Variants Human HtrA2(Omi) Protease MechanismsGieldon Artur 1Zurawa-Janicka Dorota 2Jarzab Miroslaw 2Wenta Tomasz 2Golik Przemyslaw 3Dubin Grzegorz 34*Lipinska Barbara 2Ciarkowski Jerzy 1*1 Faculty of Chemistry, University of Gdansk, Wita Stwosza 63, 80-308, Gdansk, Poland2 Department of Biochemistry, Faculty of Biology, University of Gdansk, 80-308, Gdansk, Poland3 Faculty of Biochemistry, Biophysics and Biotechnology, Jagiellonian University, Gronostajowa 7, 30-387, Krakow, Poland4 Malopolska Centre of Biotechnology, ul. Gronostajowa 7a, 30-387, Krakow, PolandSivaraman J EditorNational University of Singapore, SINGAPORECompeting Interests: The authors have declared that no competing interests exist. Conceptualization: JC AG BL. Data curation: JC GD PG. Formal analysis: AG PG MJ TW. Funding acquisition: JC GD BL. Investigation: JC AG PG BL DZ-J. Methodology: GD AG PG MJ TW DZ-J. Project administration: JC BL. Resources: JC GD BL. Supervision: JC GD BL. Validation: JC GD BL. Visualization: JC AG PG. Writing – original draft: JC GD BL. Writing – review & editing: JC GD. * E-mail: grzegorz.dubin@uj.edu.pl (GD); jerzy.ciarkowski@ug.edu.pl (JC)29 8 2016 2016 11 8 e016152623 3 2016 8 8 2016 © 2016 Gieldon et al2016Gieldon et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.HtrA2(Omi) protease controls protein quality in mitochondria and plays a major role in apoptosis. Its HtrA2S306A mutant (with the catalytic serine routinely disabled for an X-ray study to avoid self-degradation) is a homotrimer whose subunits contain the serine protease domain (PD) and the regulatory PDZ domain. In the inactive state, a tight interdomain interface limits penetration of both PDZ-activating ligands and PD substrates into their respective target sites. We successfully crystalized HtrA2V226K/S306A, whose active counterpart HtrA2V226K has had higher proteolytic activity, suggesting higher propensity to opening the PD-PDZ interface than that of the wild type HtrA2. Yet, the crystal structure revealed the HtrA2V226K/S306A architecture typical of the inactive protein. To get a consistent interpretation of crystallographic data in the light of kinetic results, we employed molecular dynamics (MD). V325D inactivating mutant was used as a reference. Our simulations demonstrated that upon binding of a specific peptide ligand NH2-GWTMFWV-COOH, the PDZ domains open more dynamically in the wild type protease compared to the V226K mutant, whereas the movement is not observed in the V325D mutant. The movement relies on a PDZ vs. PD rotation which opens the PD-PDZ interface in a lid-like (budding flower-like in trimer) fashion. The noncovalent hinges A and B are provided by two clusters of interfacing residues, harboring V325D and V226K in the C- and N-terminal PD barrels, respectively. The opening of the subunit interfaces progresses in a sequential manner during the 50 ns MD simulation. In the systems without the ligand only minor PDZ shifts relative to PD are observed, but the interface does not open. Further activation-associated events, e.g. PDZ-L3 positional swap seen in any active HtrA protein (vs. HtrA2), were not observed. In summary, this study provides hints on the mechanism of activation of wtHtrA2, the dynamics of the inactive HtrA2V325D, but does not allow to explain an increased activity of HtrA2V226K. Ministry of Science and Higher Education of the Republic of Poland (MNiSW)DS-530-530-8370-D498-16Artur GiedoMinistry of Science and Higher Education of the Republic of Poland (MNiSW)DS-530-530-8370-D498-16Ciarkowski Jerzy European Union Structural FundsPOIG.02.01.0012064/08 and POIG.02.01.0012167/08Dubin Grzegorz Ministry of Science and Higher Education of the Republic of Poland (MNiSW)DS/530-L130-D241-15Lipinska Barbara Ministry of Science and Higher Education of the Republic of Poland (MNiSW)DS/530-L130-D241-15Zurawa-Janicka Dorota This work was supported by Ministry of Science and Higher Education of the Republic of Poland, http://www.nauka.gov.pl/ DS-530-530-8370-D498-16, to AG, JC; DS/530-L130-D241-15, to DZ-J, BL; National Science Centre Poland, https://www.ncn.gov.pl/ 2012/07/E/NZ1/01907 and UMO-2011/01/D/NZ1/01169, to GD; European Union Structural Funds, http://ec.europa.eu/, POIG.02.01.0012–064/08 and POIG.02.01.00–12–167/08, to Jagiellonian University. The funders had no role in study design, data collection and analysis, decision to publish, or preparation the manuscript. Data AvailabilityAll relevant data are within the paper and its Supporting Information files.Data Availability All relevant data are within the paper and its Supporting Information files. ==== Body Introduction The human HtrA2 (high-temperature requirement A2) protease controls protein quality in mitochondria [1, 2]. It is involved in cell death (apoptosis, both caspase dependent and independent [3–5]) and consequently plays a role in oncogenesis [5, 6]), and in Parkinson [4, 7, 8] and Alzheimer diseases [9–13]. These properties allow to consider HtrA2 as a potential drug target in cancer [5, 14] and/or neurodegenerative disorders [4]. The HtrA proteases can be distinguished from other serine proteases by the presence of one or two C-terminal PDZ (Postsynaptic density of 95 kDa, Disk large, Zonula occludens 1) domain(s). The protease domain (PD) of the chymotrypsin type consists of two perpendicularly arranged β-barrel lobes β1-β6 and β7-β12. The lobes are delimited by the N- and C-terminal α-helices, arranged parallel in space, in a C2-pseudosymmetrical manner. These secondary-structure elements are connected by loops, several of which are important for proteolytic activity [15]. The loops are named according to the chymotrypsin nomenclature: LA(β1-β2), LB(β3-β4), LC(β5-β6), LE(β4-β5) in the N-terminal barrel and LD(β7-β8), L1(β9-β10), L2(β11-β12), L3(β8-β9) in the C-terminal one [16] (Fig 1). The active site and the catalytic triad (His, Asp, Ser) are embedded in the interface of the β-barrels. 10.1371/journal.pone.0161526.g001Fig 1 Human HtrA2 topology. (A) Protease domain: the N- and C-terminal helices, α1/α2 and α4, respectively, are omitted for clarity. The catalytic triad His, Asp, Ser and the loops are labeled according to the chymotrypsin nomenclature [16]. (B) PDZ domain. The carboxylate binding loop is indicated. The PDZ domains participate in the regulation of HtrA activity by recognizing and binding hydrophobic C-terminal sequences of substrates or regulatory peptides. The HtrA monomers form higher-order oligomers. The common structural unit is a pyramid-shaped trimer with PDs constituting the pyramid walls and their N-termini adding to a central core around the pyramid apex, while the PDZ domains protruding outward laterally at the pyramid base. Two bacterial HtrAs, the E. coli DegP and DegQ proteases, which contain two PDZ domains, further oligomerize forming up to 24-mers. Regardless the oligomerization status, at low temperatures or in the absence of a substrate or activating peptides HtrAs adopt inactive conformations characterized by non-catalytic architecture of the triad and/or a restricted access to the active site. Hence, it is generally accepted that HtrA proteases are activated [15, 17–20] by binding of a specific activator at the PDZ domain and a substrate at the active site. HtrAs share ~60% sequence homology. A number of crystal structures of pro- and eukaryotic HtrAs have been solved to date [17, 20–34], demonstrating that the enzymes exist in at least two distinct conformations: resting and active states. There are several recent reviews on structure-function relationships within HtrAs [15, 19, 35], including a minireview in the introduction to our previous work [36]. Human HtrA2 has a single PDZ domain and in the crystal structure forms a pyramid-shaped homotrimer. The catalytic triad located in an interior (concave) of the trimer—a feature shared by the whole HtrA proteins family—comprises H198, D228, and S306 (S306A mutation in the crystal [31]) in a noncanonical, catalytically inactive orientation [19]. PDZ and PD are connected by a flexible linker. PDZ contains a typical peptide binding groove between β14 and α7 starting from a peptide-recognition motif 361YIGV364, the PDZ-specific carboxylate-binding loop β13-β14 [37]. In HtrA2 the groove is buried in the tight interface between PDZ and PD, and so is the PD active site; this being a unique structural feature of this protein, not observed in other family members, letting the trimer compare to a budding flower or—its monomer unit—to a lid-enclosed jug. PDZ packs against PD through van der Waals contacts, involving two clusters of hydrophobic residues in (β11-L2-β12)PD-(β14-α5)PDZ (hereafter hinge A) and (β5-LC-β6)PD-(β13/α7)PDZ (hinge B). With PDZ taken as a lid to PD, the hinges are located at the C- and N-barrels within the PD domain, respectively [31]. According to Li et al. [31], HtrA2 activation requires an activating peptide to bind at hydrophobic groove of PDZ, which leads to the opening of the PDZ-PD interface and enables protease activity. The model is supported by the fact that PDZ-deleted HtrA2 variant is more active than the full-length protein [31, 38]. Moreover, specific peptides binding to the PDZ domain increase the HtrA2 activity [9, 39]. HtrA2 activity is also up-regulated in response to a heat-shock [40]. Available evidence jointly suggests that activation of HtrA2 may resemble that of DegS [24, 26]. However, while in DegS the PDZ domain only stabilizes the inactive form of the protease, in HtrA2 it physically blocks the access of the substrate to the active site. Recently, we have investigated temperature-dependent (20–45°C) conformational changes in HtrA2 using single-tryptophan mutants at the PD/PDZ interface and monitoring fluorescence of this residue. Correlation of temperature dependence of fluorescence and catalytic rates in selected mutants allowed us to elucidate the role of particular amino acids in regulation of protease activity [36]. We have concluded that when the HtrA2 structure relaxes and the PDZ-PD interface opens, the PD part of hinge A retains its structural integrity. As such it does not tolerate any mutations in the (β11-L2-β12)PD segment, e.g. V325D, without complete loss of protease activity. On the contrary, hinge B also retains major interactions upon HtrA2 activation, but may accommodate favorably selected mutations, e.g. V226K in LCPD, that increase the proteolytic activity of the protein compared to the wild type HtrA2 [36]. In our most recent study we combined a temperature-dependent tryptophan-induced quenching (TrIQ) with selected mutation-activity analyses. TrIQ is a modern technique complementary to FRET and efficient at smaller distances (≤10Å) than the latter [41]. We found that in activation, PDZ changed its position versus PD of own and of the adjacent subunit PD* (the asterisk indicates the adjacent subunit). These changes included (β14-α5)PDZ slightly closing distance to “own” LC and L2 and to LD* [38] while simultaneously increasing distances between all other not contributing to the “hinges” (see above) parts of bulk PDZ and PD, including in particular the regulatory L3 [35, 38] moving away of own PDZ, compare Fig 1 in [38]. Simultaneously, expanding our former structure-activity mutation study [36], we have found that any mutations weakening inter-domain interactions at the PD-PDZ-PD* interfaces increase HtrA2 proteolytic activity, see e.g. Figs 3 and 4 and Table 2 in Ref. [38], including the V226K mutation, see above. The aim of this study was to examine in details likely routes of HtrA2 activation, compatible with the above observations. We used the wild type HtrA2, the inactivating mutant in hinge A (V325D) and the activating mutant in hinge B (V226K) as models. Our successful crystallization and structure solution of HtrA2S306A/V226K (having the catalytic Ser routinely disabled to avoid self-degradation) revealed that S306/A/V226K mutant has the same (inactive) architecture, earlier found for HtrA2S306A [31]. Subsequent molecular dynamics (MD), both unrestrained and restrained (RMD), have led us to a hypothesis on basic structural requisites to HtrA2 activation, pertinent to the structure relation between: 1) On the one hand, a unique tight PDZ-PD interface in the inactive HtrA2 (see above) that confines the regulatory L3 to a flap on a convex side of the trimer; and 2) On the other, our observation that an active form turns up to be common for the HtrA family [17, 20, 23, 24, 26, 28–30, 32], featured by opened PD-PDZ interface, having L3 slid between L2PD and α7PDZ onto the concave side of the trimer, thus destroying hinge A. Consequently, bringing inactive HtrA2 to this “canonical active architecture” would require an extensive rearrangement of L3, equivalent with a crack of hinge A and replacement it with a new (L2-L3)PD- (β13-α7)PDZ interactions, with simultaneous whole opening of the PD-PDZ interface. An alternative would mean an entirely different activated HtrA2 architecture than one typical of all active HtrAs resolved thus far. Methods Preparation of HtrA2V226K/S306A protein E. coli strain BL21(DE3) (Novagen, San Diego, USA) transformed with the pET-derived pDZ5 V226K plasmid [36], carrying HtrA2V226K/S306A gene was used to overproduce mutant protein (amino acids 134–458) with His6-tag at the C-terminal end. The protein was purified by affinity chromatography on Ni-NTA according to the manufacturers' instructions (Qiagen, Wroclaw, Poland). The concentration of HtrA2V226K/S306A was estimated using Amido Black as described before [42]. The purity of the HtrA2V226K/S306A preparation was estimated at more than 95% by SDS-polyacrylamide gel electrophoresis. Crystallization, data collection and structure solution Immediately prior to crystallization screening the buffer was exchanged to 5 mM Tris-HCl pH 8.0 containing 150 mM NaCl and 200 mM imidazole by gel filtration on Superdex s75 (GE Healthcare), the protein was concentrated to 18 mg/ml and screening was performed using sitting drop vapor-diffusion method. Crystals appeared after several days at room temperature in Crystal Screen 2 formulation 21 (Hampton Research). The initial conditions were optimized. The crystals used for measurements were obtained from 0,15 M MES pH 6.5 containing 2 M NaCl, 0,13 mM KH2PO4 and 0,1 M NaH2PO4. The crystals were cryoprotected in 25% glycerol in mother liquor and cooled in liquid nitrogen. The diffraction data were collected using in house rotating anode copper source (MicroMax-007 HF; Rigaku). Data were indexed and integrated with MOSFLM [43]. The following steps were performed using software collected in the CCP4 package [44]. Data were scaled with Scala [45, 46]. Molecular replacement was performed using Phaser [47] with alanine search model based on the structure of HtrA2 (PDB ID: 1LCY). Model building was performed manually using Coot [48]. Water molecules were added using Coot and were inspected manually. Restrained refinement was performed with Refmac5 [49]. Throughout the refinement 5% of reflections were used for cross-validation analysis [50] and the behavior of Rfree was utilized to monitor the refinement strategy. The data collection and refinement statistics are presented in Table 1. The structure was deposited in PDB with the accession code 5FHT. 10.1371/journal.pone.0161526.t001Table 1 Data collection and refinement statistics. PDB ID 5FHT Wavelength (Å) 1.54 Resolution range (Å) 14.57–1.95 (2.019–1.95)a Space group H3 Unit cell 86.01 86.01 126.70; 90.0 90.0 120.0 Total reflections 121190 (16150) Unique reflections 25396 (3720) Multiplicity 4.8 (4.3) Completeness (%) 99.6 (99.5) Mean I/sigma(I) 7.4 (3.1) Wilson B-factor 38.41 R-merge 0.116 (0.405) R-work 0.1749 R-free 0.2275 Number of atoms 2500  macromolecules 2285  water 200 Protein residues 300 RMS(bonds) 0.019 RMS(angles) 1.97 Ramachandran favored (%) 97.9 Ramachandran outliers (%) 0.35 Average B-factor 34.37  macromolecules 33.64  solvent 42.66 aValues in parentheses represent the highest-resolution shell. Molecular dynamics of the trimers of HtrA2 and its mutants As differences between the first HtrA2S306A [31] and the current HtrA2S306A/V226K structure (5FHT, this work vide infra) have turned out to be negligible, the crystal structure of the former (HtrA2S306A, pdb code 1LCY [31]) was used as a template. 1LCY and 5FHT have 3 unresolved regions, viz: the N-terminal IAP-binding motif 134AVPSP138 and 134AVPSPPPA141; L3β8~β9 282ARDLGLPQT290 and 281PARDLG286; and the PD~PDZ linker 344RGEKKNSSSGISGSQ358 and 345GEKKNSSSGISGSQ358, respectively. We ignored the IAP-binding sequence as one of no importance in protease function [31], while L3 and the PD~PDZ linker structures were restored using the BIOPOLYMER and LOOP SEARCH modules in the SYBYL software [51]. The obtained model was optimized using the minimization protocol implemented in the AMBER11 package [52, 53]. PDZ domain in complex with a selectively-binding peptide was modeled using the crystal structure of the lone (i.e. lacking PD) PDZ-GWTMFWV complex (pdb code 2PZD) [54] because the PDZ domains in 1LCY and 2PZD overlapped perfectly (RMS<0.1Å). Several minimizations and short low-temperature MD simulations were performed in repetitive cycles to reduce all steric clash between the introduced peptide and PD and to simultaneously preserve the initial structure as much as possible. Based on the above modeled structure the V226K and V325D mutants were constructed. Additionally their equivalents with the catalytic triad serine replaced by alanine were also built: S306A, V226K/S306A and V325D/S306A. Each of the latter three was modelled with and without the selective ligand GWTMFWV. All these models were optimized as described above. Homotrimers compatible with 1LCY crystal structure were made from optimized starting monomers by imposing the C3 symmetry typical of the 1LCY crystal. Nine constructed starting homotrimers were separately immersed in a rectangular TIP3P [55] water box of 125x125x95 Å3 size. To neutralize the negative charge on the protein system, 6, 9 and 12 Na+ ions were added to HtrA2V226K, HtrA2 and HtrA2V325D trimers (and to the matching S306A mutants), respectively. Each trimer was initially optimized using similar methodology as described for the monomers. MD simulations for each of the 9 systems were carried out using the following parameters: 1 fs time step, constant pressure (1 atm) and periodic boundary conditions. SHAKE algorithm [56] was used for treatment of hydrogen atoms. Long-range electrostatic interactions were simulated using the Particle Mesh Ewald (PME) method [57] with the cutoff equal to 8Å. Non-bonding interactions were updated every 25 steps. The simulation temperature was set to 313K because it is known that the thermally induced activation of HtrA2 is efficient at this temperature [36, 38]. Each of the simulated systems consisted of about 43,000 water molecules, 3x320 or 3x327 amino acid residues (apo and ligand containing models, respectively), ~145.000 atoms in total. The productive runs were 50 ns long, with snapshots taken every 1 ps. To obtain the results in a reasonable time, all molecular systems were simulated using AMBER PMED v2.2 software [58, 59] and NVIDIA GPU-CUDA [60] card hardware. Restrained (steered) molecular dynamics of the HtrA2 monomer To this aim, the crystal structure of HtrA2S306A (1LCY [31]) monomer was used as a template. The L3 (282ARDLGLPQT290) and the linker (344RGEKKNSSSGISGSQ358) gaps in the sequence were preserved. Moreover, by trial-and-error we found crucial to truncate both the β8 inlet to and the β9 outlet from the L3 gap, each by extra 8 residues, 274IVSSAQRP281 and 291NVEYIQTD298, respectively. Otherwise both trunks would sterically hinder optimization of the 1LCY-like start structure to a target meeting the restraints deduced from TrIQ and having a more open PD-PDZ interface (see below). All above conditions held, an alternative start, compatible with the completely open (no PD-PDZ1 interface) structure of the active DegP (PDB file 3CS0) [17] was used as a control. Nine PD-PDZ distance restraints were imposed in agreement with Fig 1B,C and Table S3 in Ref. [38]. As TrIQ results provided only tendencies (increase/decrease/keeping of a Cα-Cα distance) in HtrA2 thermal activation [38, 41], we affixed to these tendencies, specific, though arbitrarily guessed restraint distances, see Table 2. The harmonic restraints were very soft, with force constants equal to 1 kcal/Å. To maintain the complete shape integrities within the PD and PDZ domains, the backbone angles (φ,ψ,ω) were fixed. Successful restrained MD was carried out at 45K using AMBER11 [52, 53]. 10.1371/journal.pone.0161526.t002Table 2 The restraints used in restrained MD of 1LCY monomer. The values in the middle column are inferences affixed to tendencies given in Fig 1 B,C in Ref. [38]. Distance, [Å] Cα- Cα 1LCY [31] Restraint Optimized, see Results I179- L398 7.96 26.00 23.5 A201-Y361 8.44 13.00 14.6 P225-Y361 8.47 13.00 13.0 P225-V364 8.79 13.00 13.6 M323-P384 17.19 25.00 21.2 M323-M365 8.93 9.00 10.7 F331-M365 12.31 12.00 12.1 V226-L367 9.71 8.00 9.5 F331-I373 11.13 10.00 12.0 Results Crystal structure of HtrA2V226K/S306A To analyze the structural consequences of activating mutation V226K in hinge B of HtrA2 we obtained a crystal structure of the mutant HtrA2V226K/S306A. Its overall architecture resembles that of HtrA2S306A [31]. No structural rearrangement of PDZ domains vs. PD domains is noted compared to HtrA2S306A (RMS<0.4Å, based on 290 [98%] common Cα atoms). Minor changes are observed only around the mutated residue and on L3. The mutated K226 forms a hydrogen bond with E425, absent in the HtrA2S306A, while on L3 the unresolved gap reduces from 9 in HtrA2S306A [31] to only 6 residues 281RARDLG286, more firmly manifesting L3 to flap on a convex side of the trimer than in 1LCY. The additional hydrogen bond K226-E225 further stabilizes the interaction between PDZ and PD domains in hinge B region (Fig 2). Previous studies have demonstrated that the HtrA2E425L exhibits activity comparable to HtrA2V226K and the activities of both mutants are higher compared to that of the wild type [36]. While V226K mutation creates an additional hydrogen bond adding to the PD-PDZ interaction, the E425L mutation results in hydrophobic interaction of V226 and L425 within the same region of the protein, with a like effect in the inactive structure of HtrA2(Omi). 10.1371/journal.pone.0161526.g002Fig 2 Interdomain interactions in hinge B area. Left: HtrAS306A (1LCY) [31] and Right: HtrAV226K/S306A (5FHT; this work). Proteolytic domain is colored magenta, PDZ domain—blue. In the V226K mutant (Right) the distance 2,8Å indicates an H-bond/ionic pair Lys226(LC)PD -E425(α7)PDZ, absent in HtrA2 (Left). Otherwise, both structures overlap perfectly on each other, see text. Preliminaries of molecular dynamics Since the gaps in L3 and the PD-PDZ linker, not defined in the crystal structure, were restored computationally (see Materials and Methods), all the models were subject to initial optimization to relax any unfavorable contacts. This optimization only minutely adjusted the PD-PDZ arrangement compared to that seen in the crystal structure (Fig 3). This is evidenced by small values of root-mean-square (RMS) deviation calculated for the positions of Cα atoms between the optimized trimers and those found within the crystal structure of HtrA2S306A (1LCY) [31] (Table 3). The most pronounced adjustments concerned: i) unresolved fragments added computationally, including the ligand, and ii) adjustments of monomers at their interfaces within the trimer. None of the adjustment steps exceeded 1 Å of total RMS. 10.1371/journal.pone.0161526.g003Fig 3 Optimized starting trimer of HtrA2. Upper panel: The view perpendicular to the C3 axis onto the concave of the trimer. Units A-C are distinguished by color of protease domain ribbons: blue (A), green (B), yellow (C); the PDZ domains are shown uniformly as pink ribbons. The restored fragments, missing in 1LCY structure: L3, PD-PDZ-linker and the ligand, are highlighted as orange threads (plus short orange helices constituting part of the linkers). Lower Panel Left: Orientation as above, PDs are represented by a gray surface. PDZ ribbons (making 3 lids to the PDs) and modeled fragments are colored as above. Lower Panel Right. A lateral view of the latter. Selected features in the units are labeled in all views. 10.1371/journal.pone.0161526.t003Table 3 Modeling statistics of the starting structures. RMS [Ǻ]: steps i) and ii); (step i) only) [see text] HtrA2S306A b HtrA2S306A-pepc 1LCY [31] a 1.7; (0.8) 1.8; (1.0) HtrA2S306A b 1.0; (0.9) a 1LCY: Cα-based RMS calculation excludes residues unresolved in 1LCY [31] structure. b HtrA2S306A exemplifies 3 tested mutants: S306A, S306A/V226K and S306A/V325D, whose optimized starting structures fit to RMS<0.1 Ǻ. c HtrA2S306A-pep exemplifies 6 tested cases containing peptide ligand: S306A, S306A/V226K, S306A/V325D, HtrA2 wild-type, V226K and V325D, whose optimized starting structures fit mutually to RMS<0.1 Ǻ. Importantly, the ligand-binding and the apo structures have optimized similarly demonstrating the correctness of our approach to modeling the peptide in complex with the apo-HtrA2. The GWTMFWV peptide fits surprisingly well to the apoPD-PDZ interface. Notably, its insertion into 1LCY has introduced only 2 new contacts (using the 3.5 Å criterion) with PD, viz. those of W2 with F331(β12) and W6 with A197(LB). Based on the mere geometrical criterion, they appeared as modest as 7 contacts already present in the crystal structure of PDZ-ligand, viz. with (M366,L367)β14, E376α5, (H394,I397)β15 and (A424,Y428)α7 [54]. A minor adjustments of W2-T3 in the ligand and of β11-L2-β12 in PD have alleviated these unfavorable interactions. Thus, the ligand pocket of PDZ, even if seemingly inaccessible in resting HtrA2, is spacious enough to accept the heptapeptide. Productive molecular dynamics Within 50 ns MD simulations each system experienced mainly specific segmental/collective motions of PDZ and PD domains relative to each other (Table 4), but retained intra-segmental consistency within the domains. Save for the N-terminal α1 (residues 139–150), β1-LA-β2 (residues 168–177), L3 (residues 282–290) and the PD-PDZ linker (residues 346–352), which have undergone more pronounced rearrangements, inherent drifts of the domains from their original architectures have not been larger (RMS) than 1.2–1.5Å and 1.1–1.3Å for ~100 PDZ and ~160 selected PD Cα atoms, respectively. It is important to note that two of the four most dynamic structure elements, L3 (9 residues) and PD-PDZ-linker (15 residues) were those restored algorithmically [51]. Therefore they might have had their starting 3D architectures not soundly consistent with the structural context of the reported HtrA2 trimer and as such expectedly could have exhibited increased mobility to adjust to surrounding structure environments. Therefore, the only significant motions observed in the simulations have included segmental translations/rotations of PDZ relative to PD within each monomer (Figs 4 and 5). 10.1371/journal.pone.0161526.g004Fig 4 HtrA2 trimers after 50 ns MD, with reference to the starting structures: The pink PDZ (ribbons) make 3 lids to the PD (surface) of the trimer viewed perpendicular to C3 as in Fig 3. Left: The resultant wtHtrA2-ligand complex. PDZ domains, in Units A, B, C, (cyan, canary and amber ribbons, respectively) have moved equatorially (budding-flower-like) relative to their starting (pink) positons, especially in units A & C thus opening the PD-PDZ interface. Right: The resultant HtrA2S306A/V325D double mutant-ligand complex. Moving PDZ domains are colored as in the left panel. Only minor PDZ motions are seen; no opening of the PD-PDZ takes place. 10.1371/journal.pone.0161526.g005Fig 5 HtrA2-ligand complex unit C after 50 ns MD with reference to the initial structure. All but the top-right views have common orientation. PD N-terminal α1-α2 and C-terminal barrel are in the foreground (bottom, dark blue ribbon), while C-terminal α4 and N-terminal barrel are in the background (bottom, light blue ribbon). α1-α2 and α4 axes run roughly perpendicular to the figure plane. Important features of the structure and its dynamics are indicated. Top-left: Unit C as extracted from the trimer in Fig 4 left panel, and reoriented (see above). Bottom-left: As above, PD unwrapped form the surface. Top-right: Unit C is rotated 90° round the vertical axis to show relationships between the PDZ motion and exposure of the catalytic triad (olive green). Bottom-right: Visualization of a true uncovering of the catalytic triad after ca. 40° PDZ counterclockwise rotation (lid-opening) during the 50 ns MD, characterized in detail in the neighbor bottom-left panel and in Table 4. 10.1371/journal.pone.0161526.t004Table 4 Overview of simulation results. For a viewer having the N-terminus and PD C-terminal barrel in the foreground, as in Fig 5, “cc” and “c” denote, respectively, counterclockwise and clockwise rotation of PDZ versus PD. The viewer sees this rotation roughly round an axis parallel to α4 and passing for “cc” between α4 and α7, see Fig 5, and; for “c” by the peptide-binding motif 361YIGV364. “Δ”applies to “cc” only and refers to another measure of extent of the “cc” rotation, viz. to its associated arc at maximized radius. I179(β2) and L398(β15) roughly fit this radius tips, hence their Cα-Cα (vs. ~6Ǻ at the start) distance increases with “cc” rotation defining “Δ“. “i” indicates minute irregular motions. Unit A Unit B Unit C (Refer to Fig 5) containing peptide ligand HtrA2 cc ~30°/ Δ = 16.4 Ǻ cc ~0°/Δ = 10.5 Ǻ cc ~40°/Δ = 23.1Ǻ HtrA2S306A i cc ~50°/Δ = 24.4 Ǻ i HtrA2V226K c ~10° i i HtrA2V226K/S306A i i i HtrA2V325D i i i HtrA2V325D/S306A i i i no ligand (apo) HtrA2S306A i i i HtrA2V226K/S306A i i i HtrA2V325D/S306A i i i The segmental motions observed in the simulated models classify into three basic types (Table 4). The prevailing motion relies on minor and apparently non-specific rearrangements of PDZ versus PD. It is seen in all but one monomers of the HtrA2S306A and HtrA2V226K models, all monomers of HtrA2V226K/S306A, HtrA2V325D, HtrA2V325D/S306A models in the peptide containing trimers, as well as in the all apo HtrA2S306A, HtrA2V226K/S306A and HtrA2V325D/S306A trimers. Wild type ligand-bound HtrA2 and HtrA2S306A exhibit the second type of motion. With the MD progress two monomers within the former structure and one within the latter have performed a relatively consistent motion, illustrated in Fig 5. For a viewer having the N-terminus and the C-terminal barrel of PD in the foreground, PDZ rotates counterclockwise relative to PD around an axis between and roughly parallel to α4PD and α7PDZ, simultaneously passing near the N-front of α5PDZ. Some translation, hard to characterize with math rigor (see, however, Supplementary Material, containing the Principal Component Analysis, PCA, of the MD results), accompany the rotations of these units. The rotations are equivalent to opening the PD-PDZ interface in a lid-like way. The two hinges (A and B) are made of clusters of hydrophobic residues. The rotation of up to 50° (Table 4 and Fig 5) enables complete access both to the catalytic triad in PD and the peptide-binding site in PDZ. It is surprising that a similar motion is not seen in MD of HtrA2V226K, even though this mutant was demonstrated experimentally to exhibit increased proteolytic activity compared to the wild type [36]. The last type of motion, seen only in a single monomer of the ligand-bound HtrA2V226K also lifts the lid, but only slightly, about 10°, clockwise and “another way around”, as if the hinges were now fixed on the opposite side of the PD-PDZ interface. Trajectory analyses using the math rigors of PCA (see Supplementary Material) have confirmed the above conclusions. Our results demonstrate that the motions within the monomers in the trimer are not synchronous in the timescale of 50 ns (Fig 4). By selecting a pair of residues most remote to the hinges, e.g. PDZ(lid) L398(β15-α6 loop) with PD(jug) I179(β2) we measured the extent of opening of the PD-PDZ interface (Table 4 and Fig 5) in diverse units. The conformations of the catalytic triad residues were analyzed in the optimized starting and resultant product structures and have been found catalytically incompetent, as seen previously in the crystal structure. Typical distances in a catalytically ready triad in serine proteases of the HtrA type amount to 3.15±0.2 Ǻ for the Oγ(Ser)-Nε2(His) distance and to 3.3±0.4 Ǻ for the Nδ1(His)-Oδ1(Asp) distance ([19], compare also Ref. [32]). The average lengths of the sides of a triangle formed by the Cα-Cα distances within the catalytically competent triad, ΔS-H = 8,6±0.1 Ǻ, ΔD-H = 6.5±0.1 Ǻ and ΔS-D = 10.2±0.1 Ǻ) alike serve as a criterion of the catalytic competence of HtrA2. Neither any starting nor resultant structures have fitted these criteria. We speculate that substrate binding induces further rearrangements yielding a proteolytically competent conformation within the catalytic triad, though this is not reflected in our simulations where substrate has not been taken into account. Restrained Molecular Dynamics (RMD) In Figs 5 and 6 of our recent work [38] we have sketched an initial step to thermal activation of HtrA2, complying qualitatively with the TrIQ restraints. The model relies on a slight opening of the PD-PDZ(red) interface, in agreement with the MD results given here in Figs 4 and 5. In order to more rigorously resolve the PD-PDZ orientation fulfilling these restraints [38], now we have worked out the TrIQ results using RMD and arrived at the following results (see also Table 2). (i) Satisfying the TrIQ-inferred temperature-driven increases and decreases of selected 9 intra-unit distances is unfeasible, unless the already incomplete L3 is further truncated at each the β8 inlet to, and the β9 outlet from the L3 gap by extra ~8 residues, which removes steric hindrance of L3 to optimization. (ii) The optimization arrives at the same result no matter starting PD-PDZ orientation, i.g. the fully closed 1LCY-like [31] or fully open 3CS0-like, see Methods. (iii) The result, defined by the nine Cα-Cα distances given in the last column of Table 2, features a PD-PDZ opening intermediate between that in Fig 5 this work and Figs 5 and 6 in [38]. 10.1371/journal.pone.0161526.g006Fig 6 HtrA trimers; PD-PDZ(1) in surface representation. L3 and the linker are shown as orange threads, L2, L1 and LD as blue threads. (A) wtHtrA2-peptide starting structure in orientation identical to that in Fig 3 bottom-right. (B) wtHtrA2-peptide after 50 ns MD. (C) active DegP PDB (entry 3CS0 [17]) PDZ2 is omitted for clarity. In contrast to A and B, where L3 protrudes outside on the convex side of the trimer, in the active DegP and in other active HtrAs (not shown) L3 enters between PD and PDZ, onto the trimer’s concave, contributing (vide bottom) to the allosteric activation cascade: L3*-LD-L1-L2 [35]. Discussion The crystal structures of HtrA2S306A [31] and HtrA2S306A/V226K (this work) resolve to homotrimers having essentially identical architectures, characterized by a unique PD-PDZ arrangements not seen in the structures of other procaryotic and eukaryotic proteins of HtrA family. In contrast to the other family members, regardless in proteolytically more or less active forms, the catalytic and the PDZ domains in HtrA2 tightly pack against each other [31]. A major consequence of this feature is an apparent inaccessibility of the activating-peptide binding site on the PDZ part and of the substrate pocket on the PD part in both structures. As such, the structures meet a model of an inactive HtrA2(Omi) protein [31]. In our 50 ns MD simulation, two subunits in the trimers of HtrA2-ligand and one in the trimer of HtrA2S306A-ligand (Figs 4 and 5 and S1, S2 and S4 Figs) have reorganized in agreement with an opening of the tight PD-PDZ interface by lid-like rotations of ~30°-50° (Fig 5 and Table 4). The rotation was around an axis ~parallel to and located between α4PD and α7PDZ and passing by hinges A and B (Figs 4 and 5 and S4 Fig). These rotations resemble opening of the inactive HtrA2 toward structures typical of all proteolytically-competent HtrAs determined to date. The motions are in agreement with the FRET result reported by Chaganti et al. [61] because, while releasing access to the PDZ and PD binding sites, they maintain the F341Cα4-Y428Wα7 (F208Cα4-Y296Wα7 in their terminology) distance in hinge B roughly intact, as concluded from their FRET measurements [61]. These Authors also studied HtrA2 allosteric activation using MD [61–63]. They stipulated that the inactive, closed form of HtrA2 employs a non-canonical binding groove by the PD-PDZ interface, before being able to utilize the canonical peptide-recognition motif [62] i.e. the (β13-β14)PDZ carboxylate-binding loop, see above. In this work we neither pursue this issue nor the allosteric regulation of HtrA2 (caspase-dependent) via its N-terminus [63]. To date, 15 proteolytically active HtrA structures of widely diverse origin have been published, including E. coli HtrA(DegP) [17], Mycobacterium tuberculosis HtrA2 [23], E. coli DegS [20, 24, 26, 32], Legionella falloni DegQ [28], plant Arabidopsis thaliana Deg1 [29] and human HtrA1 [30]. We note that their common feature is a similar mutual location (even if diverse orientation) in space of their PDZ (PDZ1 in DegP and DegQ [17, 28]) vs. PD domains. Both domains do not form an interface. In the human HtrA1 the PDZ domain is not resolved within the crystal structure due to its high flexibility [30], however, HtrA1 can be active without PDZ domain. Similarly, human HtrA3 [34] and E. coli DegS [20, 26, 32] do not require PDZ domain for activity. Even more conservative, shared by all structures of the active proteases are the positions of the sensor loop L3. Despite its length varying by up to 6 residues between different HtrA proteases, L3 occupies in all active HtrA proteases a specific site between PD and PDZ domains, where it separates L2PD and α7PDZ(1), simultaneously being in contact with LD* of the proximal unit of the trimer (asterisk denotes a neighboring monomer). In fact, these commonalities apply to all three L3*-LD-L1-L2 “activation clusters” per trimer, mutually related by the C3-symmetry (Fig 6C) [15, 19, 35]. On the contrary, in the structure of inactive human HtrA2 and during entire MD simulation carried out for HtrA2- and HtrA2S306A-ligand system the sensor loop L3 persists in clearly different site. Fig 6 illustrates that L3 and (β13-α7)PDZ (proximal to the C-barrel side of PD, a part of hinge A) are in HtrA2 (Fig 6A) reversely positioned in space than they are in an active structure of any HtrA protein, taken DegP as an example (Fig 6C). Moreover, this “wrong” arrangement augments with the progress of MD (Fig 6B). None of the subunits exhibiting the PD-PDZ opening motion is capable of letting L3 enter between the own C-barrel and PZD as required to activate the protease. Reversely, the opening motion(s) pushes L3 more equatorial, to the exterior (convex) side of the gradually opening concave of the PD-PDZ interface (Fig 6A and 6B). In a view of the published experimental data on specific contribution of L3 in the activation process, it is clear that the (semi)open HtrA2 units do not approach true active states in MD. To achieve this, L3 and PDZ should have swapped their positions (Fig 6C), a too demanding task for any simple MD. This structural requisite is diagrammatically presented in Fig 7 for a monomeric unit. 10.1371/journal.pone.0161526.g007Fig 7 Diagrammatical scheme of structural requisite to HtrA2 proteolytic activation. N-helix and barrel are colored blue, C-barrel and helix are cyan, L3 is green, the triad is red, PDZ domain is pale yellow orthorhombic box. Left: Inactive closed form, L3 flaps on the exterior (convex in the trimer) side, opposite the location of the triad. Right: Active open form, L3 is slid between C-barrel and PDZ to the interior (concave in the trimer) side, disrupting hinge A. It is noteworthy that the HtrA2V226K/S306A mutant (Table 1, entry 5FHT), crystallizing version of HtrA2V226K more active than the wtHtrA2 [36], has crystallized in exactly the same inactive architecture as HtrA2S306A [31], with its L3 even more distinctly flapped out to the exterior (convex side) of the PD-PDZ interface than the latter. In summary, X-ray study combined with MD failed to explain higher proteolytic activity of the V226K mutant, because in the MD simulations wtHtrA2-ligand complex exhibits more pronounced PD-PDZ opening than HtrA2V226K-ligand does, while one would expect the opposite from comparing their activities. At the same time our studies explain the inactive phenotype of V325D mutant, which manifests no tendency to PD-PDZ opening in the MD simulation. Our results, combined with accumulated knowledge about structures of activated HtrAs has led to structural requirements, drawn schematically in Fig 7, that have to be met on a path from inactive to proteolytically active HtrA2(Omi) protease. While opening the PD-PDZ interface, somewhere on a way to activation an L3-PDZ positional swap is required, which could only be attained upon a crack of hinge A, the (β11-L2-β12)PD-(β14-α5)PDZ cluster, while retaining hinge B, the (β5-LC-β6)PD-(β13-α7)PDZ cluster still conserved. Clearly, the exact structural changes involved in the activation mechanism of HtrA2 remain open to debate until the crystal structure in the proteolytically active form is solved. Supporting Information S1 Fig PCA of wtHtrA2/peptide trimer. Three most significant eigenvectors, modes 1–3, are represented as square displacements of sequential MD-time-averaged Cα coordinates. The secondary-structure elements of HtrA2 monomer (1LCY pdb entry) are indicated below the abscissa for reference. Clearly, the majority of segmental motions are explained in mode 1, as modes 2 and 3 contain only residual fluctuations of LB, L3 and the PD-PDZ linker. Unit A is in red, unit B green, unit C blue. The lower-bottom panel includes respective scree plots. (PDF) Click here for additional data file. S2 Fig PCA of HtrA2S306A/peptide trimer. General legend see S1 Fig. Majority of segmental motions are explained in the first two modes, as mode 3 contains only residual fluctuations of LB, L3 and the PD-PDZ linker. (PDF) Click here for additional data file. S3 Fig PCA of apo HtrA2S306A trimer. General legend see S1 Fig. Majority of segmental motions are explained in the first two modes, as mode 3 contains only residual fluctuations of LB, L3 and the PD-PDZ linker. (PDF) Click here for additional data file. S4 Fig Visualization of motional modes 1–3 within unit C of wtHtrA2-peptide complex. Directional amplitudes (double-arrow modules ~ root squares of displacements in S1 Fig). of PCA-factorized motional modes of wtHtrA2-peptide complex are depicted on the HtrA2 mean structure. Cα-trace was interpolated to a smooth curve using VMD. The structures, in stereo, are oriented in agreement with all but top-right structures in Fig 5. Selected secondary-structure elements in mode 2 are marked. The motional arrows are not to the scale of decreasing variance (λ1/λ2/λ3 = 0.58/0.08/0.06). Instead, they are progressively scaled up by 0.87•(λ1-1/2, λ2-1/2, λ3-1/2), to visualize motions in modes 2 and 3. Double-arrows are cut off below 2Å, to expose only distinct segmental motions in each mode. (PDF) Click here for additional data file. S1 File Supporting methods and results. (PDF) Click here for additional data file. S1 Table PCA of wtHtrA2/peptide trimer: the summary of the first 30 PCA modes. Factors 3 to 30 accumulate 48% to 85%, respectively, of total variance in Unit B, fair scree; and 73% to 93%, respectively, of total variance in Unit C, steep scree. (PDF) Click here for additional data file. S2 Table PCA of HtrA2S306A/peptide trimer: the summary of the first 30 PCA modes. Factors 3 to 30 accumulate 67% to 91%, respectively, of total variance in Unit A, fair scree; and 81% to 96%, respectively, of total variance in Unit B, steep scree. (PDF) Click here for additional data file. S3 Table PCA of apo HtrA2S306A trimer: the summary of the first 30 PCA modes. Factors 3 to 30 accumulate 49% to 88%, respectively, of total variance in Unit C, fair scree; and 69% to 91%, respectively, of total variance in Unit A, moderate scree. (PDF) Click here for additional data file. ==== Refs References 1 Krick S , Shi S , Ju W , Faul C , Tsai SY , Mundel P , et al Mpv17l protects against mitochondrial oxidative stress and apoptosis by activation of Omi/HtrA2 protease . Proceedings of the National Academy of Sciences of the United States of America . 2008 ;105 (37 ):14106 –11 . Epub 2008/09/06. 10.1073/pnas.0801146105 18772386 2 Moisoi N , Klupsch K , Fedele V , East P , Sharma S , Renton A , et al Mitochondrial dysfunction triggered by loss of HtrA2 results in the activation of a brain-specific transcriptional stress response . Cell death and differentiation . 2009 ;16 (3 ):449 –64 . Epub 2008/11/22. 10.1038/cdd.2008.166 .19023330 3 Vande Walle L , Lamkanfi M , Vandenabeele P . The mitochondrial serine protease HtrA2/Omi: an overview . Cell death and differentiation . 2008 ;15 (3 ):453 –60 . Epub 2008/01/05. 10.1038/sj.cdd.4402291 .18174901 4 Bhuiyan MS , Fukunaga K . Mitochondrial serine protease HtrA2/Omi as a potential therapeutic target . Current drug targets . 2009 ;10 (4 ):372 –83 . Epub 2009/04/10. .19355862 5 Zurawa-Janicka D , Skorko-Glonek J , Lipinska B . HtrA proteins as targets in therapy of cancer and other diseases . Expert opinion on therapeutic targets . 2010 ;14 (7 ):665 –79 . Epub 2010/05/18. 10.1517/14728222.2010.487867 .20469960 6 Hartkamp J , Carpenter B , Roberts SG . The Wilms' tumor suppressor protein WT1 is processed by the serine protease HtrA2/Omi . Molecular cell . 2010 ;37 (2 ):159 –71 . Epub 2010/02/04. 10.1016/j.molcel.2009.12.023 20122399 7 Dagda RK , Chu CT . Mitochondrial quality control: insights on how Parkinson's disease related genes PINK1, parkin, and Omi/HtrA2 interact to maintain mitochondrial homeostasis . Journal of bioenergetics and biomembranes . 2009 ;41 (6 ):473 –9 . Epub 2009/12/17. 10.1007/s10863-009-9255-1 20012177 8 de Castro IP , Martins LM , Loh SH . Mitochondrial quality control and Parkinson's disease: a pathway unfolds . Molecular neurobiology . 2011 ;43 (2 ):80 –6 . Epub 2010/12/02. 10.1007/s12035-010-8150-4 21120708 9 Gupta S , Singh R , Datta P , Zhang Z , Orr C , Lu Z , et al The C-terminal tail of presenilin regulates Omi/HtrA2 protease activity . The Journal of biological chemistry . 2004 ;279 (44 ):45844 –54 . Epub 2004/08/06. 10.1074/jbc.M404940200 .15294909 10 Behbahani H , Pavlov PF , Wiehager B , Nishimura T , Winblad B , Ankarcrona M . Association of Omi/HtrA2 with gamma-secretase in mitochondria . Neurochemistry international . 2010 ;57 (6 ):668 –75 . Epub 2010/08/14. 10.1016/j.neuint.2010.08.004 .20705111 11 Park HJ , Kim SS , Seong YM , Kim KH , Goo HG , Yoon EJ , et al Beta-amyloid precursor protein is a direct cleavage target of HtrA2 serine protease. Implications for the physiological function of HtrA2 in the mitochondria . The Journal of biological chemistry . 2006 ;281 (45 ):34277 –87 . Epub 2006/09/14. 10.1074/jbc.M603443200 .16968707 12 Park HJ , Seong YM , Choi JY , Kang S , Rhim H . Alzheimer's disease-associated amyloid beta interacts with the human serine protease HtrA2/Omi . Neuroscience letters . 2004 ;357 (1 ):63 –7 . Epub 2004/03/24. 10.1016/j.neulet.2003.11.068 .15036614 13 Kooistra J , Milojevic J , Melacini G , Ortega J . A new function of human HtrA2 as an amyloid-beta oligomerization inhibitor . Journal of Alzheimer's disease: JAD . 2009 ;17 (2 ):281 –94 . Epub 2009/06/09. 10.3233/JAD-2009-1037 19502709 14 Chien J , Campioni M , Shridhar V , Baldi A . HtrA serine proteases as potential therapeutic targets in cancer . Current cancer drug targets . 2009 ;9 (4 ):451 –68 . Epub 2009/06/13. 19519315 15 Clausen T , Kaiser M , Huber R , Ehrmann M . HTRA proteases: regulated proteolysis in protein quality control . Nature reviews Molecular cell biology . 2011 ;12 (3 ):152 –62 . Epub 2011/02/18. 10.1038/nrm3065 .21326199 16 Perona JJ , Craik CS . Evolutionary divergence of substrate specificity within the chymotrypsin-like serine protease fold . The Journal of biological chemistry . 1997 ;272 (48 ):29987 –90 . Epub 1997/12/31. .9374470 17 Krojer T , Sawa J , Schafer E , Saibil HR , Ehrmann M , Clausen T . Structural basis for the regulated protease and chaperone function of DegP . Nature . 2008 ;453 (7197 ):885 –90 . Epub 2008/05/23. 10.1038/nature07004 .18496527 18 Krojer T , Sawa J , Huber R , Clausen T . HtrA proteases have a conserved activation mechanism that can be triggered by distinct molecular cues . Nature structural & molecular biology . 2010 ;17 (7 ):844 –52 . Epub 2010/06/29. 10.1038/nsmb.1840 .20581825 19 Singh N , Kuppili RR , Bose K . The structural basis of mode of activation and functional diversity: a case study with HtrA family of serine proteases . Archives of biochemistry and biophysics . 2011 ;516 (2 ):85 –96 . Epub 2011/10/27. 10.1016/j.abb.2011.10.007 .22027029 20 de Regt AK , Kim S , Sohn J , Grant RA , Baker TA , Sauer RT . A conserved activation cluster is required for allosteric communication in HtrA-family proteases . Structure . 2015 ;23 (3 ):517 –26 . Epub 2015/02/24. 10.1016/j.str.2015.01.012 25703375 21 Krojer T , Garrido-Franco M , Huber R , Ehrmann M , Clausen T . Crystal structure of DegP (HtrA) reveals a new protease-chaperone machine . Nature . 2002 ;416 (6879 ):455 –9 . Epub 2002/03/29. 10.1038/416455a .11919638 22 Jiang J , Zhang X , Chen Y , Wu Y , Zhou ZH , Chang Z , et al Activation of DegP chaperone-protease via formation of large cage-like oligomers upon binding to substrate proteins . Proceedings of the National Academy of Sciences of the United States of America . 2008 ;105 (33 ):11939 –44 . Epub 2008/08/14. 10.1073/pnas.0805464105 18697939 23 Mohamedmohaideen NN , Palaninathan SK , Morin PM , Williams BJ , Braunstein M , Tichy SE , et al Structure and function of the virulence-associated high-temperature requirement A of Mycobacterium tuberculosis . Biochemistry . 2008 ;47 (23 ):6092 –102 . Epub 2008/05/16. 10.1021/bi701929m .18479146 24 Wilken C , Kitzing K , Kurzbauer R , Ehrmann M , Clausen T . Crystal structure of the DegS stress sensor: How a PDZ domain recognizes misfolded protein and activates a protease . Cell . 2004 ;117 (4 ):483 –94 . Epub 2004/05/13. .15137941 25 Sawa J , Malet H , Krojer T , Canellas F , Ehrmann M , Clausen T . Molecular adaptation of the DegQ protease to exert protein quality control in the bacterial cell envelope . The Journal of biological chemistry . 2011 ;286 (35 ):30680 –90 . Epub 2011/06/21. 10.1074/jbc.M111.243832 21685389 26 Sohn J , Grant RA , Sauer RT . Allosteric activation of DegS, a stress sensor PDZ protease . Cell . 2007 ;131 (3 ):572 –83 . Epub 2007/11/06. 10.1016/j.cell.2007.08.044 .17981123 27 Kim DY , Kim DR , Ha SC , Lokanath NK , Lee CJ , Hwang HY , et al Crystal structure of the protease domain of a heat-shock protein HtrA from Thermotoga maritima . The Journal of biological chemistry . 2003 ;278 (8 ):6543 –51 . Epub 2002/11/30. 10.1074/jbc.M208148200 .12458220 28 Wrase R , Scott H , Hilgenfeld R , Hansen G . The Legionella HtrA homologue DegQ is a self-compartmentizing protease that forms large 12-meric assemblies . Proceedings of the National Academy of Sciences of the United States of America . 2011 ;108 (26 ):10490 –5 . Epub 2011/06/15. 10.1073/pnas.1101084108 21670246 29 Kley J , Schmidt B , Boyanov B , Stolt-Bergner PC , Kirk R , Ehrmann M , et al Structural adaptation of the plant protease Deg1 to repair photosystem II during light exposure . Nature structural & molecular biology . 2011 ;18 (6 ):728 –31 . Epub 2011/05/03. 10.1038/nsmb.2055 .21532594 30 Truebestein L , Tennstaedt A , Monig T , Krojer T , Canellas F , Kaiser M , et al Substrate-induced remodeling of the active site regulates human HTRA1 activity . Nature structural & molecular biology . 2011 ;18 (3 ):386 –8 . Epub 2011/02/08. 10.1038/nsmb.2013 .21297635 31 Li W , Srinivasula SM , Chai J , Li P , Wu JW , Zhang Z , et al Structural insights into the pro-apoptotic function of mitochondrial serine protease HtrA2/Omi . Nature structural biology . 2002 ;9 (6 ):436 –41 . Epub 2002/04/23. 10.1038/nsb795 .11967569 32 Sohn J , Grant RA , Sauer RT . Allostery is an intrinsic property of the protease domain of DegS: implications for enzyme function and evolution . The Journal of biological chemistry . 2010 ;285 (44 ):34039 –47 . Epub 2010/08/27. 10.1074/jbc.M110.135541 20739286 33 Kim S , Grant RA , Sauer RT . Covalent linkage of distinct substrate degrons controls assembly and disassembly of DegP proteolytic cages . Cell . 2011 ;145 (1 ):67 –78 . Epub 2011/04/05. 10.1016/j.cell.2011.02.024 21458668 34 Glaza P , Osipiuk J , Wenta T , Zurawa-Janicka D , Jarzab M , Lesner A , et al Structural and Functional Analysis of Human HtrA3 Protease and Its Subdomains . PloS one . 2015 ;10 (6 ):e0131142 Epub 2015/06/26. 10.1371/journal.pone.0131142 26110759 35 Hansen G , Hilgenfeld R . Architecture and regulation of HtrA-family proteins involved in protein quality control and stress response . Cellular and molecular life sciences: CMLS . 2013 ;70 (5 ):761 –75 . Epub 2012/07/19. 10.1007/s00018-012-1076-4 .22806565 36 Zurawa-Janicka D , Jarzab M , Polit A , Skorko-Glonek J , Lesner A , Gitlin A , et al Temperature-induced changes of HtrA2(Omi) protease activity and structure . Cell stress & chaperones . 2013 ;18 (1 ):35 –51 . Epub 2012/08/02. 10.1007/s12192-012-0355-1 22851136 37 Lee HJ , Zheng JJ . PDZ domains and their binding partners: structure, specificity, and modification . Cell communication and signaling: CCS . 2010 ;8 :8 Epub 2010/06/01. 10.1186/1478-811X-8-8 20509869 38 Jarzab M , Wenta T , Zurawa-Janicka D , Polit A , Gieldon AJ , Wysocka M , et al Intra- and intersubunit changes accompanying thermal activation of the HtrA2(Omi) protease homotrimer . Biochimica et biophysica acta . 2016 ;1864 (3 ):283 –96 . Epub 2015/12/26. 10.1016/j.bbapap.2015.12.002 .26702898 39 Martins LM , Turk BE , Cowling V , Borg A , Jarrell ET , Cantley LC , et al Binding specificity and regulation of the serine protease and PDZ domains of HtrA2/Omi . The Journal of biological chemistry . 2003 ;278 (49 ):49417 –27 . Epub 2003/09/27. 10.1074/jbc.M308659200 .14512424 40 Gray CW , Ward RV , Karran E , Turconi S , Rowles A , Viglienghi D , et al Characterization of human HtrA2, a novel serine protease involved in the mammalian cellular stress response . European journal of biochemistry / FEBS . 2000 ;267 (18 ):5699 –710 . Epub 2000/09/06. .10971580 41 Mansoor SE , Dewitt MA , Farrens DL . Distance mapping in proteins using fluorescence spectroscopy: the tryptophan-induced quenching (TrIQ) method . Biochemistry . 2010 ;49 (45 ):9722 –31 . Epub 2010/10/05. 10.1021/bi100907m 20886836 42 Lipinska B , Zylicz M , Georgopoulos C . The HtrA (DegP) protein, essential for Escherichia coli survival at high temperatures, is an endopeptidase . Journal of bacteriology . 1990 ;172 (4 ):1791 –7 . Epub 1990/04/01. 2180903 43 Battye TG , Kontogiannis L , Johnson O , Powell HR , Leslie AG . iMOSFLM: a new graphical interface for diffraction-image processing with MOSFLM . Acta crystallographica Section D, Biological crystallography . 2011 ;67 (Pt 4 ):271 –81 . Epub 2011/04/05. 10.1107/S0907444910048675 21460445 44 The CCP4 suite: programs for protein crystallography . Acta crystallographica Section D, Biological crystallography . 1994 ;50 (Pt 5 ):760 –3 . Epub 1994/09/01. 10.1107/S0907444994003112 .15299374 45 Evans P . Scaling and assessment of data quality . Acta crystallographica Section D, Biological crystallography . 2006 ;62 (Pt 1 ):72 –82 . Epub 2005/12/22. 10.1107/S0907444905036693 .16369096 46 Evans PR , Murshudov GN . How good are my data and what is the resolution? Acta crystallographica Section D, Biological crystallography . 2013 ;69 (Pt 7 ):1204 –14 . Epub 2013/06/26. 10.1107/S0907444913000061 23793146 47 McCoy AJ , Grosse-Kunstleve RW , Adams PD , Winn MD , Storoni LC , Read RJ . Phaser crystallographic software . Journal of applied crystallography . 2007 ;40 (Pt 4 ):658 –74 . Epub 2007/08/01. 10.1107/S0021889807021206 19461840 48 Emsley P , Cowtan K . Coot: model-building tools for molecular graphics . Acta crystallographica Section D, Biological crystallography . 2004 ;60 (Pt 12 Pt 1 ):2126 –32 . Epub 2004/12/02. 10.1107/S0907444904019158 .15572765 49 Murshudov GN , Vagin AA , Dodson EJ . Refinement of macromolecular structures by the maximum-likelihood method . Acta crystallographica Section D, Biological crystallography . 1997 ;53 (Pt 3 ):240 –55 . Epub 1997/05/01. 10.1107/S0907444996012255 .15299926 50 Brunger AT . Free R value: a novel statistical quantity for assessing the accuracy of crystal structures . Nature . 1992 ;355 (6359 ):472 –5 . Epub 1992/01/30. .18481394 51 Tripos International. SYBYL-X 1.2. 52 Pearlman DA , Case DA , Caldwell JW , Ross WS , Cheatham TE III, DeBolt S , et al AMBER, a package of computer programs for applying molecular mechanics, normal mode analysis, molecular dynamics and free energy calculations to simulate the structural and energetic properties of molecules . Comp Phys Commun . 1995 ;91 :1 –41 . 53 Liu DC , Nocedal J . On the limited memory method for large scale optimization . Math Programming B . 1989 ;45 :503 –28 . 54 Zhang Y , Appleton BA , Wu P , Wiesmann C , Sidhu SS . Structural and functional analysis of the ligand specificity of the HtrA2/Omi PDZ domain . Protein science: a publication of the Protein Society . 2007 ;16 (8 ):1738 –50 . Epub 2007/07/28. 10.1110/ps.072833207 17656586 55 Jorgensen WL , Chandrasekhar J , Madura JD , Impey RW , Klein ML . Comparison of simple potential functions for simulating liquid water . J Chem Phys . 1983 ;79 :926 –35 . 56 Ryckaert JP , Ciccotti G , Berendsen HJC . Numerical integration of the cartesian equetions of motion of a system with constraints: molecular dynamics of n-alkenes . J Comp Physiol . 1977 ;23 :327 –41 . 57 Essmann U , Perera L , Berkowitz ML , Darden T , Lee H , Pedersen LG . A smooth particle mesh Ewald method . J Chem Phys . 1995 ;103 :8577 –93 . 58 Goetz AW , Williamson MJ , Xu D , Poole D , Grand SL , Walker RC . Routine microsecond molecular dynamics simulations with AMBER—Part I: Generalized Born . J Chem Theory Comput . 2012 ;8 :1542 –55 . 22582031 59 Pierce LCT , Salomon FR , de Oliveira CAF , McCammon JA , Walker RC . Routine access to milisecond timescale events with accelerated molecular dynamics . Journal Chem Theory Comput . 2012 ;8 :2997 –3002 .22984356 60 Nickolls J , Buck I , Garland M , Skadron K . Scalable Parallel Programming with CUDA . Queue—GPU . 2008 ;6 :40 –53 . 61 Chaganti LK , Kuppili RR , Bose K . Intricate structural coordination and domain plasticity regulate activity of serine protease HtrA2 . FASEB journal: official publication of the Federation of American Societies for Experimental Biology . 2013 ;27 (8 ):3054 –66 . Epub 2013/04/24. 10.1096/fj.13-227256 .23608143 62 Bejugam PR , Kuppili RR , Singh N , Gadewal N , Chaganti LK , Sastry GM , et al Allosteric regulation of serine protease HtrA2 through novel non-canonical substrate binding pocket . PloS one . 2013 ;8 (2 ):e55416 Epub 2013/03/05. 10.1371/journal.pone.0055416 23457469 63 Singh N , D'Souza A , Cholleti A , Sastry GM , Bose K . Dual regulatory switch confers tighter control on HtrA2 proteolytic activity . The FEBS journal . 2014 ;281 (10 ):2456 –70 . Epub 2014/04/05. 10.1111/febs.12799 .24698088

PMC005xxxxxx/PMC5003399.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757157510.1371/journal.pone.0162023PONE-D-16-19805Research ArticleBiology and Life SciencesCell BiologyCellular TypesAnimal CellsGlial CellsMacroglial CellsAstrocytesResearch and Analysis MethodsModel OrganismsAnimal ModelsMarmosetsBiology and life sciencesOrganismsAnimalsVertebratesAmniotesMammalsPrimatesMonkeysNew World monkeysMarmosetsBiology and Life SciencesAnatomyBrainCerebral CortexMedicine and Health SciencesAnatomyBrainCerebral CortexResearch and Analysis MethodsImaging TechniquesFluorescence ImagingBiology and Life SciencesAnatomyBrainCerebral CortexCerebellumMedicine and Health SciencesAnatomyBrainCerebral CortexCerebellumMedicine and Health SciencesEpidemiologyDisease VectorsViral VectorsBiology and Life SciencesMicrobiologyVirologyViral Transmission and InfectionViral VectorsBiology and Life SciencesAnatomyBrainCerebrumMedicine and Health SciencesAnatomyBrainCerebrumBiology and Life SciencesAnatomyBrainCerebellar CortexMedicine and Health SciencesAnatomyBrainCerebellar CortexViral Vector-Based Dissection of Marmoset GFAP Promoter in Mouse and Marmoset Brains Marmoset-Derived Astrocyte-Specific PromoterShinohara Yoichiro 12Konno Ayumu 1Takahashi Nobutaka 1Matsuzaki Yasunori 1Kishi Shoji 2Hirai Hirokazu 13*1 Department of Neurophysiology & Neural Repair, Gunma University Graduate School of Medicine, Maebashi, Gunma, Japan2 Department of Ophthalmology, Gunma University Graduate School of Medicine, Maebashi, Gunma, Japan3 Research Program for Neural Signalling, Division of Endocrinology, Metabolism and Signal Research, Gunma University Initiative for Advanced Research, Maebashi, Gunma, JapanNagai Yoshitaka EditorOsaka University Graduate School of Medicine, JAPANCompeting Interests: The authors have declared that no competing interests exist. Conceptualization: YS AK HH. Data curation: YS AK. Formal analysis: YS. Funding acquisition: AK HH. Investigation: YS AK. Methodology: YS AK YM HH. Project administration: HH. Resources: NT. Supervision: SK HH. Validation: YM. Visualization: HH. Writing – original draft: YS HH. Writing – review & editing: YS HH. * E-mail: hirai@gunma-u.ac.jp29 8 2016 2016 11 8 e01620238 6 2016 16 8 2016 © 2016 Shinohara et al2016Shinohara et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Adeno-associated virus (AAV) vectors are small in diameter, diffuse easily in the brain, and represent a highly efficient means by which to transfer a transgene to the brain of a large animal. A major demerit of AAV vectors is their limited accommodation capacity for transgenes. Thus, a compact promoter is useful when delivering large transgenes via AAV vectors. In the present study, we aimed to identify the shortest astrocyte-specific GFAP promoter region that could be used for AAV-vector-mediated transgene expression in the marmoset brain. The 2.0-kb promoter region upstream of the GFAP gene was cloned from the marmoset genome, and short promoters (1.6 kb, 1.4 kb, 0.6 kb, 0.3 kb and 0.2 kb) were obtained by progressively deleting the original 2.0-kb promoter from the 5’ end. The short promoters were screened in the mouse cerebellum in terms of their strength and astrocyte specificity. We found that the 0.3-kb promoter maintained 40% of the strength of the original 2.0-kb promoter, and approximately 90% of its astrocyte specificity. These properties were superior to those of the 1.4-kb, 0.6-kb (20% promoter strength) and 0.2-kb (70% astrocyte specificity) promoters. Then, we verified whether the 0.3-kb GFAP promoter retained astrocyte specificity in the marmoset cerebral cortex. Injection of viral vectors carrying the 0.3-kb marmoset GFAP promoter specifically transduced astrocytes in both the cerebral cortex and cerebellar cortex of the marmoset. These results suggest that the compact 0.3-kb promoter region serves as an astrocyte-specific promoter in the marmoset brain, which permits us to express a large gene by AAV vectors that have a limited accommodation capacity. Brain Mapping by Integrated Neurotechnologies for Disease Studies (Brain/MINDS)Hirai Hirokazu MEXT (Ministry of Education, Culture, Sports, Science and Technology) KAKENHI15K18330Konno Ayumu This research is (partially) supported by the program for Brain Mapping by Integrated Neurotechnologies for Disease Studies (Brain/MINDS) from the Ministry of Education, Culture, Sports, Science and Technology (MEXT), and the Japan Agency for Medical Research and Development, AMED, to HH and the MEXT KAKENHI (15K18330) to AK. Data AvailabilityAll relevant data are within the paper.Data Availability All relevant data are within the paper. ==== Body Introduction Astrocytes express various receptors and transporters for neurotransmitters and are actively involved in neuronal processing by modulating local synaptic functions: impairment of astrocytes affects the synaptic function, leading to associated behavioral defects. For example, in the cerebellar cortex, genetic removal of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)-type glutamate receptors expressed on Bergmann glial appendages results in impairments in fine motor coordination [1]. On the other hand, mutant mice lacking type-1 cannabinoid (CB1) receptors specifically in hippocampal astrocytes failed to demonstrate long-term depression of CA3-CA1 synapses and exhibited impairments in spatial working memory [2]. Moreover, astrocytes have been proposed to underlie the pathological states of various neurological diseases, such as spinocerebellar ataxia (SCA) type 1 [3, 4], SCA type 7 [5, 6] and multiple sclerosis [7]. Thus, selective gene modification of astrocytes is a useful approach for exploring the pathophysiological roles of astrocytes on a molecular basis and could be a therapeutic intervention for diseases associated with astrocyte impairment. Vectors derived from adeno-associated virus (AAV) and lentivirus are valuable for efficiently introducing transgenes into astrocytes [8–10]. AAV vectors have smaller diameters (~20 nm) than lentiviral vectors (~100 nm), and diffuse easily into the brain parenchyma, leading to transduction across large areas of brain tissue [8]. Therefore, AAV vectors are preferable to lentiviral vectors as mediators of gene transfer to large areas of the brain in the case of large animals such as primates. A major drawback of AAV vectors, however, is a limited capacity to accommodate a transgene. While lentiviral vectors can accommodate a transgene as large as 8 kb, AAV vectors have a capacity of only 4.7 kb, including a promoter sequence [11]. Hence, it is difficult to express a 3-kb transgene with a 2-kb promoter. Generally, cell type-specific promoters are large in size because they contain a region that restricts the promoter activity to a specific cell population. Thus, a compact astrocyte-specific promoter is valuable for AAV vector-mediated transgene expression. Common marmosets (Callithrix jacchus) have several advantages over macaques because they (1) are smaller and easier to handle, (2) have a higher efficiency in producing offspring and (3) do not carry fatal zoonotic diseases, such as the herpes b virus [12]. Marmosets can use their fingers to pinch food, quickly move upward and downward between tree limbs, live in family groups and communicate with each other through various vocalizations. These characteristics suggest that marmosets are superior to mice for examining a variety of brain functions, such as motor control, language and social behavior. Thus, marmosets have been increasingly used in the field of neuroscience. AAV vector-mediated transfer of various transgenes into astrocytes of the large marmoset brain in vivo allows us to clarify the pathophysiological roles that astrocytes play in the brain and, as a result, further extend the astrocyte-specific transgene expression capabilities available for gene therapies. In the present study, we aimed to develop a compact, astrocyte-specific promoter that could be used to transfer a large transgene (more than 3 kb) in combination with AAV vectors in the marmoset brain. Materials and Methods Animals All procedures regarding animal care and treatment were performed in accordance with the institutional and national guidelines and were approved by the Institutional Committee of Gunma University (No. 12–013; 15–038). Wild-type C57BL/6 mice aged between 3 and 4 weeks and two young adult common marmosets (Callithrix jacchus, H033: male, body weight (BW) 331g, H039: male, BW 343g) were used in this study. Mice were bred and housed at the Gunma University. Common marmosets were purchased through CLEA Japan (Tokyo, Japan) and bred at the Gunma University Bioresource Center. All animals were bred for research purpose. Common marmosets were kept in individual primate cages (375 mm × 550 mm × 762 mm) and maintained in rooms under controlled temperature (26 to 28°C), humidity (30 to 70%) on a 12-h light and 12-h dark cycle. We daily gave water ad libitum and 40 to 50 g of soaked monkey chow (CMS-1, CLEA Japan) with vitamin supplements, fresh fruit, vegetables, boiled chicken or milk powder. Wood branches and iron perches were placed in each cage for environmental enrichment. Animals were daily monitored to assess their health and symptoms, including a food intake, diarrhea, weight loss and trauma. Handling all animals was based on the Guide for the Care and Use of Laboratory Animals (8th edition). We made every effort to minimize animal suffering and reduce the number of animals used in the present study. Construction of expression plasmid with cjGFAP promoters The marmoset GFAP (Callithrix jacchus GFAP; cjGFAP) promoter region (Accession number: BR001393) upstream of the Gfap gene was amplified by polymerase chain reaction (PCR) of marmoset genomic DNA. To obtain the original 2-kb cjGFAP promoter region effectively, we performed nested PCR with 2 pairs of primers: cjGFAP (Nest)-F (forward primer), 5’-AGGGTCAGATGTGACTAGAGCC-3’ / cjGFAP (Nest)-R (reverse primer), 5’-GACGATTGTTGGACAGTGAG-3’ and cjGFAP (2.0)-F, 5’- CGGAACGCGTATGTGGGAAGATTGCTTGAGCCTAG-3’ / cjGFAP (2.0)-R, 5’-GTCGAATTCCCTGCCCTGGCTCTGCTTGC-3'. The shorter GFAP promoters were produced by 5’ deletion of the original 2-kb GFAP promoter using the following primers: cjGFAP (1.6)-F, 5’-CGGAACGCGTCGTGCCCACTGAATGACTCACC-3’, cjGFAP (1.4)-F, 5’-CGGAACGCGTGGCGCCACCGGCGGTGGAGAAC-3’, cjGFAP (0.6)-F, 5’- CGAACGCGTATCAAAAAGCTGGAAGGCAG-3’, cjGFAP (0.3)-F, 5’- CGGAACGCGTGTGGTCCAACCAACCCTTCTTGAC-3’ or cjGFAP (0.2)-F, 5’- CGGAACGCGTTGCCTCATGCAGGAGTTGGCGTG-3’ and cjGFAP (2.0)-R, 5’-GTCGAATTCCCTGCCCTGGCTCTGCTTGC-3'. The 0.3-kb mouse GFAP (mGFAP) (Accession number: BR001394) was amplified by PCR of mouse genomic DNA using mGfa2 (0.3)-F, 5’-CCGACGCGTGTGGGTCTTCATGCTTGACA-3’ and mGFAP (0.3)-R, 5’-GGAATTCCCTGCCCTGCCTCTGCTG-3’. The 0.3-kb human GFAP (hGFAP) (Accession number: BR001395) was amplified by PCR of human genomic DNA using hGFAP (0.3)-F, 5’- CGACGCGTTTCTTGACCCACCTTCCTAGAG-3’ and hGFAP (0.3)-R, 5’- GGAATTCCCTGCTCTGGCTCTGCTCG-3’. cjGFAP promoters ≤2.0 kb and the 0.3-kb mouse and human GFAP promoters were inserted into the MluI/EcoRI-digested site of the pCL20c-GFP lentiviral vector plasmid. The 0.3-kb cjGFAP promoter was inserted into XhoI/AgeI-digested pAAV-GFP AAV vectors. Proper insertion of the promoters into the viral plasmids was verified by DNA sequencing. Virus preparation Vesicular stomatitis virus glycoprotein (VSV-G)-pseudotyped lentiviral vectors were produced as previously described [13]. We transfected a mixture of four plasmids, consisting of pCAGkGP1R, pCAG4RTR2, pCAG-VSV-G and pCL20c/GFAP-GFP, into HEK293T cells using the calcium phosphate precipitation method. The supernatant containing the viral particles was harvested at 48 h post-transfection. After ultracentrifugation of the supernatant, the precipitated virus particles were resuspended in 70 μl of phosphate-buffered saline (-) (PBS). The lentiviral titers were determined using the comparative threshold cycle (CT) method by quantitative real-time PCR and with the following procedure: genomic RNA was isolated from 2 μl of viral solutions using an RNeasy Mini Kit (Qiagen, Hilden, Germany), which was subsequently reverse-transcribed using the ReverTra Ace qPCR RT Master Mix with gDNA Remover (Toyobo, Tokyo, Japan). The amounts of the synthesized cDNAs were quantified by quantitative real-time PCR using a protocol of 95°C for 1 min, 95°C for 15 s and 60°C for 30 s, 40 cycles (Takara Thermal Cycler Dice TP800, Takara Bio, Shiga, Japan) using THUNDERBIRD® SYBR® qPCR Mix (Toyobo) and the following primers: EGFP-F, 5’-TGG TGC AGA TGA ACT TCA GGG-3’ and EGFP-R, 5’-GTA AAC GGC CAC AAG TTC AGC-3’. The lentiviral solution was stored at 4°C and used within 2 weeks. Recombinant single-strand AAV9 vectors were produced by transfection of HEK293T cells (Thermo Fisher Scientific, Waltham, MA) with pAAV/cjGFAP-GFP, pAAV2/9 (kindly provided by Dr. J. Wilson) and a helper plasmid (Stratagene, La Jolla, CA) as previously described [14]. The viral particles were purified using ammonium sulfate precipitation and iodixanol continuous gradient centrifugation as previously described. The genomic titer of the purified AAV9 vectors was determined by real-time PCR. The pAAV/cjGFAP-GFP plasmids were used for the generation of standard curves of viral genomic titer. Injection of viral vectors into mouse and marmoset brains After deep anesthesia via an intra-peritoneal injection of ketamine (100 mg/kg BW) and xylazine (10 mg/kg BW), mice were placed in a stereotactic frame. For injection into the cerebellum, the skin covering the occipital bone was cut, and a burr hole was made 2 mm caudal from the lambda. The tip of a Hamilton syringe (33 gauge) with an attached micropump (UltraMicroPump II; World Precision Instrument (WPI) Sarasota, FL, USA) was inserted 1.8 mm below the pia mater of the cerebellar vermis. The viral solution (10 μl) was injected at a rate 450 nl/min using a microprocessor-based controller (Micro4; WPI). For injection into the cerebral cortex, the skin over the cerebral hemisphere was cut, and a burr hole was made 1 mm anterior-posterior (A/P), 1 mm medial-lateral (M/P) and 0.8 mm dorsal-ventral (D/V) from the bregma. The viral solution (5 μl) was injected into the cerebral cortex at a rate of 225 nl/min using a Micro4 (WPI). The syringe was left in place for 2 min following the injection. After closing the scalp, the mice were kept on a heating pad until they recovered from the anesthesia. Then, the mice were returned to standard home cages. For injection of AAV9 vectors, each marmoset received an intramuscular injection of ketamine and xylazine (20–25 mg/kg BW and 1.6–2.0 mg/kg BW, respectively) for sedation, and anesthesia was maintained by inhalation of 2.0–2.5% isoflurane to minimize their suffering and distress. The marmoset was then placed in a stereotactic frame, and heart rate and oxygen saturation were monitored throughout the procedure. A burr hole was made for cerebellar injection at 4 mm A/P, 3 mm M/L and 4 mm D/V from the external occipital protuberance. A second burr hole was made for cerebral injection at 0 ± 3 mm A/P, 2 mm M/L and 2.5 mm D/V from the bregma. Fifty (cerebellum) and 10 (cerebrum) μl of AAV9 vector suspension (2.0×1012 vector genomes/ml) was injected at rates of 5 μl/min and 1 μl/min, respectively. The syringe was left in place for 2 min after injection. Then the scalp was sutured, its vital signs were monitored until it recovered from the anesthesia. After awaking, it was returned to a standard home cage. A humane endpoints were in place during the animal experiments as the following indicators: severe pain, severe distress, suffering or impending death, at which conditions, marmosets were humanely euthanized. However, the operated marmosets showed only a slight ataxia for a couple of days after the viral injection, and returned quickly to their original condition thereafter. Thus, no mortality was observed. Histological analysis and immunohistochemistry The mice were sacrificed 7 days after viral injection. These deeply anesthetized mice were perfused intracardially with 4% paraformaldehyde in 0.1 M phosphate buffer. The whole brains were immersed in 4% paraformaldehyde in 0.1 M phosphate buffer. The cerebellum and cerebrum were cut into 50-μm sagittal sections using a microtome (Leica VT1000 S; Leica Microsystems, Wetzlar, Germany). The slices were blocked with PBS containing 2% normal donkey serum, 0.1% Triton X-100, and 0.05% NaN3 (blocking solution) and then incubated overnight at 4°C in the following primary antibodies: rabbit polyclonal anti-GFP (1:1000; Rb-Af2020; Frontier Institute, Hokkaido, Japan) and mouse monoclonal anti-S100 (1:1000; S2532; Sigma-Aldrich, St. Louis, MO, USA) for cerebellar slices or rat monoclonal anti-GFP (1:1000; 04404–84; Nacalai, Kyoto, Japan), rabbit polyclonal anti-GFAP (1:200; RB-087-A0; Thermo Fisher Scientific, Waltham, MA, USA) and mouse monoclonal anti-NeuN (1:1000; MAB377, Merk Millipore, Billerica, MA, USA) for cerebral sections. After washing two and three times with 0.5% and 0.1% Triton X-100 in PBS, respectively, at room temperature, the slices were incubated in blocking solution for 2 h at room temperature in the following secondary antibodies: Alexa Fluor 488 donkey anti-rabbit IgG (1:1000; Thermo Fisher Scientific, Waltham, MA) and Alexa Fluor 568 donkey anti-mouse IgG (1:1000; Thermo Fisher Scientific) for cerebellar slices or Alexa Fluor 488 donkey anti-rat IgG (1:1000; Thermo Fisher Scientific), Alexa Fluor 568 donkey anti-rabbit IgG (1:1000; Thermo Fisher Scientific) and Alexa Fluor 680 donkey anti-mouse IgG (1:1000; Thermo Fisher Scientific) for cerebral sections. After the secondary antibody reaction, Nissl bodies of the cerebellar slices were stained with NeuroTrace 640/660 (1:200; Thermo Fisher Scientific) in PBS for 1 h at room temperature. Immunostained sections were mounted in ProLong Gold or Diamond antifade reagents (Thermo Fisher Scientific). Fluorescence images were obtained on a fluorescence microscope (VB-7010 or BZ-X700; Keyence, Osaka, Japan) or a confocal laser-scanning microscope (LSM 5 or LSM 880; Carl Zeiss, Oberkochen, Germany). The marmosets were sacrificed 2 or 4 weeks after AAV9 vector injection. The marmosets were anesthetized with a mixture of ketamine and xylazine, and isoflurane. These deeply anesthetized marmosets were perfused intracardially with 1×PBS (pH 7.4) and 4% paraformaldehyde in 0.1 M phosphate buffer. Cerebellar and cerebral sections of 100-μm thickness were obtained using a procedure similar to that used for the mice, and marmoset sections were immunostained with the same antibodies used for the mouse sections. Quantification of GFP intensity in mouse cerebellum To measure the GFP fluorescence intensity, 15 sections from 5 mice (3 sections/mouse) for each promoter were randomly selected, except for the original 2.0-kb cjGFAP promoter, for which 36 sections from 12 mice (3 sections/mouse) were selected. The GFP fluorescence images of those sections were captured using a confocal microscope and the same settings. Then, the outline of the cerebellar section was traced, and the fluorescence intensity in the enclosed areas was measured using ImageJ. The background intensity was subtracted from the fluorescence intensity. The averaged GFP fluorescence intensity of sections treated with lentiviral vectors carrying the 2.0-kb cjGFAP promoter was taken as 100%, and the relative values were plotted on a graph. Assessment of the astrocyte specificity of the promoters We assessed the astrocyte specificity of the promoters by measuring the ratio of GFP-expressing astrocytes to all GFP-expressing cells. We counted more than 500 GFP-positive cells in 9 slices (3 slices/mouse, n = 3 mice), among which the number of GFP/S100 (or GFAP) double-positive astrocytes were examined. The ratio was determined by dividing the number of GFP/S100 (or GFAP) double-positive astrocytes by the total number of GFP-positive cells. Similarly, neuronal leakage of the promoters was assessed by calculating the ratio of GFP/NeuN (or GFP/Nissl substance) double-positive neurons to total GFP-positive cells. GFP-positive cells that failed to show immunolabeling for S100, GFAP, NeuN or NeuroTrace were classified as unknown. Statistical analysis Significant differences were analyzed using one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test. Statistical analyses were performed using the R software statistical package (www.r-project.org). Data are expressed as the mean ± SEM. Results Deletion constructs of the GFAP promoter Previous transgenic studies have shown that the 2.2-kb human GFAP promoter extending 2.2 kb upstream of the RNA start site (bp +1) serves as an astrocyte-specific promoter in the mouse brain [15]. Furthermore, promoter activity, as assessed by chloramphenicol acetyltransferase activity in U251 cells (a human glioma cell line that strongly expresses GFAP), remained almost unchanged between 2.2 kb and 1.7 kb [16]. We cloned a homologous 2.0-kb GFAP promoter region (from +14 to -1991) that displayed 58% and 88% sequence similarity with the mouse and human regions, respectively. The shorter promoter constructs with sizes of 1.6, 1.4, 0.6, 0.3 and 0.2 kb were produced by 5’ deletion of the original marmoset-derived GFAP promoter (cjGFAP promoter) (Fig 1). The B region and C1 segment were shown to carry various transcription factor-binding sites [17]. The 1.6-kb cjGFAP promoter retained all of those transcription factor-binding sites, whereas the cjGFAP promoters ≤1.4 kb lacked the B and C1 regions and, therefore, were devoid of these transcription factor-binding sites. 10.1371/journal.pone.0162023.g001Fig 1 Schema depicting various lengths of the marmoset GFAP promoters examined in this study. The longest GFAP promoter (2.0 kb) spanned from -1991 to +14 bp relative to the transcription start site [18]. The shorter promoters ranged from 0.2 kb to 1.6 kb and included different 5’ deletion. The percentage shown next to each promoter construct is the homology of the marmoset promoter with the mouse promoter. BP; basal promoter. Assessment of the promoter strength of the cjGFAP promoter fragments To minimize the number of marmosets used, we screened the promoter strength and astrocyte specificity of the deletion constructs in mouse brains using lentiviral vectors. We opted for lentiviral vectors as they can be prepared more quickly (4 days) and with less effort than AAV vectors (10 days), and were sufficient to transduce the small mouse brain. Lentiviral vectors expressing GFP under the control of the full-length cjGFAP promoter or promoter fragments were produced. These viral vectors were injected into the cerebella of mice aged 3–4 weeks to assess promoter strength, as determined by GFP fluorescence intensity. The injected mice were then sacrificed 1 week after the viral injection. We found that the whole cerebellum in mice treated with lentiviral vectors carrying the 2.0- or 1.6-kb promoter showed markedly brighter GFP fluorescence intensity than those treated with lentiviral vectors carrying the shorter cjGFAP promoters (Fig 2A–2F, upper panels). Examination of the cerebellar sagittal sections confirmed the results of the whole cerebellum analysis: a remarkably stronger GFP fluorescence was observed in sections treated with lentiviral vectors carrying the 2.0- or 1.6-kb cjGFAP promoter compared with the shorter cjGFAP promoters (Fig 2A–2F, lower panels). Notably, the GFP fluorescence intensities of cerebellar sections expressing GFP from the 0.3- and 0.2-kb cjGFAP promoters (Fig 2E and 2F) appeared to be brighter than those from the 1.4- and 0.6-kb cjGFAP promoters (Fig 2C and 2D). A quantitative analysis showed that the GFP fluorescence intensities of cerebellar sections, relative to those under the control of the original 2.0-kb cjGFAP promoter (100 ± 5.9%, n = 12 mice), were 92.5 ± 4.0% (1.6-kb promoter, n = 5 mice), 15.1 ± 1.3% (1.4-kb promoter, n = 5 mice), 19.1 ± 2.2 (0.6-kb promoter, n = 5 mice), 40.0 ± 2.7% (0.3-kb promoter, n = 5 mice) and 44.9 ± 3.0% (0.2-kb promoter, n = 5 mice). The GFP fluorescence intensities for the 0.3- and 0.2-kb cjGFAP promoters were significantly weaker than that for the original 2.0-kb cjGFAP promoter (***p<0.001), but they were approximately 2 times brighter and significantly stronger than those for the 1.4- and 0.6-kb cjGFAP promoters (†p<0.05) (Fig 2G). 10.1371/journal.pone.0162023.g002Fig 2 Mouse cerebella lentivirally expressing GFP under the control of cjGFAP promoters of different lengths. (A–F) Upper small panels are bright field (left) and GFP fluorescence (right) images of whole brains. Lower panels show GFP fluorescence images of cerebellar sagittal sections. The promoter lengths used to drive GFP expression are shown in the upper right corner. Scale bars, 500 μm. (G) Quantitative analysis of GFP fluorescence intensity. The fluorescence intensity on sagittal sections of the viral vector-treated cerebellar vermis was measured using ImageJ software. Between 5 and 12 mice in each group were used for the analysis, and the intensities relative to the control 2.0-kb cjGFAP promoter are presented. Small circles in each lane are individual measurements. Asterisks and daggers indicate statistically significant differences compared with mice expressing GFP under the control of the 2.0-kb promoter (asterisks) or the 1.4-kb promoter (daggers), as determined by one-way ANOVA followed by Tukey’s post hoc test, ***p<0.001, †p<0.05 and †††p<0.001. Assessment of the astrocyte specificity of the cjGFAP promoter fragments Cerebellar cortex is divided into the molecular layer, Purkinje cell layer and granule cell layer (Fig 3A). Purkinje cell somata make the Purkinje cell layer (Fig 3B), where cell bodies of Bergmann glia exist and extend long processes into the molecular layer (Fig 3A). Fig 3C shows a typical example of astrocyte (Bergmann glia)-specific GFP expression in mouse cerebellum that was treated with lentiviral vectors carrying mouse-derived 1.5 kb GFAP promoter: in sharp contrast to absence of GFP expression in Purkinje cells (P, in Fig 3C), robust GFP signal was observed in cell bodies and processes of Bergmann glia, which were co-immunolabeled for S100 (Fig 3C). 10.1371/journal.pone.0162023.g003Fig 3 High specificity of the 0.3-kb cjGFAP promoter for astrocytes in the mouse cerebellum. (A) Shema depicting Purkinje cell (PC) and Bergmann glia (BG) in the cerebellar cortex. Cell bodies of Bergmann glia are present in the Purkinje cell layer (PCL) and extend processes into the molecular layer (ML). GCL; Granule cell layer. (B) A cerebellar section immunolabeled for parvalbumin. Arrows indicate molecular layer interneurons. (C) Cerebellar slices lentivirally expressing GFP under the control of the 1.5 kb mouse GFAP promoter were double-immunostained for GFP and S100 (an astrocyte marker). Arrows and arrowheads indicate cell bodies and the processes of Bergmann glia, respectively. GFP expression was observed in Bergmann glia, but not in Purkinje cells (P). (D-I) Cerebellar slices lentivirally expressing GFP under the control of different lengths of the marmoset GFAP promoter were double-immunostained for GFP and S100. Arrows in (I) indicate GFP-positive interneurons in the molecular layer. Scale bars, 50 μm. (J) Quantitative analysis of astrocyte specificity of the marmoset GFAP promoter fragments. More than 500 GFP-positive cells from 3 mice (3 slices/mouse) were randomly selected in each group, and the ratio of S100-labeled astrocytes was determined in these random selections. Asterisks and daggers indicate statistically significant differences compared with the cerebella expressing GFP under the control of the 2.0-kb promoter (asterisks) or the 0.3-kb promoter (daggers), as determined by one-way ANOVA followed by Tukey’s post hoc test, ***p<0.001 and ††p<0.01. To examine the astrocyte specificity of the cjGFAP promoter fragments, cerebellar slices expressing GFP by the cjGFAP promoters (Fig 2A–2F) were double-immunolabeled for GFP and the astrocyte marker S100. Immunohistochemistry showed that 5’ deletion of the cjGFAP promoter did not compromise the astrocyte specificity. The majority of GFP-expressing cells were Bergmann glia (Fig 3D–3I). However, we found a significant number of GFP-expressing interneurons in the cerebella that were transduced by the 0.2-kb cjGFAP promoter (arrowheads, Fig 3I). To quantify the ratios of GFP-positive astrocytes to total GFP-positive cells, we examined more than 500 GFP-positive cells in 9 cerebellar slices from 3 mice and counted the S100-labeled astrocytes. The ratios of GFP-positive astrocytes to total GFP-positive cells were 92.0 ± 0.5% (2.0-kb promoter), 91.8 ± 0.8% (1.6-kb promoter), 89.1 ±1.8% (1.4-kb promoter), 89.7 ±1.1% (0.6-kb promoter), 89.1 ± 1.3% (0.3-kb promoter) and 74.1 ± 3.0% (0.2-kb promoter) (Fig 3J). The ratio was significantly decreased only in cerebellar sections expressing GFP under the control of the 0.2-kb cjGFAP promoter compared with those under the control of the 2.0-kb cjGFAP (***p<0.001) or 0.3-kb cjGFAP (††p<0.01) promoter. These results suggest that the 0.3-kb cjGFAP promoter shows promise as a compact astrocyte-specific promoter for viral vector-mediated gene expression in the marmoset brain. Loss of astrocyte specificity for the 0.3-kb cjGFAP promoter in the mouse cerebrum To verify the astrocyte specificity of the 0.3-kb cjGFAP promoter in a different brain region, lentiviral vectors expressing GFP under the control of the 0.3-kb cjGFAP promoter were injected into the mouse cerebral cortex. The cerebral sections obtained 1 week after the viral injection were triple-immunostained for GFP, GFAP and NeuN. The fluorescence microscopic examination clearly showed that a majority of GFP-expressing cells were NeuN-positive and GFAP-negative neuronal cells (arrows, Fig 4A and 4D), with some exceptions for NeuN-negative and GFAP-positive astrocytes (arrowheads, Fig 4A and 4D). This finding clearly indicates the absence of astrocyte specificity for the 0.3-kb cjGFAP promoter in the mouse cerebral cortex. A plausible explanation for the absence of astrocyte specificity is that the 0.3-kb cjGFAP promoter is too short and lacks the region necessary for suppressing neuronal expression. Although it may be less likely, another possibility is a mismatch between the species of the promoter (marmoset) and the integrated cells (mouse). We tested the latter possibility by examining the astrocyte specificity of the homologous 0.3-kb mouse-derived mGFAP as well as human-derived 0.3-kb hGFAP promoters. Lentiviral vectors expressing GFP under the control of the 0.3-kb mGFAP or hGFAP promoter were injected into the mouse cerebral cortex. Immunohistochemistry of the cerebral slices made 1 week after viral injection showed GFP expression specifically in the GFAP-positive and NeuN-negative astrocytes (Fig 4B and 4C), when mGFAP promoter, but not hGFAP promoter, was used. A quantitative analysis showed that more than 80% of the GFP-positive cells were astrocytes in the mouse cerebrum expressing GFP under the control of the mouse-derived mGFAP promoter (328 cells examined from 3 mice). In contrast, the ratio of GFP-positive astrocytes was only approximately 20–30% of all GFP-positive cells in the mouse cerebrum expressing GFP under the control of the marmoset- or human-derived GFAP promoter (402 and 330 cells examined from 3 mice, respectively) (Fig 4E). There was a statistically significant difference in the astrocyte specificity between the mGFAP promoter and cjGFAP promoter or hGFAP promoter (***p<0.001), suggesting a possibility that the loss of astrocyte specificity for the 0.3-kb cjGFAP promoter in the mouse cerebral cortex was attributed to a difference between the species. 10.1371/journal.pone.0162023.g004Fig 4 Absence of astrocyte specificity for the marmoset- and human-derived GFAP promoters in the mouse cerebrum. (A-C) Cerebral slices lentivirally expressing GFP under the control of the 0.3-kb marmoset-derived cjGFAP (A), 0.3-kb mouse-derived mGFAP (B) or 0.3-kb human-derived hGFAP (C) promoter were triple-immunostained for GFP (green), GFAP (magenta) and NeuN (a neuronal marker, cyan). Note the predominant expression of GFP in neurons (arrow) by the marmoset- and human-derived promoter, which is in sharp contrast to the astrocyte-specific expression (arrowhead) by the mouse-derived promoter. Scale bars, 50 μm. (D) Schema depicting morphology of neuron and astrocyte in the cerebral cortex. (E) Quantitative analysis of the astrocyte specificity for the cjGFAP, mGFAP and hGFAP promoters. More than 300 GFP-positive cells from 3 mice (3 slices/mouse) were randomly selected, and the ratio of GFAP-labeled astrocytes were determined in these random selections. Asterisks indicate statistically significant differences between the mouse promoter and the marmoset or human promoter, as determined by one-way ANOVA followed by Tukey’s post hoc test, ***p<0.001. Retention of astrocyte specificity for the 0.3-kb cjGFAP promoter in the marmoset brain To verify the astrocyte specificity of the 0.3-kb cjGFAP promoter in the marmoset brain, we used AAV serotype 9 (AAV9) vectors because AAV9 vectors have the potential to transduce significantly broader areas of the brain than lentiviral vectors [8] and, thus, are suitable for transduction in the larger marmoset brain. AAV9 vectors expressing GFP under the control of the 0.3-kb cjGFAP promoter were injected into both the cerebellar cortex and cerebral cortex of 1.8- and 2.4-year-old marmosets. The GFP expression profiles of the marmoset brains were examined 2–4 weeks after the viral vector injections were performed. Bright GFP fluorescence was observed in the cerebellar and cerebral hemispheres in regions surrounding the injection sites (Fig 5A and 5B). Sagittal sections of the brains confirmed robust and efficient GFP expression in the cerebellar and cerebral cortices (Fig 5C and 5D). To examine the cell types expressing GFP, the sections were triple-immunostained for GFP, Nissl substance and S100 (cerebellum) or GFAP (cerebrum). In the cerebellum, GFP was almost exclusively expressed in Bergmann glia (Fig 6A) and S100-positive astrocytes in the granule cell layer (arrows, Fig 6A). Similar to the cerebellar cortex, GFP was detected in GFAP-positive astrocytes in the cerebral cortex (Fig 6B). A quantitative analysis of more than 150 cells from both the cerebellar and cerebral cortices of 2 marmosets showed that approximately 90% of GFP-positive cells were astrocytes, whereas <10% were neurons (Fig 6C). These results show that, when introduced with viral vectors, the 0.3-kb cjGFAP promoter serves as an astrocyte-specific promoter in the marmoset cerebellar and cerebral cortices. 10.1371/journal.pone.0162023.g005Fig 5 AAV9 vector-mediated GFP expression in the marmoset brain. AAV9 vectors expressing GFP under the control of the 0.3-kb cjGFAP promoter were injected into the cerebral and cerebellar cortices. (A and B) Bright field images of the marmoset whole brain overlaid with GFP fluorescence. (C and D) Bright field images of the sagittal sections of the cerebellar (C) and cerebral (D) hemispheres are presented with GFP fluorescence. Scale bars, 2 mm. 10.1371/journal.pone.0162023.g006Fig 6 Astrocyte-specific GFP expression in the marmoset brain by AAV9 vectors carrying the marmoset-derived cjGFAP promoter. (A) Marmoset cerebellar slices virally expressing GFP under the control of the 0.3-kb cjGFAP promoter were triple-immunostained for GFP (green), S100 (magenta), and Nissl substance (cyan). (B) Marmoset cerebral slices virally expressing GFP under the control of the 0.3-kb cjGFAP promoter were double-immunostained for GFP (green), GFAP (magenta), and NeuN (cyan). (C) More than 150 GFP-positive cells in marmoset cerebellar and cerebral slices triple-immunolabeled for GFP, Nissl (cerebellum), or NeuN (cerebrum), and S100 (cerebellum) or GFAP (cerebrum) were randomly selected. The ratios of S100- or GFAP-labeled astrocytes and Nissl- or NeuN-labeled neurons to all GFP-positive cells were determined. GFP-positive cells without labeling for both S100 (or GFAP) and Nissl (or NeuN) were classified as ‘Unknown’. Note that approximately 90% of GFP-positive cells in both the cerebellar cortex and cerebral cortex were astrocytes. Scale bars (A, B), 100 μm. Discussion A series of previous transgenic studies examining the human GFAP promoter revealed critical roles for the B region and contiguous C1 segment in regulating the strength of GFAP promoter activity and astrocyte specificity [15–21]. The 5’ deletion of the original 2.2-kb GFAP promoter to 1.7 kb did not substantially compromise the promoter strength in U251 cells, a human glioma cell line that strongly expresses GFAP, whereas further deletion to approximately 1.5 kb, which removed the B and C1 regions, resulted in a drastic reduction in the promoter strength to 10% of the original 2.2-kb promoter [16]. Consistent with this finding, in our study, 5’ deletion of the cjGFAP promoter from 2.0 kb to 1.6 kb, which still carried the B and C1 regions, had little influence on promoter activity. However, deletion to 1.4 kb, thus removing the B and C1 regions, resulted in a drastic reduction in the promoter strength to <20% of that of the 2.0-kb cjGFAP promoter. Interestingly, the astrocyte specificity of the cjGFAP promoter was preserved for fragments of 0.3 kb or more in the mouse cerebellar cortex. Moreover, the 0.3-kb promoter showed significantly higher promoter strength than the 0.6-kb and 1.4-kb promoters, indicating that the 0.3-kb promoter was superior to the 0.6-kb and 1.4-kb promoters in terms of promoter strength and accommodation capacity for a transgene. In the cerebral cortex, however, the 0.3-kb cjGFAP promoter also drove transgene expression in pyramidal neurons. There are at least 2 possibilities that could account for the loss of astrocyte specificity for the 0.3-kb cjGFAP promoter in the mouse cerebral cortex. The first possibility is that the deleted region upstream of -0.3 kb may be crucial for suppressing non-astrocyte expression. The other possibility is related to the use of the marmoset-derived promoter in the mouse cerebrum. Specifically, the species mismatch may explain the loss of astrocyte specificity for the 0.3-kb cjGFAP promoter. Initially, we supposed that the first possibility was more likely than the second. Thus, to exclude the second possibility, we cloned the homologous 0.3-kb GFAP promoter from the mouse genome and examined the promoter’s characteristics in terms of its astrocyte specificity in the mouse cerebrum. Unexpectedly, the 0.3-kb GFAP promoter of mouse origin specifically transduced astrocytes, thus suggesting the second possibility was correct. This result motivated us to test the astrocyte specificity of the 0.3-kb cjGFAP promoter in the marmoset brain. We then injected AAV9 vectors expressing GFP under the control of the 0.3-kb cjGFAP promoter into the marmoset cerebral cortex and subsequently observed astrocyte-specific expression of GFP in the cerebral cortex, as well as in the cerebellar cortex, which was similar to the expression pattern for the mouse GFAP promoter in the mouse brain. Therefore, the 0.3-kb cjGFAP and mouse GFAP promoters are thought to contain sequences that are critical for suppressing promoter activity in cells other than astrocytes. Moreover, the putative sequences are likely species-specific, at least in the cerebral cortex, but the sequences may be shared by mouse and marmoset cerebellar cortices, since the astrocyte specificity of marmoset-derived 0.3-kb cjGFAP promoter was preserved in the mouse cerebellar cortex. Further study is needed to clarify the mechanism that suppresses non-astrocyte expression of the GFAP promoter in different brain regions. The packaging capacity for recombinant AAV vectors is approximately 4.7 kb, including the 2 inverted terminal repeats (ITRs). Because the total length of the 2 ITRs of AAV9 vectors is 372 bp, there is a capacity of approximately 4.3 kb for foreign DNA that can be accommodated between the 2 ITRs. Moreover, the woodchuck hepatitis post-transcriptional regulatory element (WPRE) [22, 23]–polyadenylation signal element is approximately 0.7 kb long. If a reporter gene such as GFP were to be co-expressed, it would require another 0.7 kb. Thus, the maximal allowable cDNA length is approximately 2.9 kb, including a promoter sequence. Thus, an efficient short promoter is valuable for the AAV vector-mediated expression of a large cDNA. The 0.3-kb GFAP promoter that we identified in this study allows us to express a transgene of up to 2.6 kb of cDNA together with GFP specifically in astrocytes. Our results identified a very short astrocyte-specific promoter region that extended the usability of AAV vectors for astrocyte-specific transgene expression in the marmoset brain. The authors are very grateful to the technicians Motoko Uchiyama and Minako Noguchi for raising the marmosets, Asako Ohnishi for AAV9 vector production and Junko Sugiyama for maintenance of mice. This research is (partially) supported by the program for Brain Mapping by Integrated Neurotechnologies for Disease Studies (Brain/MINDS) from the Ministry of Education, Culture, Sports, Science and Technology (MEXT), and the Japan Agency for Medical Research and Development, AMED to HH and the MEXT KAKENHI (15K18330) to AK. ==== Refs References 1 Saab AS , Neumeyer A , Jahn HM , Cupido A , Simek AA , Boele HJ , et al Bergmann glial AMPA receptors are required for fine motor coordination . Science . 2012 ;337 (6095 ):749 –53 . Epub 2012/07/07. 10.1126/science.1221140 .22767895 2 Han J , Kesner P , Metna-Laurent M , Duan T , Xu L , Georges F , et al Acute cannabinoids impair working memory through astroglial CB1 receptor modulation of hippocampal LTD . Cell . 2012 ;148 (5 ):1039 –50 . Epub 2012/03/06. 10.1016/j.cell.2012.01.037 .22385967 3 Shiwaku H , Yoshimura N , Tamura T , Sone M , Ogishima S , Watase K , et al Suppression of the novel ER protein Maxer by mutant ataxin-1 in Bergman glia contributes to non-cell-autonomous toxicity . The EMBO journal . 2010 ;29 (14 ):2446 –60 . Epub 2010/06/10. 10.1038/emboj.2010.116 20531390 4 Cvetanovic M . Decreased expression of glutamate transporter GLAST in bergmann glia is associated with the loss of Purkinje neurons in the spinocerebellar ataxia type 1 . Cerebellum . 2015 ;14 (1 ):8 –11 . Epub 2014/09/27. 10.1007/s12311-014-0605-0 .25255716 5 Custer SK , Garden GA , Gill N , Rueb U , Libby RT , Schultz C , et al Bergmann glia expression of polyglutamine-expanded ataxin-7 produces neurodegeneration by impairing glutamate transport . Nature neuroscience . 2006 ;9 (10 ):1302 –11 . Epub 2006/08/29. 10.1038/nn1750 .16936724 6 Bradford JW , Li S , Li XJ . Polyglutamine toxicity in non-neuronal cells . Cell research . 2010 ;20 (4 ):400 –7 . Epub 2010/03/17. 10.1038/cr.2010.32 20231860 7 Correale J , Farez MF . The Role of Astrocytes in Multiple Sclerosis Progression . Frontiers in neurology . 2015 ;6 :180 Epub 2015/09/09. 10.3389/fneur.2015.00180 26347709 8 Huda F , Konno A , Matsuzaki Y , Goenawan H , Miyake K , Shimada T , et al Distinct transduction profiles in the CNS via three injection routes of AAV9 and the application to generation of a neurodegenerative mouse model . Molecular therapy Methods & clinical development . 2014 ;1 :14032 Epub 2014/01/01. 10.1038/mtm.2014.32 26015973 9 Matsuzaki Y , Konno A , Mukai R , Honda F , Hirato M , Yoshimoto Y , et al Transduction Profile of the Marmoset Central Nervous System Using Adeno-Associated Virus Serotype 9 Vectors . Mol Neurobiol . 2016 10.1007/s12035-016-9777-6 .26884266 10 Jakobsson J , Ericson C , Jansson M , Björk E , Lundberg C . Targeted transgene expression in rat brain using lentiviral vectors . Journal of Neuroscience Research . 2003 ;73 (6 ):876 –85 . 10.1002/jnr.10719 12949915 11 Hirai H . Progress in transduction of cerebellar Purkinje cells in vivo using viral vectors . Cerebellum . 2008 ;7 (3 ):273 –8 . Epub 2008/04/18. 10.1007/s12311-008-0012-5 .18418690 12 Kishi N , Sato K , Sasaki E , Okano H . Common marmoset as a new model animal for neuroscience research and genome editing technology . Development, growth & differentiation . 2014 ;56 (1 ):53 –62 . Epub 2014/01/07. 10.1111/dgd.12109 .24387631 13 Torashima T , Yamada N , Itoh M , Yamamoto A , Hirai H . Exposure of lentiviral vectors to subneutral pH shifts the tropism from Purkinje cell to Bergmann glia . The European journal of neuroscience . 2006 ;24 (2 ):371 –80 . Epub 2006/07/14. 10.1111/j.1460-9568.2006.04927.x .16836635 14 Konno A , Shuvaev AN , Miyake N , Miyake K , Iizuka A , Matsuura S , et al Mutant ataxin-3 with an abnormally expanded polyglutamine chain disrupts dendritic development and metabotropic glutamate receptor signaling in mouse cerebellar Purkinje cells . Cerebellum . 2014 ;13 (1 ):29 –41 . Epub 2013/08/21. 10.1007/s12311-013-0516-5 .23955261 15 Brenner M , Kisseberth WC , Su Y , Besnard F , Messing A . GFAP promoter directs astrocyte-specific expression in transgenic mice . The Journal of neuroscience: the official journal of the Society for Neuroscience . 1994 ;14 (3 Pt 1):1030 –7 . Epub 1994/03/01. .8120611 16 Besnard F , Brenner M , Nakatani Y , Chao R , Purohit HJ , Freese E . Multiple interacting sites regulate astrocyte-specific transcription of the human gene for glial fibrillary acidic protein . The Journal of biological chemistry . 1991 ;266 (28 ):18877 –83 . Epub 1991/10/05. .1918004 17 Yeo S , Bandyopadhyay S , Messing A , Brenner M . Transgenic analysis of GFAP promoter elements . Glia . 2013 ;61 (9 ):1488 –99 . Epub 2013/07/09. 10.1002/glia.22536 23832770 18 Brenner M . Structure and transcriptional regulation of the GFAP gene . Brain Pathol . 1994 ;4 (3 ):245 –57 . Epub 1994/07/01. .7952266 19 de Leeuw B , Su M , ter Horst M , Iwata S , Rodijk M , Hoeben RC , et al Increased glia-specific transgene expression with glial fibrillary acidic protein promoters containing multiple enhancer elements . Journal of neuroscience research . 2006 ;83 (5 ):744 –53 . Epub 2006/02/24. 10.1002/jnr.20776 .16496373 20 Lee Y , Messing A , Su M , Brenner M . GFAP promoter elements required for region-specific and astrocyte-specific expression . Glia . 2008 ;56 (5 ):481 –93 . Epub 2008/02/02. 10.1002/glia.20622 .18240313 21 Lee Y , Su M , Messing A , Brenner M . Astrocyte heterogeneity revealed by expression of a GFAP-LacZ transgene . Glia . 2006 ;53 (7 ):677 –87 . Epub 2006/02/17. 10.1002/glia.20320 .16482522 22 Donello JE , Loeb JE , Hope TJ . Woodchuck hepatitis virus contains a tripartite posttranscriptional regulatory element . Journal of virology . 1998 ;72 (6 ):5085 –92 . Epub 1998/05/30. 9573279 23 Zufferey R , Donello JE , Trono D , Hope TJ . Woodchuck hepatitis virus posttranscriptional regulatory element enhances expression of transgenes delivered by retroviral vectors . Journal of virology . 1999 ;73 (4 ):2886 –92 . Epub 1999/03/12. 10074136

PMC005xxxxxx/PMC5003433.txt

==== Front PLoS OnePLoS ONEplosplosonePLoS ONE1932-6203Public Library of Science San Francisco, CA USA 2757141810.1371/journal.pone.0162058PONE-D-16-22140Research ArticleBiology and Life SciencesBiochemistryProteinsCytoskeletal ProteinsVimentinBiology and life sciencesBiochemistryProteinsDNA-binding proteinsHistonesBiology and Life SciencesCell BiologyCellular TypesAnimal CellsEpithelial CellsBiology and Life SciencesAnatomyBiological TissueEpitheliumEpithelial CellsMedicine and Health SciencesAnatomyBiological TissueEpitheliumEpithelial CellsBiology and life sciencesGeneticsGene expressionGene regulationSmall interfering RNAsBiology and life sciencesBiochemistryNucleic acidsRNANon-coding RNASmall interfering RNAsBiology and Life SciencesAnatomyRespiratory SystemNasal ConchaMedicine and Health SciencesAnatomyRespiratory SystemNasal ConchaBiology and life sciencesCell biologySignal transductionCell signalingSignaling cascadesTGF-beta signaling cascadeResearch and Analysis MethodsBiological CulturesOrgan CulturesBiology and Life SciencesBiochemistryProteinsCytoskeletal ProteinsTrichostatin A Inhibits Epithelial Mesenchymal Transition Induced by TGF-β1 in Airway Epithelium Trichostatin A and Epithelial Mesenchymal Transitionhttp://orcid.org/0000-0002-7011-6071Park Il-Ho 1Kang Ju-Hyung 2Shin Jae-Min 1Lee Heung-Man 123*1 Department of Otorhinolaryngology-Head and Neck Surgery, Guro Hospital, Korea University College of Medicine, Seoul, South Korea2 Department of Biomedical Sciences, Korea University Graduate School, Seoul, South Korea3 Medical Devices support Center, Guro Hospital, Korea University College of Medicine, Seoul, South KoreaKorc Murray EditorIndiana University School of Medicine, UNITED STATESCompeting Interests: The authors have declared that no competing interests exist. Conceptualization: IHP HML. Data curation: IHP JHK. Formal analysis: JHK. Funding acquisition: IHP HML. Investigation: JHK. Methodology: IHP JHK HML. Project administration: HML. Resources: IHP HML. Supervision: HML. Validation: JHK HML. Visualization: JHK JMS. Writing – original draft: IHP. Writing – review & editing: IHP JMS HML. * E-mail: lhman@korea.ac.kr29 8 2016 2016 11 8 e01620582 6 2016 16 8 2016 © 2016 Park et al2016Park et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Background and Objectives Tissue remodeling is believed to cause recalcitrant chronic rhinosinusitis (CRS). Epithelial-mesenchymal transition (EMT) is a novel clinical therapeutic target in many chronic airway diseases related with tissue remodeling. The aim of this study was to investigate the effect of trichostatin A (TSA) on transforming growth factor (TGF)-β1-induced EMT in airway epithelium and nasal tissue. Materials and Methods A549 cells, primary nasal epithelial cells (PNECs), or inferior nasal turbinate organ culture were exposed to TSA prior to stimulation with TGF-β1. Expression levels of E-cadherin, vimentin, fibronectin, α-smooth muscle actin (SMA), histone deacetylase 2 (HDAC2), and HDAC4 were determined by western blotting and/or immunofluorescent staining. Hyperacetylation of histone H2 and H4 by TSA was measured by western blotting. After siHDAC transfection, the effects of HDAC2 and HDAC4 silencing on expression of E-cadherin, vimentin, fibronectin, α-SMA, HDAC2, and HDAC4 in TGF-β1-induced A549 were determined by RT-PCR and/or western blotting. We assessed the change in migration capacity of A549 cells by using cell migration assay and transwell invasion assay. Results TGF-β1 altered mRNA and protein expression levels of EMT markers including E-cadherin, vimentin, fibronectin, α-SMA, slug, and snail in A549 cells. Inhibition and silencing of HDAC2 and HDAC4 by TSA and siRNA enhanced TGF-β1-induced EMT in A549 cells. TSA blocked the effect of TGF-β1 on the migratory ability of A549 cells. In experiments using PNECs and inferior turbinate organ cultures, TSA suppressed expression of EMT markers induced by TGF-β1. Conclusions We showed that EMT is induced by TGF-β1 in airway epithelial cells and nasal tissue via activation of HDAC2 and HDAC4, and that inhibition of HDAC2 and HDAC4 by TSA reduces TGF-β1-induced EMT. This observation indicates that histone deacetylase inhibitors such as TSA could be potential candidates for treatment of recalcitrant CRS related with tissue remodeling. This work was supported by Korean Health Technology R&D Project, Ministry of Health & Welfare, Republic of Korea (HI15C1512), HML, https://www.htdream.kr/index.do and Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (NRF-2015R1C1A1A02037312), IHP, http://www.nrf.re.kr/nrf_tot_cms/index.jsp?pmi-sso-return2=none. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Data AvailabilityAll relevant data are within the paper.Data Availability All relevant data are within the paper. ==== Body Introduction Chronic rhinosinusitis (CRS) is an inflammation of the nose and paranasal sinuses characterized by nasal blockage, nasal discharge, and olfactory dysfunction lasting more than 12 weeks [1]. Control of CRS can be defined as a resolution of symptoms combined with the recovery of mucosa. Medical treatment options for CRS include oral antibiotics, topical corticosteroids, systemic steroids, and other medications such as antihistamines, mucolytics, and decongestants. It is known that about one-third of the symptoms of CRS are relieved by medical treatment [2]. Endoscopic sinus surgery is considered an option after failure of above medical treatment. However, the disease persists in one-third of patients one year after surgery [3]. In spite of considerable effort to identify factors related to disease recalcitrance, such factors are still not clearly understood. Epithelial-mesenchymal transition (EMT) is a phenotype conversion that turns a polarized epithelial cell into a mesenchymal cell. In the process of EMT, epithelial cells lose cell-to-cell adhesion and apical-basal polarity, reorganize their cytoskeletal protein, and acquire the characteristics of mesenchymal cells, such as enhanced motility, invasiveness, and fibrogenesis [4,5]. EMT is known as a feature of embryogenesis, organ development, and cancer progression [6]. It is also activated in wound healing and inflammation, and dysregulation of EMT by repeated stress caused by them may lead to organ fibrosis [7,8]. Additionally, evidence has shown that CRS is related to EMT [9,10]. In a previous study, we showed that histone deacetylase (HDAC) inhibition by trichostatin A (TSA) is associated with extracellular matrix accumulation in nasal polyp-derived fibroblasts [11]. As extracellular matrix accumulation is one of the main features of mesenchymal cells, we hypothesized that epigenetic regulation by TSA can also be associated with suppression EMT of airway epithelium. The purposes of this study were to investigate whether EMT is induced by activation of HDACs in airway epithelial cells and nasal tissue, and to evaluate the effect that histone deacetylase inhibitors such as TSA have on EMT. We stimulated cells and tissues with transforming growth factor (TGF)-β1, which is known to induce EMT, according to several studies [12,13]. Materials and Methods Materials Human recombinant TGF-β1 was obtained from R&D Systems (Minneapolis, MN). TSA was purchased from Sigma (St. Louis, MO, USA). Cells or tissues were previously exposed to TGF-β1 (5mg/mL) after pretreatment for 1 hour with TSA (100nM) Cell culture A549 (human carcinomic alveolar basal epithelial cells, type II) cells were obtained from the American Type Culture Collection (Manassas, VA). A549 cells were grown in RPMI-1640 medium containing 10% (v/v) heat-inactivated fetal bovine serum (Invitrogen, Carlsbad, CA), 1,000 unit/mL penicillin, and 1,000 μg/mL streptomycin (Invitrogen). Inferior turbinate mucosa specimens were obtained from six patients during endoscopic sinus surgery for benign tumors at the Department of Otorhinolaryngology, Korea University Medical Center. None of the patients had a history of allergies, asthma, or aspirin sensitivity, nor had any of them received steroids, nonsteroidal anti-inflammatory drugs, antihistamines, or antibiotics during the 4 weeks prior to the biopsy. For the primary culture of the nasal epithelial cells, the nasal tissues were washed with phosphate buffered saline and immersed in Dispase (Stem cell technologies, Vancouver, Canada) for 4 h. Then, the tissue was filtered through a mesh. Primary nasal epithelial cells (PNECs) were incubated with Bronchial Epithelial Cell Growth Medium (Lonza, Basel, Switzerland). Written informed consent was obtained from each patient, and the study was approved by the Korea University Medical Center Institutional Review Board (KUGGR-12041-001). Organ culture of nasal polyps Inferior turbinates from the patients were cut into three pieces of 2 to 3 mm with scissors under sterile conditions. Tissue fragments were washed three times with phosphate buffered saline. The washed tissue fragments were placed on a prehydrated gelatin sponge (10 mm × 10 mm × 1 mm; Spongostan, Johnson & Johnson, San Angelo, TX) in 6-well plates. Then, each well was filled with 1.5 mL of culture medium containing Dulbecco’s Modified Eagle Medium (Invitrogen) supplemented with 2% fetal bovine serum (Invitrogen). Inferior turbinate tissues were stimulated with TGF-β1 (5 ng/mL) with or without TSA. The plates were maintained at 37°C in 5% CO2. 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide assay A549 cells were seeded on 96-well tissue culture plates at a concentration of 4 x 105 cells/mL with various concentrations (0–1600 nM) of TSA with or without TGF-β1 (5 ng/mL) for 72 h. Then, cells were incubated with MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide, Sigma) for 4 h, and the reaction was interrupted by the addition of acidified isopropanol. A fluorescence microplate reader (F2000; Hitachi, Ltd., Tokyo, Japan) was used to determine the results (570 nm). Immunofluorescent staining Cells were incubated with TGF-β1 (5 ng/mL) alone or in conjunction with TSA for 72 h. Images were obtained with a microscope (Olympus BX51; Olympus, Tokyo, Japan). Cells were fixed with 4% paraformaldehyde, then permeabilized with 0.2% TritonX-100 in 1% bovine serum albumin for 10 min, blocked with 5% bovine serum albumin for 1 h at room temperature, and incubated overnight at 4°C with monoclonal antibodies including vimentin, α-SMA, and snail, and polyclonal antibodies including E-cadherin, fibronectin, and slug (Santa Cruz, CA). Cells were then incubated with Dy-Light 549 horse anti-mouse IgG antibody or DyLight 488 horse anti-rabbit IgG Antibody (Vector Labs, Burlingame, CA). Finally, cells were counterstained with 4’,6-diamidino-2-phenylindole (Invitrogen, Carlsbad, CA). Immunostained cells were captured and visualized using a confocal microscope (LSM700; Zeiss, Oberkochen, Germany). Reverse transcription-polymerase chain reaction (RT-PCR) Total RNA was isolated according to the manufacturer’s recommendations using Trizol reagent (Invitrogen). Two micrograms of RNA were reverse-transcribed using MMLV reverse transcriptase (Invitrogen) according to the manufacturer’s protocol. PCR was performed using the following primers: HDAC2 (sense sequence 5′- CATCCCATGAAGCCTCATAGAATC -3′, anti-sense sequence 5′- GCACCAATATCCCTCAAGTCTCC -3′, 566 bp), HDAC4 (sense sequence 5′- CTG CAAGTGGCCCCCTCGG -3′, anti-sense sequence 5′- CTCGTGCTGTTGCCTCTGGA -3′, 179 bp), Snail (sense sequence 5′- TCTAGGCCCTGGCTGCTACAA -3′, anti-sense sequence 5′- GCCTGGCACTGGTACTTCTTGAC -3′, 152 bp), Slug (sense sequence 5′- ATGCATATTCGGACCCACACATTA -3′, anti-sense sequence 5′- AGAATTTGACCTGTCTGCAAATGCT -3′, 158 bp), and GAPDH (sense sequence 5′- GTGGATATTGTTGCCATCAATGACC -3′, anti-sense sequence 5′- GCCCCAGCCTTCTTCATGGTGGT -3′, 271 bp). The gels were captured and visualized using Molecular Imager ChemiDoc XRS+ (Bio-Rad, Hercules, CA). Western blot analysis A549 cells were lysed in PRO-PREPTM protein extraction solution (iNtRON Biotechnology, Seongnam, Korea). Lysates were separated by 10% sodium dodecyl sulfate polyacrylamide gel electrophoresis and transferred onto polyvinyl difluoride membranes (Millipore Inc., Billerica, MA). Membranes were blocked with a 5% skim milk solution and incubated with the following antibodies: E-cadherin, vimentin, α-SMA, fibronectin, snail, slug (Santa Cruz, CA), HDAC2, HDAC4, ac-histone H3, histone H3, ac-histone H4, histone H4 (Upstate, Millipore Inc.), and β-actin (Santa Cruz, CA). The blots were visualized with HRP-conjugated secondary antibodies and an ECL system (Pierce, Rockford, IL). Transfection with small interference RNA (siRNA) of HDAC2 and HDAC4 A549 cells were pelleted by centrifugation at 13,000 rpm for 3 min; thereafter, the cells were suspended in 1 mL phosphate buffered saline and dispersed using a pipette. The cells were pelleted at 1,000 rpm for 1 min. The supernatant was discarded, and the cells were suspended in Neon Resuspension buffer (Invitrogen) at a concentration of 6 × 105 cells/mL. Universal negative control siRNA (siControl; Santa Cruz) and small interference oligonucleotide RNA directed against siHDAC2 and siHDAC4 (Santa Cruz) were used as controls. Neon Electrolytic buffer (Invitrogen) was added into the Neon transfection tubes, and the tubes were then placed in the Neon transfection system device (Invitrogen) that was set to 1400 V and 30 pulses. Gold tips were used to aspirate 100 μL RNA cell mixture and place it in the device station. After electroporation, an appropriate amount of complete medium was immediately added to each cell aliquot, and the cells were re-plated onto culture dishes. Cell migration scratch assays A549 cells were plated and grown to confluence in 6-well tissue culture dishes. A straight scratch was made in the cells using a pipette tip. Scratched cells were immediately rinsed with phosphate buffered saline, and RPMI-1640 medium containing 10% (v/v) heat-inactivated fetal bovine serum (Invitrogen, Carlsbad, CA, USA), 1,000 unit/mL penicillin, and 1,000 μg/mL streptomycin (Invitrogen) was added. Cells were incubated with TGF-β1 (5 ng/mL) alone or in conjunction with TSA for 48 h. Images were obtained with a microscope (Olympus BX51; Olympus, Tokyo, Japan). Transwell migration assay The cells were seeded to the upper chamber of transwell chambers (Corning Life Sciences, MA, USA). Then, RPMI-1640 medium containing 10% (v/v) heat-inactivated fetal bovine serum (Invitrogen, Carlsbad, CA, USA), 1,000 unit/mL penicillin, and 1,000 μg/mL streptomycin (Invitrogen) was treated with TGF-β1 (5 ng/mL) alone or in conjunction with TSA to the lower chamber of transwell chambers for 48 h. The cells on the upper surface of the membrane were removed by cotton swabs. Then, the cells on the lower surface of the membrane were stained with Diff-Quik stain (Sysmex, Kobe, Japan). Images of the stained cells from five selected views were captured under a microscope at 400x magnification. Statistical analysis Results were obtained from at least three independent experiments. The statistical significance of the differences between control and experimental data was analyzed with unpaired two-way analysis of variance (ANOVA) test or one-way ANOVA followed by Tukey’s test (GraphPad Prism, version 5, Graph Pad Software, San Diego, CA). Significance was established at the 95% confidence level; p values less than 0.05 were accepted as statistically significant. Results TSA inhibits TGF-β1-induced EMT in A549 cells An MTT assay was performed to examine the effects of TSA on survival of A549 cells. Serial dilutions of A549 cells and MTT reagent were used to generate a cell titration curve. The standard curve indicated a linear relationship between number of cells and absorption at 570 nm. Concentrations of TSA ranging from 0 to 1600 nM were examined. TSA did not affect cell survival at concentrations below 400 nM regardless of the presence of TGF- β1 (Fig 1). 10.1371/journal.pone.0162058.g001Fig 1 Cytotoxicity of histamine determined by MTT assay. MTT, 3-(4,5-dimethylthiazol-2yl)-2,5-diphenyl tetrazolium bromide, *P < 0.05 vs. control. TGF-β1 induces EMT in primary airway epithelial cells [12]. To determine whether TGF-β1 induces EMT in A549 cells, cells were treated with 5 ng/mL of TGF-β1 for 48 h and change in their morphology was observed under a phase contrast microscope (Olympus japan, Dokyo, Japan). TGF-β1 treatment for 48 h resulted in the conversion from normal epithelial morphology with a cobblestone-like appearance into a migratory mesenchymal morphology with an abnormally elongated appearance. TGF-β1-stimulated A549 cells pretreated with TSA for 1 h returned to their normal epithelial morphology (Fig 2A). Expression of E-cadherin, vimentin, fibronectin, and α-SMA proteins as a marker of EMT was examined using western blotting and fluorescent immunocytochemical staining (Fig 2B and 2C). After treatment with TGF-β1 for 72 h, cells showed decreased E-cadherin and increased vimentin, fibronectin, and α-SMA expression. TSA pretreatment for 1 h inhibited the effects of TGF-β1 on EMT in A549 cells. As a last step in examining the inhibitory effect of TSA on EMT in TGF-β1-induced A549 cells, the level of EMT-related transcription factors such as snail and slug mRNA and protein was evaluated after 12 h for RT-PCR and 24 h for western blotting (Fig 3). TGF-β1 increased the mRNA and protein expression levels of slug and snail, and TSA pretreatment reversed the effect of TGF-β1. 10.1371/journal.pone.0162058.g002Fig 2 (A) Effects of trichostatin A on morphology of TGF-β1-stimulated A549 cells as observed under a phase contrast microscope. Effects of trichostatin A on expression of E-cadherin, vimentin, fibronectin, and α-smooth muscle actin protein in TGF-β1-stimulated A549 cells were determined by western blotting (B) and immunofluorescent staining (C). Representative of independent experiments. Scale bar = 50 μm. 10.1371/journal.pone.0162058.g003Fig 3 Effects of trichostatin A on expression of snail and slug mRNA and protein in TGF-β1-stimulated A549 cells were determined by RT-PCR (A) and western blotting (B) (Representative of independent experiments). Values are expressed as the mean ± standard error of the mean (SEM) of independent experiments. *P < 0.05 vs. control. †P < 0.05 vs. TGF-β1 alone. GAPDH, glyceraldehyde-3-phosphate dehydrogenase. TSA inhibits the expression of HDAC2 and HDAC4 and induces acetylation of histone H3 and H4 TSA inhibits the activity of HDAC, leading to an increase in histone acetylation. Histone acetylation is related with the enhancement of specific genes. To determine inhibition of HDAC and hyperacetylation by TSA, the expression levels of HDAC2 and HDAC4 were measured by using RT-PCR and western blotting in A549 cells. TGF-β1 induced mRNA expression of HDAC2 and HDAC4 after 24 h, and HDAC2 and HDAC4 protein expression after 72 h. TSA pretreatment blocked the effects of TGF-β1 on HDAC2 and HDAC4 expression in A549 cells (Fig 4A and 4B). Next, we investigated acetylation of histone H3 and H4 with western blotting in A549 cells. TSA induces hyperacetylation of histone H3 and H4 after 72 h, regardless of TGF-β1 stimulation (Fig 4C). These results showed that TSA suppresses HDAC2 and HDAC4 and induces histone acetylation in A549 cells. 10.1371/journal.pone.0162058.g004Fig 4 Effects of trichostatin A on expression of HDAC2 and HDAC4 mRNA and protein in TGF-β1-stimulated A549 cells were determined by RT-PCR (A) and western blotting (B) (representative of independent experiments). Effects of trichostatin A on hyperacetylation of histone H3 and H4 were determined by western blotting (C) (representative of independent experiments). Values are expressed as the mean ± SEM of independent experiments. *P < 0.05 vs. control. †P < 0.05 vs. TGF-β1 alone. GAPDH, glyceraldehyde-3-phosphate dehydrogenase. Silencing of HDAC2 and HDAC4 enhances EMT in TGF-β1-induced A549 cells Next, the effects of silencing of HDAC2 by siRNA on TGF-β1-induced EMT were examined in A549 cells. After transfection of cells with siControl or siHDAC2, the mRNA and protein expression levels of HDAC2, E-cadherin, vimentin, fibronectin, and α-SMA were determined by RT-PCR and western blotting after 24 h and 72 h, respectively. In siHDAC2 pretreated cells, stimulation with TGF-β1 did not affect the expression levels of HDAC2, E-cadherin, vimentin, fibronectin, and α-SMA that were observed in siControl cells (Fig 5A, 5B and 5C). We also investigated the effects of silencing of HDAC4 in same manner. The results of silencing HDAC4 mirrored those of silencing HDAC2 (Fig 5D, 5E and 5F). These data indicated that epigenetic regulation by HDAC2 and HDAC4 is related with TGF-β1-stimulated EMT in A549 cells. 10.1371/journal.pone.0162058.g005Fig 5 Effects of siHDACs on expression of HDAC2 and HDAC4 mRNA and protein in TGF-β1-stimulated A549 cells were determined by RT-PCR (A, D) and western blotting (B, E) (representative of independent experiments). Effects of siHDACs on expression of E-cadherin, vimentin, fibronectin, and α-smooth muscle actin protein in A549 cells were determined by western blotting (C, F) (representative of independent experiments). Values are expressed as the mean ± SEM of independent experiments. *P < 0.05 vs. control. †P < 0.05 vs. TGF-β1 alone. GAPDH, glyceraldehyde-3-phosphate dehydrogenase. TSA inhibits the migration of TGF-β1-induced A549 cells As increased migratory ability is a functional characteristic of mesenchymal cells, we assessed the change in migration capacity of A549 cells by using a cell migration assay. A straight scratch was made in adherent cells with a pipette tip. Then, we measured the distance that cells had migrated from the initial boundary after treatment with TGF-β1 with or without TSA. After 48 h, compared with the controls, cells migrated significantly further from the boundary of the initial wound area in TGF-β1-treated samples. However, pretreatment with TSA inhibited cell migration in TGF-β1-treated A549 cells (Fig 6A). To confirm the inhibitory effect of TSA on increased migratory ability of TGF-β1-induced A549 cells, we performed a transwell invasion assay. After treatment with TGF-β1 with or without TSA for 48 h, we counted the number of cells that had spread through the filter and adhered to the underside. The results from the transwell invasion assay also showed that pretreatment with TSA blocks the increased cell invasion in TGF-β1-treated cells (Fig 6B). 10.1371/journal.pone.0162058.g006Fig 6 Effects of trichostatin A on migration ability of TGF-β1-stimulated A549 cells were measured using cell migration assay (A) and transwell invasion assay (B). Values are expressed as the mean ± SEM of independent experiments. *P < 0.05 vs. control. †P < 0.05 vs. TGF-β1 alone. Scale bar = 50 μm. TSA inhibits TGF-β1-induced EMT in PNACs and organ culture To assess whether the inhibitory effects of TSA on TGF-β1-induced EMT in A549 cells are also seen in nasal tissue, we repeated several experiments in primary PNECs and in inferior turbinate organ culture. To determine whether TGF-β1 induces EMT in PNECs, we treated cells with 5 ng/mL of TGF-β1 for 72 h and observed E-cadherin, vimentin, fibronectin, and α-SMA protein expression using a fluorescence microscope. To determine protein expression of snail and slug, we treated cells with TGF-β1 for 24 h. After the treatment with TGF-β1, cells showed decreased E-cadherin and increased vimentin, fibronectin, α-SMA, snail, and slug expression. TSA pretreatment for 1 h inhibited the effects of TGF-β1 on EMT in PNECs (Fig 7A and 7B). To identify whether EMT is induced by TGF-β1 and inhibited by TSA in nasal inferior turbinate organ cultures, organ cultures were exposed to TGF-β1 for 72 h with or without TSA, and were checked for E-cadherin, vimentin, fibronectin, and α-SMA protein expression levels using western blotting (Fig 7C). Expression levels of vimentin, fibronectin, and α-SMA were increased and E-cadherin expression level was decreased in TGF-β1-treated inferior turbinate organ cultures compared with the control. Pretreatment with TSA reversed the effect of TGF-β1 on all mentioned EMT markers. These results indicate that TSA also suppresses EMT induced by TGF-β1 in nasal cells and tissue. 10.1371/journal.pone.0162058.g007Fig 7 Effects of trichostatin A on expression of E-cadherin, vimentin, fibronectin, α-smooth muscle actin, snail, and slug proteins in TGF-β1-stimulated primary nasal epithelial cells were determined by immunofluorescent staining (A). Effects of trichostatin A on expression of E-cadherin, vimentin, fibronectin, and α-smooth muscle actin protein in TGF-β1-stimulated inferior turbinate tissue were determined by western blotting (B). Representative of independent experiments. Scale bar = 50 μm. Discussion The present study showed that TSA inhibits TGF-β1-induced EMT in A549 cells, PNECs, and inferior turbinate organ culture. TGF-β1 altered the mRNA and protein expression levels of EMT markers including E-cadherin, vimentin, fibronectin, α-SMA, slug, and snail, and pretreatment with TSA reversed the effect of TGF-β1. TSA inhibited the expression of HDAC2 and HDAC4, and induced histone acetylation in A549 cells. Silencing of HDAC2 and HDAC4 by siRNA enhanced TGF-β1-induced EMT in A549 cells. When we investigated the migratory ability of A549 cells after TGF-β1 stimulation via cell migration assay and transwell invasion assay, we found that they migrated significantly further than the control. However, TSA blocked the effect of TGF-β1 on the migratory ability of cells. In the experiments using PNECs and inferior turbinate tissue, TSA suppressed the expression of EMT markers induced by TGF-β1. Remodeling is an important feature of wound healing. It is a dynamic process involving matrix production and degradation in response to inflammatory insult. Tissue remodeling can lead to a normal reconstruction process [14]. However, when remodeling causes many alterations in the composition, content, and organization of constituents of the organs, thereby causing morphological or functional disabilities, it can be considered pathological [15]. As in numerous other chronic inflammatory diseases, CRS patients develop persistent chronic inflammation of the mucosa. In a study examining histological specimens from 22 patients with refractory CRS undergoing endoscopic sinus surgery, epithelial damage such as epithelial shedding and basement membrane thickening was observed in all cases [16]. Because such remodeling processes are somewhat irreversible, there is a growing consensus that functional endoscopic sinus surgery that targets restoration of function by improving ventilation and allowing mucociliary clearance to normalize is not perfect answer to the refractory CRS [17]. Prevention of EMT is now considered an effective measure for the inhibition of tissue remodeling. Airway epithelium is a barrier between the host and the environment, and represents the first line of defense against microorganisms and allergens [18]. Airway epithelium acts as a physical barrier and is composed of apical tight junctions and underlying adherens junctions [19]. It is now recognized that airway epithelium attends a variety of immunological mechanisms by releasing cytokines and interacting with immune cells. EMT is a process essential in wound healing and tissue remodeling after injury [20]. However, in an unsuccessful attempt to repair the injured tissue that is can be happened in constant damage caused by chronic inflammation, EMT can lead to the destruction of the functions of the epithelium as a physical barrier and immune regulator. For this reason, EMT is a novel clinical therapeutic target in many chronic airway diseases. In fact, EMT was observed in several chronic inflammatory airway diseases, including asthma, COPD, and bronchiolitis obliterans syndrome [13,21,22]. There is also evidence demonstrating that epithelial cells express mesenchymal markers in CRS. We supposed that functional loss of airway epithelium caused by EMT is one of the main reasons for the unresponsiveness of recalcitrant CRS to maximal medical and surgical treatment. Epigenetic changes are changes in gene expression that do not alter the underlying DNA sequence. DNA methylation and histone modification are the most well-known mechanisms of epigenetics [23]. Histone acetylation, which is regulated by histone acetyltransferase and histone deacetylase, is one type of chromatin modification [24]. Acetylation of histones relaxes nucleosomes, thereby activating gene induction. On the contrary, histone deacetylase induces gene silencing by removal of acetyl groups from histones. Imbalance between the activities of HATs and HDACs can lead to disease states [25]. For this reason, TSA, which inhibits HDACs in a noncompetitive and reversible way, has been studied in various diseases, including cancer and fibrosis. Evidence has shown that the anti-fibrotic and anti-cancer effects of TSA are related with EMT. Wang et al. [26] showed that TSA reverses EMT in colorectal cancer cells and prostate cancer cells thereby explaining that TSA suppresses invasion and migration of cancer cells. In a study with renal cells and hepatocytes, TSA exerted anti-EMT effects[27,28]. Related with CRS, we have previously shown that HDAC2 is elevated in nasal polyps, suggesting that they may serve as potential targets of treatment and that TSA inhibits extracellular matrix production in nasal polyps [29,30]. Based on the above evidence, we can draw a hypothesis that HDAC inhibition by TSA is associated with EMT in airway epithelial cells. In conclusion, we demonstrated that EMT is induced by TGF-β1 in airway epithelial cells and nasal tissue via activation of HDAC2 and HDAC4, and that inhibition of HDAC2 and HDAC4 by TSA reduces TGF-β1-induced EMT. This observation indicates that histone deacetylase inhibitors such as TSA could be considered as candidates for treatment of recalcitrant CRS related with tissue remodeling. This study was supported by a grant from the Korean Health Technology R&D Project, Ministry of Health & Welfare, Republic of Korea (HI15C1512) and Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Science, ICT & Future Planning (NRF-2015R1C1A1A02037312) ==== Refs References 1 Fokkens WJ , Lund VJ , Mullol J , Bachert C , Alobid I , Baroody F , et al EPOS 2012: European position paper on rhinosinusitis and nasal polyps 2012. A summary for otorhinolaryngologists . Rhinology . 2012 ; 50 : 1 –12 . 10.4193/Rhino50E2 22469599 2 Baguley C , Brownlow A , Yeung K , Pratt E , Sacks R , Harvey R . The fate of chronic rhinosinusitis sufferers after maximal medical therapy . Int Forum Allergy Rhinol . 2014 ; 4 : 525 –532 . 10.1002/alr.21315 24610673 3 Smith TL , Kern R , Palmer JN , Schlosser R , Chandra RK , Chiu AG , et al Medical therapy vs surgery for chronic rhinosinusitis: a prospective, multi-institutional study with 1-year follow-up . Int Forum Allergy Rhinol . 2013 ; 3 : 4 –9 . 10.1002/alr.21065 22736422 4 Kalluri R , Neilson EG . Epithelial-mesenchymal transition and its implications for fibrosis . J Clin Invest . 2003 ; 112 : 1776 –1784 . 10.1172/jci20530 14679171 5 Lamouille S , Xu J , Derynck R . Molecular mechanisms of epithelial-mesenchymal transition . Nat Rev Mol Cell Biol . 2014 ; 15 : 178 –196 . 10.1038/nrm3758 24556840 6 Nawshad A , Lagamba D , Polad A , Hay ED . Transforming growth factor-beta signaling during epithelial-mesenchymal transformation: implications for embryogenesis and tumor metastasis . Cells Tissues Organs . 2005 ; 179 : 11 –23 . 10.1159/000084505 15942189 7 Kalluri R , Weinberg RA . The basics of epithelial-mesenchymal transition . J Clin Invest . 2009 ; 119 : 1420 –1428 . 10.1172/jci39104 19487818 8 Li M , Luan F , Zhao Y , Hao H , Zhou Y , Han W , et al Epithelial-mesenchymal transition: An emerging target in tissue fibrosis . Exp Biol Med (Maywood) . 2016 ; 241 : 1 –13 . 10.1177/1535370215597194 26361988 9 Hupin C , Gohy S , Bouzin C , Lecocq M , Polette M , Pilette C . Features of mesenchymal transition in the airway epithelium from chronic rhinosinusitis . Allergy . 2014 ; 69 : 1540 –1549 . 10.1111/all.12503 25104359 10 Shin HW , Cho K , Kim DW , Han DH , Khalmuratova R , Kim SW , et al Hypoxia-inducible factor 1 mediates nasal polypogenesis by inducing epithelial-to-mesenchymal transition . Am J Respir Crit Care Med . 2012 ; 185 : 944 –954 . 10.1164/rccm.201109-1706OC 22323302 11 Cho JS , Moon YM , Park IH , Um JY , Kang JH , Kim TH , et al Effects of histone deacetylase inhibitor on extracellular matrix production in human nasal polyp organ cultures . Am J Rhinol Allergy . 2013 ; 27 :18 –23 . 10.2500/ajra.2013.27.3827 .23406592 12 Willis BC , Liebler JM , Luby-Phelps K , Nicholson AG , Crandall ED , du Bois RM , et al Induction of epithelial-mesenchymal transition in alveolar epithelial cells by transforming growth factor-beta1: potential role in idiopathic pulmonary fibrosis . Am J Pathol . 2005 ; 166 : 1321 –1332 . 15855634 13 Hackett TL , Warner SM , Stefanowicz D , Shaheen F , Pechkovsky DV , Murray LA , et al Induction of epithelial-mesenchymal transition in primary airway epithelial cells from patients with asthma by transforming growth factor-beta1 . Am J Respir Crit Care Med . 2009 ; 180 : 122 –133 . 10.1164/rccm.200811-1730OC 19406982 14 Vignola AM , Kips J , Bousquet J . Tissue remodeling as a feature of persistent asthma . J Allergy Clin Immunol . 2000 ; 105 : 1041 –1053 . 10856134 15 Sumi Y , Hamid Q . Airway remodeling in asthma . Allergol Int . 2007 ; 56 : 341 –348 . 10.2332/allergolint.R-07-153 17965577 16 Ponikau JU , Sherris DA , Kephart GM , Kern EB , Gaffey TA , Tarara JE , et al Features of airway remodeling and eosinophilic inflammation in chronic rhinosinusitis: is the histopathology similar to asthma? J Allergy Clin Immunol . 2003 ; 112 : 877 –882 . 10.1016/j.jaci.2003.08.009 14610473 17 Bassiouni A , Naidoo Y , Wormald PJ . When FESS fails: the inflammatory load hypothesis in refractory chronic rhinosinusitis . Laryngoscope . 2012 ; 122 : 460 –466 . 10.1002/lary.22461 22252862 18 Xiao C , Puddicombe SM , Field S , Haywood J , Broughton-Head V , Puxeddu I , et al Defective epithelial barrier function in asthma . J Allergy Clin Immunol . 2011 ; 128 : 549 –556.-e512 . 10.1016/j.jaci.2011.05.038 21752437 19 Georas SN , Rezaee F . Epithelial barrier function: at the front line of asthma immunology and allergic airway inflammation . J Allergy Clin Immunol . 2014 ; 134 : 509 –520 . 10.1016/j.jaci.2014.05.049 25085341 20 Weber CE , Li NY , Wai PY , Kuo PC . Epithelial-mesenchymal transition, TGF-beta, and osteopontin in wound healing and tissue remodeling after injury . J Burn Care Res . 2012 ; 33 : 311 –318 . 10.1097/BCR.0b013e318240541e 22561306 21 Milara J , Peiro T , Serrano A , Cortijo J . Epithelial to mesenchymal transition is increased in patients with COPD and induced by cigarette smoke . Thorax . 2013 ; 68 : 410 –420 . 10.1136/thoraxjnl-2012-201761 23299965 22 Hodge S , Holmes M , Banerjee B , Musk M , Kicic A , Waterer G , et al Posttransplant bronchiolitis obliterans syndrome is associated with bronchial epithelial to mesenchymal transition . Am J Transplant . 2009 ; 9 : 727 –733 . 10.1111/j.1600-6143.2009.02558.x 19344464 23 Blumenthal MN . Genetic, epigenetic, and environmental factors in asthma and allergy . Ann Allergy Asthma Immunol . 2012 ; 108 : 69 –73 . 10.1016/j.anai.2011.12.003 22289722 24 Bhavsar P , Ahmad T , Adcock IM . The role of histone deacetylases in asthma and allergic diseases . J Allergy Clin Immunol . 2008 ; 121 : 580 –584 . 10.1016/j.jaci.2007.12.1156 18234319 25 Camelo S , Iglesias AH , Hwang D , Due B , Ryu H , Smith K , et al Transcriptional therapy with the histone deacetylase inhibitor trichostatin A ameliorates experimental autoimmune encephalomyelitis . J Neuroimmunol . 2005 ; 164 : 10 –21 . 10.1016/j.jneuroim.2005.02.022 15885809 26 Wang X , Xu J , Wang H , Wu L , Yuan W , Du J , et al Trichostatin A, a histone deacetylase inhibitor, reverses epithelial-mesenchymal transition in colorectal cancer SW480 and prostate cancer PC3 cells . Biochem Biophys Res Commun . 2015 ; 456 : 320 –326 . 10.1016/j.bbrc.2014.11.079 25434997 27 Kaimori A , Potter JJ , Choti M , Ding Z , Mezey E , Koteish AA . Histone deacetylase inhibition suppresses the transforming growth factor beta1-induced epithelial-to-mesenchymal transition in hepatocytes . Hepatology . 2010 ; 52 : 1033 –1045 . 10.1002/hep.23765 20564330 28 Yoshikawa M , Hishikawa K , Marumo T , Fujita T . Inhibition of histone deacetylase activity suppresses epithelial-to-mesenchymal transition induced by TGF-beta1 in human renal epithelial cells . J Am Soc Nephrol . 2007 ; 18 : 58 –65 . 10.1681/asn.2005111187 17135397 29 Cho JS , Moon YM , Park IH , Um JY , Kang JH , Kim TH , et al Effects of histone deacetylase inhibitor on extracellular matrix production in human nasal polyp organ cultures . Am J Rhinol Allergy . 2013 ; 27 : 18 –23 . 10.2500/ajra.2013.27.3827 23406592 30 Cho JS , Moon YM , Park IH , Um JY , Moon JH , Park SJ , et al Epigenetic regulation of myofibroblast differentiation and extracellular matrix production in nasal polyp-derived fibroblasts . Clin Exp Allergy . 2012 ; 42 : 872 –882 . 10.1111/j.1365-2222.2011.03931.x 22239687

PMC005xxxxxx/PMC5003434.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays1030107microarrays-01-00107ReviewIntegrated Amplification Microarrays for Infectious Disease Diagnostics Chandler Darrell P. 1*Bryant Lexi 1Griesemer Sara B. 2Gu Rui 2Knickerbocker Christopher 1Kukhtin Alexander 1Parker Jennifer 1Zimmerman Cynthia 1George Kirsten St. 2Cooney Christopher G. 11 Akonni Biosystems, Inc., 400 Sagner Avenue, Suite 300, Frederick, MD 21701, USA; Email: lbryant@akonni.com (L.B.); cknickerbocker@akonni.com (C.K.); akukhtin@akonni.com (A.K.); jparker@akonni.com (J.P.); czimmerman@akonni.com (C.Z.); cooney@akonni.com (C.G.C)2 Laboratory of Viral Diseases, Wadsworth Center, New York State Dept of Health, 120 New Scotland Avenue, Albany, NY 12208, USA; Email: sbg03@health.state.ny.us (S.B.G); rxg11@health.state.ny.us (R.G.); kxs16@health.state.ny.us (K.S.G.)* Author to whom correspondence should be addressed; Email: dchandler@akonni.com; Tel.: +1-734-428-0713; Fax: +1-301-698-0202.09 11 2012 12 2012 1 3 107 124 24 9 2012 31 10 2012 07 11 2012 © 2012 by the authors; licensee MDPI, Basel, Switzerland.2012This article is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).This overview describes microarray-based tests that combine solution-phase amplification chemistry and microarray hybridization within a single microfluidic chamber. The integrated biochemical approach improves microarray workflow for diagnostic applications by reducing the number of steps and minimizing the potential for sample or amplicon cross-contamination. Examples described herein illustrate a basic, integrated approach for DNA and RNA genomes, and a simple consumable architecture for incorporating wash steps while retaining an entirely closed system. It is anticipated that integrated microarray biochemistry will provide an opportunity to significantly reduce the complexity and cost of microarray consumables, equipment, and workflow, which in turn will enable a broader spectrum of users to exploit the intrinsic multiplexing power of microarrays for infectious disease diagnostics. microfluidicsdiagnosticsgel element arraysasymmetric PCRRT-PCRreverse transcriptasemultiplexintegrated microarrays ==== Body 1. Introduction While every infectious disease presents specific diagnostic challenges, there are several themes that emerge from clinical needs that eventually impact the design and development of diagnostic technologies themselves. These include but are not limited to: the number and types of microorganisms that may result in a particular assemblage of symptoms; an abundance of phylogenetically or phenotypically related microorganisms on or within the human host, and in the local (non-human) environment; rapid evolution and the acquisition of new traits, either through mutation or horizontal gene transfer; an ability to move or survive between and within the human host, non-human vectors, and/or environmental reservoirs; the capacity for colonization or dormancy without causing disease; and an aptitude to be transmitted or cause disease at low levels of infection. Even if an infection is not life-threatening, there are some universal principles and clinical user needs associated with sensitivity, specificity, total analysis time, and ease of use that have technical implications for a diagnostic platform. Diagnostic tests based on metabolic activity (e.g., growth or secondary metabolites) or molecular building blocks (e.g., lipids, proteins, nucleic acids) are developed for a very specific clinical context. Of these, nucleic acid tests based on real-time polymerase chain reaction (PCR) technology address many of the sensitivity and specificity challenges for infectious disease diagnostics, leading to widespread adoption of nucleic acid diagnostics in surveillance, epidemiology, and clinical practice over the last decade (e.g., [1,2,3,4,5,6,7,8,9,10,11]). The advent of integrated PCR systems [12,13,14,15] that address user needs for rapid analysis times and ease-of-use are likewise expected to expand and accelerate the adoption of molecular diagnostics in clinical practice, including point-of-care and point-of-use settings [16,17,18,19]. Integrated real-time PCR systems, however, typically comes at the cost of mid- to high-level multiplexing, or the ability to detect multiple microorganisms, nucleotide polymorphisms or drug resistance mutations from a single sample. Microfluidic PCR systems tend to address the multi-analyte detection problem by splitting a purified nucleic acid sample into spatially isolated analysis channels or reaction wells, where each analysis chamber contains target-specific primers and detection probes (e.g., Idaho Technologies FilmArray). Some of these systems can now support thousands of discrete real-time PCR tests in a single run (BioTrove OpenArray, and related life sciences tools). A split assay is applicable provided that the purified, target nucleic acids can be subdivided but still amplified and detected by the end-point detector (i.e., not split into extinction). The manual or automated manipulations intrinsic to parallel amplification systems, however, are still subject to a basic requirement for ≥102 copies per reaction well or vessel to avoid molecular sampling error [20,21,22] and its potential to cause false negative results, unless one employs a nested PCR strategy (as in the FilmArray). Splitting a sample (or target DNA) to extinction is actually desirable in limiting dilution PCR [23], which has now been translated into digital PCR [24], microfluidic chips, and various commercial products (e.g., Fluidigm Digital Array, BioRad ddPCR). While digital PCR is currently used for primarily quantifying DNA and analyzing copy-number variations, multiplexed digital detection is on the horizon [25], as are isothermal digital PCR systems [26]. In contrast to these PCR technologies, microarrays interrogate hundreds to millions of genetic signatures across multiple genes in a homogenous assay, and have therefore emerged as a potentially useful diagnostic platform when sample or total nucleic acid concentration is limiting [2,27,28,29,30,31]). A typical microarray workflow may involve numerous steps, including nucleic acid purification, nucleic acid amplification (gene-specific, or whole genome amplification), a second round of amplified target purification, target fragmentation, target labeling, possibly a third round of target purification, dilution with hybridization buffer, target hybridization, microarray washing, imaging, and data analysis (see e.g., [32,33,34,35,36,37,38]). In the absence of a fully automated and enclosed system and because of a continued reliance on amplification chemistry, there is significant potential for cross-contamination between samples, or contamination of the workspace with amplified nucleic acids. These workflow concerns were a primary impediment to the clinical adoption of PCR, a situation that was only surmounted with the advent of real-time PCR. In addition, the hybridization and detection steps may be further complicated depending upon whether or not the test uses allele-specific extension, ligation, or signal amplification chemistries. Some microarray methods have been fully automated, either as prototype instruments or commercial products ([39,40,41,42,43,44,45]; see also products from Autogenomics, Nanosphere; ClonDiag, and Luminex). However, microarray sensitivity and specificity are maximized with extended (>16 h) hybridizations that drive the hybridization reactions to thermodynamic equilibrium, which is generally inconsistent with the requirements for an infectious disease diagnostic test. Re-circulating flow and sample agitation are some of the methods used to improve hybridization kinetics and reduce total hybridization time, which either adds to the equipment infrastructure needed to perform a microarray test, or adds to the complexity (and cost) of an integrated microarray system. Thus, microarrays have not yet found widespread use beyond the research community due to labor, time intensive protocols, instruments and analysis software that typically require very advanced training to operate or interpret. Several commercial microarray manufacturers have simplified microarray work flow by engineering fluidic hybridization, washing, and imaging stations, and providing semi-automated software and data analysis. Sample preparation and amplification chemistries are also now being incorporated into the workflow either through robotic [43,46] or microfluidic transfer steps [15,41,44,47,48,49,50]. In spite of these advances, engineering and manufacturing challenges scale with the complexity of the integrated, microfluidic microarray systems, while robotic systems must contend with an open-amplicon workflow that plagued the early adoption of PCR technologies in clinical practice. An alternative approach to simplifying microarray workflow, especially for lower-resource settings, is to re-evaluate the biochemistry and processing steps that precede and interface with the microarray itself. By simplifying the biochemistry and analytical steps in the process, there is a corollary opportunity to simplify the complexity and cost of the microarray-based consumables and instrumentation, while meeting user needs for total analysis time, sensitivity, and specificity. This overview summarizes recent methods for combining target amplification, labeling, and microarray hybridization into a single, closed-amplicon reaction chamber, and provides examples of an integrated biochemical approach for amplification array-based analysis of DNA and RNA genomes. We also illustrate a simple consumable architecture for incorporating post-amplification wash steps with an amplification microarray while retaining an entirely closed system, which exemplifies one approach to a consumable architecture for producing low-cost, easy-to-use, closed-amplicon microarrays for point-of-use applications. 2. Solid Phase Amplification Early attempts to combine target amplification and array detection involved amplifying target nucleic acids on a solid support, where the amplification primers were cross-linked to a surface [51,52,53,54,55,56,57,58]. In most cases, these experiments showed that solid-phase PCR is less efficient than conventional solution-phase reactions with relatively poor limits of detection (approximately 105–106 genomes per reaction). These limits of detection have significant negative implications for infectious disease diagnostics. Supplementing the reaction mixture with unbound, gene-specific primers and allowing the PCR to simultaneously proceed in the liquid and solid phases was one approach to increase product yield and analytical sensitivity [59,60,61,62,63,64]. These schemes represent an analog of nested PCR, where the first stage of the reaction amplifies the nucleic acid of interest in solution using free-floating primers, and the second stage results in attachment of amplified fragments to immobilized PCR-primers with subsequent chain extension. As with any nested PCR procedure where the products of the first amplification phase are not purified before the second amplification phase, however, primer artifacts and primer interference tended to restrict multiplexing capacity and amplification efficiency. More recent variants of solid-phase or microarray-based PCR are described in several publications. For example, Tillib et al. [62] attempted to overcome primer interference by creating a microarray of monoplex PCR chambers separated from each other by mineral oil. Pemov et al. describe a gel element array where multiplex PCR occurs on and within gel element arrays and is enhanced by pseudo-monoplex PCR in solution [65]. Li et al. [66], designed a microarray of hydrophilic microwells patterned on a hydrophobic chip, where primer pairs tagged with a universal sequence were physically separated in the individual hydrophilic microwells. This construct enabled many unique PCR reactions to be proceeded simultaneously during the first step of the procedure, similar to the approach described by Tillib [62]. However, Li et al. isolated the first-stage amplification products from the PCR array for subsequent analysis by gel electrophoresis or conventional DNA microarray. Sun et al. describes an approach to influenza RNA amplification and detection, where RNA is reverse transcribed in solution over the biochip, PCR is initiated in solution with free-floating reverse primers, and the resulting cDNA is then extended from nested, immobilized primers on the array [67]. A related method used gene-specific, immobilized reverse primers to directly interrogate the mRNA transcriptome of mouse muscle fibroblasts [68]. Isothermal, helicase-dependent, solid-phase amplification has also been demonstrated, which provides an opportunity to simplify the attendant instrumentation by eliminating the need for a thermal cycler [69]. Unfortunately, solid-phase amplification is limited by the kinetics of (low-copy) target hybridization to the array surface during the initial rounds of the amplification reaction. 3. Integrated Biochemistry for Single-Step, Closed-Amplicon Microarrays There are reports of highly multiplexed, solution-phase amplification techniques that precede microarray detection, including isothermal amplification reactions, whole genome amplification systems, and highly multiplexed, gene-specific reactions (e.g., [70,71]). The underlying principles rely on driving the amplification reaction to the plateau phase and using microarray hybridization to detect specific amplicons from amplification artifacts that may arise as a consequence of the high multiplexing. Figure 1 Integrating amplification and hybridization chemistry within an amplification microarray and single microfluidic chamber. (A) Gene-specific reverse amplification primers (pA-R and pB-R) are labeled with a fluorophore and provided in excess relative to the forward primers (pA-F and pB-F). The microarray may contain one or more probes for each target gene (GeneA and Gene B) and resulting amplicon. Two probes are shown for Gene A and one probe is shown for Gene B. (B) The initial rounds of the amplification create both double-stranded and single-stranded amplicon, but hybridization to the microarray is kinetically limited because single stranded amplicon has not yet accumulated. (C) Towards the final rounds of amplification, single stranded amplicons abound, so hybridization to the microarray is kinetically favorable. Hybridization times can be extended beyond the amplification reaction to achieve thermodynamic equilibrium, if desired. In one incarnation of this approach, microarrays are printed within a gel-lined cap of a microcentrifuge tube. Nucleic acid amplification occurs within the bottom of the tube, and after amplification the tube is inverted, a hybridization buffer is released from an internal chamber, and microarray hybridization occurs within the PCR vessel [72]. The array tube format is intrinsically user friendly, but the limited surface area of a microcentrifuge cap constrains the microarray-based multiplexing capacity of the test and the arrays cannot be washed without removing the cap. A fully integrated, single cartridge for PCR amplification, array hybridization, signal enhancement, washing, and imaging has recently been described for in situ synthesized arrays [41], but the system uses off-board reagents and a fluidic processing station to complete all processing steps. The general concept of integrating quantitative, real-time PCR with microarray readouts has likewise been reported. Khodakov et al. used a first-stage, multiplex PCR in solution, and then performed a real-time, quantitative PCR amplification on gel element arrays [73]. SYBR Green I dye intercalation was used to detect target nucleic acids that were extended from gel-immobilized primers during the final 3 sec of the elongation step. Pierik et al. [74] describe a similar system and approach, except that a Cy5 fluorescent tag is incorporated into one of the solution-phase amplification primers, and real-time quantitation is based on target hybridization to the array rather than chain extension from the array. In this case, only a single-plex reaction was demonstrated. In both cases, customized instruments that integrate a thermal cycler with optical detection around a planar substrate are required, the amplification efficiency from the array is rather low, and relatively large number of cycles are required to achieve limits of detection that are useful for infectious disease diagnostics. From the foregoing literature review, it is clear that the development of closed-amplicon, microarray-based diagnostics still requires advances in integrated biochemistry, microfluidics, hardware, and software to meet user needs and requirements of infectious disease diagnostics. Towards that end, we are exploiting advances in solution-phase multiplexed amplification chemistry, the solution-phase properties and high probe immobilization capacity of gel element arrays (e.g., [75]), and microfluidic consumables [76] to significantly simplify microarray workflow for infectious disease diagnostics. The principles for simplifying the biochemical steps are to utilize multiplex, asymmetric PCR or reverse-transcriptase PCR in solution with Cy3-labeled “reverse” primers, perform thermal cycling in the presence of the gel element array, and allow the predominantly single-stranded, labeled amplicons to hybridize to the array during (or after) on-chip thermal cycling (Figure 1). Microarrays are washed (after thermal cycling) in bulk solution or with a bolus of self-imbibing wash solution pipetted into a flow cell that contains an integrated waste chamber. Three examples of the integrated biochemistry approach and one method for integrating wash steps within an amplification microarray consumable are illustrated here. Specific technical details for each of the tests can be provided upon request. 3.1. One-Step RT-PCR Amplification Microarray Influenza viruses are highly contagious, segmented, negative-sense RNA viruses that cause approximately 114,000 hospitalizations and 20,000 deaths in the U.S. annually [77]. Seventeen hemagglutinin (HA) subtypes and 10 neuraminidase (NA) subtypes are recognized, although until recently only viruses of the H1N1, H2N2, and H3N2 subtypes have been associated with widespread epidemics in humans [78,79]. Both HA and NA glycoproteins undergo antigenic drift as a result of sequence changes driven by multiple evolutionary pressures. Antigenic shift occurs when viral RNA segments re-assort during co-infection of cells by different influenza A subtypes. Rapid influenza tests are generally unable to identify influenza strains that have undergone shift or drift, or determine the match between circulating influenza viruses and those viruses contained in vaccines. Surveillance data from more sophisticated tests are therefore needed to monitor for the emergence of antiviral resistance or new influenza A subtypes that might pose a pandemic threat. The diagnostic challenge, then, is for a test to detect many different influenza signatures in trace amounts, from relatively complex and diverse sample matrices and biological backgrounds. PCR primers and microarray probes were designed from real-time RT-PCR assays developed and used by the Laboratory of Viral Diseases at the Wadsworth Center, New York State Department of Health. Gel element microarrays were manufactured essentially as described in [80] and targeted various portions of the M, NS, HA, NP and NA genes. Control probes include Cy3 beacons for positional reference, and a human GAPDH internal positive control for sample collection, extraction, reverse transcription, amplification, and detection. The microarray was surrounded by a single 50 uL gasket, and the RT-PCR amplification microarray master mix was applied to the RT-PCR amplification microarray, followed by RNA template. Arrays were sealed with a plastic cover slip and the microarray substrates mounted on a flat block thermal cycler, processed for 40 thermal cycles, and then hybridized at room temperature for up to 2 h. There was no post-PCR target labeling, fragmentation, purification, quantitation, or transfer into hybridization buffer, as typically required for conventional microarray procedures. After hybridization, the cover slips were removed, microarrays transferred to a histology slide holder, and then gently washed in buffer. Slides were then dipped in de-ionized water, air dried, and imaged on an Akonni portable, prototype imager with a 0.5 s exposure time. Local background was subtracted from the signal of each gel element and integrated, background-corrected signal intensity data exported into Excel spreadsheets for data analysis. Signal was calculated as the median background-corrected integral intensity for each probe, where each probe was printed in quadruplicate per array. Noise was calculated as 3 times the average standard deviation of all local backgrounds. Signal-to-noise ratios (SNR) >3 were considered detectable over all sources of noise. The total analysis time for the prototype RT-PCR amplification microarray test was approximately 6 h using a 2 h post-PCR hybridization time. Estimated limits of detection and amplification microarray specificity for influenza A/H3N2 and influenza B and their respective, perfectly-matched target probes are shown in Figure 2. The influenza A/H3 probe was detectable at an input of 102 genomes per reaction while the influenza A probe was detectable at 103 genomes per reaction. In contrast, influenza B was readily detected at 10 genomes·rxn−1. These results are consistent with those obtained with a conventional microarray amplification and hybridization approach and identical microarray probes (not shown), and demonstrate the potential for sensitive influenza detection within a single-step, closed amplicon reverse transcriptase amplification microarray. The bulk washing procedure described here was used primarily to increase throughput during the development and optimization of the multiplex RT-PCR and integrated hybridization chemistry, and such a manipulation may or may not be an acceptable practice in some clinical work environments or low-resource settings. One possible solution to this challenge is a valve-less, integrated flow cell that incorporates an inlet port and waste chamber, as described in detail in [76] and illustrated below. Figure 2 Sensitivity and specificity of a prototype influenza RT-amplification microarray. Results are the average from two technical replicates, where positive detection is defined as a signal to noise threshold ≥3. 3.2. Two-Step RT-PCR Amplification Microarray for DNA and RNA Genomes Encephalitis and meningitis are potentially fatal diseases defined by acute inflammation of the brain or protective membranes covering the brain and spinal cord of the central nervous system. These diseases can be caused by viruses, bacteria, fungi and parasites, and disproportionately affect children, the elderly, or the immunocompromised. Viruses are the most common cause of encephalitis, with the majority of aseptic encephalitis cases caused by enteroviruses, human herpesviruses, and arboviruses. Unfortunately, clinical presentation and initial laboratory findings in most cases of meningitis and encephalitis are overly nonspecific to permit an etiologic diagnosis. Because complications arising from CNS infection and appropriate treatment strategies can often depend on the organism involved, it is important to differentiate between those cases for which a specific treatment has been shown to be effective, and those for which only supportive treatment is indicated. PCR primers and microarray probes for herpes simplex viruses 1 and 2 (HSV-1, HSV-2), varicella zoster virus (VZV), cytomegalovirus (CMV), human herpesvirus 6 (HHV-6), enterovirus (EVU) and West Nile virus (WNV) were designed from real-time PCR assays originally developed and used by the Laboratory of Viral Diseases at the Wadsworth Center, New York State Department of Health. Control probes included Cy3 beacons for positional reference, and the green fluorescent protein (GFP) gene as an internal positive control for amplification and detection. In this case, the multiplex, asymmetric master mix consists of 8 primer pairs. Viral RNA, DNA and cDNA templates were quantified by clinically validated real-time PCR assays before use. Reaction mixtures were assembled and transferred to the gel element array chambers as described above. Amplification microarrays were processed for 45 cycles, and then hybridized at for 1–3 h, with no user intervention after applying the amplification master mix to the microarray. After hybridization, microarrays were washed, dried, and imaged as described above. The total analysis time for a test that included a 1 h, post-PCR hybridization was approximately 4 h. microarrays-01-00107-t001_Table 1Table 1 Average signal to noise ratios (n = 2) for a multiplex, closed-amplicon amplification microarray. Bold type indicates specific amplification microarray signals. Target (gene copies per reaction) Probe Specificity HSV1 (5,000) HSV2 (500) VZV (500) GFP (5,500) EVU (500) CMV (500) HHV6 (500) WNV (500) HSV1 34.56 0.72 0.67 0.44 0.68 0.55 0.54 0.17 HSV2 1.70 66.94 0.14 0.21 −0.08 0.16 0.64 −0.01 VZV 0.86 0.64 24.82 0.76 0.44 −0.21 0.90 0.05 GFP 0.76 1.10 0.43 244.26 0.52 0.53 0.04 0.21 EVU 1.03 0.57 0.63 0.27 198.73 −0.03 1.21 0.14 CMV 0.81 0.92 0.57 0.62 4.31 346.28 0.36 0.20 HHV6 1.65 0.67 1.16 0.22 0.72 0.70 132.13 −0.31 WNV 0.74 0.64 0.75 0.34 0.65 0.52 1.72 472.55 Results for 500 to 5,000 gene copies of each nucleic acid target per reaction are shown in Table 1. For these experiments, EVU and WNV RNA were first reverse-transcribed with random primers for 60 min before adding cDNA to the amplification microarray. Amplification and detection specificity was 100%. No template controls were all negative (not shown). When the same two-step asymmetric, multiplex amplification microarray was tested against >105 gene copies of non-target microorganisms, there was no detectable signal and no visible amplification product as determined by agarose gel electrophoresis (not shown). While a test-tube reverse transcriptase step (Table 1, RNA viruses) does not evoke the same level of cross-contamination concerns as transferring amplified material from one reaction tube to another, the user work flow is certainly simplified with a single-step RT-PCR amplification microarray procedure. Table 2 shows results for single-plex, on-chip, asymmetric reverse transcriptase amplification microarray detection of EVU and WNV RNA to at least 100 copies per reaction, an analytical sensitivity comparable with conventional, test tube, one-step RT-PCR. microarrays-01-00107-t002_Table 2Table 2 One-step, asymmetric RT-amplification microarray detection of West Nile virus and enterovirus RNA in single-plex reactions. Data are the average signal to noise ratios from n = 3 technical replicates. Bold type indicates specific amplification microarray signals. No template controls were all negative (not shown). Target (gene copies per reaction) Probe Specificity Enterovirus RNA West Nile virus RNA 104 103 102 104 103 102 HSV1 0.21 0.05 0.09 0.18 0.39 0.12 HSV2 −0.25 0.08 −0.08 −0.46 −0.20 −0.30 VZV 0.76 0.38 0.39 0.20 0.07 −0.01 GFP −0.25 0.25 0.01 0.13 −0.27 0.07 EVU 405.00 198.73 29.57 −0.16 0.08 −0.44 CMV 3.49 1.38 0.74 0.13 0.15 0.11 HHV6 0.03 0.57 0.11 0.27 0.22 −0.35 WNV −0.26 0.03 0.01 1,093.61 824.32 1,263.02 3.3. Integrated Waste Chamber and Entirely Closed-Amplicon Consumable Conventional drug susceptibility testing of M. tuberculosis isolates may take weeks or months, and while molecular methods are more rapid, there are numerous genes and single nucleotide substitutions within those genes that confer antibiotic resistance (see review of multi-drug resistant TB mutations in [81]). In 2006, approximately 5% of all new TB cases were MDR, an increase of 12% since 2004 and a 56% increase since 2000. From 2008 until 2010, the number of global MDR-TB cases grew an astounding 48% from 440,000 to 650,000 [82]. Molecular tests are therefore needed that can span the totality of known drug resistant mutations for both epidemiological investigations and clinical diagnostics. Low-resource settings, where tuberculosis is a significant public health problem, present some technological, manufacturing, and operational challenges for multiplexed tests. Cold-chain storage, intermittent power, complex workflows, challenging sample matrices (i.e., sputum), and biosafety concerns are several of the operational realities that exist in these locations. Even a relatively simple post-amplification wash as described above is operationally problematic. While integrated systems overcome a number of these challenges, many of the integrated, microfluidic, array-based detection systems that are expensive to manufacture and operate, and often require significant financial subsidies to reach affordable levels for governments and users where TB is endemic. For these and related reasons, then, there is now significant interest in developing “resource-appropriate” molecular platforms for TB and other infectious diseases. We have developed a valve-less, integrated amplification microarray consumable that addresses some of these work-flow and operational concerns [76]. The simplifying fluidic principle uses some of the concepts of lateral flow devices for post-PCR washing and drying the array chamber. Samples are introduced into the integrated flow cell with a pipette, thermal cycled, and then washed by injecting wash solution through the amplification/array chamber. All solutions (unbound amplicons, wash solutions) are automatically imbibed into the waste chamber, mimicking lateral flow fluidics. The feasibility for such a valve-less amplification microarray and consumable is shown here for a prototype MDR-TB microarray. Primers and probes were included for rpoB (five mutations), katG (one mutation), inhA (four mutations), IS6110 (M. tuberculosis complex), and IS1245 (M. avium complex). A matched pair of microarray probes (wildtype and single-nucleotide mutant) was designed for each mutation of interest. Nucleic acids from NALC-NaOH treated, heat-killed M. tuberculosis were isolated and assembled into a 5-plex, asymmetric amplification microarray reaction mixture before introducing the solution through the sample inlet port (Figure 3(A)). The port was then sealed with a foil disk, the array processed on a flat block thermocycler (Figure 3(B)) for 50 cycles, and then hybridized for 3 h. After thermal cycling and hybridization, the array chamber was washed by piercing the inlet port with a pipettor and sequentially dispensing buffer, Milli-Q water, and acetone through the chamber. Microarrays were imaged on Akonni’s portable analyzer (Figure 3(C,D)), and analyzed as described above. None of these experiments required disassembly of the flow cell or reaction chamber (as for the influenza and encephalitis examples described above), thereby maintaining a truly closed-amplicon workflow that is appropriate for lower-resource settings and most CLIA-certified molecular diagnostic laboratories. Figure 3 (A) Closed-amplicon amplification microarray with integrated waste chamber. (B) Amplification microarray flow cells in a Quanta flat block thermocycler. (C) Akonni portable microarray analyzer used to image amplification microarrays. (D) Image of a prototype MDR-TB array following a 50-cycle closed-amplicon asymmetric PCR protocol. microarrays-01-00107-t003_Table 3Table 3 Amplification microarray flow cell genotyping of MDR and wild-type M. tuberculosis samples (250 pg and 15 pg). Wild-type to mutant probe ratios <1 (bold, shaded) indicate a mutation at the designated position within the gene(s) of interest. DST Phenotype and DNA amounts MDR 250 pg MDR 15 pg WT 250 pg WT 15 pg Gene Mutation Wildtype to Mutant Ratios rpoB D516V 5.8 9.1 5.4 8.4 H526D 3.5 7.9 4.4 8.5 H526Y1 3.9 5.1 2.7 3.4 S531L 0.1 0.1 2.2 3.1 L533P 9.1 7.9 7.2 13.4 katG S315T 0.2 0.3 12.0 11.5 inhA T8A 9.1 4.9 8.2 4.5 T8C 5.1 2.8 4.7 2.5 C15T 7.0 7.7 7.4 7.8 G17T 8.0 3.1 7.1 2.9 Genotyping results for two isolates (250 pg and 15 pg) are shown in Table 3 relative to the MDR-TB drug susceptibility phenotype, where a wild-type to mutant probe ratio >1 is indicative of a wild-type sequence at that gene position (hence, drug susceptibility), and ratios <1 are indicative of a drug-resistance mutation at that gene position. The isolates were correctly typed and the genotyping results were consistent with a more traditional procedure that involves test-tube multiplex PCR amplification followed by stand-alone, microarray hybridization (not shown). Similar on-chip PCR results were obtained with both 15 pg genomic DNA per reaction (~3.5 × 103 genomes) and 250 pg (~5.8 × 104 genomes). From this basic prototype, we can now significantly expand the microarray content to include all known rpoB, katG and inhA mutations that reside between the primer sequences, begin incorporating additional gene targets associated with antibiotic resistance to first- and second-line drugs, and optimize reaction conditions to genotype MDR-TB from as few as 10 genomes. 4. Discussion and Future Prospects The example amplification microarrays described here illustrate that multiplex reverse transcriptase or PCR-to-microarray tests can be significantly simplified through integrated biochemistry and still achieve clinically relevant limits of detection. This represents a significant advance in clinical ease-of-use for microarray-based diagnostics. Translating the prototype tests into clinically useful research or diagnostic tools will require continued development of appropriate controls, prospective analysis of clinical samples, and a number of formal analytical studies to establish limits of detection and repeatability across arrays, users, and laboratories. Nevertheless, integrated biochemistry enables the development and manufacture of simple, low-cost consumables that can be integrated into current clinical environments and workflows without new infrastructure or expensive equipment. The extent of multiplexing for an amplification microarray appears to be dictated by the number of primer pairs in the master mix rather than the number of probes on the array, for the same reasons that tend to limit the number of genes that can be co-amplified by conventional solution-phase, multiplex PCR. Whether or not whole-genome amplification (WGA), multiple displacement amplification (MDA), or related approaches can be utilized within the context of an amplification microarray, the confines of a single chamber, and a single buffer condition is to be determined. Whether a microarray is used independent of the amplification reaction, for solid-phase PCR, or as an amplification microarray, limits of detection are primarily governed by the interrelationship between starting template concentration, amplification efficiency, microarray hybridization kinetics, and the total number of amplification cycles. At low target concentrations that typify the early cycles of a solid-phase PCR or amplification microarray test, hybridization kinetics are slow. Even after amplicons begin to accumulate in solution over the biochip, microarray hybridization is most sensitive and specific at thermodynamic equilibrium (which is typically reached at >16 h, and with mixing/agitation). Because of thermodynamic and kinetic constraints of solid-phase hybridization, “rapid” microarray tests (inclusive of gel element arrays) therefore often come at the potential expense of detection sensitivity or specificity. One approach to this kinetic constraint is to increase the total number of amplification cycles in order to “drive” microarray hybridization to completion, a method that is especially attractive for asymmetric (or log-linear) amplification reactions (see [83] for an 85-cycle example). In this case, the total analysis time is primarily based on the speed of thermal cycling and heat transfer rates to the on-chip solution. Flat block, peltier-driven, in situ thermal cyclers such as the one used here are much slower than the rapid cycling available in today’s most advanced tube-based thermal cyclers, and are unlikely to achieve 40–60 cycles within a 1–2 h time frame. Overcoming this limitation for PCR microarray-based consumables will therefore require new thermal cycler designs that conform to the unique shapes, sizes and heat-transfer properties of microfluidic, array-based consumables. In this regard, translating to thinner microarray substrates is expected to improve heat transfer rates and thermal cycling efficiency, so that the rate limiting step for amplification microarrays becomes enzyme processivity rather than thermal cycling ramp rates. A potential biochemical approach to the thermal cycling challenge is isothermal amplification methods (e.g., [13,84,85,86,87,88,89]). Isothermal amplification systems are especially attractive for low-resource settings, but currently available enzymes have relatively slow processivity, the primer design strategy is complicated and not conducive to higher-levels of multiplexing, or the methods require multiple steps that would otherwise lead to a complex and expensive microfluidic consumable. Advances in basic enzymology or assay conditions are therefore needed to develop single-step isothermal methods that exploit the multiplexing power of microarrays. In the interim, we have demonstrated that single-step, integrated, closed-amplicon gel element microarrays that are as simple and uncomplicated as many real-time PCR tests can address microarray work flow challenges and clinically relevant infectious disease diagnostic problems. Acknowledgments This work was supported in part by National Institutes of Health (NIH) grants R44 AI072784, RC3 AI089106, and R43 AI085650 to DPC, and R43 EB011274 to CGC. This results presented here would not be possible without the contributions of Amy Dean and Daryl Lamson from the Laboratory of Viral Diseases at the Wadsworth Center, New York State Department of Health. We thank Julia Golova, Amine Lambarqui, Yvonne Linger, Alexander Perov, George Rudy, and Cory Wiles from Akonni Biosystems for microarray manufacture and technical assistance. ==== Refs References 1. Lyon E. Wittwer C.T. LightCycler technology in molecular diagnostics J. Mol. Diagn. 2009 11 93 101 10.2353/jmoldx.2009.080094 19196999 2. Liu Y.T. A technological update of molecular diagnostics for infectious diseases Infect. Disord. Drug Targets 2008 8 183 188 18782035 3. Dong J. Olano J.P. McBride J.W. Walker D.H. Emerging pathogens: Challenges and successes of molecular diagnostics J. Mol. Diagn. 2008 10 185 197 10.2353/jmoldx.2008.070063 18403608 4. Yang S. Rothman R.E. PCR-based diagnostics for infectious diseases: Uses, limitations, and future applications in acute-care setting Lancet Infect Dis. 2004 4 337 348 10.1016/S1473-3099(04)01044-8 15172342 5. Dumler J.S. Valsamakis A. Molecular diagnostics for existing and emerging infections. Complementary tools for a new era of clinical microbiology Am. J. Clin. Pathol. 1999 112 S33 S39 10396299 6. Millar B.C. Xu J. Moore J.E. Molecular diagnostics of medically important bacterial infections Curr. Issues Mol. Biol. 2007 9 21 39 17263144 7. Robertson B.H. Nicholson J.K. New microbiology tools for public health and their implications Annu. Rev. Public Health 2005 26 281 302 10.1146/annurev.publhealth.26.021304.144522 15760290 8. O’Connor L. Glynn B. Recent advances in the development of nucleic acid diagnostics Expert Rev. Med. Devices 2010 7 529 539 10.1586/erd.10.22 20583889 9. Kaltenboeck B. Wang C. Advances in real-time PCR: Application to clinical laboratory diagnostics Adv. Clin. Chem. 2005 40 219 259 10.1016/S0065-2423(05)40006-2 16355924 10. Mackay I.M. Real-time PCR in the microbiology laboratory Clin. Microbiol. Infect. 2005 10 190 212 10.1111/j.1198-743X.2004.00722.x 15008940 11. Procop G.W. Molecular diagnostics for the detection and characterization of microbial pathogens Clin. Infect. Dis. 2007 45 S99 S111 10.1086/519259 17683022 12. Easley C.J. Karlinsey J.M. Bienvenue J.M. Legendre L.A. Roper M.G. Feldman S.H. Hughes M.A. Hewlett E.L. Merkel T.J. Ferrance J.P. Landers J.P. A fully integrated microfluidic genetic analysis system with sample-in-answer-out capability Proc. Natl. Acad. Sci. USA 2006 103 19272 19277 17159153 13. Mahalanabis M. Do J. Al Muayad H. Zhang J.Y. Klapperich C.M. An integrated disposable device for DNA extraction and helicase dependent amplification Biomed. Microdevices 2010 12 353 359 10.1007/s10544-009-9391-8 20066496 14. Chen D. Mauk M. Qiu X. Liu C. Kim J. Ramprasad S. Ongagna S. Abrams W.R. Malamud D. Corstjens P.L. Bau H.H. An integrated, self-contained microfluidic cassette for isolation, amplification, and detection of nucleic acid Biomed. Microdevices 2010 12 705 719 10.1007/s10544-010-9423-4 20401537 15. Njoroge S.K. Chen H.W. Witek M.A. Soper S.A. Integrated microfluidic systems for DNA analysis Top. Curr. Chem. 2011 304 203 260 10.1007/128_2011_153 21607848 16. Bissonnette L. Bergeron M.G. Diagnosing infections—Current and anticipated technologies for point-of-care diagnostics and home-based testing Clin. Microbiol. Infect. 2010 16 1044 1053 10.1111/j.1469-0691.2010.03282.x 20670286 17. Park S. Zhang Y. Lin S. Wang T.H. Yang S. Advances in microfluidic PCR for point-of-care infectious disease diagnostics Biotechnol. Adv. 2011 29 830 839 10.1016/j.biotechadv.2011.06.017 21741465 18. Niemz A. Ferguson T.M. Boyle D.S. Point-of-care nucleic acid testing for infectious diseases Trends Biotechnol. 2011 29 240 250 10.1016/j.tibtech.2011.01.007 21377748 19. Peeling R.W. Mabey D. Point-of-care tests for diagnosing infections in the developing world Clin. Microbiol. Infect. 2010 16 1062 1069 10.1111/j.1469-0691.2010.03279.x 20670288 20. Taylor T.B. Winn-Deen E.S. Picozza E. Woudenberg T.M. Albin M. Optimization of the performance of the polymerase chain reaction in silicon-based microstructures Nucl. Acids Res. 1997 25 3164 3168 10.1093/nar/25.15.3164 9224619 21. Irwin P.L. Nguyen L.H. Chen C.Y. The relationship between purely stochastic sampling error and the number of technical replicates used to estimate concentration at an extreme dilution Anal. Bioanal. Chem. 2010 398 895 903 10.1007/s00216-010-3967-2 20635079 22. Walsh P.S. Erlich H.A. Higuchi R. Preferential PCR amplification of alleles: Mechanisms and solutions PCR Methods Appl. 1992 1 241 250 10.1101/gr.1.4.241 1477658 23. Sykes P.J. Neoh S.H. Brisco M.J. Hughes E. Condon J. Morley A.A. Quantitation of targets for PCR by use of limiting dilution Biotechniques 1992 13 444 449 1389177 24. Vogelstein B. Kinzler K.W. Digital PCR Proc. Natl. Acad. Sci. USA 1999 96 9236 9241 10.1073/pnas.96.16.9236 10430926 25. Zhong Q. Bhattacharya S. Kotsopoulos S. Olson J. Taly V. Griffiths A.D. Link D.R. Larson J.W. Multiplex digital PCR: Breaking the one target per color barrier of quantitative PCR Lab Chip 2011 11 2167 2174 10.1039/c1lc20126c 21584334 26. Shen F. Davydova E.K. Du W. Kreutz J.E. Piepenburg O. SIsmagilov R.F. Digital isothermal quantification of nucleic acids via simultaneous chemical initiation of recombinase polymerase amplification reactions on SlipChip Anal. Chem. 2011 83 3533 3540 21476587 27. Weile J. Knabbe C. Current applications and future trends of molecular diagnostics in clinical bacteriology Anal. Bioanal. Chem. 2009 394 731 742 10.1007/s00216-009-2779-8 19377839 28. Shen Y. Wu B.L. Microarray-based genomic DNA profiling technologies in clinical molecular diagnostics Clin. Chem. 2009 55 659 669 10.1373/clinchem.2008.112821 19233918 29. Miller M.B. Tang Y.-W. Basic concepts of microarrays and potential applications in clinical microbiology Clin. Microbiol. Rev. 2009 22 611 633 10.1128/CMR.00019-09 19822891 30. Wu L. Williams P.M. Kock W. Clinical applications of microarray-based diagnostic tests Biotechniques 2005 39 S577 S582 18957036 31. Clerc O. Greub G. Routine use of point-of-care tests: Usefulness and application in clinical microbiology Clin. Microbiol. Infect. 2010 16 1054 1061 10.1111/j.1469-0691.2010.03281.x 20670287 32. Metzgar D. Myers C.A. Russell K.L. Faix D. Blair P.J. Brown J. Vo S. Swayne D.E. Thomas C. Stenger D.A. Single assay for simultaneous detection and differential identification of human and avian influenza virus types, subtypes, and emergent variants PLoS One 2010 5 e8995 20140251 33. Chen E.C. Miller S.A. DeRisi J.L. Chiu C.Y. Using a pan-viral microarray assay (Virochip) to screen clinical samples for viral pathogens J. Vis. Exp. 2011 50 10.3791/2536 34. Mahony J. Chong S. Merante F. Yaghoubian S. Sinha T. Lisle C. Janeczko R. Development of a respiratory virus panel test for detection of twenty human respiratory viruses by use of multiplex PCR and a fluid microbead-based assay J. Clin. Microbiol. 2007 45 2965 2970 10.1128/JCM.02436-06 17596360 35. Kessler N. Ferraris O. Palmer K. Marsh W. Steel A. Use of the DNA flow-thru chip, a three-dimensional biochip, for typing and subtyping of influenza viruses J. Clin. Microbiol. 2004 42 2173 2185 10.1128/JCM.42.5.2173-2185.2004 15131186 36. Palacios G. Quan P.-L. Jabado O.M. Conlan S. Hirschberg D.L. Liu Y. Zhai J. Renwick N. Hui J. Hegyi H. Panmicrobial oligonucleotide array for diagnosis of infectious diseases Emerg. Infect. Dis. 2007 13 73 81 10.3201/eid1301.060837 17370518 37. Lodes M.J. Suciu D. Wilmoth J.L. Ross M. Munro S. Dix K. Bernards K. Stover A.G. Quintana M. Iihoshi N. Lyon W.J. Danley D.L. McShea A. Identification of upper respiratory tract pathogens using electrochemical detection on an oligonucleotide array PLoS ONE 2007 2 e924 17895966 38. Caoili J.C. Mayorova A. Sikes D. Hickman L. Plikaytis B.B. Shinnick T.M. Evaluation of the TB-biochip oligonucleotide microarray system for rapid detection of rifampin resistance in Mycobacterium tuberculosis J. Clin. Microbiol. 2006 44 2378 2381 10.1128/JCM.00439-06 16825352 39. Liu R.H. Yang J. Lenigk R. Bonanno J. Grodzinski P. Self-contained, fully integrated biochip for sample preparation, polymerase chain reaction amplification, and DNA microarray detection Anal. Chem. 2004 76 1824 1831 10.1021/ac0353029 15053639 40. Liu R.H. Lodes M.J. Nguyen T. Siuda T. Slota M. Fuji H.S. McShea A. Validation of a fully integrated microfluidic array device for influenza A subtype identification and sequencing Anal. Chem. 2006 78 4184 4193 16771549 41. Summerer D. Hevroni D. Jain A. Oldenburger O. Parker J. Caruso A. Stähler C.F. Stähler P.F. Beier M. A flexible and fully integrated system for amplification, detection and genotyping of genomic DNA targets based on microfluidic oligonucleotide arrays New Biotechnol. 2010 27 149 155 42. Trau D. Lee T.M.H. Lao A.I.K. Lenigk R. Hsing I.-M. Ip N.Y. Carles M.C. Sucher N.J. Genotyping on a complementary metal oxide semiconductor silicon polymerase chain reaction chip with integrated DNA microarray Anal. Chem. 2002 74 3168 3173 10.1021/ac020053u 12141679 43. Regan J. Létant S. Adams K. Nguyen N. Derlet R. Cohen S. Vitalis E. Tammero L. Ortiz J. McBride M. Birch J. A sample-in-answer-out instrument for the detection of multiple respiratory pathogens in unprepared nasopharyngeal swab samples Analyst 2010 135 2316 2322 10.1039/c0an00223b 20596587 44. Teo J. Pietro P.D. Biagio F.S. Capozzoli M. Deng Y.M. Barr I. Caldwell N. Ong K.L. Sato M. Tan R. Lin R. VereFlu™: An integrated multiplex RT-PCR and microarray assay for rapid detection and identification of human influenza A and B viruses using lab-on-chip technology Arch. Virol. 2011 156 1371 1378 10.1007/s00705-011-0999-7 21503642 45. Yeung S.-W. Lee T.M.-H. Cai H. Hsing I.M. A DNA biochip for on-the-spot multiplexed pathogen identification Nucl. Acids Res. 2006 34 10.1093/nar/gkl702 46. Dorris D.R. Ramakrishnan R. Trakas D. Dudzik F. Belval R. Zhao C. Nguyen A. Domanus M. Mazumder A. A highly reproducible, linear, and automated sample preparation method for DNA microarrays Genome Res. 2002 12 976 984 10.1101/gr.227402 12045151 47. Liu R.H. Dill K. Fuji H.S. McShea A. Integrated microfluidic biochips for DNA microarray analysis Expert Rev. Mol. Diagn. 2006 6 253 261 10.1586/14737159.6.2.253 16512784 48. Raymond F. Carbonneau J. Boucher N. Robitaille L. Boisvert S. Wu W.-K. De Serres G. Boivin G. Corbeil J. Comparison of automated microarray detection with real-time PCR assays for detection of respiratory viruses in specimens obtained from children J. Clin. Microbiol. 2009 47 743 750 10.1128/JCM.01297-08 19158263 49. Kumar S. Wang L. Fan J. Kraft A. Bose M.E. Tiwari S. Van Dyke M. Haigis R. Luo T. Ghosh M. Detection of 11 common viral and bacterial pathogens causing community-acquired pneumonia or sepsis in asymptomatic patients by using a multiplex reverse transcription-PCR assay with manual (enzyme hybridization) or automated (electronic microarray) detection J. Clin. Microbiol. 2008 46 3063 3072 10.1128/JCM.00625-08 18650351 50. Foglieni B. Brisci A. San Biagio F. Di Pietro P. Petralia S. Conoci S. Ferrari M. Cremonesi L. Integrated PCR amplification and detection processes on a Lab-on-Chip platform: A new advanced solution for molecular diagnostics Clin. Chem. Lab. Med. 2010 48 329 336 20020819 51. Stamm S. Brosius J. Sanchored PCR: PCR with cDNA coupled to a solid phase Nucl. Acid. Res. 1991 19 1350 10.1093/nar/19.6.1350 52. Erdogan F. Kirchner R. Mann W. Ropers H.-H. Nuber U.A. Detection of mitochondrial single nucleotide polymorphisms using a primer elongation reaction on oligonucleotide microarrays Nucl. Acid. Res. 2001 29 10.1093/nar/29.7.e36 53. Shapero M.H. Leuther K.K. Nguyen A. Scott M. Jones K.W. SNP genotyping by multiplexed solid-phase amplification and fluorescent minisequencing Genome Res. 2001 11 1926 1934 11691857 54. Lockley A.K. Jones C.G. Bruce J.S. Franklin S.J. Bardsley R.G. Colorimetric detection of immobilised PCR products generated on a solid support Nucl. Acid. Res. 1997 25 1313 1314 10.1093/nar/25.6.1313 55. Adessi C. Matton G. Ayala G. Turcatti G. Mermod J.-J. Mayer P. Kawashima E. Solid phase DNA amplification: characterisation of primer attachment and amplification mechanisms Nucl. Acid. Res. 2000 28 e87 10.1093/nar/28.20.e87 56. Adams C.P. Kron S.J. Method for Performing Amplification of Nucleic Acid with Two Primers Bound to a Single Solid Support U.S. Patent 5,641,658 24 6 1997 57. Onodera K. d’Offay J. Melcher U. Nylon membrane-immobilized PCR for detection of bovine viruses Biotechniques 2002 32 74 80 11808702 58. Westin L. Xu X. Miller C. Wang L. Edman C.F. Nerenberg M. Anchored multiplex amplification on a microelectronic chip array Nat. Biotechnol. 2000 18 199 204 10.1038/72658 10657128 59. Strizhkov B.N. Drobyshev A.L. Mikhailovich V.M. Mirzabekov A.D. PCR amplification on a microarray of gel-immobilized oligonucleotides: detection of bacterial toxin- and drug-resistant genes and their mutations Biotechniques 2000 29 844 857 11056816 60. Turner M.S. Penning S. Sharp A. Hyland V.J. Harris R. Morris C.P. van Daal A. Solid-phase amplification for detection of C282y and H63D hemochromatosis (HFE) gene mutations Clin. Chem. 2001 47 1384 1389 11468226 61. Huber M. Losert D. Hiller R. Harwanegg C. Mueller M.W. Schmidt W.M. Detection of single base alterations in genomic DNA by solid phase polymerase chain reaction on oligonucleotide microarrays Anal. Biochem. 2001 299 24 30 11726180 62. Tillib S.V. Strizhkov B.N. Mirzabekov A.D. Integration of multiple PCR amplifications and DNA mutation analyses by using oligonucleotide microchip Anal. Biochem. 2001 292 155 160 11319830 63. Mitterer G. Huber M. Leidinger E. Kirisits C. Lubitz W. Mueller M.W. Schmidt W.M. Microarray-based identification of bacteria in clinical samples by solid-phase PCR amplification of 23S ribosomal DNA sequences J. Clin. Microbiol. 2004 42 1048 1057 10.1128/JCM.42.3.1048-1057.2004 15004052 64. Mitterer G. Schmidt W.M. Microarray-based detection of bacteria by on-chip PCR Methods Mol. Biol. 2006 345 37 51 16957345 65. Pemov A. Modi H. Chandler D.P. Bavykin S. DNA analysis with multiplex microarray-enhanced PCR Nucl. Acid. Res. 2005 33 e11 10.1093/nar/gnh184 66. Li Y. Guo S.J. Shao N. Tu S. Xu M. Ren Z.R. Ling X. Wang G.Q. Lin Z.X. Tao S.C. A universal multiplex PCR strategy for 100-plex amplification using a hydrophobically patterned microarray Lab Chip 2011 11 3609 3618 10.1039/c1lc20526a 21909519 67. Sun Y. Dhumpa R. Bang D.D. Handberg K. Wolff A. DNA microarray-based solid-phase RT-PCR for rapid detection and identification of influenza virus type A and subtypes H5 and H7 Diagn. Microbiol. Infect. Dis. 2011 69 432 439 10.1016/j.diagmicrobio.2010.11.008 21396541 68. von Nickisch-Rosenegk M. Marschan X. Andresen D. Bier F.F. Reverse transcription-polymerase chain reaction on a microarray: The integrating concept of “active arrays” Anal. Bioanal. Chem. 2008 391 1671 1678 10.1007/s00216-008-2154-1 18506429 69. Andresen D. Von Nickisch-Rosenegk M. Bier F.F. Helicase-dependent amplification: Use in onchip amplification and potential for point-of-care diagnostics Expert Rev. Mol. Diagn. 2009 9 645 650 10.1586/erm.09.46 19817549 70. Vora G.J. Meador C.E. Stenger D.A. Andreadis J.D. Nucleic acid amplification strategies for DNA microarray-based pathogen detection Appl. Environ. Microbiol. 2004 70 3047 3054 10.1128/AEM.70.5.3047-3054.2004 15128566 71. Fredriksson S. Banér J. Dahl F. Chu A. Ji H. Welch K. Davis R.W. Multiplex amplification of all coding sequences within 10 cancer genes by Gene-Collector Nucl. Acid. Res. 2007 35 10.1093/nar/gkm078 72. Borel N. Kempf E. Hotzel H. Schubert E. Torgerson P. Slickers P. Ehricht R. Tasara T. Pospischil A. Sachse K. Direct identification of chlamydiae from clinical samples using a DNA microarray assay: A validation study Mol. Cell. Probe 2008 22 55 64 10.1016/j.mcp.2007.06.003 73. Khodakov D.A. Zakharova N.V. Gryadunov D.A. Filatov F.P. Zasedatelev A.S. Mikhailovich V.M. An oligonucleotide microarray for multiplex real-time PCR identification of HIV-1, HBV, and HCV Biotechniques 2008 44 241 248 18330353 74. Pierik A. Boamfa M. van Zelst M. Clout D. Stapert H. Dijksman F. Broer D. Wimberger-Friedl R. Real time quantitative amplification detection on a microarray: Towards high multiplex quantitative PCR Lab Chip 2012 12 1897 1902 10.1039/c2lc20740k 22473033 75. Yershov G. Barsky V. Belgovskiy A. Kirillov E. Kreindlin E. Ivanov I. Parinov S. Guschin D. Drobishev A. Dubiley S. Mirzabekov A. DNA analysis and diagnostics on oligonucleotide microchips Proc. Natl. Acad. Sci. USA 1996 93 4913 4918 8643503 76. Cooney C.G. Sipes D. Thakore N. Holmberg R. Belgrader P. A plastic, disposable microfluidic flow cell for coupled on-chip PCR and microarray detection of infectious agents Biomed. Microdevices 2012 14 45 53 10.1007/s10544-011-9584-9 21909803 77. Brammer T.L. Murray E.L. Fukuda K. Hall H.E. Klimov A. Cox N.J. Surveillance for influenza—United States, 1997–98, 1998–99, and 1999–00 seasons MMWR Surveillance Summaries 2002 51 1 10 78. Murphy B.R. Webster R.G. Orthomyxoviruses Virology 3rd Fields B.N. Knipe D.M. Howley P.M. Lippincott-Raven Press Philadelphia, PA, USA 1996 1337 1420 79. Webster R.G. Bean W.J. Gorman O.T. Chambers T.M. Kawaoka Y. Evolution and ecology of influenza A viruses Microbiol. Rev. 1992 56 152 179 1579108 80. Golova J.B. Chernov B.K. Perov A.N. Reynolds J. Linger Y.L. Kukhtin A. Chandler D.P. Non-volatile copolymer compositions for fabricating gel element microarrays Anal. Biochem. 2012 421 526 533 22033291 81. Sandgren A. Strong M. Muthukrishnan P. Weiner B.K. Church G.M. Murray M.B. Tuberculosis drug resistance mutation database PLoS Med. 2009 6 e1000002 82. WHO Global Tuberculosis Control 2011; Publication No. WHO/HTM/TB/2011.16 World Health Organization, WHO Press Geneva, Switzerland 2011 248 83. Sanchez J.A. Pierce K.E. Rice J.E. Wangh L.J. Linear-After-The-Exponential (LATE)–PCR: An advanced method of asymmetric PCR and its uses in quantitative real-time analysis Proc. Natl. Acad. Sci. USA 2004 101 1933 1938 14769930 84. Edman C.F. Mehta P. Press R. Spargo C.A. Walker G.T. Nerenberg M. Pathogen analysis and genetic predisposition testing using microelectronic arrays and isothermal amplification Amer. J. Inv. Med. 2000 48 93 101 85. Detter J.C. Jett J.M. Lucas S.M. Dalin E. Arellano A.R. Wang M. Nelson J.R. Chapman J. Lou Y. Rokhsar D. Hawkins T.L. Richardson P.M. Isothermal strand-displacement amplification applications for high-throughput genomics Genomics 2002 80 691 698 12523365 86. Hataoka Y. Zhang L. Mori Y. Tomita N. Notomi T. Baba Y. Analysis of specific gene by integration of isothermal amplification and electrophoresis on poly(methyl methacrylate) microchips Anal. Chem. 2004 76 3689 3693 10.1021/ac035032u 15228342 87. Parida M. Horioke K. Ishida H. Dash P.K. Saxena P. Jana A.M. Islam M.A. Inoue S. Hosaka N. Morita K. Rapid detection and differentiation of dengue virus serotypes by a real-time reverse transcription-loop-mediated isothermal amplification assay J. Clin. Microbiol. 2005 43 2895 2903 10.1128/JCM.43.6.2895-2903.2005 15956414 88. Takakura S. Tsuchiya S. Fujihara N. Kudo T. Iinuma Y. Mitarai S. Ichiyama S. Yasukawa K. Ishiguro T. Isothermal RNA sequence amplification method for rapid antituberculosis drug susceptibility testing of Mycobacterium tuberculosis J. Clin. Microbiol. 2005 43 2489 2491 15872291 89. Demidov V.V. Rolling-circle amplification in DNA diagnostics: the power of simplicity Expert Rev. Mol. Diagn. 2002 2 89 95

PMC005xxxxxx/PMC5003435.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2010024microarrays-02-00024CommunicationProfiling Pre-MicroRNA and Mature MicroRNA Expressions Using a Single Microarray and Avoiding Separate Sample Preparation Gan Lin Denecke Bernd *Interdisciplinary Centre for Clinical Research Aachen, RWTH Aachen University, Pauwelstr. 30, 52074 Aachen, Germany; E-Mail: lgan@ukaachen.de* Author to whom correspondence should be addressed; E-Mail: bernd.denecke@rwth-aachen.de; Tel.: +49-241-80-89918; Fax: +49-241-80-82124.14 3 2013 3 2013 2 1 24 33 31 1 2013 26 2 2013 12 3 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Mature microRNA is a crucial component in the gene expression regulation network. At the same time, microRNA gene expression and procession is regulated in a precise and collaborated way. Pre-microRNAs mediate products during the microRNA transcription process, they can provide hints of microRNA gene expression regulation or can serve as alternative biomarkers. To date, little effort has been devoted to pre-microRNA expression profiling. In this study, three human and three mouse microRNA profile data sets, based on the Affymetrix miRNA 2.0 array, have been re-analyzed for both mature and pre-microRNA signals as a primary test of parallel mature/pre-microRNA expression profiling on a single platform. The results not only demonstrated a glimpse of pre-microRNA expression in human and mouse, but also the relationship of microRNA expressions between pre- and mature forms. The study also showed a possible application of currently available microRNA microarrays in profiling pre-microRNA expression in a time and cost effective manner. mature microRNApre-microRNAprofilingmicroarray ==== Body 1. Introduction MicroRNAs (miRNA) are short conservative endogenous non-coding RNAs with a length of around 22 nucleotides, which have diverse functions [1]. Since its discovery in 1993 [2], miRNA has been recognized as one of the key players in the transcript regulatory network of eukaryotes [3]. Despite intensive scientific research, many miRNAs still need to be explored or validated in different biological or pathological contexts. 1.1. Pre-miRNA, Mature miRNA and miRNA Biogenesis miRNA genes are transcribed by RNA polymerase II (Pol II) [4] in the nucleus. This process yields long primary transcripts of miRNA (pri-miRNAs). Nuclear RNase III Drosha cleaves pri-miRNA into a precursor of miRNA (pre-miRNA) [5]. Pre-miRNAs are then exported from the nucleus into the cytoplasm by exportin-5 (Exp5) [6]. After further processing by cytoplasmic RNase III Dicer [7,8], mature miRNA is integrated into the miRNA-containing RNA-induced silencing complex (miRISC) [9,10]. miRNA transcription and maturation is a precise controlled and collaborated process [11]. 1.2. Microarray Application in miRNA Expression Profiling Diverse techniques are available for the purpose of miRNA expression profiling, e.g., Northern blotting [12], dot blotting [13], primer extension analysis [14], RT-qPCR [15] and next generation sequencing (NGS) [16]. RT-qPCR, microarray and NGS are the three most state-of-the-art methods applied in miRNA profiling [17]. RT-qPCR offers the most sensitive profiling signal with a reasonable cost. With a considerate design, RT-qPCR can also achieve high throughput miRNA profiling. The recent progress of NGS has led to rapid cost reduction, which lowered the impact of NGS application in miRNA profiling. Nonetheless, microarrays with immobilized DNA probes on solid substrate are still the choice of many laboratories with a moderate budget [18]. Hence, microarrays are widely applied also in miRNA profiling [19]. One of the challenges of miRNA expression profiling is to distinguish between mature miRNAs and their precursors and to detect the miRNA expression signal with high specificity. This problem can be solved with a precise and genuine probe design on microarray platforms. Much more attention has been dedicated to mature miRNA expression profiling, although pri-miRNA and pre-miRNA are important intermediates during miRNA biogenesis [20]. Pre-miRNAs have also been acknowledged as useful disease biomarkers [21,22]. RT-qPCR has been, so far, the gold standard for pre-miRNA expression evaluation [23]. Recently, a global pre-miRNA landscape has also been successfully provided by applying a deep sequencing technique [24]. Since pre-miRNAs and mature miRNAs can both be detected in cytoplasm, we demonstrated in this paper the possibility of exploring the expression of mature and pre-miRNAs on a single microarray platform, simultaneously. 2. Experimental Section Affymetrix GeneChip® miRNA 2.0 arrays (Affymetrix, Santa Clara, California, USA) contain probes that interrogate the mature miRNA and pre-miRNA of a wide spectrum of species. In this study, profiling data of human and mouse samples (GSE39015 [25], GSE42915 and GSE33809 [26] focused on human samples; GSE33413 [27], GSE32352 [28] and GSE36257 [29] profiled mouse samples) on Affymetrix GeneChip miRNA 2.0 arrays (see Table 1) were retrieved from the Gene Expression Omnibus (GEO) database [30]. Raw data were normalized with the robust multi-array average (RMA). Present calls were made by using the detection above background (DABG) algorithm with a p-value cutoff of 0.05. Pre-miRNA and mature miRNA probes were identified according to annotations provided by Affymetrix. A paired t-test was applied on the median expression of miRNAs and pre-miRNAs for significancy calculations (see Figure 1). microarrays-02-00024-t001_Table 1Table 1 Information of data sets involved in this paper. Study Species No. of samples Short Description of the Study RNA extraction method GSE34413 M. musculus 18 whole brain tissue from day 70, fetal alcohol exposed males and matched controls TRIzol (Invitrogen) GSE32352 M. musculus 6 EC/NSPC co-cultures were incubated with FGF/VEGF receptor inhibitor and TGF-receptor inhibitor mirVana miRNA (AB/Ambion) GSE36257 M. musculus 24 heart, quadriceps femoris and diaphragm from 8-week-old male WT C57/B10 and male mdx C57/B10. TRIzol (Invitrogen) GSE39015 H. sapiens 18 2 cervix cell lines and 16 clinical tumor samples stored in FFPE RecoverAll Total Nucleic Acid Isolation kit (Ambion) GSE42915 H. sapiens 12 12 placentas (6 from first trimester and 6 from third trimester) TRIzol (Invitrogen) GSE33809 H. sapiens 24 SET2 cells were incubated with increasing concentrations of INC424/Ruxolitinib for 3-6 h RNeasy Micro kit (Qiagen) EC, endothelial cell; NSPC, neuro stem/progenitor cell; FGF, fibroblast growth factor; VEGF, vascular endothelial growth factor; TGF, transforming growth factor; FFPE, formalin-fixed, paraffin-embedded. Figure 1 Distribution of normalized expression values of pre-miRNAs and mature miRNAs: shown are boxplots of miRNA expression values after normalization for all arrays in 3 human studies (left part) and 3 murine studies (right part). Yellow color represents pre-mature miRNA expression signals, and green color stands for mature miRNA expression signals. The ordinate shows transformed signal intensities in arbitrary units. Significance: n.s. = not significant; *** = significant, with a p-value < 0.001. The correlation of pre- and mature miRNA expressions were studied under two different aspects: (i) for study, i, (i = a GEO study re-analyzed in this manuscript: GSE39015, GSE42915, GSE33809, GSE33413, GSE32352 and GSE36257), the expression values of pre-miRNAs can be described as Vi,pre = {v1,pre; v2,pre; …; vj,pre|j = the array number in study i} and the expression values of mature miRNAs as Vi,mature = {v1,mature; v2,mature; …; vj,mature|j = the array number in study i}. Pearson correlation coefficients were pairwise calculated between Vi,pre and Vi,mature for all arrays in studies, i, and then presented as a bar plot (see Figure 4). (ii) The expression matrix, M, of pre-miRNA and mature miRNA in study, i, was defined as Mi,pre = {m1,pre; m2,pre; …; mk,pre|k = the number of pre-miRNAs on one array} and Mi,mature = {m1,mature; m2,mature; …; mk,mature|k = the number of mature miRNAs on one array}. Correlation between Mi,pre and Mi,mature was also calculated by pairwise Pearson correlation coefficients. The distribution of pairwise correlation coefficients of all miRNAs in every study were shown as a histogram (Figure 5). Statistical analysis were completed with basic functions in R [31]. 3. Results and Discussion GeneChip® miRNA 2.0 has 4,592 probe sets for human and 1,412 probe sets for mouse. Seven hundred sixty seven pre-miRNAs and 919 mature miRNAs are annotated for human samples. For the mouse samples, 510 pre-miRNAs and 601 mature miRNAs are available on the platform. Eight hundred thirty six pre-miRNA/mature-miRNA pairs can be mapped in human miRNA annotation, while there are 579 according to the mouse annotation. 3.1. Mature miRNAs Are More Abundantly Detected Pre-miRNAs showed lower expression in these data sets than their matured products. Boxplots of expression values of pre- and mature miRNAs in three human and three mouse studies, respectively, are shown (Figure 1). In all murine studies, median expressions of pre-miRNA are significantly lower than those of mature miRNA, wherein one of the studies exhibits a minor difference (GSE32352). For human samples, the differences are not so pronounced; however, they are also significant for two of the three studies analyzed (GSE42915, GSE33809). At the same time, the upper quartiles of mature miRNAs distribution are located more in a higher expression value region. It is possible that mature miRNA is more enriched by sample preparation. Besides, the high stability observed by mature miRNAs could also contribute to their higher expression signal [32]. Moreover, the rapid processing of the intermediate product pre-miRNA might have an impact on the fewer copy number of pre-miRNA compared to the stable accumulating signal of mature miRNA. The lower expression value of pre-miRNA compared to its corresponding mature miRNA was also reported in another study using the RT-qPCR method [33]. Independent to the variety of individual samples, three data sets that used the same RNA isolation kit (GSE34413, GSE36257 and GSE42915) showed a similar pattern in expression value distribution. This observation indicates that sample preparation, especially RNA isolation, is a crucial factor for successful miRNA profiling. 3.2. Much More Mature miRNAs than Pre-miRNA Were Detected as Present The general higher expression level of mature miRNA was also reflected in the detection call of present miRNAs. Present calls were made for pre-miRNA and mature miRNAs compared to background signals. Mature miRNAs have almost doubled present calls with respect to pre-miRNAs in all data sets (Figure 2). Most of the pre-miRNAs and mature miRNAs share the same present/absent calls. That means, present pre-miRNAs have present mature counterparts, and absent pre-miRNAs have also absent mature corresponding miRNAs (Figure 3, red bars). However, there are also present pre-miRNAs, which have absent mature miRNA, and vice versa (Figure 3, blue and black bars). miRNAs that were detected as present in mature form, but as absent in pre-miRNA form, could support the hypothesis that mature miRNAs are more stable than their precursor counterparts. As demonstrated, there was a few number of miRNAs that were present in their premature form, but absent in their mature form. This phenomenon indicates a possible miRNA transcription regulation by degrading mature miRNA [34]. The reduction of specific mature miRNAs compared to their pre-miRNAs was also reported in human colorectal neoplasia by Northern analysis [35]. Figure 2 Number of pre-/mature miRNAs with present call in human (left panel) and murine studies (right panel): number of pre-miRNA and mature miRNA with present call on individual arrays in human and murine data sets. Dashed lines represent numbers of pre-miRNAs with a present call. Solid lines represent number of mature miRNAs with a present call. Colors of the lines distinguish different studies. Numbers on the abscissa indicate the numbers of arrays involved in corresponding data set. Figure 3 Number of miRNAs in three categories: average number of miRNAs with the standard deviation in the category, (i) pre-miRNA is called as present, while corresponded mature miRNA is called absent (blue bars), (ii) pre-miRNA and mature miRNA are called both present or absent (red bars) and (iii) pre-miRNA is called as absent, but mature miRNA is called present (black bars). 3.3. Correlation between Pre-miRNA Expression and Mature miRNA Expression The expression of mature miRNA and pre-miRNA measured on the same arrays showed a positive interdependency (Figure 4). Notwithstanding different expression levels, a positive correlation did exist between expression of pre- and mature miRNAs on the same array. This is consistent with the results reported previously [36]. Correlation coefficients of pre- and mature miRNA on the same array in one data set were consistent, while the values clearly varied between different data sets. For example, GSE32352 demonstrated a much lower correlation coefficient between pre- and mature miRNA expression. GSE34413, GSE36257 and GSE42915 showed similar correlation coefficients. While in the studies, GSE34413, GSE36257 and GSE42915, the same RNA isolation kit (TRIzol from Invitrogen) was used; in the GSE32352 study, the mirVana RNA isolation kit (Ambion) was used. For this reason, we believe that the observed correlation coefficients were strongly affected by the choice of sample preparation. The influence of the kind of probe labeling can be excluded, because all probes were labeled with the same kit (FlashTag biotin HSR kit from Genisphere). Figure 4 Correlation of pre-/mature miRNA expression: shown are the correlations of pre-miRNA and mature miRNA expression on the same arrays of murine (upper part) and human (lower part) data sets. The abscissa indicates correlation coefficients. In the ordinate axis, the number of arrays involved in the corresponding study are shown. Correlation coefficients were also calculated for the expression profile of every mature miRNA to its corresponding pre-miRNA across the arrays in the data sets. These coefficients support the evidence of the dependency between pre-miRNA and mature miRNA expression regulation, which differs according to different tissue sources and/or treatments. Correlation coefficients for all miRNAs in each study have been calculated between the pre-miRNA expression pattern and the mature miRNA expression pattern (Figure 5). In all data sets analyzed in this paper, correlation coefficients were widely distributed between −1 and 1. In particular, not only positive coefficients were observed, but also extreme negative ones. That means that upregulation of mature miRNA can be observed in a study even when the corresponding pre-miRNA is downregulated. This result indicates that during maturation of pre-miRNA to mature miRNA, complex factors are involved in regulation, depending on the kind of treatment, tissue-specific development or different biological/pathological contexts. Figure 5 The correlation of pre-/mature miRNA expression according to expression pattern: histograms of correlation coefficients of all pre-miRNA and its corresponding mature miRNA expression patterns in three human (upper part with light blue bars) and three murine (lower part with blue bars) studies. Abscissae indicate correlation coefficients. 4. Conclusions Six publicly available miRNA profiling raw data sets generated from human and mouse samples on Affymetrix GeneChip® miRNA 2.0 array platforms were re-analyzed in this study. Pre-miRNA and mature miRNA expression signals were retrieved and normalized to gain comparable signals. miRNA precursors exhibited a lower expression level in most of the analyzed data sets. Present calls of individual mature miRNAs and pre-miRNAs showed a great part consistency, with minor discrepancies. On the same array, mature miRNA expressions are positively correlated to pre-miRNA expressions. The distribution of expression values of pre-miRNA and mature miRNA, as well as the correlation coefficients between them, seem to be influenced by the RNA isolation methods applied in the studies. Because of the sample size limitation and the lack of other experimental verifications, we are not able to judge the pros and cons of RNA isolation methods for miRNA profiling at this point. The expression regulation patterns of mature miRNAs showed no clear positive correlation to the expression regulation patterns of the corresponding pre-miRNAs. Therefore, we believe in the existence of regulation factors of miRNA maturation according to treatment, tissue type and biological/pathological contexts involved in individual studies. The results presented in this paper require definitely further research and verification with more data sets and with other techniques, like RT-qPCR. Nevertheless, our study demonstrated the possibility of profiling pre-miRNA and mature miRNA simultaneously without separate sample preparation. Compared to methods like RT-qPCR, the microarray platform applied in this paper has the advantage of data normalization without the bias caused by specific reference RNA(s). At the same time, we can obtain a genome-wide overview for pre-miRNA and miRNA expression. An additional advantage of this platform is that only one simple labeling procedure is sufficient for both pre-miRNA and mature miRNA profiling. Acknowledgements The authors thank Arian Aryani Kashani for careful revision of the manuscript. Conflict of Interest The authors declare no conflict of interest. ==== Refs References 1. Bushati N. Cohen S.M. MicroRNA functions Annu. Rev. Cell Dev. Biol. 2007 23 175 205 10.1146/annurev.cellbio.23.090506.123406 17506695 2. Wightman B. Ha I. Ruvkun G. Posttranscriptional regulation of the heterochronic gene lin-14 by lin-4 mediates temporal pattern formation in C. elegans Cell 1993 75 855 862 10.1016/0092-8674(93)90530-4 8252622 3. He L. Hannon G.J. MicroRNAs: Small RNAs with a big role in gene regulation Nat. Rev. Genet. 2004 5 522 531 10.1038/nrg1379 15211354 4. Lee Y. Kim M. Han J. Yeom K.H. Lee S. Baek S.H. Kim V.N. MicroRNA genes are transcribed by RNA polymerase II Embo. J. 2004 23 4051 4060 10.1038/sj.emboj.7600385 15372072 5. Lee Y. Ahn C. Han J. Choi H. Kim J. Yim J. Lee J. Provost P. Radmark O. Kim S. Kim V.N. The nuclear RNase III Drosha initiates microRNA processing Nature 2003 425 415 419 10.1038/nature01957 14508493 6. Murchison E.P. Hannon G.J. miRNAs on the move: miRNA biogenesis and the RNAi machinery Curr. Opin. Cell Biol. 2004 16 223 229 10.1016/j.ceb.2004.04.003 15145345 7. Bernstein E. Caudy A.A. Hammond S.M. Hannon G.J. Role for a bidentate ribonuclease in the initiation step of RNA interference Nature 2001 409 363 366 11201747 8. Hutvagner G. McLachlan J. Pasquinelli A.E. Balint E. Tuschl T. Zamore P.D. A cellular function for the RNA-interference enzyme Dicer in the maturation of the let-7 small temporal RNA Science 2001 293 834 838 10.1126/science.1062961 11452083 9. Lelandais-Briere C. Sorin C. Declerck M. Benslimane A. Crespi M. Hartmann C. Small RNA diversity in plants and its impact in development Curr. Genomics 2010 11 14 23 20808519 10. Bartel D.P. MicroRNAs: Target recognition and regulatory functions Cell 2009 136 215 233 10.1016/j.cell.2009.01.002 19167326 11. Shukla G.C. Singh J. Barik S. MicroRNAs: Processing, maturation, target recognition and regulatory functions Mol. Cell. Pharmacol. 2011 3 83 92 22468167 12. Varallyay E. Burgyan J. Havelda Z. MicroRNA detection by northern blotting using locked nucleic acid probes Nat. Protoc. 2008 3 190 196 10.1038/nprot.2007.528 18274520 13. Porkka K.P. Pfeiffer M.J. Waltering K.K. Vessella R.L. Tammela T.L. Visakorpi T. MicroRNA expression profiling in prostate cancer Cancer Res. 2007 67 6130 6135 10.1158/0008-5472.CAN-07-0533 17616669 14. Raymond C.K. Roberts B.S. Garrett-Engele P. Lim L.P. Johnson J.M. Simple, quantitative primer-extension PCR assay for direct monitoring of microRNAs and short-interfering RNAs RNA 2005 11 1737 1744 10.1261/rna.2148705 16244135 15. Fiedler S.D. Carletti M.Z. Christenson L.K. Quantitative RT-PCR methods for mature microRNA expression analysis Methods Mol. Biol. 2010 630 49 64 10.1007/978-1-60761-629-0_4 20300990 16. Pritchard C.C. Cheng H.H. Tewari M. MicroRNA profiling: Approaches and considerations Nat. Rev. Genet. 2012 13 358 369 10.1038/nrg3198 22510765 17. Baker M. MicroRNA profiling: Separating signal from noise Nat. Methods 2010 7 687 692 10.1038/nmeth0910-687 20805796 18. Schena M. Shalon D. Davis R.W. Brown P.O. Quantitative monitoring of gene expression patterns with a complementary DNA microarray Science 1995 270 467 470 7569999 19. Liu C.G. Calin G.A. Volinia S. Croce C.M. MicroRNA expression profiling using microarrays Nat. Protoc. 2008 3 563 578 18388938 20. Viswanathan S.R. Daley G.Q. Gregory R.I. Selective blockade of microRNA processing by Lin28 Science 2008 320 97 100 10.1126/science.1154040 18292307 21. O’Hara A.J. Chugh P. Wang L. Netto E.M. Luz E. Harrington W.J. Dezube B.J. Damania B. Dittmer D.P. Pre-micro RNA signatures delineate stages of endothelial cell transformation in Kaposi sarcoma PLoS Pathog. 2009 5 10.1371/journal.ppat.1000389 22. O’Hara A.J. Vahrson W. Dittmer D.P. Gene alteration and precursor and mature microRNA transcription changes contribute to the miRNA signature of primary effusion lymphoma Blood 2008 111 2347 2353 10.1182/blood-2007-08-104463 18079361 23. Chugh P. Tamburro K. Dittmer D.P. Profiling of pre-micro RNAs and microRNAs using quantitative real-time PCR (qPCR) arrays J. Vis. Exp. 2010 10.3791/2210 24. Burroughs A.M. Kawano M. Ando Y. Daub C.O. Hayashizaki Y. Pre-miRNA profiles obtained through application of locked nucleic acids and deep sequencing reveals complex 5'/3' arm variation, including concomitant cleavage and polyuridylation patterns Nucleic Acids Res. 2012 40 1424 1437 10.1093/nar/gkr903 22058130 25. Hall J.S. Taylor J. Valentine H.R. Irlam J.J. Eustace A. Hoskin P.J. Miller C.J. West C.M. Enhanced stability of microRNA expression facilitates classification of FFPE tumour samples exhibiting near total mRNA degradation Br. J. Cancer 2012 107 684 694 10.1038/bjc.2012.294 22805332 26. Bortoluzzi S. Bisognin A. Biasiolo M. Guglielmelli P. Biamonte F. Norfo R. Manfredini R. Vannucchi A.M. Characterization and discovery of novel miRNAs and moRNAs in JAK2V617F-mutated SET2 cells Blood 2012 119 e120 e130 10.1182/blood-2011-07-368001 22223824 27. Kleiber M.L. Laufer B.I. Wright E. Diehl E.J. Singh S.M. Long-term alterations to the brain transcriptome in a maternal voluntary consumption model of fetal alcohol spectrum disorders Brain Res. 2012 1458 18 33 10.1016/j.brainres.2012.04.016 22560501 28. Roitbak T. Bragina O. Padilla J.L. Pickett G.G. The role of microRNAs in neural stem cell-supported endothelial morphogenesis Vasc. Cell 2011 10.1186/2045-824X-3-25 29. Roberts T.C. Blomberg K.E. McClorey G. Andaloussi S.E. Godfrey C. Betts C. Coursindel T. Gait M.J. Edvard Smith C. Wood M.J. Expression analysis in multiple muscle groups and serum reveals complexity in the microrna transcriptome of the mdx mouse with implications for therapy Mol. Ther. Nucleic Acids 2012 1 10.1038/mtna.2012.26 30. Barrett T. Wilhite S.E. Ledoux P. Evangelista C. Kim I.F. Tomashevsky M. Marshall K.A. Phillippy K.H. Sherman P.M. Holko M. Yefanov A. Lee H. Zhang N. Robertson C.L. Serova N. Davis S. Soboleva A. NCBI GEO: Archive for functional genomics data sets—Update Nucleic Acids Res. 2013 41 D991 D995 10.1093/nar/gks1193 23193258 31. Team R.D.C. R: A Language and Enviroment for Statistical Computing R Foundation for Statistical Computing Vienna, Austria 2010 32. Bail S. Swerdel M. Liu H. Jiao X. Goff L.A. Hart R.P. Kiledjian M. Differential regulation of microRNA stability RNA 2010 16 1032 1039 10.1261/rna.1851510 20348442 33. Schmittgen T.D. Lee E.J. Jiang J. Sarkar A. Yang L. Elton T.S. Chen C. Real-time PCR quantification of precursor and mature microRNA Methods 2008 44 31 38 10.1016/j.ymeth.2007.09.006 18158130 34. Kai Z.S. Pasquinelli A.E. MicroRNA assassins: Factors that regulate the disappearance of miRNAs Nat. Struct. Mol. Biol. 2010 17 5 10 10.1038/nsmb.1762 20051982 35. Michael M.Z. SM O.C. van Holst Pellekaan N.G. Young G.P. James R.J. Reduced accumulation of specific microRNAs in colorectal neoplasia Mol. Cancer Res. 2003 1 882 891 14573789 36. Jiang J. Lee E.J. Gusev Y. Schmittgen T.D. Real-time expression profiling of microRNA precursors in human cancer cell lines Nucleic Acids Res. 2005 33 5394 5403 10.1093/nar/gki863 16192569

PMC005xxxxxx/PMC5003436.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2010001microarrays-02-00001ArticleTesting a Microarray to Detect and Monitor Toxic Microalgae in Arcachon Bay in France Kegel Jessica U. 1*Del Amo Yolanda 2Costes Laurence 2Medlin Linda K. 1*1 Marine Biological Association of the United Kingdom, The Laboratory, Citadel Hill, Plymouth, PL1 2PB, UK2 Université Bordeaux 1, UMR CNRS EPOC 5805, Station Marine d’Arcachon, 2 rue du Prof. Jolyet F 33120 Arcachon, France; E-Mails: y.delamo@epoc.u-bordeaux1.fr (Y.D.A.); l.costes@epoc.u-bordeaux1.fr (L.C.)* Authors to whom correspondence should be addressed; E-Mails: kegel@obs-banyuls.fr (J.U.K.); lkm@mba.ac.uk (L.K.M.); Tel.: +44-1752-633207 (L.K.M.); Fax: +44-1752-633102 (L.K.M.).05 3 2013 3 2013 2 1 1 23 27 11 2012 24 1 2013 26 2 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Harmful algal blooms (HABs) occur worldwide, causing health problems and economic damages to fisheries and tourism. Monitoring agencies are therefore essential, yet monitoring is based only on time-consuming light microscopy, a level at which a correct identification can be limited by insufficient morphological characters. The project MIDTAL (Microarray Detection of Toxic Algae)—an FP7-funded EU project—used rRNA genes (SSU and LSU) as a target on microarrays to identify toxic species. Furthermore, toxins were detected with a newly developed multiplex optical Surface Plasmon Resonance biosensor (Multi SPR) and compared with an enzyme-linked immunosorbent assay (ELISA). In this study, we demonstrate the latest generation of MIDTAL microarrays (version 3) and show the correlation between cell counts, detected toxin and microarray signals from field samples taken in Arcachon Bay in France in 2011. The MIDTAL microarray always detected more potentially toxic species than those detected by microscopic counts. The toxin detection was even more sensitive than both methods. Because of the universal nature of both toxin and species microarrays, they can be used to detect invasive species. Nevertheless, the MIDTAL microarray is not completely universal: first, because not all toxic species are on the chip, and second, because invasive species, such as Ostreopsis, already influence European coasts. oligonucleotide microarraysmolecular monitoringharmful algal bloomsHABstoxic microalgae18S/28S ribosomal RNALSU/SSURNA hybridizationenvironmental water samples ==== Body 1. Introduction Worldwide every year, fisheries, aquaculture, human health and tourism are threatened by harmful algal blooms (HABs) in marine, brackish, as well as continental waters. Although most phytoplankton species are benign, about 2% of them can cause harm through the production of toxins or by an excessive accumulated biomass, which can affect co-occurring organisms and alter food-web dynamics [1,2]. In addition to the ecological and economic damages, public health is also at risk: the consumption of shellfish that have fed on toxic phytoplankton and accumulated toxins, and exposure to the aerosols of HAB toxins can cause illness or even mortality. Depending on the species, it can take only a few toxic cells per liter to poison shellfish and make them unsuitable for human consumption [3]. Monitoring of microalgae is therefore required by all countries with a marine coastline. HAB monitoring programs are currently based on cells identified and counted by light microscopy and on the mouse bioassay for detecting biotoxins. The mouse bioassay for the detection of phytoplankton toxins in shellfish has recently been banned by the European Commission (July 2011), and there is a mandatory replacement by chemical methods (Liquid Chromatography-Mass Spectrometry, LC-MS) in the next three years. However, the effectiveness of monitoring programs using light microscopic identification is limited by the fact that it is time consuming and that morphology, as determined by light microscopy, may be insufficient to give definitive species and toxin attribution. Thus, there is a need to implement molecular methods to ensure a fast and reliable species identification. Monitoring for toxic species using molecular techniques advances the state of knowledge for detection of harmful species because more samples can be analyzed in a shorter time period and with greater accuracy. Within the actual context of the dramatic decreasing number of taxonomic experts of phytoplankton [4], which are, notwithstanding, essential for other ecological studies, such techniques offer important advances. This is of particular interest for potentially toxic algae, because the difficulty in determining their exact identification by using light microscopy can have disastrous consequences for human health. Microarrays offer a near real-time ecosystem analysis and offer broader ecological interpretation of how key species, such as toxic algae, can extend their geographical distribution with climate change or can become invasive after introduction from remote areas [5]. Microarrays offer the most expeditious method to have high sample throughput with highly accurate species detection, in a universal approach [6,7,8]. In the FP7 EU project, MIDTAL (Microarrays for the Detection of Toxic Algae), an earlier protocol for detection of toxigenic microalgae by Gescher et al. [9] was optimized. Microarrays (as phylochips) detect multiple species simultaneously using species-specific probes that have been applied primarily for the detection of bacteria [10,11,12,13]. At present, 140 probes for various toxic algal species at various taxonomic levels are spotted onto the current generation of the MIDTAL microarray. As part of the MIDTAL project, the primary goal was to be able to infer cell numbers from the molecular signal to provide an early warning system for toxic algae. Because the MIDTAL microarray is a universal array that can be used globally, it offers a real possibility of detecting invasive species, especially in view of global warming where warm temperate species are moving northward, e.g., Gamberiodiscus. In this study, we show the effectiveness of using microarrays for the detection of toxic algae and its combination with toxin detection. We compare these results with light microscopy data from a regular French monitoring network of toxic phytoplankton. The microarray used in this study represents the third generation array developed within the EU-MIDTAL project. In generation one, probes (18–22 nt) developed for Fluorescent in situ Hybridization (FISH) were used directly; in generation two, these FISH probes, and any newly designed probes, were lengthened to 25 or more nt; in generation three, an additional poly-T spacer to lift the probes farther above the surface was tested and optimized (Figure 1). At each generation, minor changes in the hybridization protocol were made and a final optimized protocol can be found in Lewis et al. [14]. Figure 1 Scheme of the development of the MIDTAL microarray. The scheme pictures the different microarray generations with its different probes, tests and enhancements of protocols (RNA and hybridization). (* Higher temperature during 3rd washing step). 2. Experimental Section 2.1. Field Sampling In 2011, water samples from the sub-surface (1 m depth) were collected at Arcachon Bay in France (Figure 2) between July and October for microarray analysis (Table 1). The sampling site termed Tès (1°10'00 W, 44°40'00 N), is directly located in front of the town of Arcachon inside the bay. Data of toxic, harmful, and other phytoplankton abundances is provided by IFREMER (Ifremer/Quadrige2/Rephy DATA) from the paired station named Teychan (1.5 km from Tès). Cell counts were done as previously described by Medlin and Schmidt [15] and Kegel et al. [16]. microarrays-02-00001-t001_Table 1Table 1 Information about field samples taken at Arcachon Bay like sample name, sample date, filtered volume, total extracted RNA and degree of labeling (DoL). Sample name Sample date Volume filtered (L) Total RNA extracted (µg) DoL 1A 24/07/2011 3.3 7.46 2.2 2A 08/08/2011 3.0 9.48 2.0 3A 22/08/2011 3.3 9.52 1.9 4A 04/10/2011 3.25 10.66 2.2 6A 20/10/2011 3.3 13.82 2.2 Figure 2 Sampling sites in Arcachon Bay (France): the station Tès (Teychan). For the microarray analysis, a minimum of three liters (Table 1) were filtered onto 3 µm nitrocellulose filters (47 mm) in triplicate. For each sampling date, the first and second replicated filter was transferred into cryogenic vials containing 1 mL of TRI Reagent (Sigma-Aldrich). Those samples were snap frozen and stored at –80 °C until further process for RNA extraction. Toxicity was measured by one of the Partners (Queens University Belfast, UK) with a newly developed multiplex optical Surface Plasmon Resonance biosensor (Multi SPR) in parallel with the enzyme-linked immunosorbent assay (ELISA) [17]. The target toxins are domoic acid (DA) for amnesic shellfish poisoning (ASP), okadaic acid (OA) and dinophysistoxins (DTXs) for diarrhetic shellfish poisoning (DSP) and saxitoxin (STX) for paralytic shellfish poisoning (PSP) toxin analogs. Therefore, the third replicated filter was transferred into cryogenic vials without TRI Reagent and sent frozen to Queens University Belfast who was responsible for the toxin measurements. 2.2. RNA Extraction RNA extraction was done with minor changes to that presented in Kegel et al. [16]. Briefly, acid-washed glass beads (300 µm) and 500,000 cells of Dunaliella tertiolecta-strain UIO226 (stored in TRI Reagent) as a control were added to the samples and the samples were bead-beaten twice for 1 min at 4,800 oscillations/min (BioSpec Mini Bead Beater). Cell-TRI Reagent mixture was transferred into a new microcentrifuge tube, vortexed for 15 s and left to stand at room temperature (RT) for 10 min. After another 15 s of vortexing, samples were incubated at 60 °C for 10 min in a Thermoshaker at maximum speed. Samples were vortexed again for 15 s and then transferred into pre-spun Phase Lock Gel Heavy 2 mL tubes (5 Prime; 12,000 g for 30 s). After the addition of 100 µL of 1-bromo-3-chloropropane (BCP) to the samples, the tubes were shaken thoroughly for 15 s. Samples were incubated at RT for 5 min and centrifuged (12,000 ×g) for 15 min at 4 °C. The upper phase was mixed gently with 200 µL chloroform and centrifuged (12,000 ×g) for 2 min at 4 °C. The aqueous phase was then transferred to a fresh 2 mL RNase-free tube. Equal volumes of isopropanol were added, vortexed for 15 s and incubated for one hour at −20 °C. After incubation, samples were centrifuged (12,000 ×g) for 15 min at 4 °C. Supernatant was quickly removed and pellets were washed three times with 1 mL ethanol (75%): ethanol was added, vortexed for 5 s, centrifuged (12,000 ×g) for 10 min at 4 °C, and supernatant carefully removed. Following the third wash, the supernatant was completely removed and the pellet was air-dried for 5 min. The pellet was dissolved in 100 µL of RNase-free water. To get rid of TRI Reagent residuals, samples were precipitated with 0.5 volume of 7.5 M NH4Ac and 2 volumes of ice-cold ethanol (absolute, stored at −20 °C). The mixture was vortexed and incubated at −80 °C for 1.5 h. Immediately after incubation, samples were centrifuged at 4 °C and max. speed for 20 min. The supernatant was removed; the pellet was washed in 500 µL of 70% ice-cold ethanol (stored at −20 °C) and centrifuged for 5 min at max. speed. The washing step was repeated and the pellet was air-dried for 30–60 min. The RNA was re-suspended in 50 µL nuclease-free water and its concentration and integrity was measured by NanoVue spectrophotometer (GE Healthcare) and Agilent Bioanalyzer 2100 (Agilent Biotechnologies). Samples were snap-frozen in liquid nitrogen and stored at −80 °C until further use. 2.3. RNA Labeling and Fragmentation The PlatinumBright Infrared Labeling Kit from KREATECH (Amsterdam, Netherlands) was used to label 1.5 µg RNA of field sample using 2 µL ULS dye and 2 µL 10× labeling solution in a total volume of 20 µL. Samples were labeled by incubation for 30 min at 85 °C. After incubation, samples were placed on ice and spun down and then purified with KREApure columns (KREATECH) according to the manufacturer’s instructions. Concentration and incorporation of the dye was measured by a NanoVue (GE Healthcare). The DoL (degree of labeling) was calculated and was between 1.9 and 2.2% (Table 1). RNA was fragmented by adding 1/10 volume of RNA fragmentation buffer (100 mM ZnCl2 in 100 mM Tris-HCL, pH 7.0) and an incubation of 15 min at 70 °C. The reaction was stopped with the addition of 1/10 volume of 0.5 M EDTA (pH 8.0) and the samples were placed on ice. The RNA was fragmented to reduce the effect of the secondary structure on the accessibility of the probe. Despite this fragmentation, we still have heterogeneous probe sensitivity, which reflects the influence of the secondary structure and we can only partially overcome this by fragmenting the RNA to remove the strongest secondary structure formations. 2.4. Microarray Design Probe design was done with the open software package ARB [18]. All oligonucleotides including the positive and negative controls were synthesized by Thermo Fisher Scientific (Ulm, Germany) with a C6 aminolink at the 5' end of the molecule. The probes had a length between 18 and 25 nt and a 15 nT-long poly (dT) tail following the NH2 link at the 5' end. Table 2 shows a list of the probes and their targets. The complete hierarchy for each probe can be found in the GPR-Analyzer which is available online at http://folk.uio.no/edvardse/gpranalyzer. The probe sequences are patent pending and a commercial kit will soon be available from Kreatech containing the array and all reagents for hybridization. Epoxy-coated slides (Genetix or Schott) of MIDTAL version 3.2 were printed using a pin printer VersArray ChipWriter Pro (Bio-Rad Laboratories GmbH, Munich, Germany) and split pins (Point Technologies, Inc., CO) as described by Kegel et al. [16]. One array contained 136 different probes and 4–8 replicates, as well as three negative (NEGATIVE1_dT, NEGATIVE2_dT, NEGATIVE3_dT), one positive control (TBP = TATA-box binding protein), Poly-T-Cy5 (spotting control), and two internal controls (DunGS02_25_dT and DunGS05_25_dT for Dunaliella tertiolecta) (MIDTAL ver3.2). After spotting, slides were incubated for 30 min at 37 °C and then stored at −20 °C. Printing of MIDTAL slides version 3.3 was done by Scienion AG using a sciFlexarrayer S11 and epoxy-coated slides from Genetix. One array contained eight replicates of 140 different probes including the seven controls stated above. After printing, the slides were transferred to a 75% humidity chamber, kept there overnight at RT, and stored afterwards in a sealed aluminum bag refilled with argon at 4 °C. microarrays-02-00001-t002_Table 2Table 2 Summary of probes designed or modified from published FISH probes and used to form the third generation of the MIDTAL microarray, including the targeted species, and whether it was made from the 18S or 28S rRNA gene. Probe sequences are not provided because the microarray is patent pending and will soon be commercially available from Kreatech, Amsterdam, The Netherlands. A complete taxonomic ordering of the probes can be seen in the GPR-Analyzer program and the MIDTAL hierarchy file that comes with that program. Probe Name Targeted Taxon Gene Controls DunGS02_25_dT Dunaliella spp. 18S DunGS05_25_dT Dunaliella spp. 18S Higher group-level probes EukS_328_25_dT Eukaryotes 18S EukS_1209_25_dT Eukaryotes 18S HeteroS01_25_dT Heterokonta 18S PrymS01_25_dT Prymnesiophyta 18S Class-level probes PrymS03_25_dT Prymnesiophyceae 18S DinoB_25_dT Dinophyceae (incl. Apicomplexa) 18S DinoE12_25_dT Dinophyceae (incl. Apicomplexa) 18S Clade-level probes DphyexacutaFS01_25_dT Dinophysiaceae (Dinophysis + Phalacroma) 18S DphyFS02_25_dT Dinophysiaceae (Dinophysis + Phalacroma) 18S PdeliD02_25_dT P. delicatissima all clades 28S Clade 01new_25_dT Prymnesium B1 clade 18S Clade01old_25_dT Prymnesium 18S ProroPKD01_25_dT Prorocentrum planktonic clade 28S ProroFPS01 Prorocentrum planktonic clade 18S ProroFBS02_25_dT Prorocentrum benthic clade 18S ProroFBS01 Prorocentrum benthic clade 18S Genus-level probes PsnGS01_25_dT Pseudo-nitzschia 18S PsnGS02_25_dT Pseudo-nitzschia + Fragilariopsis 18S PSN+FRAGS02-25new_dT Pseudo-nitzschia + Fragilariopsis 18S PSN no pungens_25_dT Pseudo-nitzschia no pungens 18S PSN + some Frags_25_dT Pseudo-nitzschia + some Fragilariopsis 18S KareGD01_25_dT Karenia 28S AlexGD01_25_dT Alexandrium 28S DphyGD01_25_dT Dinophysis in part 28S DphyGD02_25_dT Dinophysis 28S PschGS01_25_dT Pseudochattonella (genus) 18S PschGS04_25_dT Pseudochattonella (genus) 18S PschG05_25_dT Pseudochattonella (genus) 18S DphyGS01_25_dT Dinophysis genus sensu stricto 18S DphyGS02_25_dT Dinophysis genus sensu stricto 18S DphyGS03_25_dT all Dinophysis and Phalacroma 18S DphyGS04_25_dT all Dinophysis 18s KargeD01_25_dT Karlodinium genus 28S AzaGD01_dT Azadinium genus 28S AzaGD03_dT Azadinium genus 28S AzaGS01_dT Azadinium genus 18S AzaGS02_dT Azadinium genus 18S Species-level probes AtamaS01_25_dT Alexandrium NA,WE,TA, species complex 18S AminuS01_25_dT Alexandrium minutum 18S ATNA_D01_25_dT A. tamarense (North America) 28S ATNA_D02_25_dT A. tamarense (North America) 28S ATTA _D01_25_dT A. tamarense (Temperate Asian) 28S AostD01 _25_dT A. ostenfeldii 28S AostS02 _25_dT A. ostenfeldii 18S CpolyS01_25_dT Chrysochromulina polylepis 18S PparvD01_25_dT Prymnesium parvum 28S Prymparv01_25_dT Prymnesium parvum 18S KbreD03_25_dT Karenia mikimotoi and brevis 28S KbreD04_25_dT K. mikimotoi and brevis 28S KmikiD01_25_dT K. mikimotoi 28S KbreD05_25 K. brevis 28S LSKbre0548A25_dT K. mikimotoi and brevis 28S KmGcS06_25_dT K. mikimotoi, Gymnodinium catenatum, cf. Chatonella sp. 18S KbreD03c_25_dT Competitor K. mikimotoi and brevis 18S KbreD04_25c_dT Competitor K. mikimotoi and brevis 28S SSKbre1448A25_dT K. brevis 18S SSKbre1448A25c_dT K. brevis 18S LSKBre0548A25c_dT K. brevis 28S SSGcat0826A27_dT Gymnodinium catenatum 18S LSGcat0270A24_dT G. catenatum 28S GcateS01_25_dT G. catenatum 18S KveneD01_25_dT Karlodinium veneficum 28S KveneD02_25_dT Karlodinium veneficum 28S KveneD03_25_dT Karlodinium veneficum 28S KveneD04_25_dT Karlodinium veneficum 28S KveneD05_25_dT Karlodinium veneficum 28S KveneD06_25_dT Karlodinium veneficum 28 PlimaS01_25_dT Prorocentrum lima 18S PlimaFD01_2_dT5 P. lima 28S PmicaD02_25_dT P. micans 28S PminiD01_25_dT P. minimum 28S PmacuS01 P. maculosum and belizeanum 18S PmacuD01 P. maculosum 28S PmacuD02 P. maculosum 28S PrathD01 P. rathymum and mexicanum 28S PrathD02 P. rathymum and mexicanum 28S DacumiD02_25_dT Dinophysis acuminata, dens and sacculus 28S DacutaD02_25_dT Dinophysis acuta and fortii 28S DacumiS01_25_dT Dinophysis acuminata 18S DacutaS01_25_dT Dinophysis acuta 18S DnorvS01_25_dT Dinophysis norvegica 18S PausserD01_25_dT Pseudo-nitzschia australis and seriata 28S PmulausD01_25_dT P. australis and multistriata 28S PcaserausD02_25_dT P. australis, seriata, deli2 28S PcaserausD03_25_dT P. australis, seriata, calliantha 28S PfraucalD02_25_dT P. fraudulenta, subfraudulenta, calliantha 28S PcaciD01_25_dT P. caciantha 28S PcaciD02_25_dT P. caciantha 28S PcaciD04_25_dT P. caciantha 28S Pcal1D01_25_dT P. calliantha 28S PmanD01_25_dT P. manii 28S Pman2D02_25_dT P. manii 28S Pman2D03_25_dT P. manii 28S Pman2D05_25_dT P. manii 28S Pdel4D01_25_dT P. cf. delicatissima Clade4 28S Pdel4D02_25_dT P. cf. delicatissima Clade4 28S Pdel3B_25_dT P. delicatissima clade 3 + micropora 28S Pdel3A_25_dT P. delicatissima clade 3 + micropora 28S CompPdel3_25_dT Competitor Pdel3A 28S Pdel1D01_25_dT P. delicatissima Clade1 28S Pcaldel2D01_25_dT P. delicatissima Clade2 28S PcaldelD03_25_dT P. delicatissima Clade2 and calliantha 28S Pdel4D03_25_dT P. delicatissima Clade4 28S PgalaD01_25_dT P. galaxiae 28S PgalaD02_25_dT P. galaxiae 28S PgalaD04_25_dT P. galaxiae 28S PmultS01_25_dT P. multiseries 18S PmultD02_25_dT P. multiseries 28S PmultcalD01_25_dT P. multiseries and calliantha 28S PmultcalD03_25_dT P. multiseries and calliantha 28S PmultcalD04_25_dT P. multiseries and calliantha 28S PcalfrauD04_25_dT P. fraudulenta and multistriata 28S PmulaD03_25_dT P. multistriata 28S PmulacalD02_25_dT P. multistriata and calliantha 28S PpdeD01_25_dT P. pseudodelicatissima and cuspidata 28S PpdeD02_25_dT P. pseudodelicatissima and cuspidata 28S PpungcalS01_25_dT P. pungens and calliantha 18S PpungcalD02_25_dT P. pungens and calliantha 28S PpungcalD04_25_dT P. pungens and calliantha 28S PsercalD01_25_dT P. seriata and calliantha 28S CtoxS05_25_dT cf. Chatonella sp. 18S CtoxiS07_25_dT cf. Chatonella sp. 18S CtoxiS09_25_dT cf. Chatonella sp. 18S PfarD01_25_dT Pseudochattonella farcimen 28S PverD01_25_dT Pseudochattonella verruculosa 28S SSHaka0193A25_dT Heterosigma akashiwo 18S SSHaka0200A25_dT H. akashiwo 18S LSHaka0544A25b_dT H. akashiwo 28S LSHaka0268A25_dT H. akashiwo 28S LSHaka0544A25c_dT H. akashiwo 28S LSHaka0548A25_dT H. akashiwo 28S LSHaka0329A25_dT H. akashiwo 28S LSHaka0358A24_dT H. akashiwo 28S 2.6. Microarray Hybridization Before use, slides were blocked by incubating the DNA chips in a blocking solution (0.02% SDS, 2× SSC) for 20 min at 50 °C and ~70 rpm in the dark. The slides were washed once in ddH2O for 10 min at 50 °C and twice always in fresh ddH2O for 15 min at RT and ~70 rpm in the dark. The slides were dried by centrifugation in a glass dish for 3 min at 900 rpm and stored in the fridge (possible for up to two month). Labeled field samples (1 µg RNA) were mixed with 30 µL of 2× hybridization buffer, 3 µL Poly-dA (1 µM), 10 ng TBP-control and adjusted with nuclease-free water to 45 µL. Poly-dA is added to block the poly-T spacer on the probe and TBP is the TATA box gene fragment added as the positive hybridization control. The labeled RNA was then denatured for 5 min at 94 °C. After denaturation, the samples were shortly placed on ice and 15 µL of KREAblock (background blocker from KREATECH) was added. Slides were placed into an array holder; coverslips (LifterSlips, Erie Scientific, USA) were cleaned and placed onto the microarrays. Half of the hybridization mixture (30 µL) was added to one microarray. Hybridization was carried out for 1 h at 65 °C in a 50 mL Falcon tube containing a wet Whatman paper. The DNA chips were washed three times and shaken (~70 rpm) in the dark under stringent conditions. The washings were always undertaken for 10 min. The incubation in the first washing buffer (2× SSC/10 mM EDTA/0.05% SDS) and the second washing buffer (0.5× SSC/10 mM EDTA) was done at room temperature. The incubation in the third washing buffer (0.2× SSC/10 mM EDTA) was done at 50 °C. 2.7. Data Analysis Obtained fluorescent signals and the surrounding background intensity were calculated by superimposing a grid of circles (midtal_ver32_20110429.gal or MIDTAL_V3.3.gal) onto the scanned image using the GenePix 6.0 software. First results were processed through the phylochip analyzer program to generate a hierarchy file to establish the hierarchical levels of the probes on the chip [19]. The hierarchy file and hybridization results were then progressed with the GPR-Analyzer version 1.27 and the hierarchy file version 1.06 [20]. A signal-to-noise ratio (S/N ratio) above two was taken as a cutoff for a positive signal. To compare values from different hybridizations, signals were normalized using the internal control DunGS02_25_dT (corresponds to Dunaliella tertiolecta), and replicates averaged. The mean of the total signal intensity and its standard deviation (SD) for the replicates of each probe, which are depicted in the graphs below, can be found in supplementary S2. All microarray results were uploaded to the MIDTAL database at http://www.mba.ac.uk/midtal. Specific instructions can be found in the MIDTAL manual [14] to open a new account from this site. 3. Results and Discussion 3.1. Species Composition during Sampling Period Based on Cell Counts The samples were characterized by a mixed assembly of species (Table S1) and dominated mainly by diatoms and cryptomonads (Table 3). Dominant taxa in the five samples were Chaetoceros spp., Cryptomonadales, Asterionellopsis glacialis and Cylindrotheca closterium. The last sample (6A) by the end of October showed also a bloom of Nitzschia spp. With respect to potentially toxic algae (Table 4, Table S1), it was possible to observe several developments of Pseudo-nitzschia species. Based on morphological characteristics of the valves and on previous distinctions made by other authors [21,22], species of Pseudo-nitzschia were grouped and counted using four identification groups: the “slender” (seriata complex, i.e., P. multiseries + pungens), the “thin” (valve < 3 µm, delicatissima complex, i.e., P. calliantha + delicatissima + pseudodelicatissima), the “wide” (valve > 3 µm, seriata complex, i.e., P. australis + fraudulenta + seriata + subpacifica), and the “sigmoid” (P. multistriata). The sigmoid group was observed with low abundances in August (sample 3A, 400 cells·L−1) and higher abundances in October (samples 4A and 6A) with a maximum of 40,600 cells·L−1 in the last sample. In July (sample 1A), it was possible to observe a high concentration of the “wide” Pseudo-nitzschia (P. australis, fraudulenta, seriata, and subpacifica) with 30,200 cells·L−1, as well as 2 cells of Alexandrium spp.. At the beginning of August (sample 2A) and the beginning of October (sample 4A), species of Prorocentrum (P. cf. minimum, balticum, and cordatum) were detected with 400 cells·L−1. Furthermore, 7,000 cells·L−1 of Pseudo-nitzschia (mainly from the “thin” group) were observed in sample 2A. Except for the aforementioned Pseudo-nitzschia multistriata (400 cells·L−1), no potentially toxic species were observed at the end of August (sample 3A). In both October samples (4A and 6A), it was possible to identify Heterosigma akashiwo with 600 and 400 cells·L−1, respectively. In the late October sample (6A), 30 cells·L−1 of Dinophysis caudata were also counted. Despite few events of potentially toxic algae blooms during our study period, we can point out the presence of five genera in our samples (Pseudo-nitzschia, Heterosigma, Prorocentrum, Alexandrium, and Dinophysis) that are all represented by probes on the MIDTAL microarray. microarrays-02-00001-t003_Table 3Table 3 Non-toxic cells in high abundance at the Arcachon site over the sampling period in cells·L−1. Species 1A 2A 3A 4A 6A Cryptomonadales 50.700 331.500 181.400 194.100 36.400 Chaetoceros spp. 59.000 1.841.200 629.100 4.400 40.600 Asterionellopsis glacialis 19.000 32.400 0 27.800 446.400 Nitzschia spp. 1.200 400 600 11.000 73.200 microarrays-02-00001-t004_Table 4Table 4 Cell counts of potentially harmful cells at the Arcachon site over the sampling period. Species 1A 2A 3A 4A 6A Pseudo-nitzschia spp. 30,200 7,000 0 0 800 Pseudo-nitzschia multistriata 0 0 400 1,700 40,600 Prorocentrum minimum, balticum, cordatum 0 400 0 400 0 Heterosigma akashiwo 0 0 0 600 400 Alexandrium spp. 20 0 0 0 0 Dinophysis caudata 0 0 0 0 30 3.2. Relations between Microarray Signal, Cell Counts and Detection of Toxins The insertion of a taxonomic hierarchy file in the GPR-Analyzer [20] gave us the advantage to distinguish false positives among the species-specific probes in the microarray analysis and exclude them prior to data interpretation. Briefly, for a species to be present, the entire taxonomic hierarchy leading to that species must also be present. The slopes of culture calibration curves of each species incorporated into the GPR-Analyzer allow for the transformation of microarray signals into cell abundances. 3.2.1. Pseudo-nitzschia and ASP Toxins Pseudo-nitzschia was observed throughout the sampling period and is the only potentially toxic phytoplankton genus that formed a dominant bloom according to the cell counts. The microarray detected three of five Pseudo-nitzschia genus-level probes (PSN + some Frags_25_dT, PSN + FRAGS02-25new_dT and PsnGS02_25_dT) throughout the sampling period (Figure 3(a)). The other two generic-level probes (PsnGS01_25_dT, and PSN no pungens_25_dT) were excluded because the S/N ratio was not always above two. These two are not as strong as the other three probes, which are positioned at the top of the hierarchy file and thus do not cause the hierarchy test to fail. Weaker probes are always placed inside stronger probes to prevent such failure of true positives. Domoic acid (DA) was detected with ELISA [23] (Table 5) in both October samples (4A and 6A) but not in sample 3A (22.08.2011) where 400 cells·L−1 of P. multistriata were counted. This result suggests that the threshold for detecting DA with ELISA is somewhere between 400 and 1,700 cells·L−1 for the species P. multistriata. Furthermore, the Multi SPR gave no signal even though the last October sample had 40,600 cells·L−1. In general, it was found that the ELISA was more sensitive to lower amounts of toxin than the Multi SPR [23]. Because it is quite arduous to identify Pseudo-nitzschia multistriata to the species-level with light microscopy, and because some of our species-specific probes are still being optimized, we focused our comparison on P. multistriata (i.e., the “sigmoid” group) with two genus-level probes and three species-level probes on the array. The October bloom of 40,600 cells·L−1 of Pseudo-nitzschia multistriata matched the microarray with positive hits (S/N ratio above 2) of the two genus-level probes (PSN + some Frags_25_dT and PSN+FRAGS02-25new_dT) and the three species-level probes (PmulausD01_25_dT, PmulacalD02_25_dT, and PmulaD03_25_dT) (Figure 3(b)). The probe PcalfrauD04_25_dT (now interpreted to be a genus-level probe because it cross-reacted with all Pseudo-nitzschia spp. tested) showed consistent high signals for all Pseudo-nitzschia spp. in calibration curves (data not shown) and field samples. Figure 3 Microarray signals of (a) the Pseudo-nitzschia spp. Genus-level probes (PSN + some Frags_25_dT, PSN + FRAGS02-new_dT and PsnGS02_25_dT) and (b) P. multistriata species-level probes (PmulausD01_25_dT, PmulacalD02_25_dT, PmulaD03_25_dT) normalized against Dunaliella tertiolecta (DunGS02_25_dT) for the field samples taken in Arcachon Bay, France and compared to cell counts. The graphs show only probes that yielded a signal above the detection limit (signal/noise ratio > 2), except for PmulaD03_25_dT, which is only in sample 6A above the S/N ratio. The sampling dates (24.07.2011, 08.08.2011, 22.08.2011, 04.10.201 and 20.10.2011) correspond to the sampling names: 1A, 2A, 3A, 4A and 6A. Cell counts are depicted in log10 on the secondary y-axis and as columns. microarrays-02-00001-t005_Table 5Table 5 Toxins measured by Multi SPR and ELISA during the sampling period in Arcachon Bay, France, adapted from [23]. STX Okadaic Acid, DTXS Domoic Acid (PSP) (DSP) (ASP) Sampling Date Multi SPR ELISA Multi SPR ELISA Multi SPR ELISA 24.07.2011 − − − − − − 08.08.2011 − − − + − − 22.08.2011 − + − + − − 04.10.2011 − + − + − + 20.10.2011 + + − + − + 3.2.2. Dinophysis and Prorocentrum and DSP Toxins The non-toxic species Dinophysis tripos was counted in sample 1A (20 cells·L−1) and the toxic species D. caudata in sample 6A (30 cells·L−1, Table 4). No other Dinophysis species was identified by using light microscopy. Only the top genus-level probe in the hierarchy for Dinophysis (DphyGS03_25_dT) was detected with the microarray in sample 6A, but no species-specific probes were detected with the microarray in sample 6A. This result suggests that the microarray threshold for D. caudata species probe is above 30 cells. Cells from the potentially toxic genus Prorocentrum (group of P. minimum, balticum, and cordatum) were counted in samples 2A and 4A (both with 400 cells·L−1, Table 4). In addition, two planktonic usually considered harmless species, P. micans (sample 2A) and P. triestinum (sample 3A, 4A and 6A), were also identified by light microscopy with abundances ≤800 cells·L−1 (Table S1). No Prorocentrum species were counted in sample 1A. None of the planktonic clade-level probe for Prorocentrum ProroFBS02_25_dT and the species-specific probes for P. minimum (PminiD01_25_dT) and P. micans (PmicaD02_25_dT) of the microarray detected the presence of these taxa. It is likely that they require higher cell numbers to achieve a signal. With the third generation of the MIDTAL microarray new probes for Prorocentrum (two clade-level and six species-level probes) were tested, but without the poly dT_15 spacer region to raise the probes higher above the surface because they were still under testing for specificity. The new planktonic Prorocentrum probe ProroFPS01 was detected in samples 1A, 2A and 6A whereas the benthic Prorocentrum probe ProroFBS01 was detected in samples 4A and 6A (Figure 4(a)). New species-specific probes were made for the benthic species P. belizeanum, maculosum, rathymum and mexicanum. The probe PbeliS01 specific for P. belizeanum was detected with the microarray in samples 4A and 6A and the probe PrathD01 specific for P. rathymum and mexicanum was detected in sample 6A (Figure 4(a)). As for both samples, the higher probe ProroFBS01 (benthic Prorocentrum) was detected; the species-specific probes are not false positives and point out the limitation of microscopic cell counting. The specificity of theses Prorocentrum species has only been tested against a limited number of species and it is also likely that these probes are cross-reacting to another species present in the sample. P. rathymum is found in Malaysia and in the Mediterranean and P. mexicanum has a Caribbean distribution. More work is needed to clarify the taxon that is reacting with this probe. One way to achieve this is to use the probe as a FISH probe and sort the labeled cells or look at them in the microscope. Okadaic acid was detected by ELISA in all samples except for the first (sample 1A), and the Multi SPR gave no signal at all. We presume that because Prorocentrum was more abundant than Dinophysis; its species is the source of this toxin. Figure 4 (a) Normalized signal of Prorocentrum-level probes (ProroFPS01 and ProroFBS01) and the species-level probes PrathD01 and PbeliS01. (b) Normalized signal of the Alexandrium genus-level probe AlexGD01_25_dT. 3.2.3. Alexandrium and PSP Toxins Two cells of the genus Alexandrium (ca. 20 cells·L−1) were counted only in sample 1A, whereas the microarray detected it throughout the sampling period (Figure 4(b)) with the highest signal at the end of October (sample 6A). Furthermore, PSP toxins were detected with ELISA in late August (sample 3A) and the remaining sampling period, as well as with the Multi SPR in sample 6A (Table 5, see [23] for more discussion on toxin found in these samples). If toxin probes are efficient and therefore PSP toxins are indeed present, there are two different ways to explain the absence of Alexandrium in cell counts: either Alexandrium cells have effectively been missed with the microscope, or there are other PSP-containing microorganisms in the water that are not identified. Neither A. ostenfeldii, A. minutum nor A. tamarense probes were detected by the microarray and their calibration curves for each specific probe have a detection limit of 200 cells [24]. Thus, we are unsure as to which species could be contributing to the PSP toxin profile. It could be Gymnodinium catenatum (see below) or another member of the genus Alexandrium. A. pseudogonyaulax could be a potentially missed Alexandrium species. There are no A. pseudogonyaulax-specific probes on the microarray. There are also many species that are not well investigated for toxin production. However, our data underlines the importance of including additional genus- and species-level probes for Alexandrium, in order to capture the full variability found in this genus. In any case, the detection of Alexandrium and its PSP toxins shows the advantage of the combination of the two methods (species and toxins) to detect harmful species, as well as to detect new invasive species as climate changes and tropical species move into temperate regions. 3.2.4. Heterosigma akashiwo The heterokont Heterosigma akashiwo was identified by microscopic cell counts in sample 4A (600 cells·L−1) and 6A (400 cells·L−1). The microarray detected this taxon with the species-specific probe LSHaka054425b_dT in all samples except 3A. In addition, two more species-specific probes gave positive signals in sample 4A (LSHaka0268A25_dT and LSHaka0358A24_dT), and four in sample 6A (LSHaka0268A25_dT, LSHaka0544A25c_dT, LSHaka0329A25_dT, and SSHaka0200A25_dT). This species can be difficult to identify, especially once preserved in Lugol’s. Two species-specific probes were designed from the 18S region (SSHaka) and six more from the 28S region (LSHaka) for H. akashiwo (Table 2, [25]). Their calibration curves show the sensitivity of each probe and point out a low affinity with the H. akashiwo RNA. Some probes showed no sensitivity below 5 or even 25 ng of RNA, i.e., more than 700 cells are required to get a S/N ratio above two. This means that, in our case, we had around 230 cells·L−1 of H. akashiwo because not all probes were detected. 3.2.5. Species Unfound by Cell Counts but Identified with Microarray and Hierarchy File Fish Killing Species Lugol’s-fixed cells of Pseudochattonella are difficult to identify by light microscopy because the cell shape changes and the discharge of mucocysts gives them a warty appearance [26]. It is possible to distinguish the two sister species P. farcimen and P. verruculosa molecularly, because the two differ in several bases in the large ribosomal subunit [26]. In sample 6A, all genus-level probes of Pseudochattonella (PschGS01_25_dT, PschGS04_25_dT, PschGS05_25_dT) and the two species-level probes PfarD01_25_dT (P. farcimen) and PverD01_25_dT (P. verruculosa) were detected with a signal-to-noise above 2 (Figure 5(a)). The integrated calculation of cells L−1 in the GPR-Analyzer [20] revealed for Pseudochattonella farcimen 19,463 cells·L−1 and for Pseudochattonella verruculosa 48,428 cells·L−1, which is likely to be overestimated. Indeed, this species has only been identified in fjords and open waters of the North Sea, Skagerrak, and Kattegat, whose temperatures are below 10 °C [27]. During the summer–fall season, waters in Arcachon Bay are typically >25 °C [28]. Our results suggest perhaps another very closely related species as yet undetected could be in Arcachon Bay if the distribution of this species is exclusively in cold temperate waters. If this probe continues to show positive results, the probe could be used as a FISH probe to retrieve the cells giving the signal on the microarray for further investigations. Cells hybridized by the probe could be sorted by flow cytometry and investigated morphologically or molecularly. Once identified, the cells could later be brought into culture and their toxicity tested with bioassays. Figure 5 (a) Normalized signal intensity of the genus-level probes (PschGS01_25_dT, PschGS04_25_dT, PschGS05_25_dT) of Pseudochattonella and the two species-level probes PfarD01_25_dT (Pseudochattonella farcimen) and PverD01_25_dT (Pseudochattonella verruculosa) for sample 6A (20.10.2011) only. (b) Normalized signal intensity of the class-level probes (PrymS03_25_dT, PrymS01_25_dT) and the clade-level probe (Clade01old_25_dT) of Prymnesium spp. (c) Normalized signal intensity of the genus-level probe of Karlodinium spp. (KargeD01_25_Dt). No Prymnesiophyta were identified by cell counts, but the higher group probe for Prymnesiophyta (PrymS01_25_dT), the class level for Prymnesiophyceae (PrymS03_25_dT) and the clade-level probe for Prymnesium clade B1 (Clade01old_25_dT) were detected throughout the sampling period (Figure 5(b)). The second clade-level probe for Prymnesium clade B1 (Clade01new25_dT) was detected in samples 3A, 4A and 6A. Furthermore, the species-level probe for P. polylepis (CpolyS01_25_dT) was detected in sample 6A and in sample 4A the species-level probe for P. parvum (PparvD01_25_dT). This indicates the potential for a fish-killing event in Arcachon Bay under the appropriate conditions for growth. Although the Prymnesiophyta group is taken into account within the harmful phytoplankton monitoring program, the small size of this genus (<10 µm) as well as the smaller volume of water used for Utermöhl sedimentation and observation (100 mL maximum) than the volume of filtered seawater for RNA extraction, avoid any faithful microscopy identification and counting. The microarray can detect Prymnesium above 5 ng, which is equivalent to 3,800 cells for P. polylepis and 8,800 cells for P. parvum [29]. In our case (3 L filtered) it means 1,100 and 2,500 cells·L−1, respectively, which are high enough to be counted in a 10- or 100-mL sedimented subsample. The genus Karlodinium (KargeD01_25_dT) was first detected in sample 3A and then with decreasing signals onwards (Figure 5(c)). Karlodinium veneficum is a high-biomass producer and the collapse of a bloom leads to the production of a surface scum that is visible as an oily, brownish discoloration of the water and kills fish and other gill-breathing animals [30]. However, no signals were detected for the six species-specific probes of K. veneficum present on the microarray. Based on their calibration curves (data not shown), the detection limit for four of the six probes is around 247 cells. The species-specific level probes are more sensitive than the genus-level probe. Therefore, we can exclude this species as a potential candidate being present in the bay. This is another example of how the microarray can detect potentially toxic species that are not counted or identified as being potentially toxic. Azaspiracid Shellfish Poisoning (AZP) Toxins Producer Azadinium spp. (AzaGS01_25_dT) was detected in sample 4A but only in two out of five spots on two different microarray slides. This may not be a genuine signal, but this species has only recently described [31] and it is also a relatively arduous species to identify based on light microscopy. Not all monitoring agencies are able to adjust their cell counts routinely to account for this toxic species. At least three more toxic species have been recently isolated and described [32,33,34]. Other PSP Toxins One species-level probe out of four for Gymnodinium catenatum (LSGcat0270A24_dT) was detected in samples 1A, 4A and 6A. In samples 4A and 6A, the microarray also detected another species-level probe for G. catenatum (SSGcat0826A27_dT). The signals were not very high (S/N ratio between 2.2 and 4.9). G. catenatum is known to cause PSP and could therefore contribute, besides Alexandrium, to its detection via the ELISA and Multi SPR. 4. Conclusions The third generation of the MIDTAL microarray with its improved protocols has great potential to be used as a monitoring tool for toxic algae, even in non-bloom situations, although improvements and tests are still needed. The probes on the MIDTAL microarray have been designed from a global database and the specificity tests done on the probes were made from global isolates. Thus, the MIDTAL microarray can be regarded as a universal microarray that can be used globally. Its specificity has been tested at eight sites across Europe within the MIDTAL project over a two-year period and in no case did it fail to detect the presence of a toxic species when cross-validated with the toxin array. Our results show the advantage of combining the MIDTAL microarray with toxin detection, especially for detecting species either not counted in the cell counts because of low volume or poor preservation, or because they are new to the area, i.e., invasive species, such as new toxin-producing species (Azadinium and the causative species producing the signal for PSP) that might be new to the area or not yet routinely counted in any monitoring program. We also found several species with the microarray that were difficult to identify using light microscopy, such as Prymnesium parvum, Pseudochattonella, and Azadinium. A more specific identification requires electron microscopy. In other cases, such as the recording of Karlodinium, it is likely that the volume difference between the species filtered and settled for counting reflects the potential of the microarray to be more sensitive for the detection of rare events. Acknowledgments YDA thanks her crew members (Laurent Letort and Francis Prince) for sampling in Arcachon Bay. Cell counts were supplied by Nadine Masson Neaud and Danièle Maurer from IFREMER/LER Arcachon (Ifremer/Quadrige²/Rephy DATA). This work was funded by the EU’s 7th Framework Program (FP7-ENV-2007-MIDTAL-201724). Appendix microarrays-02-00001-t006_Table S1Table S1 Cell densities (cell·L−1)of field samples taken at Arcachon Bay in France between July and October 2011 (1A = 26.07, 2A = 09.08, 3A = 23.08; 4A = 06.10; 6A = 20.10) and identified by light microscopy. The identified genera and species are ordered into higher taxon groups. Toxic species are indicated with an *. Dinoflagellates 1A 2A 3A 4A 6A * Alexandrium 20 0 0 0 0 * Dinophysis caudata 0 0 0 0 30 Dinophysis tripos 20 0 0 0 0 Gymnodiniaceae 0 400 3,800 1,800 Katodinium 100 0 0 0 0 Gyrodinium 0 100 100 200 200 Gyrodinium spirale 0 200 0 100 0 Prorocentrum micans 0 100 0 0 0 * Prorocentrum minimum + balticum + cordatum 0 400 0 400 0 Prorocentrum triestinum 0 0 200 800 400 Protoperidinium 0 200 0 800 600 Protoperidinium bipes 0 400 5,400 600 0 Protoperidinium steinii + pyriforme 0 0 100 0 0 Protoperidinium diabolus 0 0 200 0 0 Heterocapsa niei 0 100 1000 0 0 Gonyaulax 0 0 100 0 0 Scrippsiella + Ensiculifera + Pentapharsodinium + Bysmatrum 0 200 2,600 1,000 1,800 Torodinium 0 0 100 100 200 Amphidinium 0 0 0 200 0 Heterocapsa triquetra 0 0 200 0 0 Peridiniales 400 0 0 1,200 1,000 Peridiniaceae 0 0 1,400 0 0 Peridinium quinquecorne 0 400 3,200 0 0 Euglena 1A 2A 3A 4A 6A Euglenaceae 0 600 400 0 0 Eutreptiaceae 0 0 0 2,200 0 Eutreptiella 2,400 5,000 4,000 0 3,600 Cryptomonads 1A 2A 3A 4A 6A Cryptomonadales 50,700 331,500 181,400 194,100 36,400 Diatoms 1A 2A 3A 4A 6A Centrales 0 0 0 0 400 Rhizosolenia imbricata + styliformis 0 100 0 0 0 Rhizosolenia setigera + pungens 0 0 0 600 1000 Proboscia alata 0 100 0 0 0 Corethron 0 0 0 100 0 Paralia sulcata 600 0 0 400 0 Thalassiosira 0 0 0 2,400 7,000 Thalassiosira rotula 0 0 0 0 400 Corethron 0 0 0 0 1,200 Chaetoceros 56,200 1,821,400 627,900 0 13,400 Chaetoceros decipiens 2,800 19,800 1,200 3,200 6,800 Chaetoceros curvisetus + debilis + pseudocurvisetus 0 0 0 1200 19,800 Chaetoceros danicus 0 0 0 0 600 Lithodesmium 200 5,300 7,000 3,600 6,800 Leptocylindrus danicus 6,200 1,600 0 3,800 12,200 Leptocylindrus minimus 0 4,800 0 0 6200 Cerataulina pelagica 400 1,000 5,400 2,800 Dactyliosolen fragilissimus 1,000 4,600 1200 1,000 4,400 Guinardia striata 1,100 0 0 0 1,800 Guinardia flaccida 100 0 0 0 0 Guinardia delicatula 0 0 0 0 1,400 Skeletonema costatum 0 3,600 500 0 1,000 Odontella regia 0 0 0 0 600 Biddulphia alternans 0 0 0 0 600 Eucampia zodiacus 0 0 0 0 1,400 Hemiaulus 400 1,100 0 0 0 Lauderia 400 0 0 0 0 Pennales 3,800 600 200 2,000 4,200 * Pseudo-nitzschia large width, seriata complex (australis + fraudulenta + seriata + subpacifica) 30,200 200 0 0 0 * Pseudo-nitzschia narrow width, delicatissima complex (calliantha + delicatissima + pseudodelicatissima) 0 6,800 0 0 0 * Pseudo-nitzschi, slender group, seriata complex (multiseries + pungens) 0 0 0 0 800 * Pseudo-nitzschia sigmoid group (multistriata) 0 0 400 1,700 40,600 Thalassionema nitzschioides 0 2,200 0 1,200 12,600 Pleurosigma + Gyrosigma 0 0 100 600 600 Nitzschia 0 0 0 1,1000 72,800 Nitzschia longissima 1200 400 600 0 400 Cylindrotheca closterium 6,600 16,400 25,350 0 2,600 Asterionellopsis glacialis 19,000 32,400 0 2,7800 446,400 Licmophora 0 0 0 100 0 Cocconeis 0 0 0 400 200 Heterokonts 1A 2A 3A 4A 6A *Heterosigma akashiwo 0 0 0 600 400 Dictyocha 0 0 100 400 200 microarrays-02-00001-t007_Table S2Table S2 Mean of the total signal intensity (TI) and its standard deviation (STDEV) of each microarray-probe used in the graphs of the publication. 1A (24.07.2011) 2A (08.08.2011) 3A (22.08.2011) 4A (04.10.2011) 6A (20.10.2011) TI STDEV TI STDEV TI STDEV TI STDEV TI STDEV PSN + some Frags_25_dT 571,661 283,278 462,375 261,002 376,328 136,949 328,723 64,836 1,234,899 246,612 PSN+FRAGS02-25new_dT 570,012 278,217 483,590 235,672 374,378 107,819 328,689 73,975 1,236,907 237,510 PsnGS02_25_dT 366,160 122,407 269,644 59,512 176,231 40,406 154,926 102,545 527,093 47,698 PmulacalD02_25_dT 201,918 30,724 189,923 53,609 35,294 18,226 35,985 14,234 253,037 27,753 PmulaD03_25_dT 66,635 5,760 71,895 9,951 37,380 15,172 18,590 10,183 212,033 22,236 PmulausD01_25_dT 240,310 36,151 168,879 19,588 154,874 31,009 181,673 30,966 784,727 91,653 ProroFPS01 169,269 31,916 160,601 35,392 58,362 14,735 78,983 19,111 ProroFBS01 113,540 137,560 67,434 189,17 PrathD01 108,027 8,344 PbeliS01 60,137 58,069 21,412 24,719 AlexGD01_25_dT 477,054 82,900 555,744 276,647 270,780 352,356 154,715 37,017 389,966 29,238 PschGS01_25_dT 63,769 9,284 PschGS04_25_dT 547,515 98,442 PschGS05_25_dT 47,371 12,185 PverD01_25_dT 70,207 10,052 PfarD01_25_dT 51,096 8,260 PrymS01_25_dT 1,062,712 223,123 1,208,456 400,071 1,285,690 372,848 1,592,062 501,853 1,127,413 97,605 PrymS03_25_dT 184,896 64,536 262,141 132,584 204,940 51,019 231,062 63,987 285,495 49,415 Clade01old_25_dT 935,860 146,759 707,441 135,777 487,530 194,120 521,791 172,448 353,814 29,352 Clade 01new25_dt 217,043 82,294 293,107 86,528 141,925 34,971 CpolyS01_25_dT 112,327 9,400 122,730 29,900 PparvD01_25_dT 133,275 251,500 KargeD01_25_dT 138,096 195,135 90,368 92,796 17,867 11,278 ==== Refs References 1. Hallegraeff G.M. Harmful algal blooms: A global overview Manual on Harmful Marine Microalgae IOC-UNESCO Paris, France 2003 Volume 11 25 49 2. Moestrup Ø. Akselman R. Cronberg G. Elbraechter M. Fraga S. Halim Y. Hansen G. Hoppenrath M. Larsen J. Lundholm N. Nguyen L.N. Zingone A. IOC-UNESCO Taxonomic Reference List of Harmful Micro Algae Available online:http://www.marinespecies.org/HAB (accessed on 24 December 2012) 3. Yasumoto T. Murata M. Oshima Y. Sano M. Matsumoto G.K. Clardy J. Diarrhetic shellfish toxins Tetrahedron 1985 41 1019 1025 10.1016/S0040-4020(01)96469-5 4. Cotterill F.P.D. Systematics, biological knowledge and environmental conservation Biodivers. Conserv. 1995 4 183 205 10.1007/BF00137784 5. Gómez F. Phytoplankton invasions: Comments on the validity of categorizing the non-indigenous dinoflagellates and diatoms in European seas Mar. Pollut. Bull. 2008 56 620 628 10.1016/j.marpolbul.2007.12.014 18295804 6. DeSantis T.Z. Stone C.E. Murray S.R. Moberg J.P. Andersen G.L. Rapid quantification and taxonomic classification of environmental DNA from both prokaryotic and eukaryotic origins using a microarray FEMS Microbiol. Lett. 2005 245 271 278 10.1016/j.femsle.2005.03.016 15837382 7. DeSantis T.Z. Brodie E.L. Moberg J.P. Zubieta I.X. Piceno Y.M. Andersen G.L. High-density universal 16S rRNA microarray analysis reveals broader diversity than typical clone library when sampling the environment Microbial Ecol. 2007 53 371 383 8. Yergeau E. Arbour M. Brousseau R. Juck D. Lawrence J.R. Masson L. Whyte L.G. Greer C.W. Microarray and real-time PCR analyses of the responses of high-arctic soil bacteria to hydrocarbon pollution and bioremediation treatments Appl. Environ. Microbiol. 2009 75 6258 6267 19684169 9. Gescher C. Metfies K. Medlin L.K. The ALEX CHIP—Development of a DNA chip for identification and monitoring of Alexandrium Harmful Algae 2008 7 485 494 10.1016/j.hal.2007.11.001 10. Brodie E.L. DeSantis T.Z. Joyner D.C. Baek S.M. Larsen J.T. Andersen G.L. Hazen T.C. Richardson P.M. Herman D.J. Tokunaga T.K. Wan J.M. Firestone M.K. Application of a high-density oligonucleotide microarray approach to study bacterial population dynamics during uranium reduction and reoxidation Appl. Environ. Microbiol. 2006 72 6288 6298 16957256 11. Brodie E.L. DeSantis T.Z. Parker J.P. M. Zubietta I.X. Piceno Y.M. Andersen G.L. Urban aerosols harbor diverse and dynamic bacterial populations PNAS 2007 104 299 304 17182744 12. Delehanty J.B. Ligler F.S. A microarray immunoassay for simultaneous detection of proteins and bacteria Anal. Chem. 2002 74 5681 5687 10.1021/ac025631l 12433105 13. Lee D.-Y. Seto P. Korczak R. DNA microarray-based detection and identification of waterborne protozoan pathogens J. Microbiol. Meth. 2010 80 129 133 10.1016/j.mimet.2009.11.015 14. Lewis J. Medlin L.K. Raine R. MIDTAL (Microarrays for the Detection of Toxic Algae): A Protocol for a Successful Microarray Hybridisation and Analysis 1st Gantner Verlag Liechtenstein, Germany 2012 15. Medlin L.K. Schmidt K. Molecular probes improve the taxonomic resolution of cryptophyte abundance in Arcachon Bay Vie et Milieu 2010 60 9 15 16. Kegel J.U. Del Amo Y. Medlin L.K. Introduction to project MIDTAL: Its methods and samples from Arcachon Bay, France Environ. Sci. Pollut. Res. 2012 10.1007/s11356-012-1299-9 17. Campbell K. McGrath T. Sjölander S. Hanson T. Tidare M. Jansson O. Moberg A. Mooney M. Elliott C. Buijs J. Use of a novel micro-fluidic device to create arrays for multiplex analysis of large and small molecular weight compounds by surface plasmon resonance Biosens. Bioelectron. 2011 26 3029 3036 10.1016/j.bios.2010.12.007 21185716 18. Ludwig W. Strunk O. Westram R. Richter L. Meier H. Yadhukumar H. Buchner A. Lai T. Steppi S. Jobb G. ARB: A software environment for sequence data Nucl. Acid. Res. 2004 32 1363 1371 19. Metfies K. Borsutzki P. Gescher C. Medlin L.K. Frickenhaus S. Phylochipanalyser—A program for analysing hierarchical probe sets Mol. Ecol. Resour. 2008 8 99 102 10.1111/j.1471-8286.2007.01927.x 21585726 20. Dittami S.M. Edvardsen B. GPR-Analyzer: A simple tool for quantitative analysis of hierarchical multispecies microarrays Environ. Sci. Pollut. Res. 2012 10.1007/s11356-012-1051-5 21. Hasle G.R. Nitzschia and Fragilariopsis Species Studied in the Light and Electron Microscopes. II. The Group Pseudonitzschia Universitetsforlaget Oslo, Norway 1965 22. Hasle G.R. Syvertsen E.E. Marine diatoms Identifying Marine Phytoplankton Academic Press San Diego, CA, USA 1997 5 385 23. McNamee S. Elliott C. Delahaut P. Campbell K. Multiplex biotoxin surface plasmon resonance method for marine biotoxins in algal and seawater samples Environ. Sci. Pollut. Res. 2012 10.1007/s11356-012-1329-7 24. Taylor J. Kegel J.U. Lewis J. Medlin L.K. Validation of the detection of Alexandrium spp using specific RNA probes tested in a microarray format: Calibration of signal based on variability of RNA content with environmental conditions Harmful Algae 2013 submitted 25. Blanco E.P. Hagström J. Salomon P.S. Graneli E. Detection of Heterosigma akashiwo (Hada) using specific RNA probes: Variability of RNA content with environmental conditions Harmful Algae 2013 submitted 26. Dittami S.M. Hostyeva V. Egge E.S. Kegel J.U. Eikrem W. Edvardsen B. Seasonal dynamics of harmful algae in outer Oslofjorden monitored by microarray, qPCR, and microscopy Environ. Sci. Pollut. Res. 2013 10.1007/s11356-012-1392-0 27. Naustvoll L.-J. NOBANIS—Invasive Alien Species Fact Sheet Pseudochattonella farcimen —From: Online Database of the North Eurpean and Baltic Network on Invasive Alien Species—NOBANIS Available online:http://www.nobanis.org (accessed on 24 December 2012) 28. Glé C. Del Amo Y. Sautour B. Laborde P. Chardy P. Variability of nutrients and phytoplankton primary production in a shallow macrotidal coastal ecosystem (Arcachon Bay, France) Estuar. Coast. Shelf Sci. 2008 76 642 656 10.1016/j.ecss.2007.07.043 29. McCoy G.R. Touzet N. Fleming G.T. Raine R. An evaluation of the applicability of microarrays for monitoring toxic algae in Irish coastal waters Environ. Sci. Pollut. Res. 2012 10.1007/s11356-012-1294-1 30. Swan River Trust Karlodinium Micrum Algal Bloom, March 2005. Fact Sheet Available online:http://www.wa.canoe.org.au/?Page=7808 (accessed on 24 December 2012) 31. Tillmann U. Elbrächter M. Krock B. John U. Cembella A. Azadinium spinosum gen. et sp. nov. (Dinophyceae) identified as a primary producer of azaspiracid toxins Eur. J. Phycol. 2009 44 63 79 32. Tillmann U. Elbrächter M. John U. Krock B. Cembella A. Azadinium obesum (Dinophyceae), a new nontoxic species in the genus that can produce azaspiracid toxins Phycologia 2010 49 169 182 10.2216/PH09-35.1 33. Tillmann U. Elbrächter M. John U. Krock B. A new non-toxic species in the dinoflagellate genus Azadinium : A. poporum sp. nov Eur. J. Phycol. 2011 46 74 87 10.1080/09670262.2011.556753 34. Tillmann U. Salas R. Gottschling M. Krock B. O’Driscoll D. Elbrächter M. Amphidoma languida sp. nov. (Dinophyceae) reveals a close relationship between Amphidoma and Azadinium Protist 2012 163 701 719 22130577

PMC005xxxxxx/PMC5003437.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays1020095microarrays-01-00095ArticleDevelopment and Optimization of a Thrombin Sandwich Aptamer Microarray Meneghello Anna 1Sosic Alice 2Antognoli Agnese 3Cretaio Erica 1Gatto Barbara 2*1 Associazione CIVEN, Via delle Industrie 9, I-30175 Venezia-Marghera, Italy; Email: meneghello@civen.org (A.M.); cretaio@civen.org (E.C.)2 Dipartimento di Scienze del Farmaco, University of Padova, Via F. Marzolo 5, I-35131 Padova, Italy; Email: alice.sosic@studenti.unipd.it3 Veneto Nanotech S.C.p.A., Via S. Crispino 106, I -35129 Padova, Italy; Email: agnese.antognoli@venetonanotech.it* Author to whom correspondence should be addressed; Email: barbara.gatto@unipd.it; Tel.: +39-049-827-5717; Fax: +39-049-827-5366.08 8 2012 9 2012 1 2 95 106 28 6 2012 26 7 2012 07 8 2012 © 2012 by the authors; licensee MDPI, Basel, Switzerland.2012This article is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).A sandwich microarray employing two distinct aptamers for human thrombin has been optimized for the detection of subnanomolar concentrations of the protein. The aptamer microarray demonstrates high specificity for thrombin, proving that a two-site binding assay with the TBA1 aptamer as capture layer and the TBA2 aptamer as detection layer can ensure great specificity at times and conditions compatible with standard routine analysis of biological samples. Aptamer microarray sensitivity was evaluated directly by fluorescent analysis employing Cy5-labeled TBA2 and indirectly by the use of TBA2-biotin followed by detection with fluorescent streptavidin. Sub-nanomolar LODs were reached in all cases and in the presence of serum, demonstrating that the optimized aptamer microarray can identify thrombin by a low-cost, sensitive and specific method. aptamerthrombinfluorescencebioassaymultiplexing microarray ==== Body 1. Introduction Thrombin, a serine protease involved in the last step of the coagulation cascade, produces insoluble fibrin through the proteolytic cleavage of soluble fibrinogen [1,2], and is characterized by two different exosites, one binding fibrinogen (fibrinogen binding domain, FBD) and the other binding heparin (heparin binding domain, HBD). The concentration of thrombin in blood varies considerably: it can be almost absent in the blood of healthy subjects but can reach low-micromolar concentrations during the coagulation process [3]. Thrombin, in addition to its direct actions on the coagulation system, has other functions as a potent signaling molecule that regulates physiologic and pathogenic responses: it represents, for example, a potent chemotactic agent for monocytes and leukocytes involved in the atherosclerotic and thrombotic processes [4]; it is a potential pro-inflammatory mediator in neurotrauma and neurodegenerative disorders [5] and it plays a role in the development, plasticity and pathology of the nervous system [6]. Considering thrombin’s physiological and pathological roles, it is not surprising that simple, low cost and sensitive methods for its detection are actively pursued, and human thrombin is one of the preferred targets for the development of aptamer-based assays. Aptamers are oligonucleotide ligands selected in vitro through the SELEX procedure (Systematic Evolution of Ligands by EXponential enrichment), and represent alternative options to antibodies for the development of diagnostics: besides their lower production costs, advantages of aptamers over antibodies are their relative ease of isolation and modification, tailored binding affinity and reversible denaturation, making them suitable candidates for use as detection systems [7,8,9]. A variety of aptamer-based protein detection strategies have been described in the past few years, employing in most cases two aptamers selected for targeting thrombin at its exosites: TBA1, a 15-nucleotide DNA aptamer able to bind FBD [10] and TBA2, a 29-nucleotide DNA aptamer able to recognize HBD [11]. These aptamers are known to assume, in presence of monovalent cations like K+ and Na+, a G-quadruplex structure [10,12,13]. Aptamer systems described in literature that employ both TBA1 and TBA2 present different formats and detection strategies including fluorescence, electrochemistry, inductively coupled plasma mass spectrometry (ICP-MS) and surface plasmon resonance (SPR), yielding variable limits of detection (LOD), ranging from nanomolar [14,15,16,17,18,19,20,21,22] to picomolar [23,24,25]. Among the aptamers detection platforms described, microarray represents an appropriate sensing format for high-throughput analysis, allowing to analyze a great number of samples at the same time and to scale up the system in order to obtain a multi-sensing platform [7]. In addition, a two-site binding approach is suitable for high-sensitivity detection of a target protein with two or more distinct target domains [26], allowing the setup of a sandwich on a solid support: this approach ensures higher assay specificity because the detection takes place only if the analyte is simultaneously recognized by two different ligands. After a thorough analysis of the ternary complex formation in solution we recently demonstrated the feasibility of the microarray strategy in a sandwich format for thrombin detection employing post-SELEX chemically modified TBA1 as capture layer and TBA2 as detection layer [27]. In the present work we describe the further development and optimization of the thrombin sandwich aptamer microarray for fluorescent analysis, evaluating the system specificity toward related and unrelated proteins. The microarray capability to detect thrombin was evaluated at different times and in complex biological matrices, and the limit of detection (LOD) and limit of quantification (LOQ) of the system were established. Finally, we compared the direct detection method employing the fluorescently labeled TBA2 with an indirect detection system using a biotin-labeled TBA2 subsequently recognized by fluorescent streptavidin. In both cases the aptamer arrays showed a sub-nanomolar detection limit even when tested in complex matrices containing serum, supporting the system capability to identify thrombin in a convenient, low-cost, sensitive and specific way amenable to multiplexing systems. 2. Materials and Methods 2.1. Aptamers and Proteins Solutions The sequence of the unmodified 15-mer TBA1 [10], used here as selection aptamer, is: 5'-GGT TGG TGT GGT TGG-3'. To allow immobilization on microarray slides, an amino modification plus a polyT(12) spacer were added at the 5' terminus (TBA1(12T)NH2) as suggested by other authors [23]. In this way the G-quadruplex structure of TBA1, also if anchored to the slide surface, can fold correctly to recognize the target protein. The sequence of the unmodified 29-mer TBA2 [11], used here as selection aptamer, is: 5'-AGT CCG TGG TAG GGC AGG TTG GGG TGA CT-3'. TBA2 was used as selection aptamer with a 5'-Cy5 modification (TBA2-Cy5) or with a 5'-biotin modification (TBA2-biotin). All protein molecules were purchased from Sigma-Aldrich (St. Louis, MO, USA): Human α-thrombin, Bovine Serum Albumin (BSA), Lysozyme, and Human Vascular Endothelial Growth Factor 165 (VEGF) were used to evaluate aptamer microarray specificity at a final concentration of 500 nM, and Cy3-labeled Streptavidin—at a final concentration of 30 nM—was used to perform thrombin assay in the presence of TBA2-biotin aptamer. Fetal Bovine Serum (FBS) (Gibco, Monza, Italy) was used in order to obtain a complex sample matrix. 2.2. Aptamer Preparation for Microarray Printing In the aptamer arrays, chemically modified TBA1 is used as capture layer for thrombin [27]. Prior to immobilization on the microarray slide, TBA1(12T)NH2 (80 µM in the presence of KCl 100 mM) was denatured at 95 °C for 5 min and then left to cool down to room temperature. This folding step ensures molecules to assume a G-quadruplex structure, responsible for thrombin binding. Folded aptamer was then diluted in the Printing Buffer 1.5× (Printing Buffer 6×: 300 mM sodium phosphate, 0.02% Triton, pH 8.5) to a final concentration of 20 µM. TBA1 aptamers were loaded into microarray plates and submitted to the Microarray Spotter (Versarray Chip Writer Pro System, BioRad) for slide printing, using Telechem SMP3 microspotting pins (Arrayit, Sunnyvale, CA, USA). 2.3. Aptamer Microarray Printing E-surf LifeLine slides (25 mm × 75 mm, LifeLineLab, Pomezia, Italy) were used for microarray printing: these slides allow the binding of amino modified oligonucleotides to the surface. On each slide up to 64 microarray sub-grids, made of six spots of capture aptamer were printed. Distance between spots centre was 200 μm and average spots diameter was 100 μm; distance between each sub-grid was 4.5 mm both in X and Y directions. Printed slides were incubated overnight in a 75% humidity incubation chamber (μBox, Quantifoil Instrument, Jena, Germany), blocked and stored according to protocol. Each slide has the possibility to test up to 64 samples at once, since each sub-grid can be physically isolated from the others by multi-wells hybridization chamber (Grace Bio-Labs, OR, USA) during the incubation phase of the experiment. This approach ensures to analyze several samples on the same slide and nevertheless to minimize array variation resulting from minor fluctuation of external parameters. 2.4. Detection Layer Preparation TBA2 (TBA2-Cy5 or TBA2-biotin) is used as secondary aptamer for detecting thrombin captured by the primary (capture) aptamer immobilized on the microarray. TBA2 (10 µM in KCl 100 mM) was denatured at 95 °C for 5 min and then left to cool down to room temperature in order to assume G-quadruplex structure, which is essential for recognition of the heparin binding domain of thrombin. 2.5. Sandwich Aptamer Microarray Assays Printed aptamer microarrays prepared as detailed above were immersed, just before their use, in thrombin Binding Buffer 1× (20 mM Tris, 140 mM NaCl, 5 mM KCl, 1 mM MgCl2, pH 7.5) at room temperature for 30 min. Thrombin samples were pre-incubated with TBA2-Cy5 or TBA2-biotin in solution with thrombin Binding Buffer 1× (final volume 50 μL), at 25 °C for 30 min. The pre-formed complexes thus obtained were then incubated on the microarray at 25 °C for 2 h in the case of TBA2-Cy5. Steptavidin-Cy3 at the final concentration of 30 nM was added after 1 h of incubation in the case of TBA2-biotin, and the system allowed to incubating for 1 additional hour. Finally, aptamer microarrays were washed with PBS 1× (137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 2 mM KH2PO4, pH 7.4) at room temperature to remove the unbound proteins and rapidly rinsed in MilliQ H2O. 2.6. Slides Scanning and Data Analysis Spin-dried slides were scanned using GenePix 4000B laser scanner (Molecular Devices, Sunnyvale, CA, USA) and the GenePix Pro software using both 532 nm and 635 nm wavelength. Fluorescent spots intensities were quantified using the GenePix Pro software after normalizing the data by subtracting local background from the recorded spot intensities. For each set of six spots median and standard deviation were calculated. In calibration curves experiments LOD and LOQ were calculated: LOD is defined as 3 σ whereas LOQ is defined as 10 σ, where σ is the standard deviation of the blank experiment. All data analysis was performed with Origin Pro 8 software (OriginLab, Northampton, MA, USA). 3. Results and Discussion 3.1. Thrombin Specificity We have developed an aptamer-based microarray exploiting two non-overlapping DNA aptamers recognizing different exosites of human thrombin [27]. Fluorescent labeling of the secondary aptamer TBA2-Cy5 allowed the co-localization of thrombin on TBA1 aptamer immobilized on microarray spots, thus proving the formation on a solid support of a sandwich in which the protein is simultaneously recognized by both aptamers. The recognition by aptamers was specific: a negative control represented by the aptamer OTA, selected for ochratoxin, was printed in the microarray to demonstrate the specific recognition of thrombin by TBA aptamers [27]: OTA aptamer was chosen because it can fold into a G-quadruplex after target binding [28], thus excluding non-selective recognition of G-quartets folded aptamers by our thrombin specific system. With the aim of optimizing our system in view of future applications in multiplex microarray systems, we enlarged our analysis of the aptasensor specificity incubating the TBA1 microarray with proteins related and unrelated to human thrombin. Specificity toward unrelated proteins was evaluated using BSA and lysozyme [25], while human VEGF 165 was chosen for system specificity evaluation since it has both a fibrinogen [29] and a heparin binding domain (FBD and HBD respectively) like thrombin [30]. Figure 1 shows that no significant fluorescent signal was recorded on TBA1 microarray spots after incubation of samples containing BSA or lysozyme pre-incubated with TBA2-Cy5, confirming the specificity of the two aptamers towards human thrombin. The aptasensor is able to discriminate between related target domains: experimental data in Figure 1 shows clearly that no significant binding on TBA1 spots was observed in presence of VEGF 165. Figure 1 Specificity of aptasensor for thrombin. Fluorescence signal (635 nm) recorded on thrombin aptamer microarray after the incubation of the different indicated proteins with TBA2-Cy5 (protein and TBA2 final concentration: 500 nM and 1 mM, respectively); (a) thrombin, blank (TBA2-Cy5 only), BSA, lysozyme and VEGF and (b) corresponding microarray images. These results hence confirm that TBA1 and TBA2 used in the two-site binding assay ensure high recognition specificity and are able to discriminate between HBD and FBD from different proteins. 3.2. Effect of Incubation Time Assay incubation time was optimized in order to obtain the maximum fluorescent signal from analyzed samples and at the same time to perform the assay within a time compatible with routine analysis. After a pre-incubation step between the protein and the detection aptamer TBA2-Cy5, samples were incubated on the TBA1-printed microarray slide for a time ranging from 5 min up to 180 min. As shown in Figure 2 the fluorescent signal from TBA2-Cy5 increases exponentially over time, reaching a maximum intensity after 180 min, but the difference between the relative fluorescence intensity at 2 h and 3 h incubation time is lower than 6%. To achieve high assay sensitivity and to perform the assay in an overall time compatible with that of standard commercial assays, we performed all the following experiments using a 2 h incubation time. Figure 2 Effect of incubation time. Fluorescence signal (635 nm) recorded on TBA1 microarray (a) after the incubation of thrombin and TBA2-Cy5 for different times (180–120–90–60–30–15–5 min) and (b) corresponding microarray images. 3.3. Effect of Serum Immunoassays protocols require sample dilutions prior to analysis in order to dilute possible interferents [31]. To evaluate the system sensing ability in complex matrixes, equal thrombin amounts (500 nM) were analyzed in the presence of fetal bovine serum (FBS) at different dilutions ranging from 0% to 65%. After a short pre-incubation in the presence of FBS, samples were incubated for two hours on the array slide printed with TBA1. Collected data shown in Figure 3 demonstrate that the system can accurately work in the presence of 20% as well as 10% FBS without affecting the system performance. A 20% FBS solution, corresponding to a biological sample dilution of 1:5 is therefore compatible with the sandwich aptamer microarray analysis. Figure 3 Effect of serum. Fluorescence signal (635 nm) recorded on TBA1 microarray (a) after the incubation of thrombin and TBA2-Cy5 complexes in presence of different Fetal Bovine Serum (FBS) dilutions (0%–10%–20%–30%–40%–50%–65%) and (b) corresponding microarray images. 3.4. Limit of Detection of Direct and Indirect Methods The microarray developed and described is based on the “direct” detection of the labeled secondary aptamer recognizing the analyte-aptamer complex on the slide. This system was compared to an “indirect” method based on the formation of a similar sandwich-type on the microarray, but employing a secondary aptamer conjugated with biotin, followed by a detection step with fluorescently labeled streptavidin. Cy3-streptavidin has 3–9 fluorophores (Cy3) linked per molecule of protein: thus, the use of TBA2 biotin plus Cy3-streptavidin could potentially allow amplification of the signal reaching lower LOD and LOQ. This indirect method would also allow the development of a multiplex systems employing biotinylated aptamers specific for each analyte followed by a single detection step with labeled streptavidin. We set up and compared both methods for the thrombin microarray, and evaluated the assay limit of detection (LOD) and limit of quantification (LOQ) in both cases: different thrombin concentrations (200 nM–100 nM–50 nM–10 nM–5 nM–1 nM–0 nM) were pre-incubated with the two different secondary aptamers (respectively TBA2-Cy5 and TBA2-biotin) and finally laid on the TBA1 microarray slide. In the first (direct) case, the microarray is analyzed at 635 nm fluorescent excitation to detect Cy5 directly conjugated to the aptamer, in the second (“indirect”) method the microarray is analyzed at 532 nm fluorescent excitation to detect Cy3-labeled streptavidin laid over the treated samples. For each dilution, the mean (subtracted to background values) of six spots was calculated, with corresponding standard deviation. As shown in Figure 4, the direct method (red dots) employing TBA2-Cy5 yields a LOD of 0.75 nM and LOQ of 2.85 nM, corresponding to 27 ng/mL and 102.6 ng/mL of thrombin, respectively. In the indirect method with TBA2-biotin detected by fluorescent streptavidin (green dots), a LOD of 0.17 nM and LOQ of 0.97 nM were obtained, corresponding to 6.12 ng/mL and 34.92 ng/mL of thrombin, respectively. The indirect method allows therefore to obtain a more sensitive detection system and has been applied to detect thrombin in presence of FBS, as shown in the following Paragraph. Figure 4 Direct and indirect method for thrombin detection. (a) Fluorescence signal (635 or 532 nm) plot of thrombin aptamer microarray after the incubation of thrombin and TBA2-Cy5 (direct method, red squares, 635 nm) or thrombin and TBA2-biotin plus Cy3-streptavidin (indirect method, green dots, 532 nm) and (b) relative microarray images. Tested thrombin dilutions were: 200 nM–100 nM–50 nM–10 nM–5 nM–1 nM–0 nM. TBA2-Cy5 (red squares) or TBA2-biotin (green dots) concentration was 400 nM. For the indirect method the microarray slide was incubated for 1 h with Cy3-streptavidin, for a total incubation time of two hours in all cases. 3.5. Indirect Method for Aptasensor Detection: Effect of Serum To complete our characterization of the aptamer microarray for thrombin detection based on the indirect method of fluorescent analysis, samples with decreasing concentrations of thrombin were analyzed in solutions containing serum, in order to evaluate the sensing ability to detect thrombin in a complex matrix. Samples with decreasing concentrations of thrombin (50 nM–10 nM–5 nM–1 nM–0.5 nM–0 nM) and 20% FBS, were pre-incubated in presence of 200 nM of TBA2-biotin and finally incubated on the microarray slide as previously described. After Cy3-labeled streptavidin incubation, data were collected and reported in Figure 5. The presence of 20% serum contributes to lower the overall specific detection signal, as expected, but the LOD and LOQ, respectively 0.25 nM and 1.26 nM (corresponding to a thrombin concentration of 9 ng/mL and 45.36 ng/mL, respectively) allows sub-nanomolar detection of the analyte also in these conditions. Figure 5 Thrombin calibration curves in FBS solution. (a) fluorescence signal (532 nm) plot of TBA1 microarray in 20% FBS solution after the incubation of thrombin and TBA2-biotin plus Cy3-streptavidin and (b) corresponding microarray images. Tested thrombin dilutions were: 50 nM–10 nM–5 nM–1 nM–0.5 nM–0 nM, with a fixed TBA-2 biotin concentration of 200 nM. 4. Conclusions Analytical approaches based on biosensors for protein recognition and quantification have extensively been analyzed in the past few years [32,33]. The identification of disease-related biomolecules in body fluids is useful for the diagnosis of diseases as well as for follow-up protocols in clinical settings. Analytes can be identified with different transducing and recognition strategies, including antibodies, enzymes or aptamers: each different approach needs to be optimized in order to achieve increasing selectivity and specificity, and reduced assay time and costs. The data shown here demonstrate that the microarray based on aptamer recognition of an analyte is efficient and specific, and different detection aptamers and methods can be utilized. We have shown that the use of the indirect method (TBA2-biotin plus Cy3-streptavidin) allows lower LOD and LOQ when compared to TBA2-Cy5 (direct method) even in the presence of biological fluids. Moreover, the signal amplification made possible by the use of TBA2-biotin plus Cy3-streptavidin as detection strategy allows for a stronger signal at low protein concentrations, as shown in Figure 4. The TBA1 microarray exhibits a limit of detection comparable or superior to other systems described in the literature employing the same aptamers and different technologies [17,18,34] but has the advantage of a simple set up that could be exploited for the development of a multiplex microarray for protein and genome analysis in presence of a single detection strategy. Besides, the aptasensor could be further developed with signal amplification strategies by the use of fluorescent nanoparticles conjugated with streptavidin to allow for nanotech systems, to reach lower detection limits comparable to those achieved by different assay formats or technologies [23,25,35]. Acknowledgments This work was supported by the FU-PAT (Provincia Autonoma di Trento) Project NAOMI. B.G. thanks Progetti di Ricerca di Ateneo 2008, University of Padova for financial support. ==== Refs References 1. Fenton J.W. II. Thrombin specificity Ann. N. Y. Acad Sci. 1981 370 468 495 10.1111/j.1749-6632.1981.tb29757.x 7023326 2. Shuman M.A. Thrombin-cellular interactions Ann. N. Y. Acad Sci. 1986 485 228 239 10.1111/j.1749-6632.1986.tb34585.x 3032044 3. Shuman M.A. Majerus P.W. The measurement of thrombin in clotting blood by radioimmunoassay J. Clin. Invest. 1976 58 1249 1258 993343 4. Becker R.C. Spencer F.A. Thrombin: Structure, biochemistry, measurement, and status in clinical medicine J. Thrombosis Thrombolysis 1998 5 215 229 10.1023/A:1008843925851 5. Suo Z. Citron B.A. Festoff B.W. Thrombin: A potential proinflammatory mediator in neurotrauma and neurodegenerative disorders Curr. Drug Targets Inflamm. Allergy 2004 3 105 114 10.2174/1568010043483953 15032647 6. Turgeon V.L. Houenou L.J. The role of thrombin-like (serine) proteases in the development, plasticity and pathology of the nervous system Brain Res. Rev. 1997 25 85 95 9370052 7. Cho E.J. Lee J.W. Ellington A.D. Applications of aptamers as sensors Annu. Rev. Anal. Chem. (Palo Alto Calif.) 2009 2 241 264 20636061 8. Khati M. The future of aptamers in medicine J. Clin. Pathol. (Lond.) 2010 63 480 487 9. Famulok M. Mayer G. Aptamer modules as sensors and detectors Acc. Chem. Res. 2011 44 1349 1358 10.1021/ar2000293 21819040 10. Paborsky L.R. McCurdy S.N. Griffin L.C. Toole J.J. Leung L.L. The single-stranded DNA aptamer-binding site of human thrombin J. Biol. Chem. 1993 268 20808 20811 8407909 11. Tasset D.M. Kubik M.F. Steiner W. Oligonucleotide inhibitors of human thrombin that bind distinct epitopes J. Mol. Biol. 1997 272 688 698 10.1006/jmbi.1997.1275 9368651 12. Macaya R.F. Schultze P. Smith F.W. Roe J.A. Feigon J. Thrombin-binding DNA aptamer forms a unimolecular quadruplex structure in solution Proc. Natl. Acad. Sci. USA 1993 90 3745 3749 10.1073/pnas.90.8.3745 8475124 13. Ponikova S. Antalik M. Hianik T. A circular dichroism study of the stability of guanine quadruplexes of thrombin DNA aptamers at presence of K+ and Na+ ions Gen. Physiol. Biophys. 2008 27 271 277 19202200 14. Ikebukuro K. Kiyohara C. Sode K. Novel electrochemical sensor system for protein using the aptamers in sandwich manner Biosens. Bioelectron. 2005 20 2168 2172 10.1016/j.bios.2004.09.002 15741093 15. Centi S. Tombelli S. Minunni M. Mascini M. Aptamer-based detection of plasma proteins by an electrochemical assay coupled to magnetic beads Anal. Chem. 2007 79 1466 1473 17297945 16. Tang Q. Su X. Loh K.P. Surface plasmon resonance spectroscopy study of interfacial binding of thrombin to antithrombin DNA aptamers J. Colloid Interface Sci. 2007 315 99 106 10.1016/j.jcis.2007.06.040 17689549 17. Fang L. Lü Z. Wei H. Wang E. A electrochemiluminescence aptasensor for detection of thrombin incorporating the capture aptamer labeled with gold nanoparticles immobilized onto the thio-silanized ITO electrode Anal. Chim. Acta 2008 628 80 86 10.1016/j.aca.2008.08.041 18. Wang H. Liu Y. Liu C. Huang J. Yang P. Liu B. Microfluidic chip-based aptasensor for amplified electrochemical detection of human thrombin Electrochem. Comm. 2010 12 258 261 10.1016/j.elecom.2009.12.008 19. Zhao Q. Li X.-F. Shao Y. Le X.C. Aptamer-based affinity chromatographic assays for thrombin Anal. Chem. 2008 80 7586 7593 18759461 20. Edwards K.A. Wang Y. Baeumner A.J. Aptamer sandwich assays: Human α-thrombin detection using liposome enhancement Anal. Bioanal. Chem. 2010 398 2645 2654 20596697 21. Chen J. Zhang J. Li J. Yang H.-H. Fu F. Chen G. An ultrasensitive signal-on electrochemical aptasensor via target-induced conjunction of split aptamer fragments Biosens. Bioelectron. 2010 25 996 1000 10.1016/j.bios.2009.09.015 19818593 22. Tennico Y.H. Hutanu D. Koesdjojo M.T. Bartel C.M. Remcho V.T. On-chip aptamer-based sandwich assay for thrombin detection employing magnetic beads and quantum dots Anal. Chem. 2010 82 5591 5597 20545301 23. Lao Y.-H. Peck K. Chen L.-C. Enhancement of aptamer microarray sensitivity through spacer optimization and avidity effect Anal. Chem. 2009 81 1747 1754 10.1021/ac801285a 19193102 24. Zhao Q. Lu X. Yuan C.-G. Li X.-F. Le X.C. Aptamer-linked assay for thrombin using gold nanoparticle amplification and inductively coupled plasma-mass spectrometry detection Anal. Chem. 2009 81 7484 7489 10.1021/ac900961y 19670869 25. Niu S. Qu L. Zhang Q. Lin J. Fluorescence detection of thrombin using autocatalytic strand displacement cycle reaction and a dual-aptamer DNA sandwich assay Anal. Biochem. 2012 421 362 367 22033290 26. Wang L. Li L. Xu Y. Cheng G. He P. Fang Y. Simultaneously fluorescence detecting thrombin and lysozyme based on magnetic nanoparticle condensation Talanta 2009 79 557 561 10.1016/j.talanta.2009.05.034 19576412 27. Sosic A. Meneghello A. Cretaio E. Gatto B. Human thrombin detection through a sandwich aptamer microarray: Interaction analysis in solution and in solid phase Sensors 2011 11 9426 9441 10.3390/s111009426 22163703 28. Yang C. Wang Y. Marty J.-L. Yang X. Aptamer-based colorimetric biosensing of Ochratoxin A using unmodified gold nanoparticles indicator Biosens. Bioelectron. 2011 26 2724 2727 20970980 29. Sahni A. Francis C.W. Vascular endothelial growth factor binds to fibrinogen and fibrin and stimulates endothelial cell proliferation Blood 2000 96 3772 3778 11090059 30. Keyt B.A. Berleau L.T. Nguyen H.V. Chen H. Heinsohn H. Vandlen R. Ferrara N. The carboxyl-terminal domain (111–165) of vascular endothelial growth factor is critical for its mitogenic potency J. Biol. Chem. 1996 271 7788 7795 8631822 31. Hornbeck P. Winston S.E. Fuller S.A. Enzyme-linked immunosorbent assays (ELISA) Current Protocols in Molecular Biology John Wiley & Sons, Inc. Hoboken, NJ, USA 2001 32. Lim Y.C. Kouzani A.Z. Duan W. Aptasensors: A review J. Biomed. Nanotechnol. 2010 6 93 105 10.1166/jbn.2010.1103 20738063 33. Zeng X. Shen Z. Mernaugh R. Recombinant antibodies and their use in biosensors Anal. Bioanal. Chem. 2012 402 3027 3038 22159424 34. Huang D.W. Niu C.G. Qin P.Z. Ruan M. Zeng G.M. Time-resolved fluorescence aptamer-based sandwich assay for thrombin detection Talanta 2010 83 185 189 10.1016/j.talanta.2010.09.004 21035662 35. Chen Y. Nakamoto K. Niwa O. Corn R.M. On-chip synthesis of RNA aptamer microarrays for multiplexed protein biosensing with SPR imaging measurements Langmuir 2012 28 8281 8285 22458258

PMC005xxxxxx/PMC5003438.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays1020064microarrays-01-00064ReviewData Analysis Strategies for Protein Microarrays Díez Paula 1†Dasilva Noelia 1†González-González María 1Matarraz Sergio 1Casado-Vela Juan 2Orfao Alberto 1Fuentes Manuel 1*1 Centro de Investigación del Cáncer/IBMCC (USAL/CSIC), IBSAL, Departamento de Medicina and Servicio General de Citometría, University of Salamanca, Salamanca 37007, Spain; Email: pauladg@usal.es (P.D.); noeliadf@usal.es (N.D.); mariagg@usal.es (M.G.-G.); smats@usal.es (S.M.); orfao@usal.es (A.O.)2 Translational Oncology Unit, Instituto de Investigaciones Biomédicas ‘Alberto Sols’, Spanish National Research Council (CSIC-UAM), 28029 Madrid, Spain; Email: jcasado@iib.uam.es† These authors contributed equally to this work. * Author to whom correspondence should be addressed; Email: mfuentes@usal.es; Tel.: +34-923-294-811; Fax: +34-923-294-743.06 8 2012 9 2012 1 2 64 83 13 6 2012 13 7 2012 31 7 2012 © 2012 by the authors; licensee MDPI, Basel, Switzerland.2012This article is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Microarrays constitute a new platform which allows the discovery and characterization of proteins. According to different features, such as content, surface or detection system, there are many types of protein microarrays which can be applied for the identification of disease biomarkers and the characterization of protein expression patterns. However, the analysis and interpretation of the amount of information generated by microarrays remain a challenge. Further data analysis strategies are essential to obtain representative and reproducible results. Therefore, the experimental design is key, since the number of samples and dyes, among others aspects, would define the appropriate analysis method to be used. In this sense, several algorithms have been proposed so far to overcome analytical difficulties derived from fluorescence overlapping and/or background noise. Each kind of microarray is developed to fulfill a specific purpose. Therefore, the selection of appropriate analytical and data analysis strategies is crucial to achieve successful biological conclusions. In the present review, we focus on current algorithms and main strategies for data interpretation. microarrayproteomebiomarkeralgorithmnormalizationfluorescence intensitybackground correction ==== Body 1. Introduction The human proteome comprises ~23,000 protein-coding genes leading to >100,000 protein species mainly derived after alternative splicing and post-translational modifications (over a thousand post-translational modifications are currently listed in public databases). Thus, the overwhelming size and complexity of human proteome requires the development of high-throughput techniques enabling the detection of multiple proteins in a single analysis. Despite recent advances in proteomics technologies, only a small portion of human proteomes has been unraveled at the biochemical level. After the age of genomics, proteomics might seem to be a promising start to look deeper into the mechanisms of disease progression by generating individual protein expression profiles for every patient [1]. Consequently, proteomics may bring personalized medicine closer by implementing the five “Rs” criteria: right patient/target, right diagnosis, right treatment, right drug/target and right dose/time [2]. Therefore, the field of protein microarrays has expanded during the last decade, mainly due to the possibility of analyzing hundreds to thousands of proteins in a single experiment and in a high-throughput format [3]. Furthermore, high-density protein microarrays constitute a novel analytical tool that potentially will allow for biomarker discovery. The development and standardization of protein microarrays and automated data analysis of protein expression profiles might translate into a more accurate diagnostic and prognostic stratification of patients in clinical routine [4]. Despite the promising perspectives, the availability of multiple types of protein microarrays and the lack of standardized data analysis algorithms still hamper the widespread use of this technology. Here we provide an overview of the different types of protein microarrays and their application in order to address the human proteome and main data analysis methods currently available. 2. Concept of Protein Microarrays and Current Applications The idea of using microspots of antibodies printed on solid supports to develop more sensitive and quantitative assays was initially proposed by Roger Ekins in the late 1980s [5]. However, the project began to take shape in the late 1990s, with the introduction of DNA microarray technology. Indeed, Alejandro Zaffaroni and colleagues designed the first high-density microarray of peptides and oligonucleotides through photolithographic methods [6]. Briefly, protein microarrays are defined as miniaturized 2D arrays [7], which allow performing simultaneously high throughput studies of thousands of proteins [8,9,10] and subsequently the analysis of thousands of parameters within a single experiment. Moreover, it is also possible to compare two different samples labeled with two different fluorochromes on a single microarray [9,11,12]. In addition, this technology has been fruitfully employed in the identification, quantification and functional analysis of proteins in basic and applied research of proteomes [13,14], for example in antibody profiling or enzymatic studies [8,15,16]. During the last few years protein microarrays have provided the possibility to study protein-protein interactions and, to identify several biomarkers for different diseases [8,9] including neoplastic or autoimmune diseases, as well as to characterize targets for therapy protocols [9]. Indeed biomarker validation studies have been performed by protein microarrays. Certainly, several companies developed different microarrays formats, such as bead-based and planar microarrays [1,9,11,17]. It is noteworthy to highlight that microarrays have changed and implemented pharmaceutical research with continuously increasing diagnostic applications underlying multi-parametric measurements. In this sense, an additional advantage to be considered is the small amount of sample required for a single microarray, which is particularly favorable for patient management and, obviously when the sample amount is a limitation. On the other hand protein microarrays might be particularly relevant in disease monitoring and evaluation of minimal residual disease in the near future [11,12]. It is well known that a direct correlation between gene expression and protein abundance cannot be systematically established owing to post-translational protein modifications (e.g., glycosylation, phosphorylation and acetylation) [18]. For this reason, high-density protein microarrays remain necessary as they offer multiple improvements over conventional techniques, including better resolution, selectivity and sensitivity [19,20]. 2.1. Types of Protein Microarrays According to their applications, the planar protein microarrays have been classified in three main categories: analytical microarrays, reverse phase arrays (RPA) and functional microarrays (Figure 1) [7,14,21]. On the other hand, microspheres bead based systems should also be considered, which use different size or color beads as a support of the capture agent to analyze the sample. In such microarray format, flow cytometry is coupled in order to support the identification of each specific binding according to the size, color and mean fluorescence intensity of conjugated fluorochromes [14]. Figure 1 Types of different microarrays. (a) Capture arrays. (b) Cell-based protein microarrays. (c) Reverse phase arrays. (d) Cell-free nucleic acid programmable protein array. 2.1.1. Analytical Microarrays A range of capture agents, differing in composition, origin and, thus, differing in their affinity properties may be attached on microarrays [9]. Both, antigens or antibodies may be immobilized on the surface of slices and used as baits present in the test sample [7,14]. This kind of microarray is used to determine parameters such as the binding affinity and specificity and to study protein expression levels in complex mixtures [7,22], but also they cover clinical applications such as studies in immunology or biomarkers detection [9] and they can be used to monitor differential expression profiles, such as protein patterns in response to environmental stress or differences among a healthy tissue and with respect to a pathological sample [22]. In addition, analytical microarrays imply direct labeling protocols of thousands of proteins, which might be another critical limitation. The chemical labeling of proteins can destroy epitopes by covalent combination of dyes or haptens. Moreover, only selected target proteins can be analyzed by antibody microarrays [9,11,14]. 2.1.2. Reverse Phase Arrays In this case, cellular or tissue lysate or even serum samples are immobilized on the microarray surface and the detection is completed through an antibody against the target proteins. To achieve a higher fluorescent signal to be detected, a secondary antibody conjugated with a fluorochrome is added to the first one. This ensures the signal intensity is directly related with the specificity, the binding affinity and the sterical accessibility of the antibody against the target protein [7,9,14]. The production of a functional map for cell signaling pathways from individual cells or tissues by RPA arrays has increased the interest on this kind of arrays with the objective of developing personalized therapies [7,9,23]. The proteins involved in RPA do not require labeling and only a little amount of protein is needed to produce the microarrays. However, fewer analytes can be analyzed due to the limited number of labeled antibodies for detection and also low availability of specific protein antibodies suitable for RPAs [7,10]. 2.1.3. Functional Microarrays Functional microarrays are composed of full length functional proteins or protein domains and study the biochemical characteristics and functions of native proteins, as well as peptides or domains highly purified through cell-based methods or by cell-free expression on the microarray [9,11,21,22]. They allow studying the whole proteome in a single assay. Functional microarrays are also employed to examine the diverse protein interactions: protein-protein, protein-DNA, protein-RNA, protein-phospholipids and protein-small molecules [22]. In situ expressed microarrays, one of the subtypes of functional microarrays, are based on cell-free expression systems such as Escherichia coli 30 s, rabbit reticulocyte lysates (RRL) and wheat germ extracts [9], which have to be very well purified throughout chromatography or electrophoresis [7]. A library of open reading frames is also required [9,21,24]. Cell-free based protein microarrays have been applied to immunological studies [25], vaccine development [26,27], early detection of biomarkers [27,28], biochemical activity [21] protein-protein interaction studies [10,28], such as protein-protein, protein-DNA, protein-RNA, protein-phospholipids, and protein-small molecule interactions [22] and toxin detection [29,30]. Over the last few years, several in situ expressed microarrays have been developed such as: Protein in situ arrays (PISA), printing protein arrays from DNA (DAPA) arrays and Nucleic Acids Programmable Protein Arrays (NAPPA) [9,11,17]. NAPPA is one of the most relevant microarrays in this field. The DNA templates are bound onto the slide surface; the protein of interest is encoded and a GST tag is added. This is a fusion protein with a tag, which will allow binding to the slide. The biotinylated DNA plasmid is attached through an avidin to the aminopropyltriethoxysilane (APTES)-coated surface. In addition, RRL is used to carry out protein expression. There are also anti-GST antibodies attached to the slide, where the fusion protein joins. As a result, an array with the expressed protein and its corresponding DNA is achieved all on the same slide [8]. NAPPA is a good cost-effective technique because of the small volume of cell-free extract required for protein expression. Also, the use of immobilized DNA allows storage of the array for a long time until the next procedure. The main drawback is the invested time to generate the cDNA with the protein of interest and the tag, but even this method does not achieve a pure protein. On the other hand, high yields of high quality DNA were obtained for immobilization by using a diamine-derivatized resin. It was also found that BSA improved the binding efficiency of DNA and that is why a master mix of cDNA, antibody, BS3 and BSA is used [8]. 2.2. Current Application of Protein Microarrays Protein microarray technology has been successfully applied in different biomedical areas (Table 1). microarrays-01-00064-t001_Table 1Table 1 Current applications of protein microarrays. Disease Type of microarray Object of study Reference Cancer multiplexed array CA-125; CA19-9; EGFR; C-protein; myoglobin; APOA1; APOC3; MIP-1; IL6; IL18; tenascin-C Amonkar et al. 2009 NAPPA p53 Dasilva et al. 2012 Nodular thyroid disease protein array EGF; HGF; IL5; IL8; RANTES Linkov et al. 2008 multiplexed array cytokines; growth factors; cell adhesion molecules Xiaobo et al. 2010 reverse phase array Salmonella typhimurium Cid et al. 2011 Infectious disease antigen microarray Vaccinia virus; Yersinia pestis Natesan et al. 2010 antibody array cholera; diphtheria; staphylococcal enterotoxin B; tetanus toxin; anthrax protective antigen Rucker et al. 2005 protein array B lymphocyte Wadia et al. 2011; Belov et al. 2001 Systematic rheumatic disease antibody microarray nuclear proteins; nucleoprotein complexes Dolores et al. 2001 Diabetes (type I) NAPPA Sibani et al. 2011 2.2.1. Cancer One of the most relevant applications of microarrays is the detection of biomarkers for many different diseases, including cancer, where the importance of an early detection is fundamental [11]. One example is the use NAPPA to address the detection of p53 auto-antibodies present in sera from breast cancer patients. Since the occurrence of p53 auto-antibodies is directly related to the tumor burden, detection of such auto-antibodies may lead to the recommendation of neo-adjuvant chemotherapy [30]. Also, capture microarrays have been used by Sreekumar et al. to monitor changes in protein abundance in colon carcinoma cells following exposure to ionizing radiation [31,32]. Amonkar et al. identified an 11-plex protein panel including: CA-125, CA 19-9, soluble epidermal growth factor receptor (EGFR), C-reactive protein, myoglobin, apolipoprotein A1 (APOA1), APOC3, macrophage inhibitory protein 1 (MIP-1), interleukin-6 and 18 (IL6 and IL18), and tenascin C. These authors built microarrays bearing this panel of proteins to analyze plasma samples and discussed their applicability to distinguish patients with ovarian cancer from controls [32]. On the other hand, over the last few years it has given relevance to the activity of protein kinases mainly because of its decisive function in the cell and the generation of specific treatment against them. With the growing knowledge generated about cancer biology, the new generation target drugs have different action mechanisms being more specific and causing less secondary damages for the patient. However, they have brought several challenges such as biomarker identification and developing technology to assure the drug carries out the desired function. Once these difficulties have been overcome, it is necessary to establish whether the patient outcome could be defined in function of kinase expression profile or based on the treatment. Eventually, patient monitoring determines the most accurate treatment to achieve the best prognostic. Also, it is necessary to identify the correct treatment responsive population for the precise therapy and avoid non-responsive patient groups. This was achieved for inhibitors of ERK/MAPK pathway, but a great effort still needs to be made to reveal other successful therapies [33]. 2.2.2. Immunology Protein microarrays are also a valuable tool for the development of vaccines [11]. As an example of this application, Belov et al. developed an array of the expression of ninety different clusters of differentiation (CD) antigens from different leukemia cells. The slide was incubated with a cellular suspension and the cells with the matching CD bound to the corresponding antibody. As a result, they achieved different cell patterns when a control sera or a pathological sample was used, so each leukemia patient can be specifically characterized [14,34,35]. Systematic rheumatic disease, an autoimmune disease, can be accurately diagnosed thanks to the presence of nuclear proteins and nucleoprotein complexes which can be used as targets for antibodies in order to detect the phase of the disease [20]. NAPPA represents a good alternative to detect biomarkers in autoimmune diseases, such as systemic rheumatic disease, type I diabetes or systemic lupus among others, as has been demonstrated in different studies [36]. 2.2.3. Nodular Thyroid Disease Nodular thyroid disease also can be analyzed thanks to protein arrays. Linkov et al. have defined patterns of serum/plasma biomarkers. In this way, epithelial growth factor (EGF), hepatocyte growth factor (HGF), IL5, IL8 and C-C motif chemokine 5 (CCL5, also known as RANTES) can be used as protein biomarkers of disease states [37]. Also, thyroid function can be analyzed, along with other analytes, such as cytokines, growth factors and cell adhesion molecules, on the same surface, with a microarray system which uses chemiluminescence as a readout signal. 2.2.4. Infectious Disease Understanding how pathogens are capable of modifying cell pathways can be a good strategy in order to establish preventive and therapeutic interventions for infectious diseases. Although there are not many relevant studies, RPA seems to be a good choice. In this way, Cid et al. studied the potential of RPA technology through cell cultures infected by Salmonella typhimurium [38]. In these kind of arrays, cell lysates are printed to a solid support so as to be analyzed by quantitative immunodetection. The base of the analysis consists of detecting the presence of proteins or their post-translational modifications in cell lines after exposure to pathogens under different conditions. Specifically, Cid et al. studied T3SS effectors in infected HeLa cells and have demonstrated that RPA allow quantifying the relative amounts of each marker protein in the samples [38]. Natesan et al. developed an antigen microarray to establish a signature of pathogen proteins displayed by Vaccinia virus and Yersinia pestis. The proteome microarray was adapted from a functional microarray in which genomic DNA of the vaccine strain was used as the template for PCR amplification. Then, recombinant viral proteins were generated as GST-tagged fusions. Later, GST-tagged proteins were purified to ≥90% homogeneity using affinity chromatography. Both, viral and control proteins were printed onto glass slides coated with a thin layer of nitrocellulose. Nearly 95% of the viral proteome were successfully expressed, purified and arrayed [18]. 3. Feature Aspects of Protein Microarrays Three main features of protein microarrays need to be considered during their design, such as type of surface used, molecules bound onto the surface and detection techniques selected. Depending on the type of array, the characteristics cited above can vary. For example, a protein microarray is not the same as an aptamer array. Thus, it is necessary to take into account the different designs when an array is developed, since the results may be contrary to all expectations, simply by choosing an inappropriate surface or detection format. The reasons specified above justify the need for careful selection of a number of features, as detailed in this section. 3.1. Array Capture Agents Regarding planar arrays, capture agents are immobilized in rows and columns creating a set of spots onto the microarray ready to be exposed to the test sample. The binding between the capture agent and the analyte can be detected by label or label-free detection methods such as fluorescence, chemiluminescence, mass spectrometry, radioactivity or electrochemistry [9,11,12]. Although there are several different capture agents, antibodies are the most common. However, new approximations are being developed, e.g. phage display or mRNA display are the most promising techniques, but also, highly specific oligonucleotides and aptamers (Table 2) [12]. Employing monoclonal antibodies in routine array production incurs excessive costs due to the hybridoma technology and their production. Therefore, phage display is an appropriate substitute according to the easier and cheaper methods to obtain and purify the antibody fragments (Table 2) [23,39,40]. microarrays-01-00064-t002_Table 2Table 2 List of capture agents used in protein arrays, source and technique. Capture agent Source of proteins Technique Mab * mouse Hybridoma sc Fv */Fab * diabodies antibody libraries Phage display, in vitro evolution Affinity binding agents recombinant fibronectin structures In vitro evolution Affibodies Aptamers (DNA, RNA, peptide) Receptors ligands synthetic Combinatorial chemistry Substrates of enzymes synthetic; pro-and eukaryotic organisms Protein purification, recombinant protein technology(bacterial, fusion proteins, baculovirus, peptide synthesis) * Abbreviations: Fab, antigen-binding fragment; sc Fv, single-chain variable region fragment; Mab, monoclonal antibody. On the other hand, highly purified recombinant proteins are needed to satisfy the huge demand for capture agents. Because of that, new high-throughput techniques have to be developed, as well as, improving the existence ones. Moreover, these specific proteins allow generating microarrays which are analyzed faster, with high affinity bindings, permitting more efficient screening and also, avoiding or, at least decreasing, the cross-reactivity [12,17]. 3.2. Array Surfaces The slides used to immobilize the capture agent are usually made of glass, but can also be made of plastic, metal or polymer membranes [12,41] Since microarrays were first developed, surfaces have evolved from polyvinylidene fluoride (PVDF) membranes to glass slides. The most significant challenge is obtaining a surface which can take a variety of protein structures and compositions while preserving the function, structure and binding of each single protein. Immobilization is important, but also the capability of accurately detecting the protein-protein interaction. That is why three types of surfaces have been developed, each with a specific immobilization protocol. The first category is two-dimensional (2D) plain glass slides, which are activated with a diversity of coupling chemistries such as aldehyde, epoxy or carboxylic esters and they bind proteins or antibodies through electrostatic interactions or the generation of covalent bonds. The strength of the binding and the low variation enables rapid evaporation of the liquid environment and it is suspected that this affects the three-dimensional structure due to the close protein surface contact. On the other hand, there are three-dimensional (3D) gels or membrane-coated surfaces, for instance, polyacrylamide, agarose and nitrocellulose. They bind the protein by adsorption and seem to better safeguard the initial conformation; nevertheless, they present variations in signal intensity. Finally, the third type is a mixture of the previous ones: showing a supra-molecular structure at the surface which is why it is not a 2D slide, nor has it a visible 3D structure [39]. The immobilization of proteins is usually completed using non-covalent protein surface interactions with hydrophobic (e.g., nitrocellulose and polystyrene) or positively charged (e.g., poly-lysine and aminosilane) surfaces [12]. This kind of method attachment determines a random orientation of the proteins onto the slide because of the passive adsorption [22]. Covalent attachment achieves employing chemically activated surfaces (e.g., aldehyde, epoxy or ester functional groups), also the attachment is achieved by specific bimolecular interactions (e.g., streptavidin-biotin, His-tag-nickel-chelates) [12]. The uniform orientation of the different proteins onto the chip surface can be achieved using nickel coated slides for the His-tag use or Streptavidin slides [22]. The tiny microspots on the slide surface are made using contact printing arrayers with tiny needles placing sub-nanoliter sample volumes directly on the surface. However, non-contact deposition technologies are used that apply capillaries or ink-jet technology to deposit nanoliter-picoliter droplets onto the surface [12]. Immobilizing a protein onto a slide necessitates preservation of the conformation and the function of the protein, as well as, the binding capacity [22]. Finally, it is worth mentioning that thanks to flow cytometry, a new type of array has been developed, called the color-coded microsphere array. Each microsphere is covered with different antibodies, and each antibody can be detected through different fluorochromes; thereby, monitoring the binding antibody-antigen. The latter is attained from the sample of interest, can be fast, accurate, low cost and highly-sensitive. A new system of magnetic microspheres has been designed recently with high-reproducibility and sensibility, low background noise and price, and having a wide dynamic range [9,11]. In summary, a color-coded suspension microsphere array presents the following advantages over current antibody arrays: (i) High level of multiplexing. (ii) A sensitive, accurate and wide dynamic range signal readout. (iii) Information about protein-protein interactions. (iv) Flexibility in array composition. (v) Automatic preprocessing (gating, QC) in a short time. (vi) Clinical applications. All due to the integration of flow cytometry, antibody array detection of size-resolved lysates and computer-assisted data processing [42]. 3.3. Array Detection Technologies Many different detection techniques were developed over time that enable reliable, sensitive, and specific detection of arrays in a high-throughput manner. Some of them are based label tools and others are label-free techniques. The first type involved a fluorochrome or radioisotope tag molecule for the query element [7,9,11,39] or some new tags recently have been developed, such as quantum dots (QDs), gold nanoparticles (NPs), Raman-dye label carbon nanotubes or silica NPs [7,11]. However, these techniques can interfere with the probe’s capacity to bind to the target protein [22]. To solve these problems, label-free techniques have been developed. They include surface plasmon resonance (SPR), carbon nanotubes, microelectromechanical cantilevers and surface-enhanced laser desorption ionization (SELDI)-TOF-MS. These techniques can measure the mass, dielectric or optical properties of the query molecule [9,11,22]. Currently, only SPR is a label-free technique generally available in many laboratories. Although the other tools need to be developed to become suitable high-throughput techniques, they show huge potential in protein microarrays [9]. The fluorochrome labeling technique, directly incorporated from the DNA-chip technology, has become widely used due to its effectiveness and its compatibility with different systems of laser scanning [9,22]. Both, antibodies and antigens are labeled with two different fluorochromes, mixed and concurrently incubated on the same array. Then, dual color detection systems allow the detection of both signals and measure their ratios. Finally, the analysis of data reveals whether the targets are present in different or similar concentrations on each spot (Figure 2) [12,14] and allowing differentiate between two separate samples [12,39]. The fluorochrome is bound to the antibody or to antigen depending on the type of array [12,14]. A key advantage is the possibility of comparing two samples without the need of a second independent array, since both assays are performed on the same slide [39]. Figure 2 Microarrays for differential protein displays. Proteins from controls and samples are isolated and conjugated to different fluorescent molecules, for example Cy3 and Cy5. The samples are mixed in equal amounts and incubated simultaneously on an antibody microarray. Then, target molecules will be captured by their specific antibody and the differences in protein expression are directly reflected by the overlay of the color signal. The section marked with a yellow circle reflects the difficulty of quantification because a prominent signal could be the result of a single protein, but also of a large protein complex. Achieving a more powerful signal is possible using indirect labeling, which is the most common system with serum samples, antibodies and in sandwich assays. However, cross-reactivity can be caused; hence the assortment of antibodies able to be employed is limited [39]. Anyway, these problems can be solved using different strategies, such as labeling procedures based on enzymatic signal amplification. In this way, Schweitzer et al. developed a method which requires enzymatic extension. This can be done through rolling circle amplification (RCA) which is based on the enzymatic extension of a primer-antibody conjugate followed by hybridization of labeled probes [43]. Also, tyramide signal amplification (TSA) system can be utilized [44]. On the other hand, Huang et al. have employed chemiluminescence in order to achieve the required purpose [45]. However, sometimes, these strategies are not sufficient to solve the cross-reactivity problems. When these problems appear, together with others, data analysis is the tool which allows obtaining correct conclusions from experiments, by taking the problems derived from the hybridization method into account and solving them. However, there are no simple and uniform strategies for the analysis of data obtained by different kinds of protein microarrays. Thus, the development of user-friendly and accurate algorithms still needs to be developed. 4. Data Analysis Methods The wealth of information generated by protein microarrays may provide solid evidence concerning protein functions, their interactions and even their involvement in signaling pathways. Interestingly, these data can also be applicable as a tool for clinical diagnostics. Nevertheless, the translation of data into meaningful information requires automated data processing and handling. As depicted in Figure 3, data processing and analysis is inherent to protein arrays. Thus, it becomes a crucial step in the search for solid biological conclusions [46]. There are several strategies to analyze protein data, some of which have their origin in DNA microarray analysis, such as spot-finding on slide images, Z-score calculations and significance analysis of microarrays (SAM). However, concentration dependent analysis (CDA) has been specifically developed for protein microarrays [46]. Figure 3 Workflow of protein microarray development in which the sample of interest, the type of microarray and the data analysis strategy are essential for biological conclusions. • Spot intensity determination: microarray image analysis starts with the fixing of spot intensity. Generally, for this task, GenePix Pro software (Molecular Devices, Union City, CA) is used. First of all, a grid of circles must be placed over the protein spots. Their position and size have to be adjusted in order to get reliable intensity data. Finally, an output file is created by the program. • Z-score analysis: the Z-score equation, , where Zs is the Z-score for the sth spot, Ss is the signal for that spot, µ is the mean signal across all spots and σ is the standard deviation across all spots, is an interesting tool to determine which signals are significantly different from the expected value and which are not. • Concentration-dependent analysis (CDA): due to the quantity of spotted proteins on the slide, absolute signals are affected by protein concentration. As a means to solve this issue, a different Z-score, , can be calculated to remove outliers. This novel Z-score is calculated using an iteration process that is repeated until every spot signal measured is in accordance with the mean value. In the equation above, Zs is the Z-score for the sth spot, Ss is the signal for that spot, µw is the mean signal for the spots within the window and σw is the standard deviation for spots within the window. 4.1. Dual-Color vs. Single-Color Assays Antibodies immobilized on the microarray surface can be detected through direct labeling, requiring only a single capture antibody specific for each target protein. Alternatively, a sandwich approach can be carried out, which consists of two sets of antibodies, the first one is specific for the target protein, and the second one for the first antibody [47]. Then, the signal is detected by a colorimetric reaction or a fluorescent dye. This last alternative enables a dual-color layout that is based on labeling each sample with different fluorescent dyes (e.g., Cy3 and Cy5), which competes for the binding sites of the antibodies immobilized on the array. After the incubation, intensity signals are measured for each dye using fluorescence image scanners. Dual-color assays typically display better reproducibility and discriminative power compared to single-color assays [47]. * Single-color assays: Olle et al. [47] developed a single antibody-based microarray which presents standard antigen concentration. Also, it uses an internal controlled system based on two colors, one for the amount of antibody spotted and the other for the amount of the antigen used for the quantification of the level of protein expression. To validate this microarray, levels of protein expression were compared with results obtained by western blot analysis and the data were similar, although the sensitivity was higher with the microarray. In their study, they show that this microarray has not only the potential to accurately assess proteins in complex fluids, but also a large range of linearity. * Dual-color assays: Data pre-processing protocols are usually applied to prevent undesired technical artifacts. These protocols frequently include the following steps (adapted from [48]): • Filtering, in order to remove failed and low-quality spots. • Background correction, to avoid fluorescence signal due to non-specific binding. • Data normalization, aimed to reduce variations between the two samples co-hybridized on each array and also between arrays. All the steps mentioned above are commonly used in protein arrays due to the difficulty of quantifying protein expression in a multiplexed manner. Multiple causes may lead to the occurrence of such artifacts, including the effect of electric charges, hydrophobic interaction of proteins, artifacts due to differences in protein sizes and antibody/antigen binding kinetics. For all these reasons, microarray data frequently require normalization. Several methods can be used for data normalization, such as: (i) house-keeping probes, (ii) inclusion of spike-in controls and (iii) use of algorithms to define sets of probes. In the following, these methods are described [48]. Different microarray designs may be considered (Table 3, adapted from [48]) which differ in the number of samples and type of array employed, as described below: • Reference design: the sample of interest is labeled with one fluorescent fluorochrome (e.g., Cy3), whereas the single reference sample is labeled with a different fluorescent fluorochrome (e.g., Cy5). In this type of design, it is necessary to calculate the log ratio of dyes intensities. • Balanced-block design: two samples which are hybridized, bearing two different fluorochromes (Cy3 and Cy5). Then, samples are balanced with respect to dyes. In this case, the microarray is considered as a block. • Incomplete-block design: more than two samples are co-hybridized on the microarray, whereas only two fluorochromes are used (Cy3 and Cy5). Despite this, samples are balanced. • Loop design: each sample is hybridized in a different array using a different fluorochrome. This supposes a great disadvantage because the number of arrays is duplicated. Balanced-block and loop designs allow correct dye effect normalization, which are required for the correct analysis of two-dye systems. The first one is used for comparison studies, whereas the loop design is useful for discovery studies [48]. The table below shows the relations between samples and dyes used in the experimental designs indicated above [49]. microarrays-01-00064-t003_Table 3Table 3 Types of microarray experiment designs using two colors. Ai: sample i from class A; Bi: sample i from class B; Ci: sample i from class C; R: reference sample. In reference design, Ai and Bi are labeled with Cy5, whereas R is labeled with Cy3. In the rest of the designs proposed, each class is labeled with a dye, typically Cy5 and Cy3 [49]. EXPERIMENTAL DESIGN ARRAY #1 ARRAY #2 ARRAY #3 ARRAY #4 Reference A1/R A2/R B1/R B2/R Balance block A1/B1 B2/A2 Incomplete block A1/B1 B2/C1 C2/A2 Loop A1/B1 B1/A2 A2/B2 B2/A1 4.1.1. Rank-Invariant Selection Algorithm (InvTseng) This algorithm is especially valuable in those cases where house-keeping controls are not available. Tseng et al. [50] suggested a strategy which enables selecting a set of non-differentially expressed proteins. This method, applied to dual-color arrays, is an adaptation of the invariant difference selection algorithm (IDS) used with single-channel microarrays. A protein p is considered to be rank-invariant on an array, if the difference of the ranked Cy5 and Cy3 intensities is less than a threshold d and the average of the ranked intensities is not among the highest or lowest l ranks. For each array j, the set of rank-invariant proteins is determined by the following expression [50]: where r(Cy5jp) and r(Cy3jp) are the ranks of the intensities and G is the number of spotted proteins. It has to be noted that a major limitation of the InvTseng algorithm proposed by Tseng et al. [50] is that it does not cover the entire intensity range. 4.1.2. Modified Rank-Invariant Selection Algorithm (In-vMod) The intensity range limitation mentioned above can be partially solved using the modified rank-invariant selection algorithm (In-vMod) [50] depicted as follows: The In-vMod algorithm corrects the intensity values through the extrapolation of the curve to the lower and upper intensity limits. 4.1.3. Rank Difference Weighted Global Loess (RDWGL) This next algorithm is applied to the whole probes on the array (wjg) to get a global normalization [50]. where Δig = |r(Cy5jg) − r(Cy3jg)| is the absolute difference of the ranked intensities of protein g on array j. max refers to the maximum value. On the other hand, several standard normalization methods are used in order to correct the background of the slides including: (i) global loess normalization (GL); (ii) variance stabilizing normalization (VSN) and (iii) generalized procrustes analysis (GPA). GL fits a non-linear loess curve in which equal weight is assigned to all probes. VSN is based on the stabilizing of the variance of the transformed intensities to be approximately independent of the mean intensities. Finally, GPA scales and aligns matrices with the same dimension as a mean to normalize data [50]. 4.2. Automated Analysis of Highly Complex Flow Cytometry Stuchlý et al. developed a protein-profiling tool which allows computational feasibility, hands-on time, standardization and reproducibility, quality control feedback loops, data normalization and presentation of the results in an appropriate way to be analyzed [42]. Size exclusion chromatography-resolved microsphere-based affinity proteomics (Size-MAP) is a new tool that permits obtaining information about protein sizes, protein complexes and protein profile changes through flow cytometry detection. The statistical method used was based on modifications of Partitioning Around Medoids (PAM). Previously, they have tried to resolve individual color-coded microsphere types by using k-means clustering, model-based clustering, minimum spanning tree clustering, and hierarchical clustering. But, finally, PAM was seen as the best method. In this way, they adopted the standard approach of sequential two-dimensional gating. Kernel density was used to analyze the distribution of fluorescent signal [42]. • Automated gating of color-coded microspheres: the automated tool is responsible for the specific identification and differentiation among microsphere types and the consequent allocation of the code to each one. • Analysis of size-MAP data: quantification of antibody-bound proteins amounts was determined with the medians of the fluorescence label signals. Next, these data will be processed through quality control (QC), normalization and analysis. ○ Quality control (QC): first of all, the number of microspheres of each population is checked. Next, the density function of the signal is also determined. ○ Normalization: it is necessary to remove background noise and to establish protein sample differences. With the purpose of correcting the noise, the signal, from empty microspheres (those without any antibody), is subtracted from the signal of the microsphere population of interest. ○ Analysis: each protein entity has to be established and, for this purpose, fractions constituting specific protein entities must be defined. Then, signals for each fraction are summed up, representing the final result, which is the relative amount of a particular protein entity. 4.3. Data Analysis Methods from cDNA Arrays Initially, in order to analyze antibody microarrays, strategies from cDNA arrays were implemented. Fluorescent dyes (e.g., Cy3 and Cy5) were used to label each sample and, then, log ratio of the intensities was calculated for every feature on the array. Nevertheless, this design was not the optimal one [48]. Therefore, two new methods were developed: balanced-block design and loop design. The first technique considers the arrays as a block, in such a way that samples hybridized are balanced with respect to dyes. In the second strategy, each sample is hybridized onto two different arrays, each with a different dye. Both designs allow suitable dye normalization. However, using two arrays for each sample supposes a considerable disadvantage [48]. The normalization is based on an internally normalized ratio (INR). A reference sample is labeled with fluorescent dye (e.g., Cy5), whereas the sample of interest is labeled with another dye (e.g., Cy3). In other arrays, the same samples are labeled with the opposite dyes (Cy5 for the sample of interest and Cy3 for the reference sample). Then, the ratio of both fluorescent dyes is calculated for each array. The INR is the geometric mean of the two ratios [48]. Also, ANOVA models, mixed ANOVA models and within-print tip local regression smoothing methods have been developed for the removal of systemic effects typically found in cDNA arrays [48]. 4.4. Reverse Phase Array Data Analysis RPA analysis is based on the construction of a serial dilution curve, which is characterized by two main advantages [51]: first, the signals in successive dilutions can be related to each other. In this way, protein concentration and signal intensity can be accurately established. Second, data quality can be checked due to raw data display. In RPA analysis several steps must be carried out, as follows: • Serial dilution curve: the monotonic s-shaped response curve is described by Sips model: In the algorithm displayed above, a corresponds to the background noise; b is the response rate in the linear range; M is the maximum or saturation level and x is the concentration of the protein. Also, the equation can be modified in order to avoid data about protein concentration. In this way, y can vary. Generally, y≠ 1 applies to condition in which there is some heterogeneity in the solute molecules or the surface receptors. y approaches to 1 when the range of the free energy of binding shrinks to a singular point. In this last case, the equation is equivalent to the conventional Langmuir model. • Parameterization of the serial dilution curve: a non-linear regression model is used to find the optimal parameters. • Estimating protein concentrations: first of all it is necessary to check if protein concentration is saturated. This occurs if the M/r ratio is lower than signals measured. Then, the minimum and maximum of x (xmin and xmax, respectively) are estimated with these formulas: where M/r is a threshold value in which M is the saturation level and r should be >1. K refers to the Kth dilution step. In summary, the Sips model presents physically meaningful parameters and has the optimal conditions for RPA experiments [51]. 5. Conclusions Microarray analysis includes four main steps which must be followed, such as design (surfaces, content, detection method), data preprocessing, inference, classification and validation. All these variables may significantly differ depending on the kind of microarray used. Therefore, it is important to bear in mind that different data analysis methods are also required [52]. Array design is crucial and may drastically affect data analysis. For that reason, careful design of microarrays is required, since it may significantly influence data analysis and final interpretation. It is recommendable (if not compulsory) that biological replicates are included in the microarrays, providing greater statistical confidence. Nevertheless, the introduction of replicates necessarily introduces more data to be analyzed, adding more complexity to the evaluation of the results [52]. Data preprocessing (i.e., image analysis, normalization and data transformation) is the second step. Image analysis is made using image-processing algorithms that distinguish foreground from background intensities. To date, it is not known which method is the best for this purpose [52]. Inference is based on statistical strategies, which also incorporate variability in the analysis [52]. Classification and validation: Not only the large amount of data that are generated, but also the wide variety of results obtained which can be generated according to the type of array, are the reasons that explain why different data analyses are required. Since antibody microarrays and phage display arrays are different, the results obtained also differ [52]. This fact offers some advantages but also some disadvantages. On the one hand, the development of diverse types of arrays provides a variety of tools that enable disease analysis from multiple perspectives. Nevertheless, the main drawback is that there is a lack of standard analytical strategies, including array data processing. Briefly, a range of diseases are studied using different types of arrays, which are analyzed following different strategies. This introduces complexity in the analysis and the need of data analysis strategies based on different algorithms, image processing or validation methods [52]. Despite the differences among data processing methods applicable to microarray analysis, several general recommendations need to be considered [52] as follows: (a) using Bayesian approaches to examine intersections between sets of findings and evaluate multi-component hypotheses; (b) quality-control and validation methods are required; and (c) standardized testing platforms are needed. Acknowledgements We gratefully acknowledge financial support from the Carlos III Health Institute of Spain (ISCIII, FIS PI1102114) and JCYL-SAN10. María González-González is supported by a ISCIII FIS08/00721 Ph.D. scholarship. ==== Refs References 1. Yu X. Schneiderhan-Marra N. Joos T.O. Protein microarrays and personalized medicine Ann. Biol. Clin. (Paris) 2011 69 17 29 21463992 2. Dasgupta A. Handbook of Drug Monitoring Methods: Therapeutics and Drugs of Abuse Humana Press New York, NY, USA 2007 5400 5411 3. Merbl Y. Kirschner M.W. Protein microarrays for genome-wide posttranslational modification analysis Wiley Interdiscip. Rev. Syst. Biol. Med. 2011 3 347 356 10.1002/wsbm.120 20865779 4. Hanash S. Disease proteomics Nature 2003 422 226 232 10.1038/nature01514 12634796 5. MacBeath G. Protein microarrays and proteomics Nat. Genet. 2002 32 526 532 10.1038/ng1037 12454649 6. Fodor S.P. Read J.L. Pirrung M.C. Stryer L. Lu A.T. Solas D. Light-directed, spatially addressable parallel chemical synthesis Science 1991 251 767 773 1990438 7. Chandra H. Reddy P.J. Srivastava S. Protein microarrays and novel detection platforms Expert Rev. Proteomics 2011 8 61 79 10.1586/epr.10.99 21329428 8. Chandra H. Srivastava S. Cell-free synthesis-based protein microarrays and their applications Proteomics 2010 10 717 730 10.1002/pmic.200900462 19953547 9. Gonzalez-Gonzalez M. Jara-Acevedo R. Matarraz S. Jara-Acevedo M. Paradinas S. Sayagues J.M. Orfao A. Fuentes M. Nanotechniques in proteomics: Protein microarrays and novel detection platforms Eur. J. Pharm. Sci. 2012 45 499 506 10.1016/j.ejps.2011.07.009 21803154 10. Hultschig C. Kreutzberger J. Seitz H. Konthur Z. Bussow K. Lehrach H. Recent advances of protein microarrays Curr. Opin. Chem. Biol. 2006 10 4 10 10.1016/j.cbpa.2005.12.011 16376134 11. Dasilva N. Diez P. Matarraz S. Gonzalez-Gonzalez M. Paradinas S. Orfao A. Fuentes M. Biomarker discovery by novel sensors based on nanoproteomics approaches Sensors 2012 12 2284 2308 22438764 12. Templin M.F. Stoll D. Schrenk M. Traub P.C. Vohringer C.F. Joos T.O. Protein microarray technology Drug Discov. Today 2002 7 815 822 12546969 13. LaBaer J. Ramachandran N. Protein microarrays as tools for functional proteomics Curr. Opin. Chem. Biol. 2005 9 14 19 10.1016/j.cbpa.2004.12.006 15701447 14. Poetz O. Schwenk J.M. Kramer S. Stoll D. Templin M.F. Joos T.O. Protein microarrays: Catching the proteome Mech. Ageing Dev. 2005 126 161 170 10.1016/j.mad.2004.09.030 15610775 15. Gao L. Uttamchandani M. Yao S.Q. Comparative proteomic profiling of mammalian cell lysates using phosphopeptide microarrays Chem. Commun. (Camb.) 2012 48 2240 2242 22252530 16. Uttamchandani M. Lu C.H. Yao S.Q. Next generation chemical proteomic tools for rapid enzyme profiling Acc. Chem. Res. 2009 42 1183 1192 10.1021/ar9000586 19435360 17. Matarraz S. Gonzalez-Gonzalez M. Jara M. Orfao A. Fuentes M. New technologies in cancer. Protein microarrays for biomarker discovery Clin. Transl. Oncol. 2011 13 156 161 10.1007/s12094-011-0635-8 21421460 18. Natesan M. Ulrich R.G. Protein microarrays and biomarkers of infectious disease Int. J. Mol. Sci. 2010 11 5165 5183 10.3390/ijms11125165 21614200 19. Borrebaeck C.A. Wingren C. Design of high-density antibody microarrays for disease proteomics: Key technological issues J. Proteomics 2009 72 928 935 10.1016/j.jprot.2009.01.027 19457338 20. Cahill D.J. Protein and antibody arrays and their medical applications J. Immunol. Methods 2001 250 81 91 10.1016/S0022-1759(01)00325-8 11251223 21. Chen C.S. Zhu H. Protein microarrays BioTechniques 2006 40 423, 425, 427 passim 16629388 22. Hall D.A. Ptacek J. Snyder M. Protein microarray technology Mech. Ageing Dev. 2007 128 161 167 10.1016/j.mad.2006.11.021 17126887 23. Pierobon M. Vanmeter A.J. Moroni N. Galdi F. Petricoin E.F. III. Reverse-phase protein microarrays Methods Mol. Biol. 2012 823 215 235 10.1007/978-1-60327-216-2_14 22081348 24. Ramachandran N. Srivastava S. Labaer J. Applications of protein microarrays for biomarker discovery Proteomics Clin. Appl. 2008 2 1444 1459 10.1002/prca.200800032 21136793 25. Beare P.A. Chen C. Bouman T. Pablo J. Unal B. Cockrell D.C. Brown W.C. Barbian K.D. Porcella S.F. Samuel J.E. Felgner P.L. Heinzen R.A. Candidate antigens for Q fever serodiagnosis revealed by immunoscreening of a Coxiella burnetii protein microarray Clin. Vaccine Immunol. 2008 15 1771 1779 10.1128/CVI.00300-08 18845831 26. Lopez J.E. Beare P.A. Heinzen R.A. Norimine J. Lahmers K.K. Palmer G.H. Brown W.C. High-throughput identification of T-lymphocyte antigens from Anaplasma marginale expressed using in vitro transcription and translation J. Immunol. Methods 2008 332 129 141 10.1016/j.jim.2007.12.018 18243240 27. Wong J. Sibani S. Lokko N.N. LaBaer J. Anderson K.S. Rapid detection of antibodies in sera using multiplexed self-assembling bead arrays J. Immunol. Methods 2009 350 171 182 10.1016/j.jim.2009.08.013 19732778 28. Hurst R. Hook B. Slater M.R. Hartnett J. Storts D.R. Nath N. Protein-protein interaction studies on protein arrays: Effect of detection strategies on signal-to-background ratios Anal. Biochem. 2009 392 45 53 19464993 29. Mei Q. Fredrickson C.K. Jin S. Fan Z.H. Toxin detection by a miniaturized in vitro protein expression array Anal. Chem. 2005 77 5494 5500 10.1021/ac050654w 16131058 30. Anderson K.S. Ramachandran N. Wong J. Raphael J.V. Hainsworth E. Demirkan G. Cramer D. Aronzon D. Hodi F.S. Harris L. Application of protein microarrays for multiplexed detection of antibodies to tumor antigens in breast cancer J. Proteome Res. 2008 7 1490 1499 18311903 31. Sreekumar A. Nyati M.K. Varambally S. Barrette T.R. Ghosh D. Lawrence T.S. Chinnaiyan A.M. Profiling of cancer cells using protein microarrays: Discovery of novel radiation-regulated proteins Cancer Res. 2001 61 7585 7593 11606398 32. Amonkar S.D. Bertenshaw G.P. Chen T.H. Bergstrom K.J. Zhao J. Seshaiah P. Yip P. Mansfield B.C. Development and preliminary evaluation of a multivariate index assay for ovarian cancer PLoS One 2009 10.1371/journal.pone.0004599 33. Roberts P.J. Der C.J. Targeting the Raf-MEK-ERK mitogen-activated protein kinase cascade for the treatment of cancer Oncogene 2007 26 3291 3310 10.1038/sj.onc.1210422 17496923 34. Belov L. Huang P. Barber N. Mulligan S.P. Christopherson R.I. Identification of repertoires of surface antigens on leukemias using an antibody microarray Proteomics 2003 3 2147 2154 14595814 35. Belov L. de la Vega O. dos Remedios C.G. Mulligan S.P. Christopherson R.I. Immunophenotyping of leukemias using a cluster of differentiation antibody microarray Cancer Res. 2001 61 4483 4489 11389079 36. Sibani S. LaBaer J. Immunoprofiling using NAPPA protein microarrays Methods Mol. Biol. 2011 723 149 161 10.1007/978-1-61779-043-0_10 21370064 37. Linkov F. Ferris R.L. Yurkovetsky Z. Marrangoni A. Velikokhatnaya L. Gooding W. Nolan B. Winans M. Siegel E.R. Lokshin A. Multiplex analysis of cytokines as biomarkers that differentiate benign and malignant thyroid diseases Proteomics Clin. Appl. 2008 2 1575 1585 10.1002/prca.200780095 19234619 38. Cid V.J. Kauffmann E. Molina M. Reverse protein arrays applied to host-pathogen interaction studies Methods Mol. Biol. 2011 723 37 55 10.1007/978-1-61779-043-0_4 21370058 39. Angenendt P. Progress in protein and antibody microarray technology Drug Discov. Today 2005 10 503 511 10.1016/S1359-6446(05)03392-1 15809196 40. Bratkovic T. Progress in phage display: Evolution of the technique and its application Cell Mol. Life Sci. 2010 67 749 767 10.1007/s00018-009-0192-2 20196239 41. Ramachandran N. Larson D.N. Stark P.R. Hainsworth E. LaBaer J. Emerging tools for real-time label-free detection of interactions on functional protein microarrays FEBS J. 2005 272 5412 5425 16262683 42. Stuchly J. Kanderova V. Fiser K. Cerna D. Holm A. Wu W. Hrusak O. Lund-Johansen F. Kalina T. An automated analysis of highly complex flow cytometry-based proteomic data Cytometry A 2012 81 120 129 22213549 43. Schweitzer B. Wiltshire S. Lambert J. O'Malley S. Kukanskis K. Zhu Z. Kingsmore S.F. Lizardi P.M. Ward D.C. Immunoassays with rolling circle DNA amplification: A versatile platform for ultrasensitive antigen detection Proc. Natl. Acad. Sci. USA 2000 97 10113 10119 10954739 44. Varnum S.M. Woodbury R.L. Zangar R.C. A protein microarray ELISA for screening biological fluids Methods Mol. Biol. 2004 264 161 172 15020788 45. Huang R.P. Detection of multiple proteins in an antibody-based protein microarray system J. Immunol. Methods 2001 255 1 13 10.1016/S0022-1759(01)00394-5 11470281 46. DeLuca D.S. Marina O. Ray S. Zhang G.L. Wu C.J. Brusic V. Data processing and analysis for protein microarrays Methods Mol. Biol. 2011 723 337 347 10.1007/978-1-61779-043-0_21 21370075 47. Olle E.W. Sreekumar A. Warner R.L. McClintock S.D. Chinnaiyan A.M. Bleavins M.R. Anderson T.D. Johnson K.J. Development of an internally controlled antibody microarray Mol. Cell. Proteomics 2005 4 1664 1672 10.1074/mcp.M500052-MCP200 16041058 48. Eckel-Passow J.E. Hoering A. Therneau T.M. Ghobrial I. Experimental design and analysis of antibody microarrays: Applying methods from cDNA arrays Cancer Res. 2005 65 2985 2989 15833819 49. Dobbin K. Shih J.H. Simon R. Questions and answers on design of dual-label microarrays for identifying differentially expressed genes J. Natl. Cancer Inst. 2003 95 1362 1369 10.1093/jnci/djg049 13130111 50. Sill M. Schroder C. Hoheisel J.D. Benner A. Zucknick M. Assessment and optimisation of normalisation methods for dual-color antibody microarrays BMC Bioinforma. 2010 10.1186/1471-2105-11-556 51. Zhang L. Wei Q. Mao L. Liu W. Mills G.B. Coombes K. Serial dilution curve: A new method for analysis of reverse phase protein array data Bioinformatics 2009 25 650 654 10.1093/bioinformatics/btn663 19176552 52. Allison D.B. Cui X. Page G.P. Sabripour M. Microarray data analysis: From disarray to consolidation and consensus Nat. Rev. Genet. 2006 7 55 65 16369572

PMC005xxxxxx/PMC5003439.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays1020084microarrays-01-00084ArticleQuality Visualization of Microarray Datasets Using Circos Koch Martin Wiese Michael *Pharmaceutical Institute, Rheinische Friedrich Wilhelms University Bonn, An der Immenburg 4, Bonn 53121, Germany; Email: martin.koch@uni-bonn.de* Author to whom correspondence should be addressed; Email: mwiese@uni-bonn.de; Tel.: +49-228-735-213; Fax: +49-228-737-929. 07 8 2012 9 2012 1 2 84 94 25 6 2012 25 7 2012 03 8 2012 © 2012 by the authors; licensee MDPI, Basel, Switzerland.2012This article is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Quality control and normalization is considered the most important step in the analysis of microarray data. At present there are various methods available for quality assessments of microarray datasets. However there seems to be no standard visualization routine, which also depicts individual microarray quality. Here we present a convenient method for visualizing the results of standard quality control tests using Circos plots. In these plots various quality measurements are drawn in a circular fashion, thus allowing for visualization of the quality and all outliers of each distinct array within a microarray dataset. The proposed method is intended for use with the Affymetrix Human Genome platform (i.e., GPL 96, GPL570 and GPL571). Circos quality measurement plots are a convenient way for the initial quality estimate of Affymetrix datasets that are stored in publicly available databases. circosyaqcaffyquality monitoring ==== Body 1. Introduction 1.1. Microarray Raw Data and Quality Control Microarray technology has successfully made the transition from a specialized method to a common method that is widely adopted in biological research. However, marginal overlap in gene expression profiling studies [1] and discussions about analysis approaches for identifying differentially expressed genes have prompted a microarray quality initiative. The microarray quality control project (MAQC) aims to estimate the accuracy of the technology [2] and examines in a second phase analysis methods and obtained biological models [3]. In general, the aim of the MAQC-II study was to resolve the concerns about the reproducibility and the generalization capability of microarray analysis results, which may stem from: (i) a lack of information about the analysis protocol. (ii) choosing different normalization methods. (iii) the use of defective analytical methods. It was found that reproducibility seems to rely greatly on the availability and completeness of the documentation of the analysis process; however it is just as essential to start the analysis with high quality data. Therefore, prior to the analysis of the raw data, the quality of the dataset needs to be assessed. An important goal of quality assessment is detection of outliers. However, currently there is no common quality measure to estimate the soundness of a microarray dataset, although there are a variety of methods available for the analysis of microarray quality in the Bioconductor project [4]. A remedy to this situation is given by post-normalization quality assessment, which detects systematically wrong appliance of normalization. It was earlier suggested that a quality check should be performed before the actual analysis as well, thus preventing the attempt of normalizing erroneous data in the first place [5]. In general normalization methods are specifically designed either for dual channel microarray datasets [6,7,8,9,10] or single channel microarray datasets [11,12]. However, there are also manufacturer dependent normalization methods available for the different flavors of microarray data [12,13,14,15,16,17,18,19,20,21,22]. 1.2. Aim of the Present Study Here we present a convenient visualization approach for assessment of individual microarray quality in large data sets using Circos. We have applied state of the art Bioconductor packages, to obtain quality control values. To condense the quality control analysis results, only outliers in the assessed quality measurements are highlighted, i.e., via red dots, denoting absence or deviations of probes in standard technical controls. Additionally RNA degradation assessment is presented in tiles using a gradient from blue to red for depicting potential outlier probes. Finally a principal component plot shows the actual array as a red dot and the other arrays in blue, thus facilitating the detection of potential outlier candidates. Presenting the analysis results in a circular format might be a suitable and convenient way for the initial quality estimate of datasets that are stored in publicly available databases. R code and a demo are publicly available at: https://github.com/buzzmak/circos-arrayQC. 2. Experimental Section 2.1. Publicly Available Microarray Studies in GEO Microarray raw data of publicly available studies was obtained from Gene Expression Omnibus (GEO, [23]). Table 1 lists all studies, which were evaluated for array quality; note that all studies comprise Affymetrix Human Genome U133A arrays and became available recently, i.e., 2011, except for GSE9936 [24] which was submitted to GEO in 2008. microarrays-01-00084-t001_Table 1Table 1 The following studies are publicly available as microarray raw data in the Gene Expression Omnibus (GEO) database. GEO ID Samples Reference Published on GSE9801 6 [25] 7 November 2011 GSE32700 46 [26] 7 October 2011 GSE9936 105 [24] 7 February 2008 2.2. Data Processing, Normalization and Principal Component Analysis Microarray data handling was performed in R (2.15.0), using the latest version of Bioconductor [4]. Microarray raw data was obtained using the GEOquery package [27] and batch processed using the affy package [28]. Additionally we used affy to calculate potential candidate arrays for RNA degradation. We performed quality control tests using the yaqcaffy package, which provides routines that are dedicated to Affymetrix arrays. We calculated all outliers in average background and average noise. Additionally we estimated outliers in both house-keeping probes (i.e., β-Actin and GADPH), as well as outliers in the internal spike-in probe calls and poly-A controls. Finally all microarrays were processed using quantile normalization of the RMA package without background adjustment [18]. Subsequently principal component analysis was performed using the pcaMethods package [29]. The scores of the mean centred first principal component were obtained for visualization. 2.3. Data Visualization Using Circos All quality control analysis results were summarized using R and then all dedicated Circos input files were generated by use of the proposed method (R code and a demo are publicly available at: http://github.com/buzzmak/circos-arrayQC). Inside an R command shell the proposed method can easily be executed, calling the following routine: > writeCircos.files(data, celNames, workdir, fileName, pathToCircos) Where data presents an affyBatch object containing the raw data, celNames denotes all Affymetrix cel-file names used in the analysis, workdir is the path to the current directory, filename names the resulting Circos QC plot and pathToCircos leads to the home directory of Circos. Circos [30] version 0.6 and strawberry Perl (http://strawberryperl.com/) version 5.1.16 was used to generate all circular quality plots. Note that the Circos method needs to be called by the Perl interpreter, i.e., > perl bin/circos –conf C:\Path\to\where\circosFiles\are\located\circosQCconfig.txt 3. Results and Discussion 3.1. Quality Assessment and Normalization of Diverse Studies Available in GEO Since currently the majority of microarray datasets in GEO are based on Affymetrix arrays, we focused our investigation on Human Genome U133A arrays (GPL96 and GPL570) from this platform. Three representative example datasets varying mainly in the number of comprised arrays were selected to show the visualization capability of our approach (Table 1). The principal approach is introduced by using arrays from the study GSE9801. The principal component analysis results, which are made with the pcaMethods package [29], are depicted in Figure 1 for this dataset. Here the first principal component is plotted against the second principal component. Two clusters and one potential outlier array can be seen. In Figure 2, a plot of potential RNA degradation is presented, generated with the affy package [28]. One array, which represents the topmost line, is probably an outlier. The yaqcaffy (http://www.bioconductor.org/packages/release/bioc/html/yaqcaffy.html) package in Bioconductor provides several quality assessment methods for Affymetrix arrays. In Figure 3 there is a quality analysis plot of study GSE9801 made with yaqcaffy. All quality control measurements are demonstrated and the suggested cut-off values are displayed as well. The box-plot showing the GADPH present calls reveals that array GSM247404 is probably an outlier. In addition, the same array is shown also in the Affymetrix specific spike-in and poly-A control plots and therefore most probably presents an outlier. Also there is array GSM247406, which has low values in the dap poly-A control. Figure 1 Principal component analysis was performed using the pcaMethods package. The figure depicts the first versus the second principal component in a scatter plot. There are two clusters and one potential outlier to the right in dataset GSE9801. The first PC explains 67% and the second PC explains 19% of the variance of the data. Figure 2 The affy package enables the examination of potential RNA degradation probes, i.e., eleven control probes which can reveal a potential fragmentation with high significance. Here we depict potential RNA degradation of arrays in dataset GSE9801. Note that the topmost blue line represents an outlier. Figure 3 Different quality measurements, which are available in the Bioconductor package yaqcaffy, shown for the example of dataset GSE9801. The first row contains two box-plots, which denote the average background and noise. The second row contains another two box-plots, showing the expression status of so called housekeeping genes. The third row contains box-plots of the Affymetrix specific spike-in and poly-A controls. Note that array GSM247404 is highlighted as potential outlier by black dots. Array GSM247406 is also flagged as outlier in the dap poly-A control, as seen at the bottom right. 3.2. Visualization by Use of Circos In Figure 4 we present an overview of the proposed quality measurement plot, using Circos. The plot combines all previously mentioned quality measurement methods and shows also quality measurements of individual arrays. Here we depict only two arrays, for explanatory purposes. Figure 5 depicts all arrays of a dataset in a circular view. Also we wished to condense quality information almost to the presence of outliers, in this way focusing only on erroneous outlier arrays. The first rim depicts the first principal component and highlights the actual array as a red dot, which is slightly bigger than the blue dots that represent the other arrays. Here arrays having similar principal component scores are clustered together, whereas arrays having different principal component scores are located apart. Figure 4 Plot of the different quality measurements, which are shown in Figure 1, Figure 2, Figure 3 of the dataset GSE9801 combined with Circos. The outer rim depicts the first principal component, which shows the actual array in red. The second rim displays the result of RNA degradation assessment. Here the red tiles present high values and the blue tiles denote low values of degradation. In addition, white tiles show medium degradation. The next four rims present outliers based on different quality control levels, first average noise and background, second outliers in the level of housekeeping genes and lastly two rims, which are showing outliers in the control spike-in probes. The index in the figure denotes: 1 = first principal component. Note that the actual array is presented as a red dot for better visualisation. 2 = RNA degradation, 3 = background and noise, 4 = β-Actin and GADPH, 5 = internal spike-in probe calls, 6 = poly-A controls. Note that array GSM247404 presents outliers on each level, even when the scores of the first principal component clusters are apart. In contrast to that, the array GSM247405 depicts inconspicuous values for almost all quality measurements, even GADPH and β-Actin shown on the fourth level in blue appear to be of good quality. In the next rim potential RNA degradation is depicted. A low overall quality is shown as red tile, while in the case when no degradation was measured the tiles are blue. Medium RNA degradation is depicted in white; these probes are in the range of the tolerance limits. For a better visual inspection of the whole dataset, all arrays are grouped together according to the RNA degradation measure. This way, arrays of similar RNA quality are typically positioned aside in the plot. In the third rim we find red dots if the average background or average noise measurements is off scale. The following three rims are dedicated to several affymetrix specific technical quality control probes and are introduced as follows. The fourth rim comprises the house-keeping genes β-actin and GADPH in blue and flagged in red in case the probes are off scale, i.e., induced more than threefold, these probes will be depicted in red. The fifth rim presents potential outliers of the internal probe calls, i.e., three probes as depicted in Figure 3 on the bottom left part. The inner ring reveals outliers in poly-A control probes, i.e., maximum of four probes, as depicted in the Figure 3 on the bottom right part. In Figure 4 we present only two arrays from dataset GSE9801, there is one array of low (GSM247404) and another array (GSM247405) of good quality. Array GSM247404 can easily be identified as a potential outlier on all measured quality scales. All technical spike-in probes in rim five are problematic and 50 percent of the other technical controls are outliers as well. The other array (GSM247405) in this dataset however is of good quality, since it shows blue tiles in the RNA degradation plot and no outliers in the technical controls. Figure 5 Quality measurement plots of the two datasets listed in Table 1, visualized by Circos, investigating the visualization limit of the proposed method. The plot on the left shows the results of the quality assessment from 46 arrays. The plot on the right summarizes all quality measurements for more than 100 arrays. In this plot one can only trace the quality of the whole dataset; as assessing the quality of single arrays without magnifying is difficult. Therefore we suggest inspecting the same plot in scalable vector format (i.e., svg), as provided in Figure S1 of the supplement. We wanted to assess the visual capability of our approach and therefore we generated Circos quality plots for two microarray studies as shown in Figure 5. These studies are listed in Table 1 in detail. The left plot in Figure 5 presents all arrays in study GSE37200 comprising 46 arrays and the right plot shows study GSE9936, which contains 105 arrays. In the dataset GSE32700, there are only three arrays of questionable quality. Two of them present outliers, since RNA degradation is detected and there are outliers in average background or noise levels. In addition to this, on the second innermost ring all three spike-in control probes are potential outliers. The majority of the arrays in the dataset are of good quality, there are only occasionally some outliers, mostly in the poly-A controls. The dataset GSE9936 has several RNA degradation candidate arrays, however only one of these arrays has also additional outliers on all other scales. The other RNA degradation candidates have mostly outliers in the average background intensities. The arrays, where RNA degradation is not problematic show off limit values in average background and noise levels instead. Additionally these arrays have off-scale measurements in both house-keeping control genes and also several absent probes and outliers in the technical spike-in and poly-A controls. In total we can count seven arrays, where quality measurements would imply outlier candidates. This underlines the fact that it might be important to access a plurality of quality measurement methods, to gain outlier arrays with greater confidence. In summary our method suits the majority of the dataset series based on GPL96 and GPL570, since most of these comprise less than 100 arrays. However, we find also that datasets containing more than 100 arrays are not reasonably well pictured using our method. Nonetheless, our method is applicable especially for use in web-resources, since Circos produces scalable vector graphic images in which the area of interest can be easily magnified. 4. Discussion The NCBI resource GEO contains at present over 10,000 platforms, of which 1,895 comprise human genomic sequences. Among all samples in GEO (776,566) there are 108,119 samples (13.9%), which are based on the Affymetrix Human Genome platform (i.e., GPL 96, GPL570 and GPL571). Over 13,000 samples are based on the Agilent 4x44k platform (0.94%) and there are 3550 samples based on the Illumina human-6 2.0 expression beadchip platform (0.45%). The proposed Circos method supports quality estimation for the majority of microarray-based experiments, which are based on the Affymetrix platform. However, there are quality-reporting methods for other platforms i.e., beadArray [31] for Illumina data and a proprietary method for Agilent arrays. A standard visualization routine, which also depicts individual microarray quality, could allow for an initial quality estimation of a microarray dataset. In the proposed Circos quality plots we find a promising approach providing quick quality assessment, which could in future be generalized to other microarray platforms as well. 5. Conclusion Quality control is the most important initial step in the analysis of microarray data. However, until now there has been no standard visualization routine to access individual microarray quality control values of large datasets. Here we present a convenient method for accessing the results of standard quality control test results using Circos. Currently the method works only for Microarray datasets, which are based on the Affymetrix Human Genome platform (i.e., GPL 96, GPL570 and GPL571). In future, this method could be adopted for quality estimation of genomic high throughput data. Appendix Figure S1 Larger version of the quality measurement plots of dataset GSE9936 as visualized by Circos. Acknowledgments We wish to acknowledge the kind help of Martin Krzywinski and Laurent Gatto. Martin Krzywinski did also provide the improved layout of the plots and helped substantially to improve our understanding of the circos method. ==== Refs References 1. Bachtiary B. Boutros P.C. Pintilie M. Shi W. Bastianutto C. Li J.-H. Schwock J. Zhang W. Penn L.Z. Jurisica I. Gene expression profiling in cervical cancer: An exploration of intratumor heterogeneity Clin. Cancer Res. 2006 12 5632 5640 17020965 2. Shi L. Reid L.H. Jones W.D. Shippy R. Warrington J.A. Baker S.C. Collins P.J. de Longueville F. Kawasaki E.S. Lee K.Y. The Microarray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements Nat. Biotechnol. 2006 24 1151 1161 10.1038/nbt1239 16964229 3. Shi L. Campbell G. Jones W.D. Campagne F. Wen Z. Walker S.J. Su Z. Chu T.M. Goodsaid F.M. Pusztai L. The Microarray Quality Control (MAQC)-II study of common practices for the development and validation of microarray-based predictive models Nat. Biotechnol. 2010 28 827 838 10.1038/nbt.1665 20676074 4. Gentleman R.C. Carey V.J. Bates D.M. Bolstad B. Dettling M. Dudoit S. Ellis B. Gautier L. Ge Y. Gentry J. Bioconductor: Open software development for computational biology and bioinformatics Genome Biol. 2004 10.1186/gb-2004-5-10-r80 5. McClure J. Wit E. Post-normalization quality assessment visualization of microarray data Comp. Funct. Genomics 2003 4 460 467 10.1002/cfg.317 18629006 6. Yang Y.H. Dudoit S. Luu P. Lin D.M. Peng V. Ngai J. Speed T.P. Normalization for cdna microarray data: A robust composite method addressing single and multiple slide systematic variation Nucleic Acids Res. 2002 10.1093/nar/30.4.e15 7. Edwards D. Non-linear normalization and background correction in one-channel cdna microarray studies Bioinformatics 2003 19 825 833 10.1093/bioinformatics/btg083 12724292 8. Xiong H. Zhang D. Martyniuk C.J. Trudeau V.L. Xia X. Using generalized procrustes analysis (gpa) for normalization of cdna microarray data BMC Bioinformatics 2008 10.1186/1471-2105-9-25 9. Wu Y. Yan L. Liu H. Sun H. Xie H. A new outlier removal approach for cdna microarray normalization BioTechniques 2009 47 691 692 694 700 19737130 10. Wu Z. Aryee M.J. Subset quantile normalization using negative control features J. Comput. Biol. 2010 17 1385 1395 10.1089/cmb.2010.0049 20976876 11. Geller S.C. Gregg J.P. Hagerman P. Rocke D.M. Transformation and normalization of oligonucleotide microarray data Bioinformatics 2003 19 1817 1823 10.1093/bioinformatics/btg245 14512353 12. Calza S. Valentini D. Pawitan Y. Normalization of oligonucleotide arrays based on the least-variant set of genes BMC Bioinforma. 2008 10.1186/1471-2105-9-140 13. Carvalho B. Bengtsson H. Speed T.P. Irizarry R.A. Exploration, normalization, and genotype calls of high-density oligonucleotide snp array data Biostatistics 2007 8 485 499 17189563 14. Rigaill G. Hupé P. Almeida A. Rosa P.L. Meyniel J.-P. Decraene C. Barillot E. Italics: An algorithm for normalization and DNA copy number calling for affymetrix snp arrays Bioinformatics 2008 24 768 774 10.1093/bioinformatics/btn048 18252739 15. Zeller G. Henz S.R. Laubinger S. Weigel D. Rätsch G. Transcript normalization and segmentation of tiling array data Pac. Symp. Biocomput. 2008 538 527 538 18229713 16. Autio R. Kilpinen S. Saarela M. Kallioniemi O. Hautaniemi S. Astola J. Comparison of affymetrix data normalization methods using 6,926 experiments across five array generations BMC Bioinforma. 2009 10.1186/1471-2105-10-S1-S24 17. Barbacioru C.C. Wang Y. Canales R.D. Sun Y.A. Keys D.N. Chan F. Poulter K.A. Samaha R.R. Effect of various normalization methods on applied biosystems expression array system data BMC Bioinforma. 2006 10.1186/1471-2105-7-533 18. Irizarry R. Hobbs B. Collin F. Beazer-Barclay Y.D. Antonellis K.J. Scherf U. Speed T.P. Exploration, normalization, and summaries of high density oligonucleotide array probe level data Biostatistics 2003 4 249 264 10.1093/biostatistics/4.2.249 12925520 19. Wu W. Dave N. Tseng G.C. Richards T. Xing E.P. Kaminski N. Comparison of normalization methods for codelink bioarray data BMC Bioinforma. 2005 10.1186/1471-2105-6-309 20. Du P. Kibbe W.A. Lin S.M. Lumi: A pipeline for processing illumina microarray Bioinformatics 2008 24 1547 1548 10.1093/bioinformatics/btn224 18467348 21. Zahurak M. Parmigiani G. Yu W. Scharpf R.B. Berman D. Schaeffer E. Shabbeer S. Cope L. Pre-processing agilent microarray data BMC Bioinforma. 2007 10.1186/1471-2105-8-142 22. Kerr K.F. Extended analysis of benchmark datasets for agilent two-color microarrays BMC Bioinforma. 2007 10.1186/1471-2105-8-371 23. Barrett T. Suzek T.O. Troup D.B. Wilhite S.E. Ngau W.C. Ledoux P. Rudnev D. Lash A.E. Fujibuchi W. Edgar R. NCBI GEO: Mining millions of expression profiles—Database and tools Nucleic Acids Res. 2005 33 D562 D566 15608262 24. Chang E.C. Charn T.H. Park S.H. Helferich W.G. Komm B. Katzenellenbogen J.A. Katzenellenbogen B.S. Estrogen receptors alpha and beta as determinants of gene expression: Influence of ligand, dose, and chromatin binding Mol. Endocrinol. 2008 22 1032 1043 10.1210/me.2007-0356 18258689 25. Lutter D. Ugocsai P. Grandl M. Orso E. Theis F. Lang E.W. Schmitz G. Analyzing m-csf dependent monocyte/macrophage differentiation: Expression modes and meta-modes derived from an independent component analysis BMC Bioinformatics 2008 10.1186/1471-2105-9-100 26. Aoyagi K. Minashi K. Igaki H. Tachimori Y. Nishimura T. Hokamura N. Ashida A. Daiko H. Ochiai A. Muto M. Artificially induced epithelial-mesenchymal transition in surgical subjects: Its implications in clinical and basic cancer research PLoS One 2011 6 e18196 21533028 27. Sean D. Meltzer P.S. GEOquery: A bridge between the Gene Expression Omnibus (GEO) and bioconductor Bioinformatics 2007 23 1846 1847 10.1093/bioinformatics/btm254 17496320 28. Gautier L. Cope L. Bolstad B.M. Irizarry R.A. Affy—Analysis of affymetrix genechip data at the probe level Bioinformatics 2004 20 307 315 10.1093/bioinformatics/btg405 14960456 29. Stacklies W. Redestig H. Scholz M. Walther D. Selbig J. Pcamethods—A bioconductor package providing pca methods for incomplete data Bioinformatics 2007 23 1164 1167 17344241 30. Krzywinski M. Schein J. Birol I. Connors J. Gascoyne R. Horsman D. Jones S.J. Marra M.A. Circos: An information aesthetic for comparative genomics Genome Res. 2009 19 1639 1645 10.1101/gr.092759.109 19541911 31. Ritchie M.E. Dunning M.J. Smith M.L. Shi W. Lynch A.G. BeadArray expression analysis using bioconductor PLoS Comput. Biol. 2011 7 e1002276 10.1371/journal.pcbi.1002276 22144879

PMC005xxxxxx/PMC5003440.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2040318microarrays-02-00318ReviewLung Cancer Gene Signatures and Clinical Perspectives Kuner Ruprecht 121 Unit Cancer Genome Research, German Cancer Research Center and National Center for Tumor Diseases, Heidelberg 69120, Germany; E-Mail: rkuner@t-online.de; Tel.: +49-6221-56-5958; Fax: +49-6221-56-53822 Translational Lung Research Center Heidelberg (TLRC-H), German Center for Lung Research, Heidelberg 69120, Germany13 12 2013 12 2013 2 4 318 339 16 10 2013 19 11 2013 06 12 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Microarrays have been used for more than two decades in preclinical research. The tumor transcriptional profiles were analyzed to select cancer-associated genes for in-deep functional characterization, to stratify tumor subgroups according to the histopathology or diverse clinical courses, and to assess biological and cellular functions behind these gene sets. In lung cancer—the main type of cancer causing mortality worldwide—biomarker research focuses on different objectives: the early diagnosis of curable tumor diseases, the stratification of patients with prognostic unfavorable operable tumors to assess the need for further therapy regimens, or the selection of patients for the most efficient therapies at early and late stages. In non-small cell lung cancer, gene and miRNA signatures are valuable to differentiate between the two main subtypes’ squamous and non-squamous tumors, a discrimination which has further implications for therapeutic schemes. Further subclassification within adenocarcinoma and squamous cell carcinoma has been done to correlate histopathological phenotype with disease outcome. Those tumor subgroups were assigned by diverse transcriptional patterns including potential biomarkers and therapy targets for future diagnostic and clinical applications. In lung cancer, none of these signatures have entered clinical routine for testing so far. In this review, the status quo of lung cancer gene signatures in preclinical and clinical research will be presented in the context of future clinical perspectives. lung cancerNSCLCbiomarkergene signaturetesting ==== Body 1. Introduction The ultimate goal reducing the high mortality rate in lung cancer disease is strongly linked to an increased efficacy of cancer prevention strategies and screening approaches for risk assessment and early detection of lung cancer in a curable stage. Major risk factors for lung cancer onset are smoking and increasing air pollution in metropolitan areas [1]. Despite enhanced prevention campaigns in the last two decades, lung cancer still represents the second most frequent malignancy and the highest cancer-related death rate in western countries. In 2013, about 228,190 new cases and 159,480 related deaths were estimated for cancer in the lung and bronchus in the United States [2]. About 55%–60% of patients are diagnosed at late incurable stages with distant metastases. As a consequence, the five-year survival rate is only 13%–16% for all stages. Non-small cell lung cancer (NSCLC) is the most common bronchial tumor, which is classified into the two major histological subtypes adenocarcinoma and squamous cell carcinoma. Both subtypes strongly differ in DNA copy number, DNA methylation, gene mutations, transcriptome, proteome and putative biomarkers as outlined in the following chapters. The stratification of diverse lung cancer entities based on clinico-histopathology and molecular alterations also determines disease outcome and therapy options. Despite significant progress in the development of novel targeted therapies, the high mortality rate in lung cancer strongly emphasizes the need for efficient lung cancer prevention and screening approaches, and the better stratification of patients who will benefit from a particular therapy regimen. A survey of clinical studies between 2009 and 2012 indicated enlarged activities of biomarker analyses to almost half of all interventional studies [3]. Biomarker-based patient selection for therapy decision clearly increased up from 7.9%–25.8%. The major goal is to identify and validate specific biomarkers or signatures in lung cancer tissues and patient surrogates, that will shift lung cancer diagnosis towards a curable stage, better stratify patients with resectable tumors for the need of adjuvant therapies, and guide clinicians to select the most beneficial therapy regimens for their patients after first diagnosis and disease progression. 2. Risk Assessment and Early Detection of Lung Cancer In the National Lung Screening Trial, low-dose computed tomography-based lung cancer screening reduced cancer mortality in high-risk individuals [4]. However, CT screening is also accompanied with a high rate of suspicious cases without confirming a malignancy, cancer over-diagnosis and economic challenges [5]. A future diagnostic testing scenario may also include screening approaches for cancer-related nucleic acid, peptide or metabolic molecules. So far, no molecular test for lung cancer diagnosis has been established in routine health care. Serum proteins like CEA, CYFRA 21-1 or MALDI/MS signatures might have been valid to detect lung cancer subtypes, but did not overcome clinical studies [6]. An immunobiomarker test (EarlyCDT®-Lung) measuring autoantibodies to a panel of seven antigens (p53, NY-ESO-1, CAGE, GBU4-5, SOX2, HuD, and MAGE A4) in serum has been developed [7,8], and is actually evaluated in a phase 2 trial. This commercially available assay objects the early detection of lung cancer in high-risk individuals (long standing smokers) and risk stratification in patients with pulmonary nodules detected by CT scans. Depending on cut-off criteria, sensitivity (49%) and specificity (93%) leads to the diagnosis of one lung cancer patient from seven positive tested individuals [9]. Another in vitro diagnostic test assay (Epi proLung BL Reflex Assay, Epigenomics, AG) measures SHOX2 DNA methylation in bronchial aspirates of patients which are suspected for lung cancer (78% sensitivity, 96% specificity), and is suggested as diagnostic adjunct when cytology results are negative or suspicious [10]. Additionally, a clinical trial is ongoing to test the accuracy of mediastinal staging by SHOX2 methylation level in transbronchial needle aspiration [11]. Moreover, a 4-gene methylation signature (p16, TERT, WT1, and RASSF1) was reported based on 655 bronchial washings to diagnose lung cancer with 82% sensitivity and 91% specificity [12]. However, it remains a challenge to sequentially combine different methods like CT and genomic signatures in a screening approach and to avoid a high number of false positives. Circulating miRNAs, robustly detected in serum and plasma, are suggested as promising biomarkers in cancer patients. Different abundance of specific miRNAs was detected in serum and plasma of lung cancer patients, which might improve risk group assessment for further CT and invasive diagnostics [13,14,15,16]. One study outlined a weighted linear combination of the expression levels of 34-miRNAs measured in 253 patients separated in a training, validation and an additional clinical validation cohort finally displaying 71% sensitivity and 90% specificity [13]. A 10-miRNA signature developed in a screening and validation study including serum from 620 NSCLC patients and controls proposed a better accuracy (90% sensitivity and 93% specificity) for early lung cancer detection [15]. Of note, expression of a serum miRNA pair (miR-15b and miR-27b) promised a 100% sensitivity and 84% specificity [16]. The increasing number of studies reporting circulating miRNAs as putative biomarkers in cancer patients indicates the great potential in biomarker discovery and translational research. Blood cells were also analyzed for non-invasive biomarkers. For example, a large gene classifier was identified and validated in blood of 233 patients using Illumina microarrays [17]. In contrast to serum and plasma, overall RNA expression will be strongly affected by various compositions of blood cell types. Molecular alterations have been associated with the individual risk for lung tissue damage and tumorigenesis. The comparison between lung cancer and benign tissues revealed diverse transcriptional profiles and putative diagnostic biomarkers [18,19,20]. For example, comprehensive meta-analysis of 20 studies comprising over 1100 lung tumor and benign tissues resulted in a robust tumor-associated 15-miRNA signature in NSCLC [20]. miRNA detection methods include both microarray and qPCR technology. Technical studies revealed a higher variation of miRNA quantification between different microarray platforms compared with qPCR and sequencing, and proposed a higher sensitivity and specificity for qPCR-based miRNA expression analysis [21,22]. Recently, an 8-miRNA signature (miR-96, miR-450a, miR-183, miR-9, miR-577, Let-7i, miR-27b miR-34a) was proposed to diagnose NSCLC in minimal biopsy material [23]. In non-tumor lung tissues of 853 lung cancer patients large and consistent gene expression variances caused by smoking were identified by using microarray technology [24]. Furthermore, gene expression changes along the airways were analyzed in order to investigate if easily accessible epithelial cells like in the nose, mouth or main bronchus reflect early oncogenic alteration in lung tissues caused by toxic agents like cigarette smoke [25,26]. Here, a prospective study (DECAMP-1) was started in 2013 aiming at the validation of gene, protein and cytokine signatures identified in bronchial airway and serum of cancer patients [27]. Alternative lung specimens like endobronchial epithelial lining fluid, bronchial lavage and sputum are collected for biomarker research. For example, DNA methylation changes, specific gene and miRNA expression signatures were identified in the presence of cancer cells [28,29,30]. However, standardization of sampling procedures for these approaches is much more challenging, because the general health condition and comorbidities strongly affect sample recovery and cellular components. As non-invasive approach, GC/MS or electronic noses were used to identify diverse signatures of volatile organic compounds (VOCs) in exhaled breath of cancer patients compared to healthy individuals [31,32,33]. Distinct VOC profiles could be assigned to patients suffering from lung cancer. In addition to analytical techniques, the excellent olfactory sense of dogs might contribute to early lung cancer detection [34]. Here, a phase-2 trial was terminated in 2013 because of inconsistent training status of sniffer dogs. Such diagnostic methods have the potential to improve the stratification of high-risk individuals and suspicious cases before invasive diagnostic bronchoscopy. Ongoing validation of diagnostic biomarker profiles and an earlier focus on standardization parameters of the techniques are prerequisites for application into clinical practice. 3. Molecular Stratification of Non-Small Cell Lung Cancer Subtypes and Outcome 3.1. Molecular Profiling of NSCLC Tumor Subtypes After lung cancer detection, clinicopathological parameters like tumor histology, staging and localization of metastases determine the disease outcome and current therapeutic interventions [35,36]. The clinical practice guidelines differentiate between small cell lung cancer (SCLC) and NSCLC, and between the major NSCLC subtypes squamous and non-squamous cell carcinoma assessed by standard histopathology. Concerning the status of present blood biomarkers, different serum levels and ratios of ProGRP, CEA, SCC, CA 125, CYFRA 21-1 and NSE have been proposed to distinguish between the two major NSCLC subtypes adenocarcinoma and squamous cell carcinoma, and SCLC, respectively [37]. Here, the precise determination of protein isoform signatures in lung cancer patients may further improve testing accuracy [38]. Recently, Roche Diagnostics launched Elecsys ProGRP test for a more precise diagnosis of SCLC from patients’ serum and plasma. The access of resected tumor material enables a comprehensive cancer-cell related diagnostics and molecular profiling. In the past, numerous microarray-based profiles were reported after millennium. Diagnostic and prognostic gene signatures using gene expression microarrays were outlined in Tables A1 and A2. Based on transcriptional profiles lung adenocarcinomas were stratified in molecular subgroups proposing diverse cellular characteristics and prognosis [18,39,40,41,42,43]. Gene expression patterns were associated with diverse adenocarcinoma subtypes named bronchioid, squamoid, and magnoid, dependent on transcriptomic similarities with histologically defined bronchioalveolar carcinoma, squamous cell carcinoma, and large-cell carcinoma [41]. These expression profiles were further investigated across six independent studies and about 1,000 patients [44]. Here, distinct molecular alterations, mutations, copy number variation and methylation could be assigned to these intrinsic subtypes with implications for further therapy modalities. Another study differentiated between two different molecular subtypes with respect to a prognostic 193-gene signature [42]. Recently, architectural classification of invasive pulmonary adenocarcinomas described five predominant patterns and has been shown to be a stage-independent predictor of survival [45,46]. Few molecular markers like TTF-1 have been investigated across IASLC/ATS/ERS adenocarcinoma architectures and were associated with disease recurrence [47,48]. The occurrence of different histological pattern in one tumor impedes the evaluation of the predominant architecture. Moreover, the histopathological AC patterns do not directly correspond with the molecular subtypes mentioned above. However, gene expression analysis of distinct architectural patterns upon tissue microdissection is under way to select for specific signatures and novel targets. Similar subtype analyses were done for squamous cell carcinomas of the lung predicting different survival outcomes [19,49,50]. The stratification of four different SCC gene expression subtypes named primitive, classical, secretory and basal generated by a nearest-centroid predictor from microarray data was reproducible across independent microarray and RNA sequencing datasets including about 600 SCCs [50,51]. The primitive subtype was associated with the worst survival outcome. Furthermore, gene expression patterns in distinct molecular subtypes were attributed to the activation status of biological and cellular processes, and oncogenic pathways. A better understanding of the relationship between histopathological diversification of lung tumors, molecular characteristics and disease outcome will contribute to molecular pathology and biomarker development in the future. Nowadays, comprehensive genomic and transcriptomic sequencing allows an integrative analysis of gene expression, mutations, copy number variations and DNA methylation to assign more complex signatures in tumor subgroups [44,51]. In lung cancer clinics, invasive tissue sampling challenges the evaluation of tumor histology. Especially in small biopsy samples from advanced tumors, specific biomarkers and signatures would be highly useful to guide tumor stratification and outcome prediction. For example, a histology expression predictor for adenocarcinoma, carcinoid, small cell carcinoma, and squamous cell carcinoma was developed using RT-qPCR in FFPE samples [52]. Direct comparison between molecular predictor and pathologist indicated similar accuracy and precision of the biomarker approach. Several qPCR- and microarray-based studies identified miRNA patterns specific for lung cancer subtypes. For example, stratification of adenocarcinoma and SCC histology was done by a 34-miRNA panel measured in FFPE specimen from 205 male smokers [53]. Moreover, two small miRNA panels measured in FFPE samples and bronchial brushing lung specimens have been reported to discriminate SCLC from NSCLC, and SCC from adenocarcinoma (AUC = 0.94–0.99), respectively [54]. Similar accuracy has been achieved by an overlapping 8-miRNA panel measured in preoperative cytologic samples [55]. Ongoing comprehensive screening for specific gene expression signatures, driver mutations and other genomic alterations are promising to identify biomarkers and targets for therapeutic intervention in distinct tumor architectures. 3.2. Gene Signatures Associated with Disease Prognosis and Outcome In the cited microarray studies above, differences between histological and molecular tumor subgroups were often characterized by large gene signatures, which are helpful for the understanding of tumor progression and differentiation but very challenging for the translation into molecular diagnostics. In addition, the tumor-intrinsic subclasses are not implicitly associated with disease outcome. At early tumor stage, stringent signatures are needed to predict individual relapse risk and survival, and to support further therapy decisions. Several microarray study reports stated diverse prognostic gene signatures in adenocarcinoma [39,56,57,58,59], SCC [49,60], or NSCLC in general [58,61,62,63,64,65,66,67]. Prognostic classifier was usually generated from a single microarray dataset by applying a Cox model (endpoints survival time or relapse time) to a training set, and validated in an independent test set or by leave-one-out-cross-validation. Thus, most of the signatures are linked to the local study protocols including patient cohort selection, starting biomaterial, microarray platform and statistics, variables, which explain the large divergence of contributing genes and the challenge of reproducibility. The accuracy and robustness of few gene signatures were successfully tested across independent microarray datasets and study centers. Based on multi-site microarray profiles from 442 adenocarcinomas, it has been shown that the predictive value of gene signatures can profit from the addition of clinical covariates [56]. The large collection of microarray datasets from lung cancer tissues provides a valuable source for the selection and validation of clinically relevant gene signatures. This is also reflected by numerous meta-analysis studies integrating public microarray data for searching and validating prognostic gene sets [42,68,69,70,71,72,73,74]. For example, a 64-gene signature predicting survival of stage I NSCLC patients was derived from a consensus expression set of 4,905 genes across seven different microarray datasets [72]. Small prognostic and predictive gene sets in early-stage NSCLC were described to be applicable to standardized analytical techniques like qPCR technology [63,75,76,77,78,79,80,81,82]. Chen and colleagues validated a five-gene signature (DUSP6, MMD, STAT1, ERBB3, and LCK) for relapse-free and overall survival across 271 stage I-III NSCLC tumors [77]. It was proposed to test in prospective, large-scale, multicenter studies if patients with a high-risk gene signature might benefit from adjuvant therapy. Furthermore, a 14-gene quantitative PCR assay on formalin-fixed paraffin-embedded tissue predicting survival in resected stage I adenocarcinoma was developed in a cohort of 361 patients and validated in two independent cohorts comprising nearly 1,439 patients [83]. Of note, this assay was successful in ethnically distinct cohorts including US and Chinese lung cancer patients. Moreover, several prognostic miRNA signatures in lung cancer tissues have been described [53,84,85,86]. For example, a five miRNA signature (let-7a, miR-221, miR-137, miR-372, and miR-182*) was generated from fresh frozen tumor tissues of a training set (n = 56) and tested in two independent patient cohorts (n = 118) to be associated with survival and cancer relapse in NSCLC patients [86]. Similarly, Landis and colleagues reported another five miRNA signature (miR-25, miR-34c-5p, miR-191, let-7e, and miR-34a) measured in FFPE tissues predicting survival (p = 0.017) in squamous cell carcinoma patients [53]. Two further prognostic miRNA signatures were proposed after microarray profiling of tissues from 527 stage I NSCLC patients dependent on the inclusion of AC and SCC subtypes [84]. As outline for gene signatures above, diverse miRNA signatures are likely reasoned by differences in the selected patient cohorts, biomaterials, platform technologies and statistics. Of note, five frequently reported miRNAs (miR-21, miR-29b, miR-34a/b/c, miR-155 and let-7a) could not be confirmed as prognostic or predictive biomarker in a large cohort (IALT trial) including 639 patients with resectable NSCLC receiving adjuvant chemotherapy [87]. A future diagnostic test based on tissue sections has to consider feasible tissue repository in the clinics, and valid standards to evaluate biomarker molecule quality and tumor cell content. Distinct circulating miRNA signatures were detected in serum or plasma from early-stage NSCLC patients and associated with recurrence risk and survival [88,89,90]. For example, a signature of four ‘high-risk’ serum miRNAs (miR-486, miR-1, miR-499, miR-30d) was reported to predict overall survival in NSCLC patients (n = 303) after surgery and adjuvant chemotherapy [88]. For advanced NSCLC, a combined 17-miRNA signature was able to calculate a 2.5-fold increased risk of death between low- and high-risk score patients [91]. The usage of specific blood biomarkers for risk assessment would facilitate diagnostics, especially for inoperable NSCLC patients where the access of representative tumor tissue is difficult. 4. Predictive Biomarkers for Lung Cancer Therapies 4.1. Prognostic and Predictive Biomarkers for Systemic Therapies Adjuvant therapies are recommended for patients with operable lung cancer dependent on tumor staging. However, the stratification of those patients is still not accurate with regard to the relapse risk and individual response. In other words, no clinical parameter or biomarker is available to predict complete cure after removal of early-stage lung tumor, or to calculate effectiveness of adjuvant chemotherapy in preventing disease relapse. Numerous prognostic signatures in tumor tissues or patient surrogates of early disease patients were generated to potentially improve therapy management if adapted to clinical routine. It has been shown that single biomarkers often failed to efficiently predict therapy response. The well-studied DNA repair genes ERCC1 and RRM1 as predictive biomarkers for chemotherapy did not enter clinical routine [92]. Weakness in the performance of detection methods and the intratumoral heterogeneity of biomarkers limited the value of single biomarkers like ERCC1 [93,94]. Cross-validation analysis of Mucin by immunohistochemistry staining in 780 patients indicated that the biomarker was not predictive for overall survival after chemotherapy [95]. Similarly, the presence of KRAS mutations was not recommended to select patient for adjuvant chemotherapy [96]. In advanced NSCLC patients, the immunohistochemical status of beta-3 tubulin was not predictive for the benefit of ixabepilone- or paclitaxel-containing regimens reported in a phase II study [97]. In SCLC patients, a combination of serum biomarkers like nucleosomes, NSE, ProGRP and CYFRA 21-1 achieved up to 47% sensitivity at 95% specificity to predict insufficient response to first-line chemotherapy [98]. Of note, several prognostic signatures reviewed in the paragraph before were addressed to the need of adjuvant chemotherapy in early-stage lung cancer patients. For example, a 15-gene signature in tumor tissues from stage IB and II NSCLC patients was reported to predict the benefit from adjuvant chemotherapy [66]. An ongoing NCI observation study is recorded to further validate this signature in FFPE specimen by using quantitative nuclease protection and NanoString assays. Moreover, a 12-gene signature predicting the benefit from adjuvant chemotherapy with cisplatin/ vinorelbine was identified by integrative analysis of genetic aberration, genome-wide RNAi data, and mRNA expression data, and successfully validated in two independent datasets [99]. Both retrospective studies integrated published microarray data in order to validate their predictive gene signatures. The variable validation success rates might be reasoned by different microarray platforms, gene probe qualities, and heterogeneous patient cohorts. Biomarker studies accompanying radiotherapy or immunotherapy are rare. Several studies suggested putative biomarkers to predict the response or toxicity upon radiotherapy [100,101,102,103]. For example, a gene expression classifier was calculated to predict radiosensitivity by comparing microarray expression profiles of the NCI 60 cell line panel and clonogenic survival assay outcome after 2 Gy of radiation [103]. In NSCLC patients, a blood biomarker panel (CRP, LDH, Osteopontin, CA-9 IL-6, IL-8, CEA, CYFRA 21-1, and α-2M) has been successfully tested to predict survival after (chemo-) radiotherapy [104], and a prospective clinical trial has been currently started to correlate blood biomarkers with overall survival. Similarly, a decrease of serum ProGRP has been associated with response to chemo- and radio-chemotherapy in SCLC [105]. So far, the activation status of oncogenic drivers like EGFR is most conclusive to contribute to radiotherapy efficacy [106]. It has been shown that targeting EGFR pathway increased radiosensitivity of tumor cell [107,108]. In contrast, radiation itself may activate diverse oncogenic pathways and benefit disease relapse [109]. Therefore, it is important to better understand the interaction between radiotherapy and oncogenic signaling in tumor cells. Based on phase II trials including melanoma and NSCLC patients treated with immunotherapeutic recombinant MAGE-A3 protein, an 84-gene gene signature was associated with clinical response [110]. The further validation of this gene signature was announced for two phase III trials. Diverse immune signatures were compared and further dissected for their contribution of the tumor genome, host genetic background and environmental factors [111,112]. The increasing number of clinical trials focusing on immunotherapies may strongly benefit from valid predictive biomarkers. 4.2. Prognostic and Predictive Biomarkers for Targeted Therapies Substantial progress has been achieved in the field of targeted therapies for lung cancer. At advanced inoperable tumor stage, molecular pathology plays an increasing role for tumor characterization, target identification and individualized therapy options. The era of high-throughput tumor genome sequencing and personalized medicine enables a further classification into molecular subtypes based on activated, therapeutically targetable oncogenes. So far, more than 50% of adenocarcinoma and squamous cell carcinoma can be characterized by mutations, fusion genes or amplifications leading to driver activation with potentially effective targeted drugs [113]. Tumor histology guides driver mutation testing and the ability of targeted therapy approaches. For lung adenocarcinoma, EGFR mutation and ALK rearrangement testing is recommended, KRAS mutation testing is suggested by the NCCN guidelines [35]. The biomarker testing is further specified by frequent mutations and reliable analytical techniques [114]. Tumors with EGFR mutations preferentially respond to EGFR tyrosine kinase inhibitors (TKIs), tumors with ALK rearrangements are associated with response to crizotinib [115,116]. The value of the most frequently mutated gene KRAS as predictive biomarker for EGFR-TKI insensitivity is controversially discussed [117,118]. Further targeted drugs are investigated in clinical trials for molecular subtypes harboring BRAF, PIK3CA or HER2 mutations, ROS1 or RET rearrangements, or c-MET amplification [113]. In about 35% of squamous cell carcinoma, aberrant FGFR1, PDGFRA, AKT1 or DDR2 are putative drug targets for individualized therapy schemes. In a simplified diagnostic scenario, genomic alterations in lung cancer are tested for relevant drivers in order to apply suitable drugs. Back to reality, the assessment of an individual drug scheme is impeded by limited predictive value of present biomarkers, less robust testing techniques, pressure of therapy timing, and subsequently initial and acquired resistance. For example, a significant number of EGFR wild-type tested lung cancer patients respond to EGFR-TKIs [119]. In contrast, 30%–40% of patients with EGFR-mutated tumors do not respond to this therapy, and most of the responders develop resistance after few months [35]. Based on specific peaks in mass spectrometry, a commercial serum/plasma-based assay (VeriStrat®) was developed to predict response to EGFR TKI therapy, and retrospectively tested in 441 patients of the BR.21 phase III trial comparing outcome of erlotinib versus placebo treatment [120]. In this study the test was prognostic for progression-free and overall survival, but was not able to predict for differential survival benefit from erlotinib. The stratification of patients for EGFR-TKI drug sensitivity depends on the mutation type of the target itself, the activation status of EGFR downstream actors and potential bypass signaling [121]. A better understanding of the acquired resistance mechanism can disclose novel therapy options using combinatorial treatments to prevent bypass signaling, or to assess a suitable therapy after relapse against the novel acquired molecular subtype [122]. Thus, additional predictive biomarkers are urgently needed to improve patient stratification and to suggest targets for mono- or combinatorial therapies for the primary tumor and after disease relapse. Microarrays have been used to identify gene signatures associated with driver mutations and response to targeted therapies [123,124,125,126]. A 76-gene signature associated with epithelial-mesenchymal transition was generated from gene expression profiles of cell lines and tissues of NSCLC patients, and proposed to predict resistance to EGFR and PI3K inhibitors [124]. Recently, a 47-gene signature associated with sorafenib sensitivity was retrospectively analyzed based on the BATTLE trial [123]. The signature was reported to serve as additional biomarker for the definition of a subgroup of patients with tumors wild-type for EGFR that may benefit from sorafenib treatment. Furthermore, a large gene set derived from microarrays well stratified lung adenocarcinomas in one ALK-mutated and two EGFR/KRAS/ALK-mutation negative subgroups [125]. Based on distinct profiles, novel target candidates have been identified in patients of a triple-negative subgroup with worse prognosis. Large comprehensive genomic studies in lung cancer are ongoing to precisely define clinically relevant tumor subtypes by combining histopathology, mutation status, DNA copy number variation, gene and protein expression, and disease outcome. Future lung cancer diagnosis and therapy will benefit from a continuous histological and molecular characterization of the tumor, and its diversity and mutability during an individual disease course, to be one step ahead of the beast. 5. Conclusions and Outlook In the last years, the number of clinical trials accompanied by biomarker studies has been continuously increasing [3]. The implementation of novel therapies more and more depend on a parallel development of biomarkers for patient stratification. So far, the translation of promising findings from biomarker research studies into valid test assays is an exceptional case for lung cancer. None of the aforementioned diagnostic or prognostic biomarkers and signatures is implemented in actual lung cancer clinical practice guidelines [127]. For advanced lung cancer, immunohistochemistry staining of protein markers can help to assess tumor histology in small biopsies and limited biomaterial [35]. In the case of non-squamous NSCLC EGFR and ALK testing are recommended by the NCCN guidelines to stratify patients for targeted therapy approaches, and hopefully represent the starting point for a wide range of targeted therapy options in future. The progress in chip-based molecular stratification of breast cancer patients for therapeutic intervention indicates the potential of molecular diagnostics for cancer patient care [128,129,130]. An evaluation of six different genomic tests also emphasizes the need of large prospective randomized trial and the potential benefit of integrated clinicopathological factors [130]. The controversial debates and the reservation against genomic tests may also reflect social-economic challenges and competition with well-established clinicopathological standards. In the context of intratumoral heterogeneity and clonal selection throughout disease course, it is very likely that a combination of molecular biomarkers and clinicopathological factors can increase the power of diagnostic tests and therapy decision. The knowledge about histopathological and molecular subtypes cleared the way for genomic testing of specific drivers beneficial for a small subset of affected lung cancer patients. In contrast, the statistical requirements for diagnostic and prognostic molecular biomarkers in most of the studies included both a high sensitivity and specificity in an epidemiologically or clinically defined risk population. Of course, a high specificity is very important to avoid a large number of suspicious cases, which would not be manageable in further clinical programs. However, do we really need a high sensitivity? If a biomarker approach for early diagnosis would be able to shift one third of advanced lung cancer diagnoses towards a potentially curable stage, this would have dramatic consequences on therapy options and cancer mortality. The numerous preclinical and clinical studies reporting diagnostic, prognostic, and predictive biomarkers and signatures well reflect the huge activity in the fields of tumor detection, prognostic stratification and molecular subtyping. So far, the clinical utility of many reported microarray-based prognostic gene signatures in lung cancer is questionable [131]. The future translation of genomic tests into clinical practice will strongly depend on the answer to the unambiguous clinical question, the inclusion criteria of the target population, the availability of required biomaterial, robust analytical techniques and standards, and the validation of the biomarker assay in large, prospective, randomized trials. Acknowledgments The author thanks Holger Sültmann, Helen Hülsmann and Sajo Kaduthanam from the Unit Cancer Genome Research at the DKFZ, Michael Meister and Thomas Muley from the Section Translational Research at the Thoraxklinik Heidelberg, and Rainer Hipfel for discussion about biomarker research and challenges in lung cancer. Conflicts of Interest The author declares no conflict of interest. Appendix microarrays-02-00318-t001_Table A1Table A1 Gene expression microarray studies describing gene clusters or gene signatures in lung cancer or NSCLC. Clinical focus Tumor Type Biomaterial Gene signature Screening Validation Technology References Diagnosis NSCLC Blood cells 484-feature classifier n = 77 n = 156 Illumina microarrays [17] Diagnosis Lung cancer normal large-airway epithelial cells 80-gene signature n = 77 n = 52 Affymetrix microarrays [26] Risk, Smoking Lung cancer non-tumor lung tissue 599-feature set n = 344 n = 509 Affymetrix microarrays [24] Prognosis NSCLC Tissues 6-gene signature, clinical covariates n = 56 n = 59 Affymetrix microarrays [67] Prognosis NSCLC Tissues 72-gene signature n = 103 n = 69 Agilent oligo microarray [64] Prognosis NSCLC Tissues 17-gene signature n = 91 public dataset; Potti, 2006 Affymetrix microarrays [62] Prognosis; Chemotherapy prediction NSCLC Tissues 15-gene signature n = 133 public datasets; Potti, 2006; Raponi, 2006; Shedden, 2008; Roepman, 2009; qPCR (n = 30) Affymetrix microarrays [66] Prognosis NSCLC Tissues 4-gene signature, clinical covariates n = 27 n = 138 Affymetrix microarrays [63] Prognosis NSCLC Tissues 59-gene signature n = 55 public datasets; Bhattacharjee, 2001; Bild, 2006 Affymetrix microarrays [65] Prognosis NSCLC Tissues 450-gene signature n = 196 public datasets; Bild, 2006; Raponi, 2006; Shedden, 2008; Zhu, 2010; Hou, 2010 Affymetrix microarrays [61] Prognosis NSCLC Tissues 5-gene signature n = 125 n = 60; public datasets; Beer, 2002 cDNA microarray, qPCR arrays [77] In vitro model; Pathway Lung cancer Cell lines, Tissues Oncogenic pathway signatures cell line, lung cancer (n = 111) none Affymetrix microarrays [132] microarrays-02-00318-t002_Table A2Table A2 Gene expression microarray studies describing gene clusters or gene signatures in adenocarcinoma or squamous cell carcinoma. Clinical focus Tumor Type Biomaterial Gene signature Screening Validation Technology References AC-Subtypes; Prognosis Lung cancer–AC Tissues Gene clusters n = 139 none Affymetrix microarrays [18] AC-Subtypes; Prognosis Lung cancer–AC Tissues Gene clusters n = 67 none cDNA microarray [40] AC-Subtypes; Prognosis NSCLC–AC Tissues Gene clusters n = 149 none Agilent oligo microarray [43] AC-Subtypes; Prognosis AC Tissues 50-gene signature n = 43 n = 43 Affymetrix microarrays [39] Prognosis AC Tissues 54-gene signature n = 48 n = 95 Agilent oligo microarray [57] Prognosis AC Tissues Gene classifiers; clinical covariates n = 256 n = 186 Affymetrix microarrays [56] Prognosis AC Tissues 82-feature signature n = 60 n = 57 Agilent oligo microarray [59] Prognosis AC Tissues 3-gene signature n = 82 public datasets; Bhattacharjee, 2001; Shedden, 2008 Illumina microarrays [58] Integrative analysis AC Tissues None n = 75 none Affymetrix microarrays [133] Integrative analysis AC Tissues EGFR and KRAS associated gene signatures n = 193 none Affymetrix microarrays [126] Genomic subtypes AC Tissues Gene signatures n = 226 none Affymetrix microarrays [125] SCC-Subtypes; Prognosis NSCLC–SCC Tissues Gene clusters n = 48 none cDNA microarray [19] SCC-Subtypes; Prognosis NSCLC–SCC Tissues Gene clusters, 50-gene signature n = 129 n = 36 Affymetrix microarrays [49] SCC-Subtypes; Prognosis SCC Tissues Subtype predictor public datasets; Bild, 2006; Lee, 2008; Raponi, 2006; Roepman, 2009 n =56 Agilent oligo microarray [50] SCC-Subtypes SCC Tissues Subtype predictor Wilkerson, 2010 n = 178 Agilent oligo microarray [51] Prognosis SCC Tissues 111-gene signature n = 51 n = 58 Operon oligo microarray [60] ==== Refs References 1. Alberg A.J. Brock M.V. Ford J.G. Samet J.M. Spivack S.D. Epidemiology of lung cancer: Diagnosis and management of lung cancer, 3rd ed: American College of Chest Physicians evidence-based clinical practice guidelines Chest 2013 143 e1S e29S 10.1378/chest.12-2345 23649439 2. Siegel R. Naishadham D. Jemal A. Cancer statistics, 2013 Cancer J. Clin. 2013 63 11 30 10.3322/caac.21166 3. Subramanian J. Regenbogen T. Nagaraj G. Lane A. Devarakonda S. Zhou G. Govindan R. Review of ongoing clinical trials in non-small-cell lung cancer: A status report for 2012 from the ClinicalTrials.gov Web site J. Thoracic Oncol. 2013 8 860 865 10.1097/JTO.0b013e318287c562 4. Kovalchik S.A. Tammemagi M. Berg C.D. Caporaso N.E. Riley T.L. Korch M. Silvestri G.A. Chaturvedi A.K. Katki H.A. Targeting of low-dose CT screening according to the risk of lung-cancer death N. Engl. J. Med. 2013 369 245 254 10.1056/NEJMoa1301851 23863051 5. Aberle D.R. Abtin F. Brown K. Computed tomography screening for lung cancer: Has it finally arrived? Implications of the national lung screening trial J. Clin. Oncol. 2013 31 1002 1008 10.1200/JCO.2012.43.3110 23401434 6. Hassanein M. Callison J.C. Callaway-Lane C. Aldrich M.C. Grogan E.L. Massion P.P. The state of molecular biomarkers for the early detection of lung cancer Cancer Prev. Res. 2012 5 992 1006 10.1158/1940-6207.CAPR-11-0441 7. Chapman C.J. Healey G.F. Murray A. Boyle P. Robertson C. Peek L.J. Allen J. Thorpe A.J. Hamilton-Fairley G. Parsy-Kowalska C.B. Early CDT® -Lung test: Improved clinical utility through additional autoantibody assays Tumour Biol. 2012 33 1319 1326 10.1007/s13277-012-0379-2 22492236 8. Lam S. Boyle P. Healey G.F. Maddison P. Peek L. Murray A. Chapman C.J. Allen J. Wood W.C. Sewell H.F. Early CDT-Lung: An immunobiomarker test as an aid to early detection of lung cancer Cancer Prev. Res. 2011 4 1126 1134 10.1158/1940-6207.CAPR-10-0328 9. Macdonald I.K. Murray A. Healey G.F. Parsy-Kowalska C.B. Allen J. McElveen J. Robertson C. Sewell H.F. Chapman C.J. Robertson J.F. Application of a high throughput method of biomarker discovery to improvement of the Early CDT® -Lung test PLoS One 2012 7 10.1371/journal.pone.0051002 10. Dietrich D. Kneip C. Raji O. Liloglou T. Seegebarth A. Schlegel T. Flemming N. Rausch S. Distler J. Fleischhacker M. Performance evaluation of the DNA methylation biomarker SHOX2 for the aid in diagnosis of lung cancer based on the analysis of bronchial aspirates Int. J. Oncol. 2012 40 825 832 22108652 11. Darwiche K. Zarogoulidis P. Baehner K. Welter S. Tetzner R. Wohlschlaeger J. Theegarten D. Nakajima T. Freitag L. Assessment of SHOX2 methylation in EBUS-TBNA specimen improves accuracy in lung cancer staging Ann. Oncol. 2013 24 2866 2870 10.1093/annonc/mdt365 24026539 12. Nikolaidis G. Raji O.Y. Markopoulou S. Gosney J.R. Bryan J. Warburton C. Walshaw M. Sheard J. Field J.K. Liloglou T. DNA methylation biomarkers offer improved diagnostic efficiency in lung cancer Cancer Res. 2012 72 5692 5701 10.1158/0008-5472.CAN-12-2309 22962272 13. Bianchi F. Nicassio F. Marzi M. Belloni E. Dall’olio V. Bernard L. Pelosi G. Maisonneuve P. Veronesi G. di Fiore P.P. A serum circulating miRNA diagnostic test to identify asymptomatic high-risk individuals with early stage lung cancer EMBO Mol. Med. 2011 3 495 503 10.1002/emmm.201100154 21744498 14. Boeri M. Verri C. Conte D. Roz L. Modena P. Facchinetti F. Calabro E. Croce C.M. Pastorino U. Sozzi G. MicroRNA signatures in tissues and plasma predict development and prognosis of computed tomography detected lung cancer Proc. Natl. Acad. Sci. USA 2011 108 3713 3718 10.1073/pnas.1100048108 21300873 15. Chen X. Hu Z. Wang W. Ba Y. Ma L. Zhang C. Wang C. Ren Z. Zhao Y. Wu S. Identification of ten serum microRNAs from a genome-wide serum microRNA expression profile as novel noninvasive biomarkers for nonsmall cell lung cancer diagnosis Int. J. Cancer 2012 130 1620 1628 10.1002/ijc.26177 21557218 16. Hennessey P.T. Sanford T. Choudhary A. Mydlarz W.W. Brown D. Adai A.T. Ochs M.F. Ahrendt S.A. Mambo E. Califano J.A. Serum microRNA biomarkers for detection of non-small cell lung cancer PLoS One 2012 7 10.1371/journal.pone.0032307 17. Zander T. Hofmann A. Staratschek-Jox A. Classen S. Debey-Pascher S. Maisel D. Ansen S. Hahn M. Beyer M. Thomas R.K. Blood-based gene expression signatures in non-small cell lung cancer Clin. Cancer Res. 2011 17 3360 3367 10.1158/1078-0432.CCR-10-0533 21558400 18. Bhattacharjee A. Richards W.G. Staunton J. Li C. Monti S. Vasa P. Ladd C. Beheshti J. Bueno R. Gillette M. Classification of human lung carcinomas by mRNA expression profiling reveals distinct adenocarcinoma subclasses Proc. Natl. Acad. Sci. USA 2001 98 13790 13795 10.1073/pnas.191502998 11707567 19. Inamura K. Fujiwara T. Hoshida Y. Isagawa T. Jones M.H. Virtanen C. Shimane M. Satoh Y. Okumura S. Nakagawa K. Two subclasses of lung squamous cell carcinoma with different gene expression profiles and prognosis identified by hierarchical clustering and non-negative matrix factorization Oncogene 2005 24 7105 7113 10.1038/sj.onc.1208858 16007138 20. Vosa U. Vooder T. Kolde R. Vilo J. Metspalu A. Annilo T. Meta-analysis of microRNA expression in lung cancer Int. J. Cancer 2013 132 2884 2893 10.1002/ijc.27981 23225545 21. Jang J.S. Simon V.A. Feddersen R.M. Rakhshan F. Schultz D.A. Zschunke M.A. Lingle W.L. Kolbert C.P. Jen J. Quantitative miRNA expression analysis using fluidigm microfluidics dynamic arrays BMC Genomics 2011 12 10.1186/1471-2164-12-144 22. Pradervand S. Weber J. Lemoine F. Consales F. Paillusson A. Dupasquier M. Thomas J. Richter H. Kaessmann H. Beaudoing E. Concordance among digital gene expression, microarrays, and qPCR when measuring differential expression of microRNAs BioTechniques 2010 48 219 222 10.2144/000113367 20359303 23. Bediaga N.G. Davies M.P. Acha-Sagredo A. Hyde R. Raji O.Y. Page R. Walshaw M. Gosney J. Alfirevic A. Field J.K. A microRNA-based prediction algorithm for diagnosis of non-small lung cell carcinoma in minimal biopsy material Br. J. Cancer 2013 109 2404 2411 10.1038/bjc.2013.623 24113142 24. Bosse Y. Postma D.S. Sin D.D. Lamontagne M. Couture C. Gaudreault N. Joubert P. Wong V. Elliott M. van den Berge M. Molecular signature of smoking in human lung tissues Cancer Res. 2012 72 3753 3763 10.1158/0008-5472.CAN-12-1160 22659451 25. Beane J. Vick J. Schembri F. Anderlind C. Gower A. Campbell J. Luo L. Zhang X.H. Xiao J. Alekseyev Y.O. Characterizing the impact of smoking and lung cancer on the airway transcriptome using RNA-Seq Cancer Prev. Res. 2011 4 803 817 10.1158/1940-6207.CAPR-11-0212 26. Spira A. Beane J.E. Shah V. Steiling K. Liu G. Schembri F. Gilman S. Dumas Y.M. Calner P. Sebastiani P. Airway epithelial gene expression in the diagnostic evaluation of smokers with suspect lung cancer Nat. Med. 2007 13 361 366 10.1038/nm1556 17334370 27. Brothers J.F. Hijazi K. Mascaux C. El-Zein R.A. Spitz M.R. Spira A. Bridging the clinical gaps: Genetic, epigenetic and transcriptomic biomarkers for the early detection of lung cancer in the post-National Lung Screening Trial era BMC Med. 2013 11 10.1186/1741-7015-11-168 28. Kahn N. Meister M. Eberhardt R. Muley T. Schnabel P.A. Bender C. Johannes M. Keitel D. Sultmann H. Herth F.J. Early detection of lung cancer by molecular markers in endobronchial epithelial-lining fluid J. Thoracic Oncol. 2012 7 1001 1008 10.1097/JTO.0b013e31824fe921 29. Leng S. Do K. Yingling C.M. Picchi M.A. Wolf H.J. Kennedy T.C. Feser W.J. Baron A.E. Franklin W.A. Brock M.V. Defining a gene promoter methylation signature in sputum for lung cancer risk assessment Clin. Cancer Res. 2012 18 3387 3395 10.1158/1078-0432.CCR-11-3049 22510351 30. Yu L. Todd N.W. Xing L. Xie Y. Zhang H. Liu Z. Fang H. Zhang J. Katz R.L. Jiang F. Early detection of lung adenocarcinoma in sputum by a panel of microRNA markers Int. J. Cancer 2010 127 2870 2878 10.1002/ijc.25289 21351266 31. Bajtarevic A. Ager C. Pienz M. Klieber M. Schwarz K. Ligor M. Ligor T. Filipiak W. Denz H. Fiegl M. Noninvasive detection of lung cancer by analysis of exhaled breath BMC Cancer 2009 9 10.1186/1471-2407-9-348 32. Phillips M. Altorki N. Austin J.H. Cameron R.B. Cataneo R.N. Greenberg J. Kloss R. Maxfield R.A. Munawar M.I. Pass H.I. Prediction of lung cancer using volatile biomarkers in breath Canc. Biomarkers 2007 3 95 109 33. Ulanowska A. Kowalkowski T. Trawinska E. Buszewski B. The application of statistical methods using VOCs to identify patients with lung cancer J. Breath Res. 2011 5 10.1088/1752-7155/5/4/046008 34. McCulloch M. Turner K. Broffman M. Lung cancer detection by canine scent: Will there be a lab in the lab? Eur. Respir. J. 2012 39 511 512 10.1183/09031936.00215511 22379142 35. Ettinger D.S. Akerley W. Borghaei H. Chang A.C. Cheney R.T. Chirieac L.R. D’Amico T.A. Demmy T.L. Ganti A.K. Govindan R. Nccn, non-small cell lung cancer J. Natl. Compr. Cancer Netw. 2012 10 1236 1271 36. Kalemkerian G.P. Akerley W. Bogner P. Borghaei H. Chow L.Q. Downey R.J. Gandhi L. Ganti A.K. Govindan R. Grecula J.C. Small cell lung cancer J. Natl. Compr. Cancer Netw. 2013 11 78 98 37. Molina R. Auge J.M. Bosch X. Escudero J.M. Vinolas N. Marrades R. Ramirez J. Carcereny E. Filella X. Usefulness of serum tumor markers, including progastrin-releasing peptide, in patients with lung cancer: correlation with histology Tumour Biol. 2009 30 121 129 10.1159/000224628 19506400 38. Torsetnes S.B. Nordlund M.S. Paus E. Halvorsen T.G. Reubsaet L. Digging deeper into the field of the small cell lung cancer tumor marker ProGRP: A method for differentiation of its isoforms J. Proteome Res. 2013 12 412 420 10.1021/pr300751j 23190087 39. Beer D.G. Kardia S.L. Huang C.C. Giordano T.J. Levin A.M. Misek D.E. Lin L. Chen G. Gharib T.G. Thomas D.G. Gene-expression profiles predict survival of patients with lung adenocarcinoma Nat. Med. 2002 8 816 824 12118244 40. Garber M.E. Troyanskaya O.G. Schluens K. Petersen S. Thaesler Z. Pacyna-Gengelbach M. van de Rijn M. Rosen G.D. Perou C.M. Whyte R.I. Diversity of gene expression in adenocarcinoma of the lung Proc. Natl. Acad. Sci. USA 2001 98 13784 13789 10.1073/pnas.241500798 11707590 41. Hayes D.N. Monti S. Parmigiani G. Gilks C.B. Naoki K. Bhattacharjee A. Socinski M.A. Perou C. Meyerson M. Gene expression profiling reveals reproducible human lung adenocarcinoma subtypes in multiple independent patient cohorts J. Clin. Oncol. 2006 24 5079 5090 10.1200/JCO.2005.05.1748 17075127 42. Park Y.Y. Park E.S. Kim S.B. Kim S.C. Sohn B.H. Chu I.S. Jeong W. Mills G.B. Byers L.A. Lee J.S. Development and validation of a prognostic gene-expression signature for lung adenocarcinoma PLoS One 2012 7 10.1371/journal.pone.0044225 43. Takeuchi T. Tomida S. Yatabe Y. Kosaka T. Osada H. Yanagisawa K. Mitsudomi T. Takahashi T. Expression profile-defined classification of lung adenocarcinoma shows close relationship with underlying major genetic changes and clinicopathologic behaviors J. Clin. Oncol. 2006 24 1679 1688 10.1200/JCO.2005.03.8224 16549822 44. Wilkerson M.D. Yin X. Walter V. Zhao N. Cabanski C.R. Hayward M.C. Miller C.R. Socinski M.A. Parsons A.M. Thorne L.B. Differential pathogenesis of lung adenocarcinoma subtypes involving sequence mutations, copy number, chromosomal instability, and methylation PLoS One 2012 7 10.1371/journal.pone.0036530 45. Warth A. Muley T. Meister M. Stenzinger A. Thomas M. Schirmacher P. Schnabel P.A. Budczies J. Hoffmann H. Weichert W. The novel histologic International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society classification system of lung adenocarcinoma is a stage-independent predictor of survival J. Clin. Oncol. 2012 30 1438 1446 10.1200/JCO.2011.37.2185 22393100 46. Travis W.D. Brambilla E. Riely G.J. New pathologic classification of lung cancer: Relevance for clinical practice and clinical trials J. Clin. Oncol. 2013 31 992 1001 10.1200/JCO.2012.46.9270 23401443 47. Kadota K. Nitadori J. Sarkaria I.S. Sima C.S. Jia X. Yoshizawa A. Rusch V.W. Travis W.D. Adusumilli P.S. Thyroid transcription factor-1 expression is an independent predictor of recurrence and correlates with the IASLC/ATS/ERS histologic classification in patients with stage I lung adenocarcinoma Cancer 2013 119 931 938 10.1002/cncr.27863 23096929 48. Solis L.M. Behrens C. Raso M.G. Lin H.Y. Kadara H. Yuan P. Galindo H. Tang X. Lee J.J. Kalhor N. Histologic patterns and molecular characteristics of lung adenocarcinoma associated with clinical outcome Cancer 2012 118 2889 2899 10.1002/cncr.26584 22020674 49. Raponi M. Zhang Y. Yu J. Chen G. Lee G. Taylor J.M. Macdonald J. Thomas D. Moskaluk C. Wang Y. Gene expression signatures for predicting prognosis of squamous cell and adenocarcinomas of the lung Cancer Res. 2006 66 7466 7472 10.1158/0008-5472.CAN-06-1191 16885343 50. Wilkerson M.D. Yin X. Hoadley K.A. Liu Y. Hayward M.C. Cabanski C.R. Muldrew K. Miller C.R. Randell S.H. Socinski M.A. Lung squamous cell carcinoma mRNA expression subtypes are reproducible, clinically important, and correspond to normal cell types Clin. Cancer Res. 2010 16 4864 4875 10.1158/1078-0432.CCR-10-0199 20643781 51. The Cancer Genome Atlas Research Network Comprehensive genomic characterization of squamous cell lung cancers Nature 2012 489 519 525 10.1038/nature11404 22960745 52. Wilkerson M.D. Schallheim J.M. Hayes D.N. Roberts P.J. Bastien R.R. Mullins M. Yin X. Miller C.R. Thorne L.B. Geiersbach K.B. Prediction of lung cancer histological types by RT-qPCR gene expression in FFPE specimens J. Mol. Diagn. 2013 15 485 497 10.1016/j.jmoldx.2013.03.007 23701907 53. Landi M.T. Zhao Y. Rotunno M. Koshiol J. Liu H. Bergen A.W. Rubagotti M. Goldstein A.M. Linnoila I. Marincola F.M. MicroRNA expression differentiates histology and predicts survival of lung cancer Clin. Cancer Res. 2010 16 430 441 10.1158/1078-0432.CCR-09-1736 20068076 54. Huang W. Hu J. Yang D.W. Fan X.T. Jin Y. Hou Y.Y. Wang J.P. Yuan Y.F. Tan Y.S. Zhu X.Z. Two microRNA panels to discriminate three subtypes of lung carcinoma in bronchial brushing specimens Am. J. Respir. Crit. Care Med. 2012 186 1160 1167 10.1164/rccm.201203-0534OC 23043084 55. Gilad S. Lithwick-Yanai G. Barshack I. Benjamin S. Krivitsky I. Edmonston T.B. Bibbo M. Thurm C. Horowitz L. Huang Y. Classification of the four main types of lung cancer using a microRNA-based diagnostic assay J. Mol. Diagn. 2012 14 510 517 10.1016/j.jmoldx.2012.03.004 22749746 56. Director’s Challenge Consortium for the Molecular Classification of Lung Adenocarcinoma Shedden K. Taylor J.M. Enkemann S.A. Tsao M.S. Yeatman T.J. Gerald W.L. Eschrich S. Jurisica I. Giordano T.J. Gene expression-based survival prediction in lung adenocarcinoma: A multi-site, blinded validation study Nat. Med. 2008 14 822 827 10.1038/nm.1790 18641660 57. Larsen J.E. Pavey S.J. Passmore L.H. Bowman R.V. Hayward N.K. Fong K.M. Gene expression signature predicts recurrence in lung adenocarcinoma Clin. Cancer Res. 2007 13 2946 2954 10.1158/1078-0432.CCR-06-2525 17504995 58. Li Y. Tang H. Sun Z. Bungum A.O. Edell E.S. Lingle W.L. Stoddard S.M. Zhang M. Jen J. Yang P. Network-based approach identified cell cycle genes as predictor of overall survival in lung adenocarcinoma patients Lung Cancer 2013 80 91 98 10.1016/j.lungcan.2012.12.022 23357462 59. Tomida S. Takeuchi T. Shimada Y. Arima C. Matsuo K. Mitsudomi T. Yatabe Y. Takahashi T. Relapse-related molecular signature in lung adenocarcinomas identifies patients with dismal prognosis J. Clin. Oncol. 2009 27 2793 2799 10.1200/JCO.2008.19.7053 19414676 60. Larsen J.E. Pavey S.J. Passmore L.H. Bowman R. Clarke B.E. Hayward N.K. Fong K.M. Expression profiling defines a recurrence signature in lung squamous cell carcinoma Carcinogenesis 2007 28 760 766 17082175 61. Botling J. Edlund K. Lohr M. Hellwig B. Holmberg L. Lambe M. Berglund A. Ekman S. Bergqvist M. Ponten F. Biomarker discovery in non-small cell lung cancer: Integrating gene expression profiling, meta-analysis, and tissue microarray validation Clin. Cancer Res. 2013 19 194 204 10.1158/1078-0432.CCR-12-1139 23032747 62. Hou J. Aerts J. den Hamer B. van Ijcken W. den Bakker M. Riegman P. van der Leest C. van der Spek P. Foekens J.A. Hoogsteden H.C. Gene expression-based classification of non-small cell lung carcinomas and survival prediction PLoS One 2010 5 10.1371/journal.pone.0010312 63. Mitra R. Lee J. Jo J. Milani M. McClintick J.N. Edenberg H.J. Kesler K.A. Rieger K.M. Badve S. Cummings O.W. Prediction of postoperative recurrence-free survival in non-small cell lung cancer by using an internationally validated gene expression model Clin. Cancer Res. 2011 17 2934 2946 10.1158/1078-0432.CCR-10-1803 21242119 64. Roepman P. Jassem J. Smit E.F. Muley T. Niklinski J. van de Velde T. Witteveen A.T. Rzyman W. Floore A. Burgers S. An immune response enriched 72-gene prognostic profile for early-stage non-small-cell lung cancer Clin. Cancer Res. 2009 15 284 290 10.1158/1078-0432.CCR-08-1258 19118056 65. Xie Y. Xiao G. Coombes K.R. Behrens C. Solis L.M. Raso G. Girard L. Erickson H.S. Roth J. Heymach J.V. Robust gene expression signature from formalin-fixed paraffin-embedded samples predicts prognosis of non-small-cell lung cancer patients Clin. Cancer Res. 2011 17 5705 5714 10.1158/1078-0432.CCR-11-0196 21742808 66. Zhu C.Q. Ding K. Strumpf D. Weir B.A. Meyerson M. Pennell N. Thomas R.K. Naoki K. Ladd-Acosta C. Liu N. Prognostic and predictive gene signature for adjuvant chemotherapy in resected non-small-cell lung cancer J. Clin. Oncol. 2010 28 4417 4424 10.1200/JCO.2009.26.4325 20823422 67. Lee E.S. Son D.S. Kim S.H. Lee J. Jo J. Han J. Kim H. Lee H.J. Choi H.Y. Jung Y. Prediction of recurrence-free survival in postoperative non-small cell lung cancer patients by using an integrated model of clinical information and gene expression Clin. Cancer Res. 2008 14 7397 7404 10.1158/1078-0432.CCR-07-4937 19010856 68. Boutros P.C. Lau S.K. Pintilie M. Liu N. Shepherd F.A. Der S.D. Tsao M.S. Penn L.Z. Jurisica I. Prognostic gene signatures for non-small-cell lung cancer Proc. Natl. Acad. Sci. USA 2009 106 2824 2828 10.1073/pnas.0809444106 19196983 69. Chen D.T. Hsu Y.L. Fulp W.J. Coppola D. Haura E.B. Yeatman T.J. Cress W.D. Prognostic and predictive value of a malignancy-risk gene signature in early-stage non-small cell lung cancer J. Natl. Cancer Instit. 2011 103 1859 1870 10.1093/jnci/djr420 70. Guo N.L. Wan Y.W. Tosun K. Lin H. Msiska Z. Flynn D.C. Remick S.C. Vallyathan V. Dowlati A. Shi X. Confirmation of gene expression-based prediction of survival in non-small cell lung cancer Clin. Cancer Res. 2008 14 8213 8220 10.1158/1078-0432.CCR-08-0095 19088038 71. Lu Y. Lemon W. Liu P.Y. Yi Y. Morrison C. Yang P. Sun Z. Szoke J. Gerald W.L. Watson M. A gene expression signature predicts survival of patients with stage I non-small cell lung cancer PLoS Med. 2006 3 10.1371/journal.pmed.0030467 72. Lu Y. Wang L. Liu P. Yang P. You M. Gene-expression signature predicts postoperative recurrence in stage I non-small cell lung cancer patients PLoS One 2012 7 10.1371/journal.pone.0030880 73. Sun Z. Wigle D.A. Yang P. Non-overlapping and non-cell-type-specific gene expression signatures predict lung cancer survival J. Clin. Oncol. 2008 26 877 883 10.1200/JCO.2007.13.1516 18281660 74. Van Laar R.K. Genomic signatures for predicting survival and adjuvant chemotherapy benefit in patients with non-small-cell lung cancer BMC Med. Genomics 2012 5 10.1186/1755-8794-5-30 75. Akagi I. Okayama H. Schetter A.J. Robles A.I. Kohno T. Bowman E.D. Kazandjian D. Welsh J.A. Oue N. Saito M. Combination of protein coding and noncoding gene expression as a robust prognostic classifier in stage I lung adenocarcinoma Cancer Res. 2013 73 3821 3832 10.1158/0008-5472.CAN-13-0031 23639940 76. Bianchi F. Nuciforo P. Vecchi M. Bernard L. Tizzoni L. Marchetti A. Buttitta F. Felicioni L. Nicassio F. di Fiore P.P. Survival prediction of stage I lung adenocarcinomas by expression of 10 genes J. Clin. Investig. 2007 117 3436 3444 10.1172/JCI32007 17948124 77. Chen H.Y. Yu S.L. Chen C.H. Chang G.C. Chen C.Y. Yuan A. Cheng C.L. Wang C.H. Terng H.J. Kao S.F. A five-gene signature and clinical outcome in non-small-cell lung cancer N. Engl. J. Med. 2007 356 11 20 10.1056/NEJMoa060096 17202451 78. Lau S.K. Boutros P.C. Pintilie M. Blackhall F.H. Zhu C.Q. Strumpf D. Johnston M.R. Darling G. Keshavjee S. Waddell T.K. Three-gene prognostic classifier for early-stage non small-cell lung cancer J. Clin. Oncol. 2007 25 5562 5569 10.1200/JCO.2007.12.0352 18065728 79. Raz D.J. Ray M.R. Kim J.Y. He B. Taron M. Skrzypski M. Segal M. Gandara D.R. Rosell R. Jablons D.M. A multigene assay is prognostic of survival in patients with early-stage lung adenocarcinoma Clin. Cancer Res. 2008 14 5565 5570 10.1158/1078-0432.CCR-08-0544 18765549 80. Seike M. Yanaihara N. Bowman E.D. Zanetti K.A. Budhu A. Kumamoto K. Mechanic L.E. Matsumoto S. Yokota J. Shibata T. Use of a cytokine gene expression signature in lung adenocarcinoma and the surrounding tissue as a prognostic classifier J. Natl. Cancer Instit. 2007 99 1257 1269 10.1093/jnci/djm083 81. Wistuba I.I. Behrens C. Lombardi F. Wagner S. Fujimoto J. Raso M.G. Spaggiari L. Galetta D. Riley R. Hughes E. Validation of a proliferation-based expression signature as prognostic marker in early stage lung adenocarcinoma Clin. Cancer Res. 2013 19 6261 6271 10.1158/1078-0432.CCR-13-0596 24048333 82. Skrzypski M. Jassem E. Taron M. Sanchez J.J. Mendez P. Rzyman W. Gulida G. Raz D. Jablons D. Provencio M. Three-gene expression signature predicts survival in early-stage squamous cell carcinoma of the lung Clin. Cancer Res. 2008 14 4794 4799 10.1158/1078-0432.CCR-08-0576 18676750 83. Kratz J.R. He J. van den Eeden S.K. Zhu Z.H. Gao W. Pham P.T. Mulvihill M.S. Ziaei F. Zhang H. Su B. A practical molecular assay to predict survival in resected non-squamous, non-small-cell lung cancer: Development and international validation studies Lancet 2012 379 823 832 10.1016/S0140-6736(11)61941-7 22285053 84. Lu Y. Govindan R. Wang L. Liu P.Y. Goodgame B. Wen W. Sezhiyan A. Pfeifer J. Li Y.F. Hua X. MicroRNA profiling and prediction of recurrence/relapse-free survival in stage I lung cancer Carcinogenesis 2012 33 1046 1054 10.1093/carcin/bgs100 22331473 85. Yanaihara N. Caplen N. Bowman E. Seike M. Kumamoto K. Yi M. Stephens R.M. Okamoto A. Yokota J. Tanaka T. Unique microRNA molecular profiles in lung cancer diagnosis and prognosis Cancer Cell 2006 9 189 198 10.1016/j.ccr.2006.01.025 16530703 86. Yu S.L. Chen H.Y. Chang G.C. Chen C.Y. Chen H.W. Singh S. Cheng C.L. Yu C.J. Lee Y.C. Chen H.S. MicroRNA signature predicts survival and relapse in lung cancer Cancer Cell 2008 13 48 57 10.1016/j.ccr.2007.12.008 18167339 87. Voortman J. Goto A. Mendiboure J. Sohn J.J. Schetter A.J. Saito M. Dunant A. Pham T.C. Petrini I. Lee A. MicroRNA expression and clinical outcomes in patients treated with adjuvant chemotherapy after complete resection of non-small cell lung carcinoma Cancer Res. 2010 70 8288 8298 10.1158/0008-5472.CAN-10-1348 20978195 88. Hu Z. Chen X. Zhao Y. Tian T. Jin G. Shu Y. Chen Y. Xu L. Zen K. Zhang C. Serum microRNA signatures identified in a genome-wide serum microRNA expression profiling predict survival of non-small-cell lung cancer J. Clin. Oncol. 2010 28 1721 1726 10.1200/JCO.2009.24.9342 20194856 89. Kaduthanam S. Gade S. Meister M. Brase J.C. Johannes M. Dienemann H. Warth A. Schnabel P.A. Herth F.J. Sultmann H. Serum miR-142-3p is associated with early relapse in operable lung adenocarcinoma patients Lung Cancer 2013 80 223 227 10.1016/j.lungcan.2013.01.013 23410826 90. Sanfiorenzo C. Ilie M.I. Belaid A. Barlesi F. Mouroux J. Marquette C.H. Brest P. Hofman P. Two panels of plasma microRNAs as non-invasive biomarkers for prediction of recurrence in resectable NSCLC PLoS One 2013 8 10.1371/journal.pone.0054596 91. Wang Y. Gu J. Roth J.A. Hildebrandt M.A. Lippman S.M. Ye Y. Minna J.D. Wu X. Pathway-based serum microRNA profiling and survival in patients with advanced stage non-small cell lung cancer Cancer Res. 2013 73 4801 4809 10.1158/0008-5472.CAN-12-3273 23774211 92. Besse B. Olaussen K.A. Soria J.C. ERCC1 and RRM1: Ready for prime time? J. Clin. Oncol. 2013 31 1050 1060 10.1200/JCO.2012.43.0900 23401439 93. Friboulet L. Olaussen K.A. Pignon J.P. Shepherd F.A. Tsao M.S. Graziano S. Kratzke R. Douillard J.Y. Seymour L. Pirker R. ERCC1 isoform expression and DNA repair in non-small-cell lung cancer N. Engl. J. Med. 2013 368 1101 1110 10.1056/NEJMoa1214271 23514287 94. Jakobsen J.N. Santoni-Rugiu E. Ravn J. Sorensen J.B. Intratumour variation of biomarker expression by immunohistochemistry in resectable non-small cell lung cancer Eur. J. Cancer 2013 10.1016/j.ejca.2013.04.003 95. Graziano S.L. Lacas B. Vollmer R. Kratzke R. Popper H. Filipits M. Seymour L. Shepherd F.A. Rosell R. Veillard A.S. Cross-validation analysis of the prognostic significance of mucin expression in patients with resected non-small cell lung cancer treated with adjuvant chemotherapy: Results from IALT, JBR.10 and ANITA Lung Cancer 2013 82 149 155 10.1016/j.lungcan.2013.06.015 23920379 96. Shepherd F.A. Domerg C. Hainaut P. Janne P.A. Pignon J.P. Graziano S. Douillard J.Y. Brambilla E. le Chevalier T. Seymour L. Pooled analysis of the prognostic and predictive effects of KRAS mutation status and KRAS mutation subtype in early-stage resected non-small-cell lung cancer in four trials of adjuvant chemotherapy J. Clin. Oncol. 2013 31 2173 2181 10.1200/JCO.2012.48.1390 23630215 97. Edelman M.J. Schneider C.P. Tsai C.M. Kim H.T. Quoix E. Luft A.V. Kaleta R. Mukhopadhyay P. Trifan O.C. Whitaker L. Randomized phase II study of ixabepilone or paclitaxel plus carboplatin in patients with non-small-cell lung cancer prospectively stratified by beta-3 tubulin status J. Clin. Oncol. 2013 31 1990 1996 10.1200/JCO.2012.45.3282 23589560 98. Holdenrieder S. von Pawel J. Dankelmann E. Duell T. Faderl B. Markus A. Siakavara M. Wagner H. Feldmann K. Hoffmann H. Nucleosomes, ProGRP, NSE, CYFRA 21-1, and CEA in monitoring first-line chemotherapy of small cell lung cancer Clin. Cancer Res. 2008 14 7813 7821 10.1158/1078-0432.CCR-08-0678 19047109 99. Tang H. Xiao G. Behrens C. Schiller J. Allen J. Chow C.W. Suraokar M. Corvalan A. Mao J. White M.A. A 12-gene set predicts survival benefits from adjuvant chemotherapy in non-small cell lung cancer patients Clin. Cancer Res. 2013 19 1577 1586 10.1158/1078-0432.CCR-12-2321 23357979 100. Chai Y. Lam R.K. Calaf G.M. Zhou H. Amundson S. Hei T.K. Radiation-induced non-targeted response in vivo : Role of the TGFbeta-TGFBR1-COX-2 signalling pathway Br. J. Cancer 2013 108 1106 1112 10.1038/bjc.2013.53 23412109 101. Yuan S.T. Ellingrod V.L. Schipper M. Stringer K.A. Cai X. Hayman J.A. Yu J. Lawrence T.S. Kong F.M. Genetic variations in TGFbeta1, tPA, and ACE and radiation-induced thoracic toxicities in patients with non-small-cell lung cancer J. Thoracic Oncol. 2013 8 208 213 10.1097/JTO.0b013e318274592e 102. Niu N. Qin Y. Fridley B.L. Hou J. Kalari K.R. Zhu M. Wu T.Y. Jenkins G.D. Batzler A. Wang L. Radiation pharmacogenomics: A genome-wide association approach to identify radiation response biomarkers using human lymphoblastoid cell lines Genome Res. 2010 20 1482 1492 10.1101/gr.107672.110 20923822 103. Torres-Roca J.F. Eschrich S. Zhao H. Bloom G. Sung J. McCarthy S. Cantor A.B. Scuto A. Li C. Zhang S. Prediction of radiation sensitivity using a gene expression classifier Cancer Res. 2005 65 7169 7176 10.1158/0008-5472.CAN-05-0656 16103067 104. Dehing-Oberije C. Aerts H. Yu S. de Ruysscher D. Menheere P. Hilvo M. van der Weide H. Rao B. Lambin P. Development and validation of a prognostic model using blood biomarker information for prediction of survival of non-small-cell lung cancer patients treated with combined chemotherapy and radiation or radiotherapy alone (NCT00181519, NCT00573040, and NCT00572325) Int. J. Radiat. Oncol. Biol. Phys. 2011 81 360 368 10.1016/j.ijrobp.2010.06.011 20888135 105. Ono A. Naito T. Ito I. Watanabe R. Shukuya T. Kenmotsu H. Tsuya A. Nakamura Y. Murakami H. Kaira K. Correlations between serial pro-gastrin-releasing peptide and neuron-specific enolase levels, and the radiological response to treatment and survival of patients with small-cell lung cancer Lung Cancer 2012 76 439 444 10.1016/j.lungcan.2011.12.012 22300752 106. Koh P.K. Faivre-Finn C. Blackhall F.H. de Ruysscher D. Targeted agents in non-small cell lung cancer (NSCLC): Clinical developments and rationale for the combination with thoracic radiotherapy Cancer Treatm. Rev. 2012 38 626 640 10.1016/j.ctrv.2011.11.003 107. Das A.K. Chen B.P. Story M.D. Sato M. Minna J.D. Chen D.J. Nirodi C.S. Somatic mutations in the tyrosine kinase domain of epidermal growth factor receptor (EGFR) abrogate EGFR-mediated radioprotection in non-small cell lung carcinoma Cancer Res. 2007 67 5267 5274 10.1158/0008-5472.CAN-07-0242 17545606 108. Wang M. Morsbach F. Sander D. Gheorghiu L. Nanda A. Benes C. Kriegs M. Krause M. Dikomey E. Baumann M. EGF receptor inhibition radiosensitizes NSCLC cells by inducing senescence in cells sustaining DNA double-strand breaks Cancer Res. 2011 71 6261 6269 10.1158/0008-5472.CAN-11-0213 21852385 109. Contessa J.N. Hampton J. Lammering G. Mikkelsen R.B. Dent P. Valerie K. Schmidt-Ullrich R.K. Ionizing radiation activates Erb-B receptor dependent Akt and p70 S6 kinase signaling in carcinoma cells Oncogene 2002 21 4032 4041 10.1038/sj.onc.1205500 12037685 110. Ulloa-Montoya F. Louahed J. Dizier B. Gruselle O. Spiessens B. Lehmann F.F. Suciu S. Kruit W.H. Eggermont A.M. Vansteenkiste J. Predictive gene signature in MAGE-A3 antigen-specific cancer immunotherapy J. Clin. Oncol. 2013 31 2388 2395 10.1200/JCO.2012.44.3762 23715562 111. Galon J. Angell H.K. Bedognetti D. Marincola F.M. The continuum of cancer immunosurveillance: Prognostic, predictive, and mechanistic signatures Immunity 2013 39 11 26 10.1016/j.immuni.2013.07.008 23890060 112. Morse M.A. Osada T. Hobeika A. Patel S. Lyerly H.K. Biomarkers and correlative endpoints for immunotherapy trials Am. Soc. Clin. Oncol. Educ. Book 2013 2013 287 293 113. Pikor L.A. Ramnarine V.R. Lam S. Lam W.L. Genetic alterations defining NSCLC subtypes and their therapeutic implications Lung Cancer 2013 82 179 189 10.1016/j.lungcan.2013.07.025 24011633 114. Cagle P.T. Sholl L.M. Lindeman N.I. Alsabeh R. Divaris D.X. Foulis P. Lee G. Neal J.W. Nowak J.A. Yu P.P. Template for reporting results of biomarker testing of specimens from patients with non-small cell carcinoma of the lung Arch. Pathol. Lab. Med. 2013 10.5858/arpa.2011.0232-CP 115. Neal J.W. Sequist L.V. Targeted therapies: Optimal first-line therapy for NSCLC with EGFR mutations Nat. Rev. Clin. Oncol. 2010 7 71 72 10.1038/nrclinonc.2009.191 20118973 116. Shaw A.T. Yeap B.Y. Solomon B.J. Riely G.J. Gainor J. Engelman J.A. Shapiro G.I. Costa D.B. Ou S.H. Butaney M. Effect of crizotinib on overall survival in patients with advanced non-small-cell lung cancer harbouring ALK gene rearrangement: A retrospective analysis Lancet Oncol. 2011 12 1004 1012 10.1016/S1470-2045(11)70232-7 21933749 117. Mao C. Qiu L.X. Liao R.Y. Du F.B. Ding H. Yang W.C. Li J. Chen Q. KRAS mutations and resistance to EGFR-TKIs treatment in patients with non-small cell lung cancer: A meta-analysis of 22 studies Lung Cancer 2010 69 272 278 10.1016/j.lungcan.2009.11.020 20022659 118. Pao W. Wang T.Y. Riely G.J. Miller V.A. Pan Q. Ladanyi M. Zakowski M.F. Heelan R.T. Kris M.G. Varmus H.E. KRAS mutations and primary resistance of lung adenocarcinomas to gefitinib or erlotinib PLoS Med. 2005 2 10.1371/journal.pmed.0020017 119. Laurie S.A. Goss G.D. Role of epidermal growth factor receptor inhibitors in epidermal growth factor receptor wild-type non-small-cell lung cancer J. Clin. Oncol. 2013 31 1061 1069 10.1200/JCO.2012.43.4522 23401452 120. Carbone D.P. Ding K. Roder H. Grigorieva J. Roder J. Tsao M.S. Seymour L. Shepherd F.A. Prognostic and predictive role of the VeriStrat plasma test in patients with advanced non-small-cell lung cancer treated with erlotinib or placebo in the NCIC Clinical Trials Group BR.21 trial J. Thoracic Oncol. 2012 7 1653 1660 10.1097/JTO.0b013e31826c1155 121. Martini M. Vecchione L. Siena S. Tejpar S. Bardelli A. Targeted therapies: How personal should we go? Nat. Rev. Clin. Oncol. 2012 9 87 97 22083042 122. Sequist L.V. Waltman B.A. Dias-Santagata D. Digumarthy S. Turke A.B. Fidias P. Bergethon K. Shaw A.T. Gettinger S. Cosper A.K. Genotypic and histological evolution of lung cancers acquiring resistance to EGFR inhibitors Sci. Transl. Med. 2013 7 75ra26 10.1126/scitranslmed.3002003 123. Byers L.A. Diao L. Wang J. Saintigny P. Girard L. Peyton M. Shen L. Fan Y. Giri U. Tumula P.K. An epithelial-mesenchymal transition gene signature predicts resistance to EGFR and PI3K inhibitors and identifies Axl as a therapeutic target for overcoming EGFR inhibitor resistance Clin. Cancer Res. 2013 19 279 290 10.1158/1078-0432.CCR-12-1558 23091115 124. Blumenschein G.R. Saintigny P. Liu S. Kim E.S. Tsao A.S. Herbst R. Alden C.M. Lee J.J. Tang X. Stewart D.J. Comprehensive biomarker analysis and final efficacy results of sorafenib in the BATTLE (Biomarker-Integrated Approaches of Targeted Therapy for Lung Cancer Elimination) trial Clin. Cancer Res. 2013 10.1158/1078-0432.CCR-12-1818 125. Okayama H. Kohno T. Ishii Y. Shimada Y. Shiraishi K. Iwakawa R. Furuta K. Tsuta K. Shibata T. Yamamoto S. Identification of genes upregulated in ALK-positive and EGFR/KRAS/ALK-negative lung adenocarcinomas Cancer Res. 2012 72 100 111 10.1158/0008-5472.CAN-11-1403 22080568 126. Chitale D. Gong Y. Taylor B.S. Broderick S. Brennan C. Somwar R. Golas B. Wang L. Motoi N. Szoke J. An integrated genomic analysis of lung cancer reveals loss of DUSP4 in EGFR-mutant tumors Oncogene 2009 28 2773 2783 10.1038/onc.2009.135 19525976 127. Ettinger D.S. Akerley W. Borghaei H. Chang A.C. Cheney R.T. Chirieac L.R. D’Amico T.A. Demmy T.L. Govindan R. Grannis F.W. Non-small cell lung cancer, version 2.2013 J. Natl. Compr. Cancer Netw. 2013 11 645 653 128. Muller B.M. Keil E. Lehmann A. Winzer K.J. Richter-Ehrenstein C. Prinzler J. Bangemann N. Reles A. Stadie S. Schoenegg W. The endopredict gene-expression assay in clinical practice–performance and impact on clinical decisions PLoS One 2013 8 10.1371/journal.pone.0068252 129. Rouzier R. Pronzato P. Chereau E. Carlson J. Hunt B. Valentine W.J. Multigene assays and molecular markers in breast cancer: Systematic review of health economic analyses Breast Cancer Res. Treatm. 2013 139 621 637 10.1007/s10549-013-2559-1 130. Azim H.A. Jr. Michiels S. Zagouri F. Delaloge S. Filipits M. Namer M. Neven P. Symmans W.F. Thompson A. Andre F. Utility of prognostic genomic tests in breast cancer practice: The IMPAKT 2012 Working Group Consensus Statement Ann. Oncol. 2013 24 647 654 10.1093/annonc/mds645 23337633 131. Subramanian J. Simon R. Gene expression-based prognostic signatures in lung cancer: Ready for clinical use? J. Natl. Cancer Instit. 2010 102 464 474 132. Bild A.H. Yao G. Chang J.T. Wang Q. Potti A. Chasse D. Joshi M.B. Harpole D. Lancaster J.M. Berchuck A. Oncogenic pathway signatures in human cancers as a guide to targeted therapies Nature 2006 439 353 357 16273092 133. Ding L. Getz G. Wheeler D.A. Mardis E.R. McLellan M.D. Cibulskis K. Sougnez C. Greulich H. Muzny D.M. Morgan M.B. Somatic mutations affect key pathways in lung adenocarcinoma Nature 2008 455 1069 1075 10.1038/nature07423 18948947

PMC005xxxxxx/PMC5003441.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2040304microarrays-02-00304ReviewChromosomal Microarrays in Prenatal Diagnosis: Time for a Change of Policy? Miny Peter 1*Wenzel Friedel 1Tercanli Sevgi 2Filges Isabel 131 Medical Genetics, Department of Biomedicine, University Hospital Basel, Burgfelderstr. 101, Building J, CH-4055 Basel, Switzerland; E-Mails: friedel.wenzel@usb.ch (F.W.); isabel.filges@unibas.ch (I.F.)2 Ultraschall Freie Strasse, Freie Strasse 38, CH-4001 Basel, Switzerland; E-Mail: sevgi.tercanli@unibas.ch3 Department of Medical Genetics, Box 153, BC Children’s and Women’s Hospital, 4480 Oak Street, Vancouver BC, V6H 3V4, Canada* Author to whom correspondence should be addressed; E-Mail: peter.miny@unibas.ch; Tel.: +41-61-265-3620; Fax: +41-61-265-3621.05 12 2013 12 2013 2 4 304 317 22 10 2013 19 11 2013 27 11 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Microarrays have replaced conventional karyotyping as a first-tier test for unbalanced chromosome anomalies in postnatal cytogenetics mainly due to their unprecedented resolution facilitating the detection of submicroscopic copy number changes at a rate of 10–20% depending on indication for testing. A number of studies have addressed the performance of microarrays for chromosome analyses in high risk pregnancies due to abnormal ultrasound findings and reported an excess detection rate between 5% and 10%. In low risk pregnancies, clear pathogenic copy number changes at the submicroscopic level were encountered in 1% or less. Variants of unclear clinical significance, unsolicited findings, and copy number changes with variable phenotypic consequences are the main issues of concern in the prenatal setting posing difficult management questions. The benefit of microarray testing may be limited in pregnancies with only moderately increased risks (advanced maternal age, positive first trimester test). It is suggested to not change the current policy of microarray application in prenatal diagnosis until more data on the clinical significance of copy number changes are available. microarraysarray CGHprenatal diagnosis ==== Body 1. Introduction Microarrays detecting copy number changes in genomic DNA [1] have replaced conventional microscopic chromosome analyses in the routine diagnostic work-up of children and adults with suspected unbalanced chromosome anomalies during recent years [2,3]. The conventional approach is now restricted to the confirmation of clinically distinct chromosomal syndromes (e.g., Down syndrome) and the diagnosis of balanced structural anomalies in potential carriers such as family members at risk and couples with recurrent abortions or infertility. The reasons for this rather rapid paradigm change include a greatly increased resolution of this whole genome approach which allowed the definition of a number of previously unknown microdeletion or -duplication syndromes [4,5,6], but also the prospect of an easier standardization and finally automation of lab procedures as well as diagnostic assessment. Although equal benefits may be expected for the prenatal diagnosis of chromosome anomalies the conventional microscopic approach has maintained its role as the “gold standard” for the time being [7,8]. The use of microarrays in this setting has been met with caution for various reasons [9,10]. These include the common observation of copy number variation (CNV) with unknown clinical significance (VOUS) being obviously more difficult to manage in the sensitive prenatal setting as well as the fact that the majority of all unbalanced chromosome findings in prenatal diagnosis concern common trisomies or other anomalies perfectly amenable to the conventional diagnostic approach or by the recently introduced non-invasive prenatal testing (NIPT) on a maternal blood sample [11]. This latter method uses high throughput sequencing technology on cell free DNA (cf DNA) fragments in the maternal plasma to assess the copy number of the most prevalent fetal aneuploidies currently including chromosomes 13, 18, 21, X and Y. Considerable attention, including extensive discussions in the lay press, has been paid to NIPT since the longstanding “dream” of a prenatal genetic diagnosis without physical risks for fetus and mother finally came true. NIPT currently dominates public discussions on prenatal diagnosis. This contribution will review the current clinical application of microarrays in prenatal cytogenetics and discuss aspects relevant to its formal implementation as an addition to or as a replacement of alternative diagnostic options. 2. Procedures and Methods 2.1. The Current Approach to Prenatal Diagnosis of Chromosome Anomalies: A Brief Introduction For decades advanced maternal age (in practice mostly >34 years) has been the predominant indication for amniocentesis (AC) or chorionic villus sampling (CVS) in order to obtain cells required for a cytogenetic work-up. A procedure-related risk for pregnancy loss of 0.5–1% is usually quoted for both these invasive procedures, but is likely to be significantly lower in experienced centers today [12]. Over the years the microscopic chromosome analysis has been amended with various molecular extensions for rapid aneuploidy testing or the targeted diagnosis of selected submicroscopic structural changes (fluorescence in situ hybridization (FISH) [13]; QF-PCR [14], MLPA [15]). For around 20 years, the additional assessment of maternal serum parameters and the more recent sonographic measurement of the thickness of the fetal nuchal skin (nuchal translucency) has greatly improved the pre-procedural risk assessment. The current first trimester test according to the Fetal Medicine Foundation London [16] or Germany [17] is the most popular among various proposed risk assessment schemes for all pregnancies regardless of maternal age. The sensitivity for trisomy 21 is around 90% for a false-positive rate of 5%. The implementation of this improved so-called risk “screening” has led to a significant decrease of invasive procedures [18]. The recently introduced NIPT, which is rather considered an advanced screening tool than a diagnostic test, offers a sensitivity above 99% for a false-positive rate of below 1% for trisomy 21 [19]. Referring to the still limited experience current recommendations suggest a conservative testing approach restricted to pregnancies at increased risk for common aneuploidies due to maternal age or first trimester test result [20,21,22]. Most experts agree, however, that NIPT will sooner or later replace the current risk assessment schemes except for the sonographic evaluation of the fetus and induce another dramatic decrease of invasive procedures. Invasive testing remains the primary option of choice in pregnancies at high risk for chromosome abnormalities [12]. These are mainly identified by a first trimester test result in the high positive range or by fetal malformations and other anomalies assessed by ultrasound. A total of some 5–10% of all pregnancies will probably fall into this category. Specific congenital anomalies (such as, e.g., an omphalocele or heart defects), particularly in multiple occurrence or combined with intrauterine growth restriction, may pose a high risk for unbalanced chromosome counts including unusual structural changes and microdeletion or -duplication syndromes. In parallel to experiences in postnatal diagnostics, microarrays would be the subsequent diagnostic option in pregnancies with suggestive ultrasound findings but a normal conventional karyotyping result. A subgroup of couples, however, ready to accept the small procedure related risk, will seek a comprehensive exclusion of fetal chromosomal conditions irrespective of prior risk assessments. 2.2. Technical Aspects Following the seminal paper by Pinkel et al. [23] the diagnostic application of array CGH started with BAC clones and insert sizes in the 100–150 kb range targeted to genome regions of interest in cancer samples. The number of targets steadily increased and tiling path BAC arrays with 32,000 targets covering the whole genome began to be used in constitutional cytogenetics to search for cryptic chromosome anomalies [24]. In a further evolutionary step the large BAC inserts were replaced by synthetic oligonucleotides of sizes in the 50 base range resulting in a major leap in resolution with current target numbers of several millions. In parallel single nucleotide polymorphism (SNP) arrays became available originally designed for genotyping patients in genome wide association studies for complex diseases. These platforms also detect copy number changes and due to the abundance of SNPs throughout the genome are able to provide high resolution coverage. Some current dedicated array designs for use in high resolution cytogenetics rely on a combination of SNPs and oligonucleotides in order to maintain the advantages of both approaches. For a more in-depth review of the technical aspects involved in the clinical use of microarrays see references [25,26]. 2.3. Limitations of Clinical Relevance Microarrays detect changes in copy number but not balanced chromosome anomalies such as translocations, inversions and others. While the copy number change provides the critical information relevant to the immediate management of the pregnancy it does not reveal its etiology which in a proportion of cases will be an unbalanced segregational product of a parental balanced rearrangement. This requires adequate follow-up studies using methods visualizing the chromosome structure such as conventional chromosome preparation and FISH. Ploidy changes such as triploidy or tetraploidy have been diagnosed with SNP arrays [27] but are not detectable by array CGH, a problem that can at least be partly solved by the use of abnormal control DNA (47, XXY) [28]. In principle, mosaicism will show with both approaches but the minimum level of detection is not clearly established and depends on technical as well as biological aspects [29], which is also true for conventional cytogenetics. The inadvertent detection of VOUS is the currently most significant objection against replacing conventional karyotyping with microarrays in all pregnancies undergoing invasive testing. There is an ongoing debate on the role of array design in terms of probe distribution (targeted versus genome wide) and the resolution attempting to optimize the trade-off between detection rate for clinically significant findings and the frequency of VOUS [9]. A current estimate of the mutation rates for large CNVs (>100 kbp) is 1.2 × 10−2 with a still unknown pathogenic proportion in fetuses [30]. Also mutation rates for smaller CNVs and their clinical impact have to be assessed in appropriate trials. Unsolicited findings with major clinical impact for the fetus and/or other family members such as predispositions for late onset disease as well as the management of CNVs with variable clinical expressivity and consequences are further issues of concern. 3. Results and Discussion Numerous studies differ significantly in case numbers, patient ascertainment, design, and methodology and have addressed the central questions of interest: How often do microarrays show clinically significant copy number changes not detectable by conventional cytogenetics and how frequent are variants of unknown significance (VOUS). In a large collaborative trial of 29 US centers on more than 4,000 pregnancies at increased risk for chromosomal abnormalities due to advanced maternal age, first trimester test result, or abnormal ultrasound findings [31] microarrays detected all anomalies diagnosed by conventional cytogenetics except for balanced structural changes and triploidies as expected. In the subgroup of pregnancies with abnormal ultrasound findings and a normal conventional karyotype microarrays showed microdeletion or -duplications in 6%, half of these considered to be known pathogenic and the remaining potentially clinically significant, including those with variable clinical expressivity. The respective findings in the two lower risk groups were 1.7% with approximately 2/3 of these being potentially clinically significant. The overall rate of common benign copy number variation was around 30%. From a single lab with more than 5,000 prenatal cases tested for various indications using platforms evolving with the technical progress a detection rate of 6.5% was reported for pregnancies with abnormal ultrasound findings and 8.2% after fetal demise [32]. VOUS were seen in 4.2% of all cases, a rate dropping to 0.4% if only de novo findings were included. 71% of the aberrations were not detectable by conventional karyotyping. The same group reported on risk stratification according to specific ultrasound findings in a retrospective study on almost 3,000 pregnancies [33]. Particularly high detection rates (10% or more) were observed in pregnancies with anomalies in two or more organ systems, and specific malformations including holoprosencephaly, posterior fossa defects, skeletal anomalies, ventricular septal defect, hypoplastic left heart, and cleft lip/palate occurring either isolated or combined with other anomalies. In a Spanish series of 276 pregnancies with fetal heart defects the overall rate of microscopically visible chromosomal pathology was high (15.9%) [34]. In a subset of pregnancies with a normal karyotype a targeted fluorescence in situ hybridization (FISH) test for 22q11 deletions resulted in an anomaly rate of 6.4%. Microarray testing in 51 patients with a normal karyotype and normal or no FISH result revealed one pathogenic copy number variant (2%) and no VOUS. Hillman et al. [35] reported an excess detection rate (aberrations detected in addition to conventional chromosome count) of 4.1% in their prospective cohort study on 243 pregnancies with abnormal ultrasound findings using a relatively low resolution BAC-based platform. The authors provide a systematical review and meta-analysis of relevant case series (totaling > 18,000 pregnancies) including the ones addressed above [31,32]. The overall excess detection rate was 10% and 7% for series published in more recent years (2011–2012). VOUS were observed in 2.1% of cases in the abnormal ultrasound group and in 1.4% with other indications. Another recent review of 18 series using different inclusion criteria [36] found an excess anomaly rate of 5.6% in 2,220 pregnancies with ultrasound anomalies in one anatomical system and 9.1% in 1,139 pregnancies with multiple anomalies. Similar detection rates were reported from a single center on 410 pregnancies [37]. VOUS were seen in 1.6% of all 1,115 cases. In a recent assessment of the clinical utility of microarray technologies [38] the review focused on more than 12,000 pregnancies with a normal conventional karyotype from various published series including some of those mentioned earlier. The rate of pathogenic copy number changes (pCNC) was 6.5% in pregnancies ascertained with an abnormal ultrasound, 1% with advanced maternal age and 1.1% with other indications (parental anxiety, abnormal serum screening and others). Series with abnormal ultrasound as an exclusive indication were considered separately, and an overall pCNC rate of 7% was found. The authors refrained from reporting VOUS rates because they considered the assessment conditions to be heterogeneous in the individual series. 3.1. Clinical Utility of Microarrays Drawing conclusions from the published evidence some caveats apply. Data currently available have been obtained using a variety of array platforms with low resolution BAC-based targeted arrays at one end of the spectrum and high resolution genotyping chips at the other. There is no formal agreement on a minimal level of resolution for microarrays used in prenatal diagnosis for the time being. The classification of VOUS and CNVs reported as pathogenic is not consistent in the different series and the assignment of a VOUS may be a matter of discretion (Figure 1), in particular if the parental genotype is unknown [38]. Considering the different clinical and diagnostic settings with labs serving highly specialized ultrasound units on the one hand and large commercial centers obtaining samples from a variety of sources on the other hand, patient ascertainment is likely to differ between the series as well. Regarding those heterogeneities and biases in data acquisition, we consider a meaningful comparison and stratification of reported CNVs and VOUS to be difficult and the drawing of firm conclusions premature at present. Figure 1 Artefact, variant of unknown significance (VOUS), or pathogenic copy number variation (CNV)? Markers left of the vertical line (n = 13) suggesting a small (1.2 kb) but intragenic (intronic) SOS1 deletion in a pregnancy with isolated increased nuchal translucency (>99. centile). SOS1 mutations are a known cause of a (mostly mild) Noonan syndrome. The variant was considered to be likely benign if real at all but extensively discussed with the parents in a formal counseling session. They decided against any further testing and the pregnancy is ongoing. A positive correlation between array resolution and the detection rate of pathogenic CNVs as well as of VOUS is likely and not unexpected [39,40]. Srebniak et al. [41] have demonstrated this by retrospectively reassessing high resolution genome-wide array data of 456 fetuses at different resolution levels. They consider an implemented 0.5 Mb minimal detection threshold on a high resolution platform to be a favorable trade-off between the relevant criteria and propose this approach as a first-line prenatal cytogenetic test in cases without ultrasound abnormalities. This is an interesting model which, in contrast to the use of different platforms depending on the indication as applied by Shaffer et al. [42], implies the possibility to re-evaluate the data at a higher resolution if necessary e.g., in cases of ultrasound scan anomalies manifesting later in pregnancy. Approaches like these may finally pave the way for the replacement of conventional karyotyping by microarrays, but require careful pre-test information of the patients. The use of high resolution array designs in prenatal diagnosis was stated by most laboratories in a recent meeting report of the Genetic Services Quality Committee of the European Society of Human Genetics [9]. Targeted designs were considered to be disadvantageous because pathogenic imbalances may be missed and frequent updates of such platforms are required in order to include newly reported conditions. The authors also favor a common approach to post- as well as prenatal testing for most laboratories in order to gain a maximum of experience in data interpretation with a given platform. Solid evidence accumulated in years of pre- and postnatal testing confirmed that microarrays reliably detect all copy number variation regardless of size within their technical limitations as discussed above. They will detect additional pathogenic CNVs as compared to conventional karyotyping in a proportion of cases depending on indication for testing and microarray resolution chosen. A preliminary reasonable estimate of the excess detection rate of microarrays in pregnancies with sonographic anomalies in a single anatomical system to be used for the counseling of affected parents might be around 5% rising up to 10% with multiple anomalies. Available data do not allow a meaningful and reliable stratification of detection rates according to specific sonographic findings at this time. The excess detection rate is significantly lower in pregnancies with an only moderate risk increase due to maternal age or a positive first trimester test result with 0.6% for known or potentially pathogenic CNVs in one major series [43], 1.7% (0.5% known pathogenic) in the US collaborative study [31] and 1.3% in a single center series of almost 400 patients (advanced maternal age exclusively) [37]. We are unaware of data for first trimester test results in the high positive range (e.g., >1:10). Figure 2 CNV with variable phenotype. Rare deletion (0.75 Mb) of the distal part of the 22q11 critical region for the DiGeorge/Velocardiofacial syndrome. Variable phenotypes have been reported [44]. The parents decided against further testing, the pregnancy is ongoing. For reasons indicated earlier a sensible estimate of the frequency of VOUS is difficult. Published data suggest, however, that in low risk pregnancies VOUS might be more frequently encountered than true pathogenic CNVs at least with high resolution platforms [41]. The availability of parental blood samples is obviously a critical issue for the classification of copy number changes of doubtful significance and may reduce the VOUS rate dramatically [32] implying that a variant also present in a healthy parents is likely benign. This is usually reassuring to the parents and physicians but not always applicable. Increasing experience and a careful collection of data are expected to continuously ameliorate classification issues of genomic variation in the future. Well-defined CNVs with a variable prognosis such as the duplication of the Williams syndrome critical region [45] and others [6] (Figure 2) may also pose counseling challenges but are not specific to microarrays and well-known in classical chromosomal syndromes as well monogenic conditions. Unsolicited findings such as late onset inherited disorders or cancer predispositions are rare but of particular concern in the prenatal setting. Their frequency has been estimated as 1–2 per thousand [9] or 0.6% in a large postnatal cohort for CNVs affecting cancer genes [46]. 3.2. Counseling Issues High resolution karyotyping by microarrays is currently the most comprehensive approach to test for classical chromosomal disorders as well as submicroscopic copy number changes with a proven record of diagnostic accuracy. For some time, it has already been an essential extension to established diagnostic tools available in most specialized diagnostic labs and mainly offered in pregnancies at high risk for chromosomal conditions. Respecting patient autonomy [47] obliges to address microarray testing when discussing prenatal testing options in all pregnancies just as the new non-invasive aneuploidy tests. If microarray testing is considered pretest counseling must include information on the indication related expected detection rate, resolution specific VOUS rate as well as the possible need to test the parents. A clearly written agreement on the disclosure of unexpected or uncertain test results should be mandatory. 3.3. The Local Approach to Prenatal Microarray Testing We have introduced microarrays into prenatal cytogenetics having temporarily used a low resolution BAC based platform, but soon switched to a high resolution mixed oligo- and SNP array which we also use for postnatal testing. Our setting can be described as being a small academic lab working with a limited number of obstetricians highly specialized in maternal fetal medicine and serving mostly local patients. This overall setting allows for a close contact between the professionals involved as well as an immediate access to the patients. Comprehensive pre- and post-test counseling is provided. Our current policy is to: Highly recommend microarray testing for further characterization of abnormal results obtained by conventional chromosome analysis which are of unclear clinical significance such as de novo translocations, inversions, marker chromosomes and others. Recommend microarray testing in pregnancies at high risk for chromosomal abnormalities due to abnormal ultrasound findings or first trimester test results in the high positive range. Not encourage but accept the occasional parental request for a comprehensive exclusion of fetal chromosomal conditions even without a significant risk increase. Address microarray testing routinely with all patients seeking general advice on prenatal risk screening and testing options. A general precondition for microarray testing is the parental consent to provide a blood sample if this is required to classify a CNV. Independent of insurance coverage issues the counseling approach is individualized, not strictly adhering to prefixed risk cut-offs which do serve, however, as a general orientation guide. One of these, relevant for insurance coverage, is the risk for “a fetal genetic condition” at a maternal age of 35 years. For practical implications, we consider risks smaller than this as low, risks beyond 2% as high and the range in between as intermediate. The access to microarray testing is not assigned but based on parental choice. The regular schedule includes 1 to 2 array runs per week allowing for a turn-around time of 7 to 10 days. Emergency testing within 3–4 days is available if required to provide a result before 24 completed weeks of gestation, the de facto limit for a termination of the pregnancy in this country. Microarray testing is usually carried out on uncultured amniotic fluid cells or chorionic villi if QF-PCR or direct chromosome preparation revealed a normal result. A backup culture is set up routinely. In cases with common aneuploidies a conventional chromosome analysis is initiated. We do not advocate further conventional testing if the microarray result was normal. Turn-around times for QF-PCR and direct chromosome preparation are 1–3 days and around 10 days for a conventional karyotype. In emergencies (high risk, late gestational age) microarrays are used as a first line test. Our current strategy is to report all known pathogenic deletions larger than 100 kb, duplication larger than 200 kb and findings of potential significance with regard to the individual reason for referral. Patients undergoing microarray testing are aware of the possible occurrence of VOUS or other unsolicited findings. We do not ask for parental blood samples at the time of the invasive procedure. Should further testing be required this is discussed in a formal counseling session. Most parents agree to just have the CNV in question checked which has solved most VOUS issues we have had so far. CNVs with variable phenotypic consequences are always communicated and carefully discussed with the parents in analogy to e.g., an extra Y chromosome in conventional prenatal karyotyping. Pretest counseling includes also information on the unlikely event of detecting a predisposition for certain cancers or late onset disease. Written information, including an individual agreement on how to proceed with such information, has been in preparation for some time but will not be easily implemented into practice regarding the emotionally exceptional situation of many parents discussing a test after major fetal anomalies have been diagnosed. 3.4. Should Microarrays Replace Conventional Karyotyping as a First-Tier Prenatal Diagnostic Test? We believe that an entire replacement is premature and suggest adhering to the current approach shared by a majority of centers to not actively promote microarrays in low-risk pregnancies until more experience has been accumulated. We expect that a growing body of data on the clinical significance of copy number variation will help to prevent the provision of ambiguous information to our patients. The use of low-resolution targeted platforms to avoid the detection of VOUS will also miss pathogenic CNVs and may be a temporary solution for low-risk pregnancies but does not appear to be a promising long term option in particular for testing high risk pregnancies. Our own experience is in complete accordance with the recommendations by Vetro et al. [9] that a common platform for pre- and postnatal testing is advantageous in smaller labs. In our view this and other issues addressed earlier will enhance the trend towards high resolution platforms with genome wide coverage. Indication-specific or personal choice adaptation of the resolution level in such platforms [41] may be a viable option to increase the overall acceptability of molecular cytogenetics. In principal, prenatal microarray testing does not alter the traditional dogma that invasive testing requires an increased risk for the conditions to exclude as practiced in some countries but not in others such as the US [48]. In health care systems with restricted access to invasive testing strictly adhering to risk cut-offs, the excess detection rate of microarray testing will have to be considered once reliable data are available. We do not advocate additional conventional karyotyping for economic reasons if the microarray result was normal being aware of missing an occasional balanced rearrangement some of which may pose a small risk for uniparental disomy. Abnormal microarray results must be followed-up by fetal and, if appropriate, parental karyotyping in order to exclude a parental balanced rearrangement implying a recurrence risk. Bui et al. [49] provide a more detailed account of the different views regarding the practical implementation of microarray testing. 4. Conclusions Microarrays are an established molecular tool for high resolution karyotyping in prenatal diagnosis. They are available as a variety of platforms with considerable differences in resolution and coverage of the genome. Their clinical utility is widely accepted in high risk pregnancies. The general replacement of conventional karyotyping as a first-tier clinical test also in pregnancies at low risk for chromosome anomalies is advocated by some but presently not supported by a majority of experts [7,8] mainly due to the detection of VOUS and other issues difficult to manage in a prenatal setting. While a change of policy may have been straightforward in postnatal cytogenetics, the prenatal setting proves to be incomparably more complex involving parental as well as fetal concerns, procedure related risks and method specific benefits and limitations. There is little doubt, however, that microarrays will eventually replace conventional karyotyping also in the prenatal setting in the near future, if invasive testing is required or requested. The novel non-invasive testing options, which appear to be gaining popularity, are expected to continuously restrict these requests to true high risk pregnancies. Conflicts of Interest The authors declare no conflict of interest. ==== Refs References 1. Schaaf C.P. Wiszniewska J. Beaudet A.L. Copy number and SNP arrays in clinical diagnostics Annu. Rev. Genom. Hum. Genet. 2011 12 25 51 10.1146/annurev-genom-092010-110715 2. Miller D.T. Adam M.P. Aradhya S. Biesecker L.G. Brothman A.R. Carter N.P. Church D.M. Crolla J.A. Eichler E.E. Epstein C.J. Consensus statement: Chromosomal microarray is a first-tier clinical diagnostic test for individuals with developmental disabilities or congenital anomalies Am. J. Hum. Genet. 2010 86 749 764 10.1016/j.ajhg.2010.04.006 20466091 3. Manning M. Hudgins L. Array-based technology and recommendations for utilization in medical genetics practice for detection of chromosomal abnormalities Genet. Med. 2010 12 742 745 10.1097/GIM.0b013e3181f8baad 20962661 4. Cooper G.M. Coe B.P. Girirajan S. Rosenfeld J.A. Vu T.H. Baker C. Williams C. Stalker H. Hamid R. Hannig V. A copy number variation morbidity map of developmental delay Nat. Genet. 2011 43 838 846 10.1038/ng.909 21841781 5. Vissers L.E. Stankiewicz P. Microdeletion and microduplication syndromes Methods Mol. Biol. 2012 838 29 75 22228006 6. Carvill G.L. Mefford H.C. Microdeletion syndromes Curr. Opin. Genet. Dev. 2013 23 232 239 10.1016/j.gde.2013.03.004 23664828 7. ACOG Committee Opinion No. 446: Array comparative genomic hybridization in prenatal diagnosis (Replaced by Committee Opinion No. 581) Obstet. Gynecol. 2009 114 1161 1163 10.1097/AOG.0b013e3181c33cad 20168129 8. Novelli A. Grati F.R. Ballarati L. Bernardini L. Bizzoco D. Camurri L. Casalone R. Cardarelli L. Cavalli P. Ciccone R. Microarray application in prenatal diagnosis: A position statement from the cytogenetics working group of the italian society of human genetics (sigu), November 2011 Ultrasound Obstet. Gynecol. 2012 39 384 388 22262341 9. Vetro A. Bouman K. Hastings R. McMullan D.J. Vermeesch J.R. Miller K. Sikkema-Raddatz B. Ledbetter D.H. Zuffardi O. van Ravenswaaij-Arts C.M. The introduction of arrays in prenatal diagnosis: A special challenge Hum. Mutat. 2012 33 923 929 10.1002/humu.22050 22508381 10. Stark Z. Gillam L. Walker S.P. McGillivray G. Ethical controversies in prenatal microarray Curr. Opin. Obstet. Gynecol. 2013 25 133 137 10.1097/GCO.0b013e32835ebb67 23454982 11. Benn P. Cuckle H. Pergament E. Non-invasive prenatal testing for aneuploidy: Current status and future prospects Ultrasound Obstet. Gynecol. 2013 42 15 33 10.1002/uog.12513 23765643 12. Simpson J.L. Invasive procedures for prenatal diagnosis: Any future left? Best Pract. Res. Clin. Obstet. Gynaecol. 2012 26 625 638 10.1016/j.bpobgyn.2012.05.007 22749621 13. Stumm M. Tonnies H. Fluorescence in situ hybridization techniques in medical diagnostics Expert Opin. Med. Diagn. 2008 2 1381 1390 10.1517/17530050802558899 23496784 14. Mann K. Ogilvie C.M. QF-PCR: Application, overview and review of the literature Prenat. Diagn. 2012 32 309 314 10.1002/pd.2945 22467160 15. Willis A.S. van den Veyver I. Eng C.M. Multiplex ligation-dependent probe amplification (MLPA) and prenatal diagnosis Prenat. Diagn. 2012 32 315 320 10.1002/pd.3860 22467161 16. Nicolaides K.H. Spencer K. Avgidou K. Faiola S. Falcon O. Multicenter study of first-trimester screening for trisomy 21 in 75,821 pregnancies: Results and estimation of the potential impact of individual risk-orientated two-stage first-trimester screening Ultrasound Obstet. Gynecol. 2005 25 221 226 10.1002/uog.1860 15736186 17. Merz E. Thode C. Alkier A. Eiben B. Hackeloer B.J. Hansmann M. Huesgen G. Kozlowski P. Pruggmaier M. Wellek S. A new approach to calculating the risk of chromosomal abnormalities with first-trimester screening data Ultraschall Med. 2008 29 639 645 10.1055/s-2008-1027958 19085755 18. Ekelund C.K. Jørgensen F.S. Petersen O.B. Sundberg K. Tabor A. Danish Fetal Medicine Research Group Impact of a new national screening policy for Down’s syndrome in Denmark: Population based cohort study BMJ 2008 337 a2547 10.1136/bmj.a2547 19039015 19. Morain S. Greene M.F. Mello M.M. A new era in noninvasive prenatal testing N. Engl. J. Med. 2013 369 499 501 10.1056/NEJMp1304843 23862975 20. Gregg A.R. Gross S.J. Best R.G. Monaghan K.G. Bajaj K. Skotko B.G. Thompson B.H. Watson M.S. The Noninvasive Prenatal Screening Work Group of the American College of Medical Genetics and Genomics. ACMG statement on noninvasive prenatal screening for fetal aneuploidy Genet. Med. 2013 15 395 398 10.1038/gim.2013.29 23558255 21. ACOG Committee Opinion No. 545: Noninvasive prenatal testing for fetal aneuploidy Obstet. Gynecol. 2012 120 1532 1534 10.1097/01.AOG.0000423819.85283.f4 23168792 22. Benn P. Borell A. Chiu R. Cuckle H. Dugoff L. Faas B. Gross S. Johnson J. Maymon R. Norton M. Position statement from the aneuploidy screening committee on behalf of the board of the international society for prenatal diagnosis Prenat. Diagn. 2013 33 622 629 10.1002/pd.4139 23616385 23. Pinkel D. Segraves R. Sudar D. Clark S. Poole I. Kowbel D. Collins C. Kuo W.L. Chen C. Zhai Y. High resolution analysis of DNA copy number variation using comparative genomic hybridization to microarrays Nat. Genet. 1998 20 207 211 10.1038/2524 9771718 24. Vissers L.E. de Vries B.B. Osoegawa K. Janssen I.M. Feuth T. Choy C.O. Straatman H. van der Vliet W. Huys E.H. van Rijk A. Array-based comparative genomic hybridization for the genomewide detection of submicroscopic chromosomal abnormalities Am. J. Hum. Genet. 2003 73 1261 1270 10.1086/379977 14628292 25. Brady P.D. Vermeesch J.R. Genomic microarrays: A technology overview Prenat. Diagn. 2012 32 336 343 10.1002/pd.2933 22467164 26. Pinto D. Darvishi K. Shi X. Rajan D. Rigler D. Fitzgerald T. Lionel A.C. Thiruvahindrapuram B. Macdonald J.R. Mills R. Comprehensive assessment of array-based platforms and calling algorithms for detection of copy number variants Nat. Biotechnol. 2011 29 512 520 10.1038/nbt.1852 21552272 27. Tyreman M. Abbott K.M. Willatt L.R. Nash R. Lees C. Whittaker J. Simonic I. High resolution array analysis: Diagnosing pregnancies with abnormal ultrasound findings J. Med. Genet. 2009 46 531 541 10.1136/jmg.2008.065482 19451135 28. Ballif B.C. Kashork C.D. Saleki R. Rorem E. Sundin K. Bejjani B.A. Shaffer L.G. Detecting sex chromosome anomalies and common triploidies in products of conception by array-based comparative genomic hybridization Prenat. Diagn. 2006 26 333 339 10.1002/pd.1411 16491513 29. Filges I. Kang A. Klug V. Wenzel F. Heinimann K. Tercanli S. Miny P. aCGH on chorionic villi mirrors the complexity of fetoplacental mosaicism in prenatal diagnosis Prenat. Diagn. 2011 31 473 478 10.1002/pd.2721 21351283 30. Campbell C.D. Eichler E.E. Properties and rates of germline mutations in humans Trends Genet. 2013 29 575 584 10.1016/j.tig.2013.04.005 23684843 31. Wapner R.J. Martin C.L. Levy B. Ballif B.C. Eng C.M. Zachary J.M. Savage M. Platt L.D. Saltzman D. Grobman W.A. Chromosomal microarray versus karyotyping for prenatal diagnosis N. Engl. J. Med. 2012 367 2175 2184 10.1056/NEJMoa1203382 23215555 32. Shaffer L.G. Dabell M.P. Fisher A.J. Coppinger J. Bandholz A.M. Ellison J.W. Ravnan J.B. Torchia B.S. Ballif B.C. Rosenfeld J.A. Experience with microarray-based comparative genomic hybridization for prenatal diagnosis in over 5000 pregnancies Prenat. Diagn. 2012 32 976 985 10.1002/pd.3945 22865506 33. Shaffer L.G. Rosenfeld J.A. Dabell M.P. Coppinger J. Bandholz A.M. Ellison J.W. Ravnan J.B. Torchia B.S. Ballif B.C. Fisher A.J. Detection rates of clinically significant genomic alterations by microarray analysis for specific anomalies detected by ultrasound Prenat. Diagn. 2012 32 986 995 10.1002/pd.3943 22847778 34. Mademont-Soler I. Morales C. Soler A. Martinez-Crespo J.M. Shen Y. Margarit E. Clusellas N. Obon M. Wu B.L. Sanchez A. Prenatal diagnosis of chromosomal abnormalities in fetuses with abnormal cardiac ultrasound findings: Evaluation of chromosomal microarray-based analysis Ultrasound Obstet. Gynecol. 2013 41 375 382 10.1002/uog.12372 23233332 35. Hillman S.C. McMullan D.J. Hall G. Togneri F.S. James N. Maher E.J. Meller C.H. Williams D. Wapner R.J. Maher E.R. Use of prenatal chromosomal microarray: Prospective cohort study and systematic review and meta-analysis Ultrasound Obstet. Gynecol. 2013 41 610 620 10.1002/uog.12464 23512800 36. De Wit M.C. Srebniak M.I. Govaerts L.C. van Opstal D. Galjaard R.J. Go A.T. The additional value of prenatal genomic array testing in fetuses with (isolated) structural ultrasound abnormalities and a normal karyotype: A systematic review of the literature Ultrasound Obstet. Gynecol. 2013 10.1002/uog.12575 37. Breman A. Pursley A.N. Hixson P. Bi W. Ward P. Bacino C.A. Shaw C. Lupski J.R. Beaudet A. Patel A. Penatal chromosomal microarray analysis in a diagnostic laboratory; experience with >1000 cases and review of the literature Prenat. Diagn. 2012 32 351 361 10.1002/pd.3861 22467166 38. Callaway J.L. Shaffer L.G. Chitty L.S. Rosenfeld J.A. Crolla J.A. The clinical utility of microarray technologies applied to prenatal cytogenetics in the presence of a normal conventional karyotype: A review of the literature Prenat. Diagn. 2013 10.1002/pd.4209 39. Hillman S.C. McMullan D.J. Silcock L. Maher E.R. Kilby M.D. How does altering the resolution of chromosomal microarray analysis in the prenatal setting affect the rates of pathological and uncertain findings? J. Matern. Fetal Neonatal Med. 2013 10.3109/14767058.2013.825601 40. Ganesamoorthy D. Bruno D.L. McGillivray G. Norris F. White S.M. Adroub S. Amor D.J. Yeung A. Oertel R. Pertile M.D. Meeting the challenge of interpreting high-resolution single nucleotide polymorphism array data in prenatal diagnosis: Does increased diagnostic power outweigh the dilemma of rare variants? BJOG 2013 120 594 606 10.1111/1471-0528.12150 23332022 41. Srebniak M.I. Mout L. van Opstal D. Galjaard R.J. 0.5 Mb array as a first-line prenatal cytogenetic test in cases without ultrasound abnormalities and its implementation in clinical practice Hum. Mutat. 2013 34 1298 1303 10.1002/humu.22355 23674485 42. Shaffer L.G. Dabell M.P. Rosenfeld J.A. Neill N.J. Ballif B.C. Coppinger J. Diwan N.R. Chong K. Shohat M. Chitayat D. Referral patterns for microarray testing in prenatal diagnosis Prenat. Diagn. 2012 32 344 350 10.1002/pd.3856 22467165 43. Fiorentino F. Napoletano S. Caiazzo F. Sessa M. Bono S. Spizzichino L. Gordon A. Nuccitelli A. Rizzo G. Baldi M. Chromosomal microarray analysis as a first-line test in pregnancies with a priori low risk for the detection of submicroscopic chromosomal abnormalities Eur. J. Hum. Genet. 2013 21 725 730 10.1038/ejhg.2012.253 23211699 44. Verhagen J.M. Diderich K.E. Oudesluijs G. Mancini G.M. Eggink A.J. Verkleij-Hagoort A.C. Groenenberg I.A. Willems P.J. du Plessis F.A. de Man S.A. Phenotypic variability of atypical 22q11.2 deletions not including tbx1 Am. J. Med. Genet. A 2012 158a 2412 2420 10.1002/ajmg.a.35517 22893440 45. Dixit A. McKee S. Mansour S. Mehta S.G. Tanteles G.A. Anastasiadou V. Patsalis P.C. Martin K. McCullough S. Suri M. 7q11.23 Microduplication: A recognizable phenotype Clin. Genet. 2013 83 155 161 22369319 46. Pichert G. Mohammed S.N. Ahn J.W. Ogilvie C.M. Izatt L. Unexpected findings in cancer predisposition genes detected by array comparative genomic hybridisation: What are the issues? J. Med. Genet. 2011 48 535 539 10.1136/jmg.2010.087593 21429933 47. McGillivray G. Rosenfeld J.A. McKinlay Gardner R.J. Gillam L.H. Genetic counselling and ethical issues with chromosome microarray analysis in prenatal testing Prenat. Diagn. 2012 32 389 395 10.1002/pd.3849 22467169 48. ACOG Practice Bulletin No. 77: Screening for fetal chromosomal abnormalities Obstet. Gynecol. 2007 109 217 227 17197615 49. Bui T.H. Vetro A. Zuffardi O. Shaffer L.G. Current controversies in prenatal diagnosis 3: Is conventional chromosome analysis necessary in the post-array CGH era? Prenat. Diagn. 2011 31 235 243 10.1002/pd.2722 21374637

PMC005xxxxxx/PMC5003442.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2040284microarrays-02-00284ArticleCopy Number Studies in Noisy Samples Ginsbach Philip 1†Chen Bowang 2Jiang Yanxiang 1Engelter Stefan T. 3Grond-Ginsbach Caspar 1†*1 Neurology Department, University of Heidelberg, INF 400, Heidelberg D69120, Germany; E-Mails: Philip.Ginsbach@t-online.de (P.G.); jiangyanxiang@googlemail.com (Y.J.)2 Division of Molecular Genetic Epidemiology, German Cancer Research Center, INF 280, Heidelberg D69120, Germany; E-Mail: c.bowang@dkfz-heidelberg.de3 Stroke Unit and Department of Neurology, University Hospital Basel, Petersgraben 4, Basel CH4031, Switzerland; E-Mail: Stefan.Engelter@usb.ch† These authors contributed equally to this work. * Author to whom correspondence should be addressed; E-Mail: Caspar.Grond-Ginsbach@med.uni-heidelberg.de; Tel.: +49-6221-568213; Fax: +49-6221-565461.06 11 2013 12 2013 2 4 284 303 20 9 2013 24 10 2013 25 10 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).System noise was analyzed in 77 Affymetrix 6.0 samples from a previous clinical study of copy number variation (CNV). Twenty-three samples were classified as eligible for CNV detection, 29 samples as ineligible and 25 were classified as being of intermediate quality. New software (“noise-free-cnv”) was developed to visualize the data and reduce system noise. Fresh DNA preparations were more likely to yield eligible samples (p < 0.001). Eligible samples had higher rates of successfully genotyped SNPs (p < 0.001) and lower variance of signal intensities (p < 0.001), yielded fewer CNV findings after Birdview analysis (p < 0.001), and showed a tendency to yield fewer PennCNV calls (p = 0.053). The noise-free-cnv software visualized trend patterns of noise in the signal intensities across the ordered SNPs, including a wave pattern of noise, being co-linear with the banding pattern of metaphase chromosomes, as well as system deviations of individual probe sets (per-SNP noise). Wave noise and per-SNP noise occurred independently and could be separately removed from the samples. We recommend a two-step procedure of CNV validation, including noise reduction and visual inspection of all CNV calls, prior to molecular validation of a selected number of putative CNVs. copy number variation (CNV)variancewave noiseper-SNP noisenoise-free-cnv softwarenoise reductionvalidation of CNV findings ==== Body 1. Introduction Genomic copy number variation (CNV) was associated with a variety of clinical phenotypes [1,2,3,4,5,6]. Hence, the study of CNV is of diagnostic importance. CNV identification from high-density SNP-microarrays may be unreliable, particularly in noisy data [7,8,9]. Therefore, extensive validation of CNV findings is needed. Since CNV detection software may identify hundreds of putative CNVs in each sample and since validation of CNV findings by qPCR, or by other molecular methods, is laborious, we searched for simple strategies to evaluate large numbers of CNV findings. Rigorous studies revealed that several components of system error occur in copy number data [10,11,12,13]. Here we focus on two major types of noise and present the noise-free-cnv software package for the visualization of copy number data and for the reduction of noise. This software enables large-scale inspection of CNV findings (produced by PennCNV [14], Birdview [15,16], or other specialized software packages). For illustration, we used 77 microarrays from a previous study of patients with cervical artery dissection from Switzerland and Southern Germany (age: 42.5 ± 9.8 years; 31 (40.3%) women) [17]. DNA was isolated from peripheral blood samples (no DNA from lymphoblastoid cell lines was used). DNA extraction, array hybridization, and array scanning were performed according to the manufacturer’s instructions [17]. The LRR and BAF values were obtained from the CEL files with the Affymetrix Power Tools software (APT). The quantile normalization was done in APT. The LRR and BAF can be then imported to PennCNV, to other CNV detections software packages (QuantiSNP, MAD), or to noise-free-cnv. The Affymetrix 6.0 microarrays used for CNV detection contain a total of 906,600 single nucleotide polymorphisms (SNPs) and 946,000 non-polymorphic copy number probes (CNPs) covering all human chromosomes. In the present article, the notion of SNP is used for all analyzed probe sets (SNPs as well as CNPs). 2. Noise Components Figure 1 shows two samples (visualized by noise-free-cnv), displaying signal intensity (LRR—upper panel) and B-allele frequency (BAF—lower panel) of all SNPs ordered along the chromosomes. The Log R Ratio (LRR) is a normalized measure of the total signal intensity for two alleles of the SNP. The B-Allele Frequency (BAF) is a normalized measure of the allelic intensity ratio of two alleles [18]. Signal intensities in sample ID 2355 show larger variance than in ID 1022. Moreover, a prominent pattern of waves is apparent in sample ID 2355. In many samples, we observed similar wave patterns. The noise-free-cnv software identified waves using a Gaussian filter with a large standard deviation, for instance comprising 1,000 SNPs. This filter “blurs” the values as shown in Figure 2(G,H). We called the resulting wave data the wave component of the LRR values. The variance of the blurred LRR values is a measure for the prominence of waves, the wave variance. Figure 1 Signal strength (LRR) and B-allele frequency (BAF) of samples from two male patients (ID 2355 and ID 1022). SNPs were visualized in increasing position along the chromosomes. LRR values of patient ID 2355 have larger variance and show pronounced wave noise. Figure 2 Wave noise. Ideograms of pro-metaphase (A) and metaphase (B) chromosome 7 were compared with signal intensities of SNPs of chromosome 7 of two patients (C,D) and with a human prometaphase (E) and metaphase (F) chromosome 7. Signal intensities shown in C and D were smoothed (noise-free-cnv software, function “blur” across 1,000 probe sets) to visualize genomic waves (G,H). This wave pattern was compared with the banding pattern of metaphase chromosomes (Figure 2). Human metaphase chromosomes were stained with the Giemsa-trypsine procedure, which induces a banding pattern. AT-rich regions are more frequent in Giemsa-dark bands than in Giemsa-light bands [19,20]. In our study samples, Giemsa-dark bands corresponded to genomic regions with reduced probe set signals. This pattern of noise was described by others as “genomic waves” or “CG-waves” [10,11,12,13]. The co-linearity of genomic waves with Giemsa bands illustrates that genomic waves follow a similar pattern in all samples. After subtraction of the wave component, the resulting LRR values follow an approximately normal distribution around zero. We called the resulting values per-SNP component and their variance the per-SNP variance. The decomposition of system noise in wave component and per-SNP component is shown for one sample in Figure 3. Wave variance and per-SNP variance components were calculated for all samples in Table A1. Figure 3 Noise components. LRR values of a noisy sample (A), split up in wave component (B) and per-SNP component (C). All SNPs of chromosomes 1–3 were shown (chromosomes indicated on top of panel A). The system deviations of individual SNP signal intensities are strongly correlated across samples (Figure 4). To quantify the correlation of the noise (variance) components between different samples, we computed two additional data series: for each SNP the median through all 77 per-SNP components was computed and saved as the per-SNP profile. For the wave profile the same procedure was applied to the wave components. We then computed, for each sample, the correlation between the wave profile and the (individual) wave component as well as the correlation between the per-SNP profile and the (individual) per-SNP component. Details of the algorithm are described in Appendix. The high correlations found in our 77 samples confirmed that wave noise and per-SNP noise are system noise, i.e., follow highly non-random patterns. On average, the correlation was 0.843 for the wave component and 0.568 for the per-SNP component. 3. Factors Associated with Quality of Copy Number Data The resolution of a classical chromosome study depends on the quality of the chromosomes and is expressed as the total number of visible cytogenetic bands (400 bands: low to moderate quality; 850 bands: excellent quality). According to our knowledge, no comparable quality metric for molecular karyotyping exists. Quality control in most copy number studies consists of rejecting samples with outlier numbers of CNV findings. A quality metric for the resolution of a CNV study (relating the size of a CNV and the likelihood of its detection) has not yet been defined. Figure 4 per-SNP system noise. Signal intensities in genomic region 2: 189766706–189891527 shown for four patients (ID 1020; ID 1022; ID1026; ID 1028). The lower panel shows the per-SNP median profile (median signal intensities) of all samples (n = 77). Arrows and arrowheads indicate SNPs with LRR values far above and below the mean. In the current study we propose a preliminary quality metric based on the median number of SNPs per chromosome with copy number state (CN) ≠ 2 (numbers/chromosome for all cases are shown in Table A1). Copy Number state of each SNP was determined by the Affymetrix Power Tools software package (APT). SNPs located in common CNVs were excluded from this analysis. To identify SNPs located in common CNVs, we analyzed 403 control samples without visible waves and with highest genotype call rates selected from a large German population (PopGen [21]), as described before [17]. The median number of SNPs with CN ≠ 2 per chromosome was considered as a preliminary quality metric. The quality of a sample was related to the chromosomal background of SNPs with abnormal copy number (Figure 5). We defined deliberate quality categories: samples were classified as eligible, if the median number of SNPs per chromosome with CN ≠ 2 was zero, those with >100 SNPs with CN ≠ 2 were classified as ineligible. Figure 5 Quality of copy number samples. Number of SNPs with CN ≠ 2 per chromosome were scored. Sample ID 715 is eligible for CNV studies (most chromosomes without SNPs with CN ≠ 2). Accumulation of aberrant SNPs in chromosome 7 and 18 indicates presence of rare CNVs. Sample ID 50 is of intermediate quality. Sample ID 062 was classified as ineligible for CNV studies (>100 SNPs with CN ≠ 2 in most chromosomes). Samples were classified according to the defined quality categories in Table 1. The use of freshly prepared DNA (compared to DNA samples that were used since years and had been thawed and frozen repeatedly) was a significant determinant of eligible samples (p < 0.001). Samples with high call rate (rate of successfully genotyped SNPs) were more likely to be suitable for copy number studies than those with lower call rates (p < 0.001). Low levels of wave variance as well as per-SNP variance were associated with eligibility for CNV analysis (p < 0.001). Eligibility for CNV studies was not significantly associated with the median number of calls by PennCNV (p = 0.053). However, eligible samples had between 63 and 165 calls, while the range of calls was much broader in ineligible samples. Birdview yielded significantly more calls in ineligible samples (p < 0.001). The proportion of putative false positive Birdview calls increased with decreasing confidence rates: The number of CNV findings with confidence below 2.5 was most strongly elevated. microarrays-02-00284-t001_Table 1Table 1 Characteristics of 77 analyzed samples, classified according to eligibility for copy number variation (CNV) analysis. Numbers indicate mean values and range (lowest–highest value). Mean values were compared between groups with the Chi-2 test or the Kruskal-Wallis test. Ineligible Intermediate Eligible Chi-2/ kruskal-wallis (n = 29) (n = 25) (n = 23) p Fresh DNA preparation 0 (0.0 %) 6 (20.7 %) 14 (60.9 %) <0.001 Genotyping call rate 94.7 [80.9–97.3] 96.6 [94.8–98.3] 97.7 [96.6–98.5] <0.001 Autosomal variance 0.2291 [0.115–0.706] 0.1343 [0.068–0.208] 0.0870 [0.062–0.114] <0.001 wave noise 0.0109 [0.002–0.058] 0.0034 [0.001–0.017] 0.0015 [0.001–0.013] <0.001 per–SNP noise 0.2259 [0.082–0.696] 0.1281 [0.067–0.204] 0.0811 [0.060–0.164] <0.001  PennCNV, No. of calls 238 [14–1821) 103 [34–1024] 98 [63–165] 0.053  PennCNV, % of deletions 18.6 [1.3–81.3] 27.4 [0.7–65.9] 40.0 [10.3–54.8] 0.164 Birdview No. of calls 527 [163–8,203] 225 [154–1,339] 208 [163–348] <0.001  Birdview (cf > 10) 15 [2–717] 12 [5–33] 14 [4–20] 0.048  Birdview (cf = 10) 89 [76–145] 92 [74–105] 94 [77–102] 0.209  Birdview (cf 2.5–10) 93 [14–3344] 19 [10–361] 21 [11–45] <0.001  Birdview (cf < 2.5) 370 [52–5665] 106 [35–857] 85 [42–194] <0.001 Figure 6 summarizes salient aspects of system noise in SNP microarrays. Figure 6(A) plots for each sample the variances of wave component and per-SNP component. Wave variance and per-SNP variance seem to occur independently from each other: the observed correlation between both noise components (r = 0.124) was not significant (p = 0.401). Figure 6(B) illustrates the relation between sample eligibility and noise components in the eligible (n = 23) and ineligible (n = 29) cases. Eligible samples (i.e., those that are supposed to be excellent for copy number studies) have low levels of per-SNP variance. Samples with high wave variance are inappropriate for copy number studies. Figure 6 Wave variance and per-SNP variance. (A) Noise components in all 77 samples and (B) in samples of low (O) and high (●) quality (samples of intermediate quality were not included in (B)). 4. Noise Reduction in Copy Number Samples The noise-free-cnv software package permits the visualization of samples, the isolation of noise components and the subtraction of isolated noise components. The next two examples (Figure 7 and Figure 8) illustrate noise reduction by comparing a test sample with a reference sample. We finally demonstrate the use of the noise-free-cnv-filter algorithm for the evaluation of CNVs. Figure 7 shows a deletion in chromosome 20 of patient ID 1091, which was detected by PennCNV and Birdview analysis. Due to strong waves, reduced signal intensities in the region of the putative deletion are not easily seen. Visual inspection of the LRR values of chromosome 20 after subtraction of a reference sample (A–B) suggested the presence of a true deletion in this patient. Figure 7 Signal intensities (y-axis: LRR values) of all SNPs from chromosome 18q up to chromosome 22. (A) Patient ID 1091; (B) reference sample ID 2355. After subtraction of the samples, a deletion in chromosome 20 became apparent (arrow). Figure 8 illustrates the analysis of a mosaic deletion. Although sample ID D62 was classified as ineligible for CNV studies, analysis of SNPs with CN ≠ 2 per chromosome revealed significant clustering on chromosome 5 (Table A1; Figure 5). Neither PennCNV nor Birdsuite identified a large CNV on chromosome 5. After noise reduction, LRR and BAF values were suggestive for the presence of a mosaic deletion [22,23,24] (Figure 8(B,D)). To confirm the diagnosis of a mosaic deletion, a conventional chromosome analysis was performed: Some rare 5q chromosomes were observed amongst a majority of normal chromosome sets. Interestingly, it was recently demonstrated that the identification of mosaic abnormalities by microarray analysis is unreliable [25]. We developed the noise-free-cnv-filter algorithm for optimized noise reduction (Appendix). In the samples of our study population, noise-free-cnv-filter analysis resulted in an average reduction of the wave variance by 74.2%, of per-SNP variance by 35.3% and of the overall variance by 38.1%. Noise-reduction according to this algorithm supports the evaluation of CNV findings, in particular when the putative CNVs are small (Figure 9). In patient ID 715, both Birdview and PennCNV identified a deletion on chromosome 18 (green bar in Figure 9). Noise-free-cnv-filter analysis of the sample (ID 715 nf) suggested that the deletion was true. Subsequent molecular analysis confirmed the finding: the joining segment of the deletion was identified by a case-specific PCR and the breakpoints of the deletion were identified by DNA sequencing following standard procedures [17,26]. Two putative duplications in patients ID 412 were evaluated after noise-free-cnv-filter analysis. We considered the duplication in chromosome 1 (region 222 Mb) as spurious (red bar), but the duplication in chromosome 9 as probably true. As a consequence, this putative duplication is a candidate for further validation by molecular methods. Figure 8 Sample with mosaic large deletion in chromosome 5q. (A,B) LRR- and BAF-values of SNPs of chromosomes 5 and 6 of patient. (C) LRR values of reference sample. (D) Signal intensities after subtraction of reference sample. Arrows indicate region with reduced LRR values. (E) LRR values after application of noise-free-cnv blur over 2,000 SNPs. (Bottom panel) Chromosome analysis of cultured peripheral blood lymphocytes from patient (courtesy of Johannes W.G. Janssen, Department of Human Genetics, University of Heidelberg). Arrow points to 5q-minus chromosome. Figure 9 Validation of CNV findings. Left panels show crude LRR values, left panels show LRR values after noise-free-cnv-filter analysis. Samples were renamed with suffix “nf” after noise-free-cnv-filter analysis. Bars indicate putative CNV findings. 5. Conclusions—Proposal of a Two-Step Procedure for the Validation of CNV Findings Our analysis had the following key findings: (1) Copy number samples may be noisy, which interferes—above a certain level of noise—with reliable identification of CNVs; (2) Eligible copy number samples were more likely when fresh DNA was used for microarray hybridization; (3) wave component and per-SNP component of noise are independent; (4) noise-free-cnv software enables noise reduction by subtracting wave and per-SNP noise components from samples; and (5) noise-free-cnv software supports the quality control of copy number data and the validation of copy number findings. The current noise-free-cnv version was developed for the analysis of SNP microarray samples and was not designed for noise reduction in array based comparative genomic hybridization samples. The present study highlighted the value of noise reduction for large scale CNV validation (after software-assisted CNV detection). However, the value of noise reduction before software-assisted CNV detection is to be analyzed in future studies. Based on our analysis of noise in real-life copy number samples we suggested a two-step procedure of CNV validation. As a first step of preliminary CNV validation we proposed large-scale inspection of CNV findings after noise reduction, to select putative candidate CNVs and reject false positive findings. In a second stage, this selection of putative CNV calls is analyzed further by independent molecular methods for final validation [17,26]. Acknowledgments This work was supported by a grant from the Swiss Heart Foundation. Conflicts of Interest The authors declare no conflict of interest. Appendix: Comments to the Noise-Free-CNV Software A1. Noise-Free-CNV The noise-free-cnv program package was specifically developed to analyze copy number variation in SNP-microarray samples and to manipulate the data in order to reduce noise. It was written in C++ and released as free software under the GNU General Public License version 3. Installer packages are available for Debian-based Linux systems and Windows. For the computation of the Fast Fourier Transform, we used the FFTW library [27]. Noise-free-cnv is compatible with the file format used by PennCNV [14]. The central program of the noise-free-cnv package is noise-free-cnv-gtk, a visual editor for interactive visualization and manipulation of SNP microarray data. Besides functioning as a browser for direct inspection and verification of CNV findings, it allows the user to perform many operations on the data. These include the Gaussian filters and variance computation referred to in the article. For further information, see the project homepage http://noise-free-cnv.sourceforge.net. A second program, noise-free-cnv-filter, implements a specific algorithm for system noise reduction, as described below. It is usable as a command line program to be easily applied to a batch of samples. A2. The Noise-Free-CNV-Filter Algorithm The noise reduction algorithm noise-free-cnv-filter consists of two main steps. In the first step, a genomic wave profile and a per-SNP noise profile are deduced from a batch of samples. In the second step, these profiles are used to modify the individual samples. A2.1. System Noise Assessment For each individual sample: (1) The non-autosomal data is removed and the Log R Ratio values are normalized towards an average value of zero. (2) The wave component is computed by applying a Gaussian filter with a standard deviation of 1,000 SNPs to the Log R Ratio sequence (3) The wave component is subtracted from the Log R Ratio values to calculate the per-SNP component. Subsequently, the batch-specific wave is computed by regarding each SNP throughout the wave components of all samples and taking the median value. The same is done for the per-SNP profile utilizing the per-SNP components. A2.2. System Noise Removal In the second step, we use the median profiles to adjust the original samples. For each individual: (1) The covariance of the wave component and the batch-specific wave profile is divided by the variance of the wave profile. (2) The result is used as a scaling factor for the wave profile, the scaled profile is then subtracted from the wave component The same procedure is repeated on the per-SNP components. (3) Finally, the corrected components are added together and yield the corrected Log R Ratio values. A3. Program Usage Noise-free-cnv-filter was implemented as a command-line program. In the most simple case, it receives the file names of several SNP microarray samples in the PennCNV file format (due to the nature of the algorithm, application on a single sample is pointless). It then computes the profiles (saved as “wave_profile” and “per-snp_profile”) and the cleaned versions of all provided samples, which it saves as “<original filename>.nf”. As additional options, noise-free-cnv-filter allows the use of pre-computed profile sequences and the inclusion of the sex chromosomes into the analysis. As an example, noise-free-cnv-filter—verbose individuals/* applies the algorithm to all files in the directory individuals, discards the sex chromosomes and outputs detailed information about the progress and statistical information about the samples. For further help, type: noise-free-cnv-filter—help. microarrays-02-00284-t002_Table A1Table A1 Eligibility of samples. ID call rate var. wave var. per_SNP var. Chr1 Chr2 Chr3 Chr4 Chr5 Chr6 Chr7 Chr8 Chr9 Chr10 Chr11 Chr12 Chr13 Chr14 Chr15 Chr16 Chr17 Chr18 Chr19 Chr20 Chr21 Chr22 3 98.33 0.068 0.001 0.067 2 0 0 2 2 0 0 0 0 0 0 11 0 1 0 0 2 2 0 0 0 102 15 96.02 0.144 0.001 0.142 10 4 38 8 11 2 0 50 16 11 8 150 13 0 78 22 11 21 19 0 4 0 36 96.00 0.243 0.004 0.238 506 845 678 751 1,376 503 977 974 722 385 379 639 593 232 541 752 356 397 364 225 262 203 38 95.76 0.183 0.001 0.181 179 80 102 48 140 80 91 268 33 84 122 44 27 144 20 47 114 41 155 0 40 20 48 96.16 0.174 0.007 0.167 0 39 60 141 23 42 90 69 56 5 10 59 110 18 32 361 29 32 0 94 15 55 49 94.81 0.174 0.007 0.167 203 31 35 111 40 28 33 22 14 6 22 9 8 15 41 0 0 23 12 0 7 0 50 97.14 0.103 0.002 0.101 46 11 0 14 22 242 2 0 8 0 0 65 0 0 11 2 3 0 45 0 2 2 62 97.92 0.090 0.003 0.087 0 14 0 14 3 19 15 11 2 2 0 0 0 0 0 0 6 16 0 0 0 2 71 93.92 0.202 0.002 0.200 667 490 326 78 149 40 269 252 215 116 45 231 96 163 142 266 134 148 248 38 97 50 76 93.71 0.229 0.002 0.226 511 251 200 65 229 59 336 467 352 422 457 233 85 185 170 252 285 161 462 112 167 49 97 89.52 0.291 0.008 0.282 3,123 5,467 11,613 3,795 6,729 4,870 5,374 4,898 4,455 4,169 5,721 6,492 3,486 3,020 3,709 4,031 4,120 3,177 3,581 2,466 1,775 1,281 101 97.85 0.077 0.002 0.075 0 0 0 0 0 0 0 0 0 0 0 0 0 9 0 0 0 0 0 2 0 0 111 96.51 0.175 0.003 0.172 70 76 14 49 287 187 29 76 64 266 105 47 61 17 15 20 0 74 10 136 6 21 112 94.70 0.147 0.011 0.134 2,378 2,988 2,743 2,514 2,823 4,455 2,774 3,014 2,837 2,707 4,227 3,054 2,315 1,595 1,905 1,087 2,085 2,044 1,562 650 580 602 129 96.61 0.139 0.003 0.135 7 31 51 9 19 35 3 94 30 77 23 2 3 10 0 96 0 2 74 0 22 0 131 96.45 0.199 0.002 0.196 319 229 139 314 108 125 248 304 172 363 279 68 96 93 118 212 182 71 83 28 13 28 141 96.44 0.121 0.017 0.101 266 121 174 173 147 64 195 121 258 24 43 291 45 36 59 202 105 96 208 90 86 72 144 94.72 0.176 0.005 0.170 251 746 130 464 250 328 253 410 390 139 799 55 553 63 193 178 52 286 195 9 15 45 168 94.36 0.315 0.036 0.275 5,205 5,382 5,272 4,479 4,337 5,737 6,682 4,504 3,966 2,627 2,901 5,294 2,492 3,065 2,487 3,126 2,559 2,658 2,179 1,405 961 948 182 95.02 0.316 0.002 0.313 1,189 1,988 2,051 1,096 2,301 2,991 1,814 2,365 1,408 687 1,144 1,314 839 654 689 815 569 1,953 427 517 456 377 188 90.12 0.474 0.011 0.461 14,534 15,554 28,322 10,245 10,904 16,212 12,471 14,300 6,642 6,499 8,681 9,241 5,595 6,784 7,503 5,150 6,048 7,028 3,978 2,417 3,536 2,274 189 97.32 0.097 0.012 0.084 120 34 155 12 35 41 74 69 0 29 47 34 16 6 1 0 0 0 30 0 35 2 193 97.34 0.093 0.004 0.088 53 3 0 3 0 0 0 5 71 6 2 0 47 0 0 0 0 0 0 0 0 0 412 97.51 0.092 0.006 0.086 0 0 62 2 0 0 20 0 2 14 0 10 22 0 0 0 2 2 0 0 0 4 415 98.23 0.103 0.001 0.102 5 11 0 5 3 10 35 0 2 0 0 2 0 0 0 0 0 9 0 0 0 2 421 96.73 0.165 0.055 0.102 0 7 0 0 0 5 28 0 0 0 0 0 0 2 0 0 0 4 0 0 0 0 422 96.10 0.074 0.002 0.073 37,996 36,654 32,343 29,935 29,248 29,062 23,198 24,951 21,457 22,172 22,247 21,574 15,560 14,542 14,525 12,574 10,444 14,095 8,616 10,294 7,218 4,456 430 96.39 0.160 0.019 0.138 10,614 10,096 9,196 16,278 15,383 7,959 7,758 9,465 6,757 6,295 6,291 6,559 4,955 3,487 2,491 3,459 3,074 5,404 1,525 2,399 3,569 1,187 438 89.76 0.463 0.057 0.399 45,981 56,162 42,403 45,750 55,976 37,901 44,212 34,069 30,387 26,842 26,461 38,930 23,963 20,800 18,258 19,160 16,624 18,367 9,637 13,485 8,790 5,479 442 97.82 0.084 0.008 0.076 486 382 0 0 14 0 9 57 2 0 33 4 0 0 0 0 0 0 67 0 4 3 451 95.96 0.205 0.004 0.200 109 154 49 363 75 361 107 213 124 100 85 150 124 20 63 4 60 116 50 59 0 19 461 80.86 0.706 0.007 0.696 64,822 74,302 41,655 64,993 56,987 58,825 49,369 55,433 45,357 47,372 35,753 49,377 30,870 27,158 25,448 29,590 14,637 22,231 15,372 19,289 12,431 8,971 613 97.64 0.090 0.001 0.088 2 6 3 7 4 0 49 15 0 0 22 2 10 2 0 0 0 0 7 0 2 0 647 95.49 0.157 0.002 0.154 15 1 30 166 0 47 9 0 6 45 41 58 0 5 28 41 24 0 14 4 4 11 653 98.22 0.079 0.004 0.074 12 2 14 1,618 26 7 12 7 0 9 0 0 0 0 0 2 0 0 0 0 4 4 665 97.32 0.123 0.037 0.082 5,071 8,140 6,289 5,520 6,417 6,816 5,890 6,894 4,238 5,326 4,503 5,246 2,641 2,696 2,301 1,766 1,445 4,211 1,699 1,399 2,448 932 670 96.74 0.152 0.043 0.105 3,160 4,486 3,895 3,913 3,812 3,038 3,309 3,676 2,359 2,112 3,327 2,791 1,591 1,595 1,142 969 1,028 2,043 543 1,213 1,034 286 675 97.67 0.084 0.009 0.074 0 0 3 0 4 1 3 37 0 0 58 4 0 0 0 11 0 0 48 0 0 0 676 98.15 0.078 0.001 0.077 0 0 0 0 45 2 18 0 0 0 14 4 0 0 0 0 0 0 0 0 0 0 677 95.58 0.208 0.003 0.204 25 149 161 54 77 57 267 157 38 210 63 13 52 64 34 44 57 137 121 16 56 75 693 95.72 0.133 0.005 0.128 4 73 18 6 31 12 15 27 7 0 10 8 73 14 10 0 5 3 20 2 0 12 715 97.44 0.095 0.006 0.089 3 5 0 0 4 0 471 0 2 0 0 0 0 4 0 4 0 24 0 4 0 0 717 96.60 0.134 0.001 0.132 0 2 22 21 29 41 0 17 0 6 57 10 0 2 0 15 0 0 34 0 0 0 729 94.75 0.189 0.001 0.187 115 23 115 184 49 55 84 91 42 45 101 111 16 57 103 80 57 61 192 0 22 0 733 95.84 0.198 0.009 0.188 23 28 84 218 218 0 306 180 40 34 59 11 32 3 33 62 108 6 8 38 46 0 735 95.28 0.173 0.003 0.169 94 21 111 290 81 41 116 14 220 79 0 87 47 31 34 43 16 13 181 0 0 22 742 96.83 0.114 0.013 0.099 0 0 3 0 2 0 1 0 4 0 2 0 0 15 8 2 18 0 137 0 0 0 744 97.26 0.108 0.017 0.089 0 135 126 332 73 148 263 0 135 31 52 150 180 34 15 40 0 166 114 0 94 0 746 95.72 0.253 0.004 0.248 283 895 350 378 287 388 227 723 594 712 229 559 323 206 98 478 398 689 341 73 184 173 750 97.15 0.114 0.015 0.097 2,389 642 1,750 1,501 1,370 1,792 707 1,440 997 615 674 750 559 225 544 63 187 814 458 177 184 0 752 97.80 0.103 0.004 0.099 88 121 119 155 196 83 118 187 93 206 66 147 27 35 50 41 32 70 15 36 27 25 796 94.23 0.247 0.002 0.244 2,165 482 568 1,127 539 263 360 591 440 697 908 806 106 630 256 498 262 565 127 275 67 48 1020 98.21 0.075 0.001 0.074 2 0 0 0 2 1 2 0 0 0 0 3 0 0 0 0 0 0 8 0 16 0 1022 97.77 0.082 0.001 0.081 6 16 2 30 2 2 2 0 12 0 11 0 0 0 0 0 0 1 0 0 0 0 1026 98.34 0.066 0.001 0.064 1 0 0 0 5 0 0 0 0 0 2 0 0 11 9 0 0 0 3 0 0 0 1028 97.49 0.089 0.001 0.088 9 0 0 0 40 49 0 3 0 0 3 0 0 0 0 0 0 0 0 0 0 0 1029 98.54 0.062 0.001 0.060 3 64 0 0 0 0 0 0 0 0 1 0 2 50 0 25 0 2 0 0 0 0 1033 97.58 0.087 0.001 0.085 0 0 83 0 0 0 2 0 70 4 4 0 0 0 8 0 0 0 0 0 0 0 1034 97.50 0.092 0.010 0.081 0 0 0 2 0 0 1 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1037 98.26 0.068 0.001 0.067 8 5 2 50 2 6 2 12 0 2 1 0 2 0 0 3 54 13 2 2 0 0 1040 96.56 0.114 0.001 0.113 9 0 17 53 3 1 4 0 0 0 0 0 17 0 0 0 0 0 0 0 1 0 1041 97.16 0.094 0.001 0.093 0 0 0 4 2 2 0 0 0 0 13 3 0 0 0 0 0 0 0 0 0 0 1042 97.46 0.087 0.003 0.084 15 14 19 20 11 28 10 19 4 2 2 6 8 4 18 4 2 24 2 0 2 0 1056 97.11 0.098 0.003 0.095 2 0 2 2 0 2 0 0 0 0 0 15 0 4 0 0 35 0 10 0 0 18 1063 98.31 0.075 0.002 0.072 0 2 169 10 0 0 29 0 0 0 0 0 2 0 13 0 0 0 2 0 0 0 1065 98.23 0.068 0.003 0.065 0 4 2 7 0 0 0 2 0 0 0 0 8 0 0 0 0 58 4 0 0 0 1088 97.64 0.092 0.004 0.088 53 23 61 45 8 90 26 76 32 44 50 71 34 6 15 54 0 22 13 4 9 39 1091 96.69 0.138 0.029 0.105 7,631 6,080 7,146 6,512 7,299 4,006 4,952 6,666 4,169 2,579 3,990 4,012 2,388 2,194 1,415 2,613 1,030 2,828 646 3,697 1,862 451 1147 97.96 0.079 0.002 0.077 5 76 4 8 0 0 4 4 16 0 2 3 2 0 2 2 0 7 190 0 0 0 1151 97.90 0.087 0.001 0.086 2 4 0 0 3 0 0 8 0 4 9 0 0 0 0 0 2 0 0 0 11 2 2110 93.16 0.343 0.010 0.332 671 3,307 3,614 1,414 3,548 3,035 2,775 2,070 1,581 2,718 2,153 2,054 1,573 1,426 995 1,173 1,584 1,358 662 285 971 581 2134 95.73 0.134 0.008 0.125 144 52 81 51 51 100 28 88 70 0 35 77 3 55 7 16 56 7 41 2 18 1 2144 97.12 0.093 0.004 0.089 2 5 8 1 4 0 4 0 2 0 14 4 0 0 0 3 0 0 0 2 0 0 2240 94.48 0.299 0.004 0.294 488 1,656 1,019 1,734 2,180 1,605 754 1,750 1,179 531 916 1,728 916 751 643 921 501 427 713 356 339 374 2355 94.50 0.258 0.026 0.229 1,870 1,799 1,422 741 1,491 2,829 1,663 1,353 829 1,500 1,360 1,714 654 677 1,262 767 1,420 1,061 842 826 464 452 2406 94.78 0.195 0.011 0.183 67 125 136 122 201 80 142 109 62 61 5 51 70 68 34 7 60 83 122 53 43 0 D_062 94.17 0.322 0.003 0.318 772 838 547 536 4,419 436 711 496 239 308 219 582 319 215 222 81 62 540 101 86 0 94 For each sample, genotype call rate, variance, wave variance and per-SNP variance were calculated. The remaining columns show for each chromosome (chromosome number indicated) the number of probe sets with CN ≠ 2. This analysis included only probe sets that had normal copy number (CN = 2) in 403 samples from a population based German study (for details and references see [17]). A non-random distribution of probe sets with CN ≠ 2 is highly suggestive for the existence of a rare CNV (for instance ID 1147 or ID653, in contrast to ID 1042 or ID 2034). Even in samples with high variance, non-random distribution can be detected (chromosome 5 of ID D_062, chromosomes 1 and 2 in ID 442). microarrays-02-00284-t003_Table A2Table A2 Analysis of noise components in samples. ID variance wave variance per-SNP variance wave correlation per-SNP correlation wave subtraction factor per-SNP subtraction factor 3 0.068 0.001 0.067 0.804 0.676 0.409 0.800 15 0.144 0.001 0.142 0.567 0.564 0.393 0.972 36 0.243 0.004 0.238 0.877 0.388 0.997 0.865 38 0.183 0.001 0.181 0.456 0.502 0.314 0.977 48 0.174 0.007 0.167 0.939 0.508 1.405 0.949 49 0.174 0.007 0.167 0.959 0.527 1.419 0.985 50 0.103 0.002 0.101 0.881 0.536 0.626 0.779 62 0.090 0.003 0.087 0.929 0.609 0.886 0.820 71 0.202 0.002 0.200 0.365 0.589 0.274 1.203 76 0.229 0.002 0.226 0.580 0.529 0.511 1.151 97 0.291 0.008 0.282 0.875 0.360 1.427 0.873 101 0.077 0.002 0.075 0.906 0.727 0.717 0.910 111 0.175 0.003 0.172 0.914 0.482 0.903 0.914 112 0.147 0.011 0.134 0.864 0.578 1.671 0.968 129 0.139 0.003 0.135 0.898 0.620 0.964 1.043 131 0.199 0.002 0.196 0.875 0.417 0.790 0.844 141 0.121 0.017 0.101 0.949 0.703 2.297 1.024 144 0.176 0.005 0.170 0.881 0.591 1.141 1.115 168 0.315 0.036 0.275 0.936 0.406 3.234 0.975 182 0.316 0.002 0.313 0.809 0.386 0.704 0.990 189 0.097 0.012 0.084 0.930 0.668 1.874 0.883 193 0.093 0.004 0.088 0.960 0.704 1.174 0.956 412 0.092 0.006 0.086 0.950 0.691 1.304 0.925 421 0.103 0.001 0.102 0.898 0.680 0.598 0.991 422 0.165 0.055 0.102 0.873 0.523 3.763 0.763 425 0.074 0.002 0.073 0.908 0.572 0.653 0.705 430 0.160 0.019 0.138 0.896 0.457 2.265 0.778 438 0.463 0.057 0.399 0.913 0.354 4.006 1.021 442 0.084 0.008 0.076 0.942 0.663 1.496 0.835 451 0.205 0.004 0.200 0.927 0.453 1.111 0.927 461 0.706 0.007 0.696 −0.310 0.228 −0.460 0.868 613 0.090 0.001 0.088 0.850 0.565 0.589 0.767 647 0.157 0.002 0.154 0.766 0.589 0.671 1.059 653 0.079 0.004 0.074 0.938 0.569 1.141 0.707 665 0.123 0.037 0.082 0.896 0.555 3.159 0.726 670 0.152 0.043 0.105 0.906 0.565 3.422 0.837 675 0.084 0.009 0.074 0.951 0.672 1.647 0.837 676 0.078 0.001 0.077 0.646 0.635 0.313 0.807 677 0.208 0.003 0.204 0.901 0.465 0.870 0.960 693 0.133 0.005 0.128 0.953 0.581 1.179 0.952 715 0.095 0.006 0.089 0.950 0.667 1.308 0.911 717 0.134 0.001 0.132 0.522 0.650 0.351 1.081 729 0.189 0.001 0.187 0.606 0.611 0.411 1.209 733 0.198 0.009 0.188 0.947 0.488 1.628 0.967 735 0.173 0.003 0.169 0.901 0.609 0.917 1.144 742 0.114 0.013 0.099 0.956 0.649 2.017 0.936 744 0.108 0.017 0.089 0.921 0.570 2.186 0.779 746 0.253 0.004 0.248 0.906 0.414 1.033 0.943 750 0.114 0.015 0.097 0.940 0.612 2.137 0.874 752 0.103 0.004 0.099 0.937 0.471 1.033 0.679 796 0.247 0.002 0.244 0.015 0.527 0.012 1.192 1020 0.075 0.001 0.074 0.742 0.614 0.348 0.767 1022 0.082 0.001 0.081 0.909 0.644 0.640 0.836 1026 0.066 0.001 0.064 0.861 0.664 0.557 0.770 1028 0.089 0.001 0.088 0.742 0.643 0.380 0.871 1029 0.062 0.001 0.060 0.912 0.661 0.572 0.742 1033 0.087 0.001 0.085 0.820 0.703 0.518 0.940 1034 0.092 0.010 0.081 0.963 0.709 1.732 0.924 1037 0.068 0.001 0.067 0.782 0.633 0.390 0.748 1040 0.114 0.001 0.113 0.639 0.701 0.355 1.077 1041 0.094 0.001 0.093 0.850 0.672 0.546 0.937 1042 0.087 0.003 0.084 0.947 0.695 0.884 0.921 1056 0.098 0.003 0.095 0.941 0.686 0.893 0.966 1063 0.075 0.002 0.072 0.924 0.571 0.832 0.701 1065 0.068 0.003 0.065 0.959 0.657 0.904 0.764 1088 0.092 0.004 0.088 0.918 0.497 1.046 0.675 1091 0.138 0.029 0.105 0.912 0.537 2.854 0.797 1147 0.079 0.002 0.077 0.944 0.717 0.795 0.909 1151 0.087 0.001 0.086 0.688 0.572 0.311 0.769 2110 0.343 0.010 0.332 0.948 0.408 1.715 1.075 2134 0.134 0.008 0.125 0.936 0.551 1.575 0.893 2144 0.093 0.004 0.089 0.953 0.609 1.046 0.832 2240 0.299 0.004 0.294 0.909 0.433 1.060 1.075 2355 0.258 0.026 0.229 0.960 0.530 2.826 1.161 2406 0.195 0.011 0.183 0.940 0.570 1.833 1.113 188c 0.474 0.011 0.461 0.899 0.314 1.715 0.975 D62 0.322 0.003 0.318 0.819 0.340 0.761 0.878 ==== Refs References 1. Girirajan S. Campbell C.D. Eichler E.E. Human copy number variation and complex genetic disease Annu. Rev. Genet. 2011 45 203 226 10.1146/annurev-genet-102209-163544 21854229 2. Zhang F. Gu W. Hurles M.E. Lupski J.R. Copy number variation in human health, disease, and evolution Annu. Rev. Genomics Hum. Genet. 2009 10 451 481 10.1146/annurev.genom.9.081307.164217 19715442 3. Fakhro K.A. Choi M. Ware S.M. Belmont J.W. Towbin J.A. Lifton R.P. Khokha M.K. Brueckner M. Rare copy number variations in congenital heart disease patients identify unique genes in left-right patterning Proc. Natl. Acad. Sci. USA 2011 108 2915 2920 10.1073/pnas.1019645108 21282601 4. Priebe L. Degenhardt F. Strohmaier J. Breuer R. Herms S. Witt S.H. Hoffmann P. Kulbida R. Mattheisen M. Moebus S. Copy number variants in german patients with schizophrenia PLoS One 2013 8 e64035 10.1371/journal.pone.0064035 23843933 5. Vandeweyer G. Kooy R.F. Detection and interpretation of genomic structural variation in health and disease Expert. Rev. Mol. Diagn. 2013 13 61 82 10.1586/erm.12.119 23256704 6. Southard A.E. Edelmann L.J. Gelb B.D. Role of copynumber variants in structural birth defects Pediatrics 2012 129 755 763 10.1542/peds.2011-2337 22430448 7. Zhang D. Qian Y. Akula N. Alliey-Rodriguez N. Tang J. The Bipolar Genome Study Gershon E.S. Liu C. Accuracy of CNV detection from GWAS data PLoS One 2011 6 e14511 10.1371/journal.pone.0014511 21249187 8. Dellinger A.E. Saw S.M. Goh L.K. Seielstad M. Young T.L. Li Y.J. Comparative analyses of seven algorithms for copy number variant identification from single nucleotide polymorphism arrays Nucleic Acids Res. 2010 38 e105 10.1093/nar/gkq040 20142258 9. Zheng X. Shaffer J.R. McHugh C.P. Laurie C.C. Feenstra B. Melbye M. Murray J.C. Marazita M.L. Feingold E. Using family data as a verification standard to evaluate copy number variation calling strategies for genetic association studies Genet. Epidemiol. 2012 36 253 262 10.1002/gepi.21618 22714937 10. Marioni J.C. Thorne N.P. Valsesia A. Fitzgerald T. Redon R. Fiegler H. Andrews T.D. Stranger B.E. Lynch A.G. Dermitzakis E.T. Breaking the waves: Improved detection of copy number variation from microarray-based comparative genomic hybridization Genome Biol. 2007 8 R228 10.1186/gb-2007-8-10-r228 17961237 11. Diskin S.J. Li M. Hou C. Yang S. Glessner J. Hakonarson H. Bucan M. Maris J.M. Wang K. Adjustment of genomic waves in signal intensities from whole-genome SNP genotyping platforms Nucleic Acids Res. 2008 36 e126 10.1093/nar/gkn556 18784189 12. Van de Wiel M.A. Brosens R. Eilers P.H. Kumps C. Meijer G.A. Menten B. Sistermans E. Speleman F. Timmerman M.E. Ylstra B. Smoothing waves in array CGH tumor profiles Bioinformatics 2009 25 1099 1104 10.1093/bioinformatics/btp132 19276148 13. Lee Y.H. Ronemus M. Kendall J. Lakshmi B. Leotta A. Levy D. Esposito D. Grubor V. Ye K. Wigler M. Reducing system noise in copynumber data using principal components of self-self hybridizations Proc. Natl. Acad. Sci. USA 2012 109 E103 E110 10.1073/pnas.1106233109 22207624 14. Wang K. Li M. Hadley D. Liu R. Glessner J. Grant S.F. Hakonarson H. Bucan M. PennCNV: An integrated hidden Markov model designed for high-resolution copy number variation detection in whole-genome SNP genotyping data Genome Res. 2007 17 1665 1674 10.1101/gr.6861907 17921354 15. Korn J.M. Kuruvilla F.G. McCarroll S.A. Wysoker A. Nemesh J. Cawley S. Hubbell E. Veitch J. Collins P.J. Darvishi K. Integrated genotype calling and association analysis of SNPs, common copy number polymorphisms and rare CNVs Nat. Genet. 2008 40 1253 1260 10.1038/ng.237 18776909 16. McCarroll S.A. Kuruvilla F.G. Korn J.M. Cawley S. Nemesh J. Wysoker A. Shapero M.H. de Bakker P.I. Maller J.B. Kirby A. Integrated detection and population-genetic analysis of SNPs and copy number variation Nat. Genet. 2008 40 1166 1174 10.1038/ng.238 18776908 17. Grond-Ginsbach C. Chen B. Pjontek R. Wiest T. Burwinkel B. Tchatchou S. Krawczak M. Schreiber S. Brandt T. Kloss M. Copy number variation in patients with cervical artery dissection Eur. J. Hum. Genet. 2012 20 1295 1299 10.1038/ejhg.2012.82 22617347 18. Wang K. Bucan M. Copy number variation detection via high-density SNP genotyping Cold Spring Harb. Protoc. 2008 2008 10.1101/pdb.top46 19. Niimura Y. Gojobori T. In silico chromosome staining: Reconstruction of Giemsa bands from the whole human genome sequence Proc. Natl. Acad. Sci. USA 2002 99 797 802 10.1073/pnas.022437999 11792839 20. Costantini M.L. Clay O. Federico C. Saccone S. Auletta F. Bernardi G. Human chromosomal bands: Nested structure, high-definition map and molecular basis Chromosoma 2007 116 29 40 10.1007/s00412-006-0078-0 17072634 21. Krawczak M. Nikolaus S. von Eberstein H. Croucher P.J. El Mokhtari N.E. Schreiber S. PopGen: Population-based recruitment of patients and controls for the analysis of complex genotype-phenotype relationships Community Genet. 2006 9 55 61 10.1159/000090694 16490960 22. Piotrowski A. Bruder C.E. Andersson R. Diaz de Ståhl T. Menzel U. Sandgren J. Poplawski A. von Tell D. Crasto C. Bogdan A. Somatic mosaicism for copy number variation in differentiated human tissues Hum. Mutat. 2008 29 1118 1124 10.1002/humu.20815 18570184 23. Jasmine F. Rahaman R. Dodsworth C. Roy S. Paul R. Raza M. Paul-Brutus R. Kamal M. Ahsan H. Kibriya M.G. A genome-wide study of cytogenetic changes in colorectal cancer using SNP microarrays: Opportunities for future personalized treatment PLoS One 2012 7 e31968 10.1371/journal.pone.0031968 22363777 24. Laurie C.C. Laurie C.A. Rice K. Doheny K.F. Zelnick L.R. McHugh C.P. Ling H. Hetrick K.N. Pugh E.W. Amos C. Detectable clonal mosaicism from birth to old age and its relationship to cancer Nat. Genet. 2012 44 642 650 10.1038/ng.2271 22561516 25. Bi W. Borgan C. Pursley A.N. Hixson P. Shaw C.A. Bacino C.A. Lalani S.R. Patel A. Stankiewicz P. Lupski J.R. Comparison of chromosome analysis and chromosomal microarray analysis: What is the value of chromosome analysis in today’s genomic array era? Genet. Med. 2013 15 450 457 10.1038/gim.2012.152 23238528 26. Vissers L.E. Bhatt S.S. Janssen I.M. Xia Z. Lalani S.R. Pfundt R. Derwinska K. de Vries B.B. Gilissen C. Hoischen A. Rare pathogenic microdeletions and tandem duplications are microhomology-mediated and stimulated by local genomic architecture Hum. Mol. Genet. 2009 18 3579 3593 10.1093/hmg/ddp306 19578123 27. Frigo M. Johnson S.G. The design and implementation of FFTW3 Proc. IEEE 2005 93 216 231 10.1109/JPROC.2004.840301

PMC005xxxxxx/PMC5003443.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays3020091microarrays-03-00091ReviewIdentification of New Players in Hepatocarcinogenesis: Limits and Opportunities of Using Tissue Microarray (TMA) Quagliata Luca *Schlageter Manuel Quintavalle Cristina Tornillo Luigi Terracciano Luigi M. Molecular Pathology Division, Institute of Pathology, University Hospital of Basel, CH-4031 Basel, Switzerland; E-Mails: Manuel.Schlageter@usb.ch (M.S.); cristina.quintavalle@usb.ch (C.Q.); luigi.tornillo@usb.ch (L.T.); lterracciano@uhbs.ch (L.M.T.)* Author to whom correspondence should be addressed; E-Mail: luca.quagliata@usb.ch; Tel./Fax: +41-61-265-29-66.15 4 2014 6 2014 3 2 91 102 19 2 2014 21 3 2014 © 2014 by the authors; licensee MDPI, Basel, Switzerland.2014This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Liver tumours are among the leading causes of cancer-related death worldwide and hepatocellular carcinoma (HCC) accounts for the vast majority of liver tumours. When detected at an early stage of disease, patients might still be eligible for surgical-based curative treatments. However, currently only small portion of HCC affected patients are diagnosed at an early stage. For late stage HCC no treatment option exists beside the multi-tyrosine kinase inhibitor Sorafenib. Thus new molecular targets and treatment options for HCC are urgently needed. Nevertheless, despite some improvements in diagnosis and patient management, the biology of liver tumour remains inadequately understood, mainly because these tumours have shown to harbour a highly complex genomic landscape. In addition, one major obstacle delaying the identification of new molecular targets in biomedical research is the necessity to validate them using a large collection of tissue specimens. Tissue microarray (TMA) technology allows the prompt molecular profiling of multiple tissue specimens and is therefore ideal to analyze presumptive candidate biomarkers in a fast an effective manner. The use of TMA has substantial benefits over standard techniques and represents a significant advancement in molecular pathology. For example, TMA technology reduces laboratory work, offers a high level of experimental uniformity and provides a judicious use of precious tissue. On the other hand, one potential limitation of using TMA is that the small cores sampled may not be representative of whole tumors. This issue is very critical in particularly heterogeneous cancers such as HCC. For liver focused studies, it is ideal to evaluate the staining patters of a determined marker over the structure of an entire acinus and to define staining in as many as possible anatomical regions. In this review we analyze the limits and opportunities offered by the usage of TMA technology in HCC research. In summary, TMA has revolutionized the histopathological analysis and will be of great help to further advance the knowledge in the field of hepatocarcinogenesis research. TMAliverHepatocellular carcinomaIHC ==== Body 1. Introduction: HCC an Overview Liver tumours are among the leading causes of cancer-related death worldwide and the principal cause of mortality among cirrhotic patients [1]. In contrast to other tumour entities, mortality from liver cancer has considerably increased over the past decades [2]. In addition, epidemiologic data about the prevalence of chronic hepatitis indicates that the medical and economic burden of liver cancers will still drastically increase in the next 15 years [2]. Hepatocellular carcinoma (HCC) accounts for 80% of all liver tumours, with the others being either cholangiocarcinoma (CC) or mixed forms. HCC mostly arises in patients suffering from a cirrhotic liver [3], with more than half of new cases (mainly in Eastern countries) being associated to chronic infection of either hepatitis B (HBV) or C (HCV) virus [4]. Besides established risk factors, such as male gender, older age, high levels of bilirubin, altered liver enzymes status, increased portal hypertension, ethnicity and viral genotypes [4,5,6], little is known about the mechanisms that favor HCC development and progression [6]. Notably, it is not clear why a subgroup of patients with cirrhosis will eventually develop HCC, whereas others do not. When detected at an early stage of disease, patients might be still eligible for surgical-based curative treatments. However, currently only 30% to 40% of HCC affected patients are diagnosed at an early stage and can undergo resection, local ablation or transplantation [6]. Finally, even complete tumor removal does not certainly offer a sheltered curative solution, as the underlying liver disease (mostly cirrhosis) will still persist in the remaining liver. In fact, about 80% of HCC patients experience recurrence after surgical resection [7,8,9]. Furthermore, if HCC is detected at an intermediate or advanced stage, no treatment option exists beside Sorafenib [10]. Sorafenib is multi-tyrosine kinase inhibitor that effectively blocks several receptors activity such as VEGFR (Vascular Endothelial Growth Factor Receptor), PDGFR (Platelet Derived Growth Factor Receptor) and the RAF serine/threonine kinases along the RAF/MEK/ERK pathway [10]. However, Sorafenib has shown a consistent but limited survival benefit in HCC (10 to 12 weeks increased survival) accompanied by a number of moderate to severe side effects [3,10]. Concerning HCC diagnosis, nowadays, histopathological assessment of specimens still remains the most effective and accurate option [4]. Such an approach is clearly dependent on the expertise of the pathologist revising the case and therefore suffers of inter-observer variability. While late stage HCC is usually a straightforward Haematoxylin & Eosin (H&E) diagnosis, early stage HCC is more problematic and often demands the evaluation of additional histological features, for example an assessment of the reticulin framework. Therefore, the identification of new molecular markers to specifically identify early HCC might result in an increased diagnostic output. Despite some improvements in diagnosis and patient management, the biology of liver tumour remains inadequately understood, mainly because these tumours have shown to harbour a highly complex genomic landscape [6]. Among the best-described pathways driving HCC, there are Wnt/β-catenin, MAPK, p14ARF/p53, p16INK4A/Rb, transforming growth factor-β (TGF-β) and PTEN/Akt [11]. Moreover, using comparative genomic hybridization (CGH) technology, HCC has shown frequent DNA copy number gains at chromosomes 1q, 8q and losses at 1p, 4q, 8p, 13q, 16q and 17q [12,13,14,15]. An accurate molecular classification should highlight drug-targets, such as growth factor receptors or kinases, thus allowing personalized targeted therapies. So far, several studies have attempted to establish a comprehensive HCC molecular classification, mostly based on gene expression profile [16,17,18]. However till now, none of such classifications have been validated in the clinical practice [19], mainly because of the many discrepancies between the presented models [20,21]. Thus the identification of specific molecular markers able to classify the different HCC subgroups and possibly identify early stage of disease onset is urgently needed. 2. TMA Methodology: A Historical Overview and Basic Concepts Back in 1986, original work from Hector Battifora described a ‘sausage’ block method to prepare multi tissue sections that is considered the prototype of nowadays well-known tissue microarray (TMA) [22]. Battifora’s innovative approach however suffered of several limitations, as for example the inability to identify individual tissue rods, that were subsequently addressed by Kononen et al. [23] who shaped the TMA as we know it now. Kononen and colleagues introduced a novel sampling method to produce tissues of regular shape and defined size, making them more suitable to be densely and precisely arrayed. This TMA technology allows the prompt molecular profiling of multiple tissue specimens and is therefore ideal to analyze presumptive candidate biomarkers in a fast and effective manner [24,25]. The constant expanding knowledge produced by recent research in molecular biology has identified a plethora of novel presumptive biomarkers, which might have major diagnostic, prognostic and therapeutic significance [26]. However, one major obstacle in biomedical research is the necessity to validate them on a large collection of tissue specimens. Traditional histopathological techniques are often unacceptably time consuming, extremely labor intensive and very expensive [27]. These limitations have delayed the introduction of novel markers into everyday clinical practice. TMA offers a valuable solution to overcome some of these problems. Using TMA technology many tissue specimens (in our institute, up to 1000 histology blocks) can be arrayed at the same time [28]. Therefore, TMAs are often used for the characterization of antibodies and for tissue specific expression profiling of proteins and genes via in situ hybridization [25,28]. Of note, the vast majority of TMAs are analyzed using immunohistochemistry, while a small fraction are investigated by in situ hybridisation techniques, such as interphase FISH. Immunohistochemcal analysis is often criticized due to the subjective and semiquantitative means of determining the level of protein expression hampered by the intra laboratory differences in staining procedures, as well as the inter-observer variability. Importantly, compared to the broadly available standard histopathological approach, the TMA methodology allows to analyze all the tissue specimens arrayed in the same manner, for example exposing them to the identical antigen retrieval procedure, reagent concentrations, incubation times with antibodies/probes, thus consequently resulting in an high level of standardization [28]. Thus, the use of TMA has substantial benefits over standard techniques [25]. For example, Diaz and colleagues [29] showed that the evaluation of HER-2 using TMA-IHC survey was correctly scored in over 90% of the tested laboratories. These results further underline that the use of TMAs can reduce variability during the evaluation. Furthermore, among the other major advantages of using TMA, only few quantities of reagents and substantially less laboratory work is required to perform the experiments, making TMA approach exceptionally cost-effective [28]. Additional benefits of using TMA include marginal exhaustion of donor tissue blocks (obviously considered as vital resources) and the possibility to include internal positive and negative controls (cell line materials or tissues with a known expression) while constructing the TMA [30,31]. All together TMA methodology represents a significant advancement in molecular pathology over traditional methods. 3. TMA in Hepatocarcinogenesis Research Tissue specimens represent a fundamental tool for biomedical research. Their analysis is often a crucial step towards the understanding of the molecular background of a disease. For research purposes of liver tumors, TMA technology has been proven to be a reliable and effective tool. However, it is also important to mention that the TMA technology harbors a number of limitations. One potential limitation of using TMA is that the small cores sampled may not be representative of whole tumors [24]. This issue is very critical in particularly heterogeneous cancers such as HCC. For other tumor entities, several studies have addressed this point by comparing TMA analysis results with whole mount sections data. High levels of correlation have been described comparing these procedures in a range of tumor types such as breast, prostate, bladder and human fibroblastic tumors [27,30]. For example, Kononen et al. [32] using breast-TMA found the same frequencies of HER-2, c-myc, cyclinD1 and 17q23 amplifications in breast cancer as were expected from previous published literature using whole tissue sections [33]. Interestingly, some studies demonstrated that increasing the number of cores, to compensate for heterogeneity, only slightly increased the rate of data validity [24,28,33]. Conversely, such an approach has the disadvantage of generating significant additional labor work during the arrays preparation [27]. Importantly, it should be mentioned that TMA are intended to estimate the prevalence of a selected markers within a large population of samples, rather then to provide a detailed analysis at the level of single specimen. For liver focused studies, in case of non-neoplastic liver specimens, it is ideal to evaluate the staining patters of a determined marker over the structure of an entire acinus. It is important to define staining in as many as possible anatomical regions (Figure 1). Thus, it would be optimal to select tissue punches that include at least one portal tract and one central vein. It is conceivable that choosing the 1 mm diameter (or higher) punch is the best option to be selected while constructing a liver TMA. In order to facilitate the cutting and the evaluation procedures, while constructing a TMA it is also important to have convenient spacing between cores of at least 0.15 mm [23]. Furthermore, tissue losses are observed at different percentage depending on the tissue used to construct the TMA [34], with a range from 5% to 33% of used specimens [27], and represent a frequent problem also in liver TMA. The choice of larger tissue punches (e.g., 1 mm compared to 0.6 mm) reduces the frequency of tissue loss [35]. The value of TMAs in the study of liver hepatocarcinogenesis has been proven by a number of studies that employed this technology to unravel some of the key players involved in HCC’s biology. For example, the association of Clusterin, a highly conserved glycoprotein with previously reported pro-tumorigenic function, with metastasis in HCC has been proved by Lau et al. [36] using a TMA containing 104 pairs of primary HCCs and their matched metastasis. Clusterin is overexpressed in HCC metastasis and facilities them via modulating YKL-40, a mediator of matrix remodeling processing [37]. Similarly, Hu et al. [38], using a TMA composed of 60 pairs of primary/metastatic HCCs, demonstrated that the transcriptional repressor ZHX2 (Zinc-fingers and homeoboxes-2) is altered in HCC and its levels correlate with disease stage. Furthermore, ZHX2 is highly overexpressed in metastatic prone lesions. Additional work performed using a TMA containing primary and recurrent HCCs showed that FGF3 (fibroblast growth factor 3) is associated with HCC recurrence and metastasis [39]. The analysis of liver-TMA with different characteristics, such has the one used by Chen et al. [40], generated using HCC and adjacent tissue plus cirrhotic and normal liver specimens, has shown that Heparanase overexpression is linked to HCC prognosis and grade. The expression of KisSS-1, a multi-protein producing genes involved in gonadotropin-releasing hormone and a putative metastasis suppressor in melanoma, was investigated in intra-hepatic HCC metastasis and reported to be lost in these lesions [41]. An extensive survey of putative hepatic stem/progenitor cell biomarkers, namely CK19 (cytokeratin 19), CD133, Nestin and CD44 conducted by Yang and collogues [42] using HCC-TMA revealed that high HSC/HPC profile along with high VEGFA (vascular endothelial growth factor A) levels and increased MVD (micro vascular density) had significant lower OS (overall survival) and RFS (recurrence free survival). More recently, CK19 levels together with CK7 (cytokeratin 7) have been investigated again using an HCC-TMA and found to be significantly associated with tumor grade and AFP levels (alpha-fetoprotein) [43]. Importantly, this work also shows that CK19 expression is extremely rare in pre-cancerous DNs (dysplastic nodules) while increases in small HCCs, suggest CK19 association with disease progression. Conversely, CK7 is already expressed in DNs and further augmented in HCC lesions. The Wnt/β-catenin pathway is among the best-studied and mostly altered one in HCC [11]. Using a liver TMA generated from 179 HCC and matched non-tumorous liver blocks, Cheng et al. [44] reported that expression of a histone-lysine N-methyltransferase, EZH2 (Enhancer Of Zeste Homolog 2), is significantl associated both with the nuclear and cytoplasmic β-catenin expression. Furthermore, they demonstrated that poorly differentiated HCC presents stronger EZH2 and β-catenin staining compared to both moderate and well-differentiated HCC. In contrast, neither EZH2 nor β-catenin nuclear staining is found in the surrounding liver tissue or in the normal liver specimens, suggesting the importance of Wnt/β-catenin and EZH2 in HCC biology. The use of liver specific TMAs has been reported to be a valuable tool also for the study of the endothelial compartment of the liver. Geraud and colleagues [45] could show that liver sinusoidal endothelial cells (LSEC) undergo a phenotypic switch upon development of chronic liver disease. Specifically, they observed that the typical sinusoidal cell’s fenestrae are lost and a basal membrane is formed, thus leading to the capillarization of liver sinusoids. Taking advantage of our large collection of HCC tissues, our group also contributed to liver carcinogenesis research by reporting that GPC3 (Gypican 3), a heparin sulfate proteoglycan protein with cell proliferation and apoptosis regulatory activity, could be a useful diagnostic marker to differentiate between HCC, non-neoplastic liver disease and pre-neoplastic lesions [46]. In this study, using a multi-tumor TMA, we investigated a total of 4387 tissue samples from 139 tumor types. Our data revealed that GPC3 is expressed in more than about 60% of the investigated HCC specimens, while it is observed in less than 10% of the non-neoplastic liver tissue and in about 16% of pre-neoplastic lesions [46]. Hep Par 1 (Hepatocyte paraffin 1) was proposed as valuable marker to differentiate HCC from other lesions metastasizing to the liver [47]. Using a TMA comprising 3940 tissue samples, we observed that Hep Par 1 is frequently expressed (ca. 73%) of analyzed HCC, while in non-hepatic tumors such as lung, gallbladder, pancreas, stomach, small intestine, adenoma of the colon with high-grade dysplasia, adrenal gland carcinoma, paraganglioma and malignant melanoma it is almost virtually absent. Our work suggests that Hep Par 1 is a highly specific marker for HCC [48]. Figure 1 Liver TMA construction steps and examples of different quality cores. (A) A HE-stained whole section is evaluated and specific areas are selected. (B) Matching tissue areas are punched on the corresponding FFPE block. (C) Representative picture of multiple arrayed cores. (D) Representative picture of a low quality normal liver sample containing only hepatocyte cells. (E) Example of high quality normal liver punch with portal tract and hepatocytes. (F) Low quality HCC sample with no other cells than transformed hepatocytes. (G) High quality HCC sample containing aportion of normal tissue, portal tract and an HCC area. More recently, by analyzing a TMA composed of 69 normal liver specimens, 93 cirrhotic samples and 174 HCCs, we also showed that the SH2D4A (SH2 domain containing 4A) gene is frequently down regulated in HCC, further corroborating its presumptive role as tumor suppressor gene. Finally, for HCC, the use of TMA has revealed a great potential in a comparative study analyzing Asian and American cohort of patients, underling different expression profiles of p53 and MDM2 in the different populations. This approach could represent a novel strategy to identify novel molecular targets based on patient ethnicity [49]. 4. Future Prospective The vast majority of TMAs so far employed for research purposes in liver tumor studies, have been generated starting from formalin-fixed paraffin-embedded (FFPE) embedded material. More recently fresh frozen tissue-TMAs of cores embedded in optimal cutting temperature (OCT) blocks have been described [30]. TMAs are constructed from unfixed fresh-frozen tissue that has been embedded in a recipient block of OCT media. Such types of TMA present a number of advantages; for example contrary to FFPE-generated TMA where fixatives in the embedded tissue might severely affect the quality of RNA, giving sub-optimal results for RNA hybridization, frozen TMA provide high quality material for study of RNA, DNA and proteins. Indeed, as frozen TMAs are generated from unfixed tissue with antigen preserved structures, these TMAs are extremely useful when antibodies do not work on FFPE tissue, or when FISH-based analysis is required. FISH on TMAs has been frequently used to validate findings of gene amplifications discovered by genome-wide screening [30]. Conversely, one major drawback is that the brittleness of frozen OCT makes coring procedures much more difficult and only fewer samples can be arrayed to avoid cracking of the blocks. In addition, cell morphology in frozen TMAs is of lower quality than in the FFPE counterpart [50]. Moreover, in multi-step diseases such as HCC it is of fundamental help to assess molecular changes through the different stages of tumor progression [30]. Generating progressing TMAs can face this issue [28,51]. Progressing TMAs are defined as TMAs containing from normal to hyperplastic, dysplastic lesions up to HCC specimens. Such TMAs have already proven their ability to uncover stage-specific molecular alterations, for example in prostate cancer progression, where the amplification of the Androgen Receptor (AR) gene [52] or the amplification of IGFBP2 locus (insulin-like growth factor binding protein 2) [53] were usually found in hormone refractory end-stage prostate cancers but rarely observed in untreated primary tumors. The assembly of such TMA will be of fundamental help for the study of liver carcinogenesis. 5. Conclusions The rapid and effective translation of molecular based discoveries into new therapeutic targets, useful markers to predict response to therapy, or to help diagnostic assessment is a fundamental issue in modern biomedical research. The TMA approach has proven to have valuable advantages in comparison to standard whole section analysis, as hundreds of tissue samples can be examined in a single experiment. Furthermore, the TMA method provides a judicious use of precious tissue and offers high experimental uniformity. In the current world of high-throughput technology, TMA has revolutionized the histopathological analysis and will be of great help to further advance the knowledge in the field of hepatocarcinogenesis research. Acknowledgments We acknowledge the Institute of Pathology members for their support and critical suggestions to this review. Author Contributions LQ has conceived and wrote this review, MS has performed morphological and TMA analysis, CQ has performed literature search and contributed to write this review, LT has performed morphological and TMA analysis, LMT has conceived this review and contributed to write this review. Abbreviation Tissue microarray (TMA), Hepatocellular carcinoma (HCC), hepatitis B virus (HBV), hepatitis C virus (HCV), Immunohistochemistry (IHC). Conflicts of Interest The authors declare no conflict of interest. ==== Refs References 1. Gomaa A.I. Khan S.A. Toledano M.B. Waked I. Taylor-Robinson S.D. Hepatocellular carcinoma: Epidemiology, risk factors and pathogenesis World J. Gastroenterol. 2008 14 4300 4308 10.3748/wjg.14.4300 18666317 2. Schutte K. Bornschein J. Malfertheiner P. Hepatocellular carcinoma—Epidemiological trends and risk factors Dig. Dis. 2009 27 80 92 19546545 3. D'Angelo S. Secondulfo M. de Cristofano R. Sorrentino P. Selection and management of hepatocellular carcinoma patients with sorafenib: Recommendations and opinions from an italian liver unit Fut. Oncol. 2013 9 485 491 10.2217/fon.12.208 4. Marquardt J.U. Galle P.R. Teufel A. Molecular diagnosis and therapy of hepatocellular carcinoma (hcc): An emerging field for advanced technologies J. Hepatol. 2012 56 267 275 10.1016/j.jhep.2011.07.007 21782758 5. Teufel A. Marquardt J.U. Staib F. Galle P.R. Snapshot liver transcriptome in hepatocellular carcinoma J. Hepatol. 2012 56 990 992 10.1016/j.jhep.2011.08.024 22173158 6. Forner A. Llovet J.M. Bruix J. Hepatocellular carcinoma Lancet 2012 379 1245 1255 10.1016/S0140-6736(11)61347-0 22353262 7. Cucchetti A. Piscaglia F. Cescon M. Ercolani G. Pinna A.D. Systematic review of surgical resection vs. radiofrequency ablation for hepatocellular carcinoma World J. Gastroenterol. 2013 19 4106 4118 10.3748/wjg.v19.i26.4106 23864773 8. Welker M.W. Bechstein W.O. Zeuzem S. Trojan J. Recurrent hepatocellular carcinoma after liver transplantation—An emerging clinical challenge Transpl. Int. 2013 26 109 118 10.1111/j.1432-2277.2012.01562.x 22994652 9. Cescon M. Cucchetti A. Ravaioli M. Pinna A.D. Hepatocellular carcinoma locoregional therapies for patients in the waiting list. Impact on transplantability and recurrence rate J. Hepatol. 2013 58 609 618 10.1016/j.jhep.2012.09.021 23041304 10. Llovet J.M. Ricci S. Mazzaferro V. Hilgard P. Gane E. Blanc J.F. de Oliveira A.C. Santoro A. Raoul J.L. Forner A. Sorafenib in advanced hepatocellular carcinoma New Engl. J. Med. 2008 359 378 390 10.1056/NEJMoa0708857 18650514 11. El-Serag H.B. Rudolph K.L. Hepatocellular carcinoma: Epidemiology and molecular carcinogenesis Gastroenterology 2007 132 2557 2576 10.1053/j.gastro.2007.04.061 17570226 12. Roessler S. Long E.L. Budhu A. Chen Y. Zhao X. Ji J. Walker R. Jia H.L. Ye Q.H. Qin L.X. Integrative genomic identification of genes on 8p associated with hepatocellular carcinoma progression and patient survival Gastroenterology 2012 142 957 966 957–966, e912 22202459 13. Tornillo L. Carafa V. Richter J. Sauter G. Moch H. Minola E. Gambacorta M. Bianchi L. Vecchione R. Terracciano L.M. Marked genetic similarities between hepatitis b virus-positive and hepatitis c virus-positive hepatocellular carcinomas J. Pathol. 2000 192 307 312 10.1002/1096-9896(2000)9999:9999<::AID-PATH706>3.0.CO;2-O 11054713 14. Farazi P.A. DePinho R.A. Hepatocellular carcinoma pathogenesis: From genes to environment Nat. Rev. Canc. 2006 6 674 687 10.1038/nrc1934 15. Kwon S.M. Kim D.S. Won N.H. Park S.J. Chwae Y.J. Kang H.C. Lee S.H. Baik E.J. Thorgeirsson S.S. Woo H.G. Genomic copy number alterations with transcriptional deregulation at 6p identify an aggressive hcc phenotype Carcinogenesis 2013 34 1543 1550 10.1093/carcin/bgt095 23508637 16. Oishi N. Kumar M.R. Roessler S. Ji J. Forgues M. Budhu A. Zhao X. Andersen J.B. Ye Q.H. Jia H.L. Transcriptomic profiling reveals hepatic stem-like gene signatures and interplay of mir-200c and epithelial-mesenchymal transition in intrahepatic cholangiocarcinoma Hepatology 2012 56 1792 1803 10.1002/hep.25890 22707408 17. Yang J.D. Seol S.Y. Leem S.H. Kim Y.H. Sun Z. Lee J.S. Thorgeirsson S.S. Chu I.S. Roberts L.R. Kang K.J. Genes associated with recurrence of hepatocellular carcinoma: Integrated analysis by gene expression and methylation profiling J. Kor. Med. Sci. 2011 26 1428 1438 10.3346/jkms.2011.26.11.1428 18. Lee J.S. Thorgeirsson S.S. Comparative and integrative functional genomics of hcc Oncogene 2006 25 3801 3809 10.1038/sj.onc.1209561 16799621 19. Wang X.W. Thorgeirsson S.S. Transcriptome analysis of liver cancer: Ready for the clinic? J. Hepatol. 2009 50 1062 1064 10.1016/j.jhep.2009.02.007 19328580 20. Woo H.G. Park E.S. Thorgeirsson S.S. Kim Y.J. Exploring genomic profiles of hepatocellular carcinoma Mol. Carcinog. 2011 50 235 243 10.1002/mc.20691 21465573 21. Hoshida Y. Toffanin S. Lachenmayer A. Villanueva A. Minguez B. Llovet J.M. Molecular classification and novel targets in hepatocellular carcinoma: Recent advancements Semin. Liver Dis. 2010 30 35 51 10.1055/s-0030-1247131 20175032 22. Battifora H. The multitumor (sausage) tissue block: Novel method for immunohistochemical antibody testing Lab. Investig. 1986 55 244 248 3525985 23. Kononen J. Bubendorf L. Kallioniemi A. Barlund M. Schraml P. Leighton S. Torhorst J. Mihatsch M.J. Sauter G. Kallioniemi O.P. Tissue microarrays for high-throughput molecular profiling of tumor specimens Nat. Med. 1998 4 844 847 10.1038/nm0798-844 9662379 24. Bubendorf L. Nocito A. Moch H. Sauter G. Tissue microarray (TMA) technology: Miniaturized pathology archives for high-throughput in situ studies J. Pathol. 2001 195 72 79 10.1002/path.893 11568893 25. Shergill I.S. Shergill N.K. Arya M. Patel H.R. Tissue microarrays: A current medical research tool Curr. Med. Res. Opin. 2004 20 707 712 10.1185/030079904125003412 15140337 26. Eguiluz C. Viguera E. Millan L. Perez J. Multitissue array review: A chronological description of tissue array techniques, applications and procedures Pathol. Res. Pract. 2006 202 561 568 10.1016/j.prp.2006.04.003 16782284 27. Chen W. Foran D.J. Advances in cancer tissue microarray technology: Towards improved understanding and diagnostics Anal. Chim. Acta 2006 564 74 81 10.1016/j.aca.2005.11.083 17723364 28. Kallioniemi O.P. Kononen J. Sauter G. Introducing tissue microarrays to molecular pathology Clin. Chem. 2012 58 1717 1718 10.1373/clinchem.2012.188748 23065478 29. Diaz L.K. Gupta R. Kidwai N. Sneige N. Wiley E.L. The use of tma for interlaboratory validation of fish testing for detection of her2 gene amplification in breast cancer J. Histochem. Cytochem. 2004 52 501 507 10.1177/002215540405200408 15034001 30. Tzankov A. Went P. Zimpfer A. Dirnhofer S. Tissue microarray technology: Principles, pitfalls and perspectives—Lessons learned from hematological malignancies Exp. Gerontol. 2005 40 737 744 10.1016/j.exger.2005.06.011 16125349 31. Packeisen J. Korsching E. Herbst H. Boecker W. Buerger H. Demystified...Tissue microarray technology Mol. Pathol. 2003 56 198 204 10.1136/mp.56.4.198 12890740 32. Simon R. Nocito A. Hubscher T. Bucher C. Torhorst J. Schraml P. Bubendorf L. Mihatsch M.M. Moch H. Wilber K. Patterns of her-2/neu amplification and overexpression in primary and metastatic breast cancer J. Natl. Canc. Inst. 2001 93 1141 1146 10.1093/jnci/93.15.1141 33. Schraml P. Kononen J. Bubendorf L. Moch H. Bissig H. Nocito A. Mihatsch M.J. Kallioniemi O.P. Sauter G. Tissue microarrays for gene amplification surveys in many different tumor types Clin. Canc. Res. 1999 5 1966 1975 34. Hoos A. Urist M.J. Stojadinovic A. Mastorides S. Dudas M.E. Leung D.H. Kuo D. Brennan M.F. Lewis J.J. Cordon-Cardo C. Validation of tissue microarrays for immunohistochemical profiling of cancer specimens using the example of human fibroblastic tumors Am. J. Pathol. 2001 158 1245 1251 10.1016/S0002-9440(10)64075-8 11290542 35. Skacel M. Skilton B. Pettay J.D. Tubbs R.R. Tissue microarrays: A powerful tool for high-throughput analysis of clinical specimens: A review of the method with validation data Appl. Immunohistochem. Mol. Morphol. 2002 10 1 6 10.1097/00022744-200203000-00001 11893029 36. Lau S.H. Sham J.S. Xie D. Tzang C.H. Tang D. Ma N. Hu L. Wang Y. Wen J.M. Xiao G. Clusterin plays an important role in hepatocellular carcinoma metastasis Oncogene 2006 25 1242 1250 10.1038/sj.onc.1209141 16247463 37. Kazakova M.H. Sarafian V.S. Ykl-40—A novel biomarker in clinical practice? Folia Med. 2009 51 5 14 38. Hu S. Zhang M. Lv Z. Bi J. Dong Y. Wen J. Expression of zinc-fingers and homeoboxes 2 in hepatocellular carcinogenesis: A tissue microarray and clinicopathological analysis Neoplasma 2007 54 207 211 17447851 39. Hu L. Sham J.S. Xie D. Wen J.M. Wang W.S. Wang Y. Guan X.Y. Up-regulation of fibroblast growth factor 3 is associated with tumor metastasis and recurrence in human hepatocellular carcinoma Canc. Lett. 2007 252 36 42 10.1016/j.canlet.2006.12.003 40. Chen G. Dang Y.W. Luo D.Z. Feng Z.B. Tang X.L. Expression of heparanase in hepatocellular carcinoma has prognostic significance: A tissue microarray study Oncol. Res. 2008 17 183 189 10.3727/096504008785114138 18773863 41. Shengbing Z. Feng L.J. Bin W. Lingyun G. Aimin H. Expression of kiss-1 gene and its role in invasion and metastasis of human hepatocellular carcinoma Anat. Rec. 2009 292 1128 1134 10.1002/ar.20950 42. Yang X.R. Xu Y. Yu B. Zhou J. Qiu S.J. Shi G.M. Zhang B.H. Wu W.Z. Shi Y.H. Wu B. High expression levels of putative hepatic stem/progenitor cell biomarkers related to tumour angiogenesis and poor prognosis of hepatocellular carcinoma Gut 2010 59 953 962 10.1136/gut.2008.176271 20442200 43. Bae J.S. Choi H.N. Noh S.J. Park B.H. Jang K.Y. Park C.K. Moon W.S. Expression of k19 and k7 in dysplastic nodules and hepatocellular carcinoma Oncol. Lett. 2012 4 213 220 22844356 44. Cheng A.S. Lau S.S. Chen Y. Kondo Y. Li M.S. Feng H. Ching A.K. Cheung K.F. Wong H.K. Tong J.H. Ezh2-mediated concordant repression of wnt antagonists promotes beta-catenin-dependent hepatocarcinogenesis Canc. Res. 2011 71 4028 4039 10.1158/0008-5472.CAN-10-3342 45. Geraud C. Mogler C. Runge A. Evdokimov K. Lu S. Schledzewski K. Arnold B. Hammerling G. Koch P.S. Breuhahn K. Endothelial transdifferentiation in hepatocellular carcinoma: Loss of stabilin-2 expression in peri-tumourous liver correlates with increased survival Liver Int. 2013 33 1428 1440 10.1111/liv.12262 23870052 46. Baumhoer D. Tornillo L. Stadlmann S. Roncalli M. Diamantis E.K. Terracciano L.M. Glypican 3 expression in human nonneoplastic, preneoplastic, and neoplastic tissues: A tissue microarray analysis of 4,387 tissue samples Am. J. Clin. Pathol. 2008 129 899 906 10.1309/HCQWPWD50XHD2DW6 18480006 47. Fasano M. Theise N.D. Nalesnik M. Goswami S. Garcia de Davila M.T. Finegold M.J. Greco M.A. Immunohistochemical evaluation of hepatoblastomas with use of the hepatocyte-specific marker, hepatocyte paraffin 1, and the polyclonal anti-carcinoembryonic antigen Mod. Pathol. 1998 11 934 938 9796718 48. Lugli A. Tornillo L. Mirlacher M. Bundi M. Sauter G. Terracciano L.M. Hepatocyte paraffin 1 expression in human normal and neoplastic tissues: Tissue microarray analysis on 3,940 tissue samples Am. J. Clin. Pathol. 2004 122 721 727 10.1309/KC09YTF2M4DLUYQ6 15491968 49. Song T.J. Fong Y. Cho S.J. Gonen M. Hezel M. Tuorto S. Choi S.Y. Kim Y.C. Suh S.O. Koo B.H. Comparison of hepatocellular carcinoma in american and asian patients by tissue array analysis J. Surg. Oncol. 2012 106 84 88 10.1002/jso.23036 22234941 50. Radhakrishnan R. Solomon M. Satyamoorthy K. Martin L.E. Lingen M.W. Tissue microarray—A high-throughput molecular analysis in head and neck cancer J. Oral Pathol. Med. 2008 37 166 176 18251941 51. Torhorst J. Bucher C. Kononen J. Haas P. Zuber M. Kochli O.R. Mross F. Dieterich H. Moch H. Mihatsch M. Tissue microarrays for rapid linking of molecular changes to clinical endpoints Am. J. Pathol. 2001 159 2249 2256 10.1016/S0002-9440(10)63075-1 11733374 52. Bubendorf L. Kononen J. Koivisto P. Schraml P. Moch H. Gasser T.C. Willi N. Mihatsch M.J. Sauter G. Kallioniemi O.P. Survey of gene amplifications during prostate cancer progression by high-throughout fluorescence in situ hybridization on tissue microarrays Canc. Res. 1999 59 803 806 53. Bubendorf L. Kolmer M. Kononen J. Koivisto P. Mousses S. Chen Y. Mahlamaki E. Schraml P. Moch H. Willi N. Hormone therapy failure in human prostate cancer: Analysis by complementary DNA and tissue microarrays J. Natl. Canc. Inst. 1999 91 1758 1764 10.1093/jnci/91.20.1758

PMC005xxxxxx/PMC5003444.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays3020103microarrays-03-00103ReviewOverview on Techniques to Construct Tissue Arrays with Special Emphasis on Tissue Microarrays Vogel Ulrich Institute of Pathology, University Hospital, Eberhard-Karls-University, Liebermeisterstrasse 8, 72076 Tuebingen, Germany; E-Mail: ulrich.vogel@med.uni-tuebingen.de; Tel.: +49-7071-29-82265; Fax: +49-7071-29-225817 4 2014 6 2014 3 2 103 136 09 1 2014 28 3 2014 09 4 2014 © 2014 by the authors; licensee MDPI, Basel, Switzerland.2014This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).With the advent of new histopathological staining techniques (histochemistry, immunohistochemistry, in situ hybridization) and the discovery of thousands of new genes, mRNA, and proteins by molecular biology, the need grew for a technique to compare many different cells or tissues on one slide in a cost effective manner and with the possibility to easily track the identity of each specimen: the tissue array (TA). Basically, a TA consists of at least two different specimens per slide. TAs differ in the kind of specimens, the number of specimens installed, the dimension of the specimens, the arrangement of the specimens, the embedding medium, the technique to prepare the specimens to be installed, and the technique to construct the TA itself. A TA can be constructed by arranging the tissue specimens in a mold and subsequently pouring the mold with the embedding medium of choice. In contrast, preformed so-called recipient blocks consisting of the embedding medium of choice have punched, drilled, or poured holes of different diameters and distances in which the cells or tissue biopsies will be deployed manually, semi-automatically, or automatically. The costs of constructing a TA differ from a few to thousands of Euros depending on the technique/equipment used. Remarkably high quality TAs can be also achieved by low cost techniques. tissue microarraystechniquesparaffinOCT ==== Body 1. Introduction With the advent of new histopathological staining techniques (histochemistry, immunohistochemistry, in situ hybridization) and the discovery of thousands of new genes, mRNA, and proteins by molecular biology the need grew for a technique to compare many different cells or tissues on one slide in a cost effective manner by sparing labor and staining consumables and with the advantages of equal staining conditions and the possibility to easily track the identity of each specimen: the tissue array (TA). Basically, a TA consists of at least two different specimens per slide. Two fundamental techniques to create a TA may be separated (Figure 1, Figure 2A–C): The tissue macroarray with different tissue or cellular materials being arranged directly on slides as imprints/suspensions of cells or sections of tissue or cell blocks [1], and the tissue microarray (TMA) with the construction of a tissue microarray block from which a lot of sections can be cut (Figure 2D). With the exception of the recently published patch TMA by Deng et al. by which cores of already stained and retrieved tissue sections are arranged on a slide the term TMA comprises the construction of a TMA block in the following [2]. Since the, probably, first description of a TMA for histochemistry by Lilie in 1965, several different techniques to construct a TMA and many synonyms for TMA were published until now [3]. A schematic overview on the techniques may be found in Figure 1. A more comprehensive overview in chronological order is presented in Table 1. Figure 1 Schematic overview on techniques to construct tissue arrays. Figure 2 Overview on techniques to construct tissue arrays. (A–C) Tissue macroarray: Paraffin block with one small biopsy (A). Paraffinblock with tumor of a resection specimen: The paraffin block may be scratched to get several small fragments of the tumor out of one section (B). Sections of eight different specimens are arranged on one slide to create a tissue macroarray (C). (D) A typical paraffin tissue microarray (TMA) with 561 paraffin tissue core biopsies (Computer numerical control (CNC) predrilled recipient block, manually deployed paraffin tissue core biopsies 0.6 mm in diameter). (E) Tissue rods trimmed with a razor blade as material to construct TMAs. (F) Paraffin tissue punch (1 mm in diameter, Beecher Instruments, Inc., Sun Prairie, WI, USA) with a paraffin tissue core biopsy protruding at the tip. (G) Thick paraffin section (100 µm) being cut on a rotary microtome as material to construct TMAs. (H) Tissue rods in a routine steel embedding mold to be poured with paraffin to get paraffin tissue layers, which can be stacked to produce a TMA. (I) Paraffin tissue core biopsies standing upright on a double-sided adhesive tape, which is mounted on a routine x-ray film placed in a standard steel embedding mold. (J) Many different thick paraffin sections are stacked to create a paraffin TMA. (K) A CNC predrilled recipient block, which will become a paraffin TMA (completely filled in D). (L) A predrilled paraffinized agar block (a) embedded in a standard paraffin block to function as stabilization body. TMAs may differ in: - the kind of specimens (cells (e.g., from effusions), cell lines, tissues (needle core biopsies, resection specimens)). - the technique to prepare the specimens to be installed (knife for tissue rods, tissue punches for tissue core biopsies, microtomes for sections (Figure 2E–G). - the arrangement of the specimens (haphazardly or oriented, distance of the specimens). - the number of specimens installed (2–10,000). - the dimension of the specimens (0.1 mm–5 mm). - the embedding medium depending on the physicochemical property of the specimens (Optimal cutting temperature (OCT) compound for frozen material, paraffin for formalin fixed material, resins for formalin, or glutaraldehyde fixed material). - the technique to construct the TMA itself (not preformed recipient block, preformed recipient block with or without a stabilization body). - the aim of the TMA (e.g., Battifora: multipurpose TMA, segmented TMA, theme oriented segmented TMA, clinically defined segmented TMA [4]; for more details see also Kajdacsy-Balla et al. [5]). microarrays-03-00103-t001_Table 1Table 1 Chronological compilation of techniques to construct tissue arrays (TA) and TA applications (only the first name of publications or patents cited). Year Author Name of the technique/tissue microarray 1965 Lilie Special blocking and trimming procedure for cross sections of multiple small tubular structures [3] 1986 Battifora Multitumor (sausage) tissue block (MTTB): wrapped fixed tissue rods [4] 1987 Wan Multi-tissue straw of paraffin embedded tissue cores: drinking straw as encasement [6] 1988 Kraaz Multiblock control for immunohistochemistry: 4 mm skin biopsy punch modified with a mandrin; cores placed into a warm cast [7] 1988 Rowden Histocomposites for immunohistological screening of monoclonal antibodies: alignment of the tissue sticks by standard hand-operated cigarette roller; swine casing [8] 1990 Battifora Checkerboard tissue block: stack of agar plates with embedded fixed tissue rods [9] 1991 Miller Multitumor (sausage) blocks (MTSBs) as controls for immunohistochemistry: stack of paraffin plates of tissue rods [10] 1994 Press Multitumor tissue blocks: tissue strips arranged in rows separated by a layer of parafilm [11] 1994 Rose Multiblock slides for teaching: Kraaz punching method, grid pattern by careful hand positioning [12] 1994 Sundblad Simplified multi-tissue blocks (SMB): Wedge shaped tissue rods removed from the surface of a paraffin donor block and further processed similar to the Miller technique [10,13] 1997 Petrosyan Multispecimen tissue blocks: Multichambered (“honeycomb”) plastowax dividers prepared with a rubber mold [14] 1998 Kononen Tissue microarray (TMA): punching paraffin tissue cores; arrangement in a Cartesian coordinate system; development of a manual tissue arrayer [15,16] 2000 Gillett Multiple tissue core array: Tissue cores punched with a 11-gauge core cut needle and installed into a recipient block with preformed holes, which were punched with a 13-gauge needle. Even grid by using the back of a standard R. A. Lamb processing cassette as a 34-hole template [17] 2000 Chan Multitissue spring-roll control block: On-slide multitissue controls for immunohistochemistry [18] 2001 Hoos Tissue microarrays using cell lines and frozen tissue microarrays (OCT) [19] 2001 Schoenberg Frozen tissue microarray: OCT embedding medium [20] 2002 Packeisen Tissue microarray: Kononen technique without using the tape transfer system [21] 2002 Badve Multiorgan tissue blocks [22] 2003 Mengel Tissue microarrays constructed with poured holes and a double melt procedure [23] 2003 Vogel Tissue microarrays constructed with predrilled ordinary steel embedding molds [24] 2003 Hidalgo Small format tissue array: manual construction using a bone marrow aspiration needle [25] 2003 Wang Tissue macroarray by arranging section fragments on different slides [1] 2003 Matysiak Tissue microarrays automatically constructed with a simple method: semi-automated Kononen tissue arrayer [26] 2003 Wilkens Tissue microarrays constructed with a double sided adhesive tape [27] 2004 Schnetz Robotic tissue arrayer using the punching technique of Kononen in combination with positive and negative pressurized air [28] 2004 Vogel TMA constructed with a microcompound table and a drill grinder [29] 2004 Vogel Tissue microarrys constructed with a computer numerical control (CNC) drilling machine [30] 2004 Dan Tissue microarray constructed with a common microscope [31] 2004 Pan High density tissue array (manual construction): conventional 16-gauge bone marrow biopsy trephine apparatus to puncture the paraffin blocks [32] 2005 Vogel Tissue microarrays with paraffin tissue core biopsies of 0.43 mm in diameter [33] 2005 Howat Resin tissue microarrays [34] 2005 LeBaron Ultrahigh density microarrays of solid samples: stacks of tissue sections [35,36] 2005 Montgomery Tissue microarrays from suspension cells: paraffin embedding of cell pellets in Eppendorf tubes and punching the cells out of the tube [37] 2005 Datta Microarrays from needle biopsy specimens: foil templates to reshape paraffinized needle biopsy specimens for further installment into Kononen TMAs [38] 2005 Chen Tissue microarrays without prefabricating recipient blocks: double sided adhesive tape technique with x-ray film backbone [39] 2005 Meng Tissue microarrays constructed with the ZM-1 arrayer by injecting the tissue cores in liquid paraffin [40] 2005 Mengel Tissue microarrays of agar embedded cell lines for on-slide Control [41] 2005 Song Tissue microarray made of paraffinized agar: stabilization body technique [42] 2006 Vogel Tissue microarrays using a stabilization body [43] 2006 Vogel Tissue microarrays filled with a paraffin tissue punch with a countersink [44] 2006 Pires Tissue microarrays constructed with custom-built needles and double sided adhesive tape technique [45] 2006 Wang Tissue arrays manually constructed using a hand-made paper mold [46] 2007 Vogel Tissue microarrays constructed by combining different techniques [47] 2007 Vogel Tissue microarrays with cracks cured with a soldering iron and adhesive tape [48] 2007 Zhou Tissue microarray technology for frozen pathological samples: agar as stabilization body [49] 2007 Jiang Microarray group: Different sections of small TMAs arranged on one slide [50] 2008 Vogel Tissue microarrays constructed with evenly long core biopsies created with a cutting board and a cutting board arrayer [51] 2008 Szekeres Tissue micro-array builder for pouring TMAs [52] 2009 Vogel Cast recipient blocks for paraffin tissue microarrays using conventional steel embedding molds (“top pin tissue arrayers”) [53,54] 2009 McCarthy Checkers of prostate biopsy specimens installed in wax templates for TMA construction [55] 2010 Vogel Tissue microarrays constructed from needle biopsy specimens by combining the drilling technique with the adhesive tape technique [56] 2010 Tsao Gelatin-based capsules for frozen TMA construction [57] 2011 Fridman Vertical clustering re-arrangement technique for prostate needle biopsies [58] 2011 Shebl Mechanical pencil tips as paraffin and tissue punches [59] 2011 McCarthy Improvement of the checker technique by punching the prostate biopsies out of the checkers with the Kononen technique (Beecher tissue arrayer) [60] 2012 Pilla Implementation of a barcode-driven error control of the design and execution of a TMA by using the laboratory information system [61] 2012 Yang HT-1 tissue arrayer: punching the holes of the recipient block in one action; negative pressure for removing air bubbles [62] 2012 Choi Dot grid paper on surface of a recipient block to structure the TMA [63] 2013 Deng Patch TMA: TMA on a slide using cores of retrieved already stained sections [2] 2013 Shi Tissue rods: punched parallel to the donor block surface to ensure equal length of the rods and the tissue of interest in every section to be cut [64] 2013 Foda Modification of Shebl’s technique with mechanical pencil tips as punches [65] 2013 Garcia-García Inexpensive self-made tissue punches useful in paraffin TMAs [66] 2013 Zanini Homogeneous distribution of cells or spheres in cell blocks used for TMA construction by shaking [67] A TMA can be constructed by arranging the tissue specimens in a mold and subsequently pouring the mold with the embedding medium of choice without the need for a prefabricated recipient block (Figure 1, Figure 2H–J) [4,6,9,10,35]. Tissue rods as well as tissue core biopsies (TCBs) or tissue sections may be used (Figure 2E–G) [4,6,10,35]. To prevent the TCBs from tumbling and to structure the TMA, different methods were designed like an encasement in a drinking straw [6], the use of a double sided adhesive tape (Figure 2I) [27,39], paper molds or even the injection of the cores in already liquid paraffin [40,46]. In contrast preformed so-called recipient blocks consisting of the embedding medium of choice have punched, drilled or poured holes of different diameters and distances in which the cell or tissue biopsies will be deployed manually, semi-automatically, or automatically (Figure 1, Figure 2K) [15,23,24,52,53]. Concerning paraffin TMAs (PTMAs) an additional melting process may be performed to get a strong contact between the paraffin of the PTMA and the paraffin tissue core biopsies (PTCBs) installed [4,6,23]. The use of a stabilization body preferably made of agar may facilitate this melting process by allowing a one step complete melting procedure (Figure 2L) [42,43]. The costs of constructing a TMA differ from a few to thousands of Euros depending on the technique/equipment to be used. Remarkably high quality TMAs can be also achieved by low cost techniques [27,29]. In the following, the milestones in the development of the TMAs and some minor modifications of the techniques are described chronologically. Furthermore, the criteria to choose the right technique in a certain setting are discussed and low cost methods presented. 2. The Development of the TMAs in the Course of Time 2.1. The Multitumor (Sausage) Tissue Block Invented by Battifora (1986) [4] In 1986, Battifora published his technique to construct a multitumor (sausage) tissue block to test new antibodies on about a hundred tissues on one slide [4]. Tissues of interest were excised by knife from paraffin tissue blocks, deparaffinized in xylene, and rehydrated using a descending alcohol series (100%–50%). These rehydrated tissues were trimmed with a razor blade to rods of about 10 mm in length and of an average cross-sectional area of 1 mm2 (Figure 3A). About 100 different rods were tightly wrapped in small intestine of small mammals, such as rabbits (Figure 3B), and routinely reparaffinized (Figure 3C). More than a 1000 5 µm thick sections could be prepared from such multitumor tissue blocks (Figure 3D). Although being a milestone in the development of TMAs this method appeared to be too laborious especially due to the unnecessary deparaffinization process prior to construct the TMA. Furthermore the TMA must be constructed very carefully to relocate a single probe. The wrapping of the tissue rods is not so easy (own experience). Therefore, Battifora himself improved this technique a few years later. Figure 3 The trail-blazing precursor techniques of the modern tissue microarrays. (A–D) The multitumor (sausage) tissue block by Battifora. Fixed tissue rods (r) stacked like log piles were tightly wrapped in amnion (a) (instead of Battifora’s small intestine of small mammals) (A), secured by thin thread (th) (B), paraffinized (C), trimmed and cast into a standard paraffin block (D). (E–I) The multi-tissue straws of Wan et al. Sections of an ordinary red plastic drinking straw were fixed in a standard steel embedding mold by a little amount of paraffin (E) to enhance the installment of paraffin tissue core biopsies (PTCBs) (own modification) (one PTCB on the tip of the needle) (F). These filled straws were melted to glue the PTCBs together (G). After resolidification the plastic encasement was removed (H) and the composed PTCBs embedded in a standard paraffin block (I). (J–M) The checkerboard tissue block by Battifora and Mehta. Stripes of fixed tissue were poured into agar plates (J), the plates were stacked (K), paraffinized (L), trimmed and cast into a standard paraffin block. The cut surface with the brown tissue rods and the white surrounding agar. (M) (Note: alignment of the rods not as precise as by Battifora and Mehta). 2.2. The Multi-Tissue Straws of Wan et al. (1987) [6] With their publication in 1987 Wan et al. introduced the punching technique for retrieving the paraffin tissue material (the PTCBs) to be later installed in PTMAs [6]. The tip of a 16 gauge syringe needle was removed and the new edge resharpened to get an instrument resembling a miniature cork borer (Figure 8A,B in Section 2.7). Mounted on a plastic syringe tissue cores could be removed from paraffin blocks and extruded from the needle with a wire stylet. These tissue cores were stored in vials (Figure 12I in Section 4) creating a tissue library or directly placed into an ordinary plastic drinking straw (6.3 mm in diameter) as encasement (Figure 3E,F). These straws with an average of 24 tissue cores were then melted to get a firm contact between the paraffin of the tissue cores (Figure 3G). After resolidification the plastic straw casing was removed (Figure 3H). One or more tissue straws could then be embedded in a paraffin block and sectioned (Figure 3I). By using a marker core with a certain pattern of staining or structure, the orientation in every straw was facilitated. With this technique about 120 individual tissue samples could be evaluated on a standard slide. The published method of punching the paraffin tissue of interest (the PTCBs) was trailblazing and is used in nearly every technique for PTMAs today. By using PTCBs the possibility of establishing tissue banks of a small volume was introduced. Furthermore, this technique avoided the unnecessary deparaffinization process before assembling the specimens as proposed by Battifora, therefore preceding current technology. Wan et al. also mentioned the possibility of different core sizes and addressed the problem of sampling errors due to tissue heterogeneity. The compilation of several tissue cores to elongate the core was also described. Due to the uniformity of the core size and the use of a marker core the accurate tissue identification within the straws was facilitated in contrast to the technique of Battifora. However, the stabilization of relatively few PTCBs during the melting process by the straw appeared to be too laborious for wide-spread application. Moreover, the installment of the PTCBs into the straws needs good training (own experience) that the PTCBs do not topple down in the encasement. In 2010, Tsao et al. adopted the Wan technique to construct low cost frozen tissue microarrays by using sectionable gelatin-based capsules for arranging the punched frozen tissue [57]. 2.3. The Checkerboard Tissue Blocks by Battifora and Mehta (1990) [9] Although sticking to the formerly described complex deparaffinization technique this publication of Battifora and Mehta in 1990 became a footstep by introducing the alignment of the tissue specimens in a Cartesian coordinate system (checkerboard pattern) [9]. By using a multi-blade knife of disposable microtome knives and different-sized spreaders tissue rods of uniform thickness and square cross sections were cut out of fresh or dewaxed tissue. These rods were placed into the rectangular grooves of an aluminum tissue embedding mold and covered with fluid agar 3% at 60 °C (Figure 3J; manual alignment without an aluminum mold). The solidified agar plates were stacked (Figure 3K) and placed in a perforated metal cassette for paraffin embedding. By aligning the specimens in a Cartesian coordinate system the exact relocation of a single probe was easy to manage and led the way to current techniques. Battifora and Mehta already anticipated the possibility of mass screening of tissue samples for new prognostic markers probably by the use of automatic robotic screening and the use of the TMA for interlaboratory quality assessment. They also mentioned the compilation of several short rods within the grooves to elongate the rods. However, due to the complex construction process with deparaffinization and agar embedding this technique did not get wide-spread acceptance. In 1991, Miller advanced the technique of Battifora and Mehta by omitting the dewaxing step and the agar embedding [10]. Miller cut already paraffinized tissue into rods (Figure 4B), assembled the rods in one layer in an ordinary steel mold to melt these rods to plates (Figure 4C) and stacked these plates after resolidification of the paraffin (Figure 4D) to get a PTMA. This low cost, simple and robust technique is till now in use especially to create PTMAs for positive controls in immunohistochemistry. Figure 4 The Miller technique—tissue rods. (A) Multiple paraffin blocks used as donor blocks. (B) For cutting out the paraffinized tissue rods a trimming knife, a scalpel or a razor blade may be used. (C) Alignment of the tissue rods in one layer in a routine steel embedding mold to be cast into a paraffin tissue plate. (D) Stacking of different paraffin tissue plates. (E) Trimming of the stacked plates. (F) Sectioning of the embedded stacked plates e.g., for use as positive controls in immunhistochemistry. (G) Stained section of the tissue microarray (Hematoxylin Eosin). 2.4. The Tissue Microarray by Kononen et al. (1998) [15] The breakthrough of the TMAs leading to world wide use especially in the setting of translational research and medicine was achieved by a technique coined as “tissue microarray” which was developed in a cooperation project by the groups of Guido Sauter at the Institute of Pathology of the University of Basel (Switzerland) and of Olli Kallioniemi at the National Human Genome Research Institute in Bethesda (Maryland, USA). This technique was published in 1998 in Nature Medicine by Juha Kononen as first author and may be called the Kononen technique in the following [15]. This publication may represent the numerous well-known publications of these groups. By constructing the “tissue microarray” the authors combined the punching technique of Wan et al. with the precise alignment and easy relocation of the tissue specimens (the PTCBs) of the “checkerboard tissue block” of Battifora and Mehta [6,9]. Kononen et al. developed a machine in cooperation with S.B. Leighton (inventor according to US patent), the manual tissue arrayer (Beecher Instruments Inc, Sun Prairie, WI, USA) (Figure 5A), which consisted of two slightly different sized tissue punches (Figure 5B, arrows) mounted on a vertically movable “precision guide” which itself was fixed vertically to a horizontal xy stage. With the recipient punch, which is slightly smaller in diameter than the donor punch, the holes in the recipient block, the later PTMA, were constructed. The PTCBs were punched out of so-called donor blocks, i.e., routinely fabricated paraffin tissue blocks, by using the donor punch and transferred into the holes of the recipient block. Punches of different diameter (e.g., 0.6 mm, 0.8 mm, 1.0 mm, 2.0 mm) were available. As many as 1000 PTCBs could be installed into a 45 × 20 mm recipient block in a perfect Cartesian coordinate system which made the relocation of the PTCBs very easy. A disadvantage of this method—if the word disadvantage may be even used in the light of the great success of this technique—is the missing melting procedure of the PTMA after the deployment of the PTCBs in contrast to the methods of Battifora, Mehta and Wan (Figure 5D,E) [6,9]. A strong contact between the paraffin of the PTMA and the PTCBs is decisive at cutting and floating the PTMA section on the waterbath. If the section is more than 3 µm thick (own unpublished data) there may be some rolling and folding of the PTCBs in the waterbath (Figure 5F), which miss good contact to the surrounding paraffin. In consequence, such PTCBs have a small contact area to the slide (Figure 5G), cause an increased mechanical resistance during the washing and staining procedures, may float off the slide and will be lost for evaluation therefore reducing the efficacy of the PTMA technique. Kononen et al. also mentioned melting of the PTMA surface to ensure easier sectioning, but warned of moving of PTCBs in the PTMA during melting. Kononen et al. tried to achieve the strong contact of the paraffin of the PTMA and the PTCBs by pressing PTCBs somewhat larger in diameter into smaller holes. Such a procedure may avoid the folding of the PCTBs. However, firstly, such a good fit cannot be achieved for all PTCBs (Figure 5E) and can only be ensured by using not deformed tissue punches. Secondly, especially high density PTMAs with a hundreds of PTCBs may crack during the cooling procedure before sectioning because of the high tension in the block (Figure 5H) [68]. This tension may be caused by the increased volume of the PTCBs in comparison to the smaller holes. Of course, cracking of a paraffin block is not restricted to Kononen-PTMAs. To avoid this cracking and some other general problems of sectioning like the disruption of the section on the waterbath, Kononen et al. advised the use of the paraffin tape transfer system (Instrumedics Inc., St. Louis, MO, USA) (Figure 5I–K). By mounting a tape on the surface of a PTMA fixed in a microtome clamp a section can be cut at room temperature and will adhere without any folding or cracking on the tape (Figure 5J). This section may be transferred to a slide coated with some resin. After polymerization of the resin by UV light the section will strongly adhere on the slide and the tape can be removed by a solvent (Figure 5K). The disadvantage of this tape system may be the price for the UV lamp and the consumables (about 2,500 €) and the resin itself, which may lead to some problems at fluorescence in situ hybridization [68]. Furthermore, according to Hoos and Cordon-Cardo the lowest degree of tissue damage was seen without using the adhesive transfer tape [19]. Catchpoole et al. describe a higher incidence of nonspecific staining in immunohistochemistry by using the tape method [69]. For more details on loss of cores see also the review of Pinder et al. [70]. The problems of unevenly long PTCBs (Figure 5L–N) or the submerging of PTCBs in the holes of the PTMA which might reduce the efficacy of the TMA technique may be solved by installing as many PTCBs as needed to fill the holes completely [19]. However, the PTCB may be pulled out by the microtome knife or may be bent during the cutting procedure, if the PTCB overtops the surface of the PTMA too much (own experience). Figure 5 The Kononen technique coined as tissue microarray—punching the holes of the preformed recipient block and the tissue of interest. (A) The manual tissue arrayer (MTA-1) developed by Kononen et al. and produced by Beecher Instruments Inc., Sun Prairie, WI, USA. (B) The turret of the manual tissue arrayer allowing the switch between the two paraffin punches (arrows) (the punch for the paraffin tissue core biopsies (PTCBs) has a somewhat larger inner diameter). (C) Tissue arrayer with a motorized stage (Alphelys, Plaisir, France). (D) PTCB within the hole of the recipient block: at least one half of the PTCB with a good contact to the surrounding paraffin of the recipient block. (E) PTCB within the hole of the recipient block: missing contact to the surrounding paraffin of the recipient block (arrows). (F) Paraffin section with some PTCBs with a good contact to the surrounding paraffin (asterisk). One PTCB with a missing contact is rolled (arrow). (G) Rolled PTCB after staining with limited evaluation. (H) Cracked paraffin tissue microarray at cooling before sectioning. (I) Paraffin tape transfer system (Instrumedics Inc., St. Louis, MO, USA) consisting of a tape and a special slide coated with a resin. (J) The tape is mounted on the surface of a paraffin tissue microarray block to be loaded with the section. (K) The mounted section is transferred to the slide. After UV polymerization of the resin the section sticks to the slide; the tape can be removed after incubation in a solvent. (L) Deeply cut paraffin tissue microarray with thinning out of the unevenly long PTCBs. (M) Routinely punched PTCBs demonstrating the different length, which is due to the different thickness of the donor tissue. (N) Cross section of a paraffin tissue microarray (PTMA): Holes of the recipient block filled with more than one PTCB to avoid thinning in deeper sections. The PTCBs protrude the surface of the PTMA. Surface of the paraffin block (asterisk) (O) Tissue arrayer with a reflecting microscope to facilitate the detection of the best punching location on the donor block (Veridiam Oceanside, CA, USA). (P) Fully automated tissue arrayer (Beecher Instruments Inc., Sun Prairie, WI, USA). The installment of cores into 2–3 holes per case will minimize a probable sampling problem and the problem of tissue loss due to unevenly long PTCBs and due to rolling and folding of PTCBs [19]. The original Beecher manual tissue microarrayer was improved in a little while to become a computer numerical control (CNC) arrayer with still manual (Figure 5C) or fully automatic transfer of the PTCBs (Figure 5P, automated tissue arrayer ATA-27, Beecher Instruments, Inc. [71,72]) [26]. Furthermore an arrayer was designed with a reflecting microscope to improve the selection of the tissue from the donor block (e.g., VTA-100 Tissue Arrayer (about 55,000 US$), Veridiam, Oceanside, CA, USA [73]) (Figure 5O). Based on the punching technique of Kononen, Schnetz et al. invented a robotic tissue arrayer using negative and positive pressurized air to guide and improve the automatic punching process (Oridis Biomed Forschungs- und EntwicklungsGmbH, Graz, Austria) [28]. The Galileo CK family of semiautomatic tissue arrayers also applies the Kononen technique (Integrated Systems Engineering S.R.L., Milano, Italy [74]). In 2012 Yang et al. presented their HT-1 tissue arrayer by which the holes of the paraffin recipient block are punched out in one action comparable to the technique of Song (see below) [42,62]. This may be one of the fastest methods to create a preformed recipient block. Furthermore, Yang designed a very inventive method to extrude the air between the PTCBs and the holes. 2.5. The PTMA by Mengel et al. (2003) [23] In 2003 Mengel et al. reintroduced the melting as a two step procedure into the construction process of PTMAs as originally described by Battifora, Mehta, and Wang to achieve a strong contact between the PTCBs and the surrounding paraffin and to prevent tensions in the paraffin block [6,9,23]. Moreover, Mengel et al. transferred the cost-effective pouring of the holes of the recipient block, which was already described for frozen TMABs by Hoos and Cordon-Cardo, to paraffin TMAs (PTMAs) [19,23]. This patented procedure is currently licensed by Zymed (San Francisco, CA, USA) or Zytomed (Berlin, Germany) to produce, e.g., customized PTMAs (MaxArray System) as a commercial service [23,75]. In brief, as disclosed in the patent, 60–120 cylinder pins with a diameter of e.g., 1.5 mm were driven into an aluminum block in a Cartesian coordinate system. These pins fit into the holes of the bottom of a modified conventional embedding mold (Figure 6A). After solidification of the paraffin, which was poured into the embedding mold, the cylinder pins were withdrawn resulting in a PTMA blank with up to 120 holes. This blank was inserted in a second conventional embedding mold and the holes were filled with PTCBs. By applying heat from the bottom of the mold the filled PTMA was melted up to 80% of the height of the block. After resolidification overhead heat was applied for melting the 20% rest of the paraffin to ensure a complete melting of the PTMA. The equipment for this two-step melting procedure cannot be bought commercially. Customers of Zymed receive a tissue punch to retrieve the tissue specimens of interest and send the PTCBs to Zymed to construct the PTMA. A great advantage of the Mengel technique is the strong contact between the PTCBs and the surrounding paraffin (Figure 6B) with nearly no loss of PTCBs (<1%) due to folding and rolling rendering the paraffin tape transfer system unnecessary [23]. Unaddressed by Mengel et al. the problem of unevenly long PTCBs. Probably only one PTCB may be installed in one hole due to the melting process. By routinely using PTCBs of different length some PTCBs may be lost in deeper sections of the PTMA. Moreover, the Mengel system is designed for 96 PTCBs per array in commercial service, which is less than the number of PTCBs which could be already achieved by Wan et al. (120 PTCBs) [6]. A steel mold to cast the holes of the recipient block which is slightly different to that of Mengel was patented by Szekeres et al. in 2008, and can now be purchased at Thermo Scientific as Thermo Scientific™ Lab Vision™ Tissue Microarray (TMA) Builder (Waltham, MA, USA [76]) or at 3DHISTECH (Budapest, Hungary [77]) as manual TMA kit (Figure 6C) [52]. Furthermore, a modification of the mold made of rubber is distributed by Unitma (Seoul, Korea [78]) under the brand of Quick-Ray Mold Kit (170 holes, 1.0 mm in diameter, 500 US$/each) (Figure 6D). Another modification of this technique was introduced by Vogel as top pin tissue arrayer which can be used with ordinary steel embedding molds to cast the holes of the recipient block (Figure 6E,F) [54]. Figure 6 The Mengel technique—pouring the holes of the preformed recipient block. (A) Routinely used steel embedding mold with many holes in the bottom through which steel pins are pushed to work as spacers for the holes of the recipient block (for demonstration only 20 steel pins inserted). (B) Perfectly melted paraffin tissue core biopsy (PTCB) in a paraffin tissue microarray (PTMA). (C) Tissue arrayer for pouring the holes of the recipient block as patented by Szekeres et al. (D) Tissue arrayer made of rubber for pouring the holes of the recipient block (Unitma, Seoul, South Korea). (E,F) Top pin tissue microarrayer as designed by Vogel. Steel pins (arrow) fixed to a metal plate are pushed through a perforated plate (asterisk) and inserted from above into a routinely used steel embedding mold (composite in F). 2.6. The PTMA by Wilkens and Chen et al. (2003, 2005) [27,39] Probably, the simplest technique to construct a PTMA was patented in 2003 by Wilkens and published as an apparently second independent invention in 2005 by Chen et al. [27,39]. Figure 7 Wilkens and Chen technique—double sided adhesive tape to fix the PTCBs without the need for a preformed recipient block. (A) Example for a double-sided adhesive tape to fix the paraffin tissue core biopsies (PTCBs). (B) White double-sided adhesive tape with a brown protection paper mounted on a standard black x-ray film (asterisk). (C) PTCBs arranged in a Cartesian coordinate system on the double-sided adhesive tape. (D) Strong contact between the PTCBs and the double-sided adhesive tape—no PTCB falls off. (E) Melting of the PCTB-adhesive tape-x-ray-film-sandwich in a standard steel embedding mold. Do not heat over 65 °C, otherwise the adhesive tape may shrink and destroy the PTMA. Note that one PTCB toppled down (arrow). Not paraffin, but the paraffinized tissue must be in contact with the adhesive tape. (F) PTMA after resolidification and removal of the adhesive tape-x-ray-sandwich. Note the tumbled PTCB (arrow). (G) Routinely used steel embedding mold with many holes (see also Figure 6A) in the bottom is painted with a standard permanent marker. (H) The double-sided adhesive tape is fixed to the painted bottom of the mold. (I) After removing the tape from the bottom of the mold the grid of the mold is transferred to the tape and can be used to structure the PTMA. (J) The double-sided adhesive tape is fixed to a recipient block with preformed holes, which is mounted on a microtome clamp. (K) After cutting a 5–10 µm thick section of the preformed recipient block is fixed to the adhesive tape. (L,M) PTCBs can now be transferred manually from a paraffin donor block to the grid of the double-sided adhesive tape. According to the technique of Wan et al. PTCBs were punched out of ordinary paraffin tissue blocks and transferred to a double sided adhesive tape (Figure 7A) mounted, e.g., on a piece of regular x-ray film (Figure 7B). The PTCBs were manually aligned in a Cartesian coordination system like in the Battifora/Kononen technique (Figure 7C). The glue of the adhesive tape held the PTCBs in place (Figure 7D) and in an upright position especially when the PTCB-tape-x-ray film sandwich was put in an ordinary steel embedding mold and filled with fluid paraffin to construct the PTMA (Figure 7E). A disadvantage of this technique may be the use of only one PTCB per spot whereby the PTMA may loose some cores in deeper sections as already discussed. Furthermore, the manual alignment of the cores may be not so precise as with the Beecher tissue arrayer making the evaluation of the stained sections more complicated. When using PTCBs smaller than 0.6 mm in diameter the fluid paraffin has to be poured into the mold very carefully in order not to incline the PTCBs which may also make the evaluation of the stained section difficult (own unpublished experience). Nonetheless, Wilkens and Chen proved that the construction of a PTMA with a Cartesian alignment of the PTCBs is possible without a prefabricated recipient block in a very cost-effective way. A modification of this technique was published by Wang et al. in 2006, using a hand-made paper mold instead of an adhesive tape to keep the PTCBs in line [46]. Furthermore, different systems to define a grid on the tape were designed. Pires et al. used a translucent adhesive tape and put a piece of paper with a printed grid under the tape [45]. Vogel developed two different systems to get some kind of a grid on the tape: One system with a marker-painted metal grid to transfer the ink to the tape (Figure 7G–I); the other system to glue sections of predrilled recipient blocks on the tape (Figure 7J–M). The technique of Wilkens and Chen is also well suited for curing cracked PTMAs by arranging the broken parts onto the adhesive tape and consecutive melting. 2.7. The Predrilled PTMA by Vogel (2004) [29,79] In 2004 Vogel presented a method to construct PTMAs by using a conventional drill grinder, a microcompound table and a drill stand which could be purchased in every hardware store for less than 300 € (Proxxon GmbH, Föhren, Germany) [29,79]. In brief, the tips of routinely used hypodermic needles were cut with a cutting disk using the drill grinder and resharpened as proposed by Wan et al. (Figure 8A,B) [6]. Skin biopsy punches (Kai Industries, Seki, Japan) (Figure 8C) or commercially available paraffin tissue punches (Figure 8D) were also used to retrieve the PTCBs. The holes of the prefabricated recipient block were drilled in a Cartesian alignment into a standard paraffin block (Figure 8E). The recipient block was fixed in a water bath of polyvinylchloride (PVC) (Figure 8E). The PTCBs were punched from donor blocks and manually transferred to the holes of the recipient block with the optional use of an illuminated magnifying glass, which may be found in every laboratory of pathology. With this low cost equipment high densitiy PTMAs could be constructed with more than 600 precisely arranged PTCBs per standard paraffin block. Like with the Kononen technique a melting step was primarily not included causing some rolling and folding of PTCBs at sectioning when the paraffin tape transfer system of Instrumedics Inc. was not used. Figure 8 The Vogel technique—drilling of the holes of the recipient block. (A) Drill grinder in a drill stand with a cutting disk (Proxxon GmbH, Föhren, Germany). (B) Tissue punches of different inner diameters (0.3 mm to 1.0 mm) constructed out of routine needles. In case needles are commercially not provided with a stylet, wires or other needles may be used as stylets. Infusion caps (red pieces) (C) Skin punches of different inner diameters (1–5 mm; Kai Industries). The stylet of a bone marrow biopsy needle (asterisk) can help to push the PTCBs out of the skin punches after the narrow (arrows) was widened manually with a drill bit (arrowhead). Nowadays, the skin punches are also provided with a built-in stylet; however, this stylet may be easily jammed by paraffin and may break (own unpublished observation). (D) Resharpened commercial paraffin tissue punch (Beecher Instruments, Inc.), which was waste material of a TMA core facility after breakage of the tip of the cannula. (E) A water bath made of polyvinylchloride mounted on a microcompound table (x-y table) which is fixed to a drill stand equipped with a drill grinder (Proxxon GmbH, Föhren, Germany). A standard paraffin block is fixed within the water bath for drilling of the holes. (F–H) Computer numerical control (CNC) drilling machine. The water bath is fixed to a bench vice of the CNC drilling machine. The holes of the recipient block are drilled under water (cooling effect and floating off the paraffin debris). (I) Paraffin recipient block perfectly drilled by the CNC drilling machine. (J) Section of a paraffin recipient block with a honeycomb pattern to enlarge the number of installed PTCBs. Perfect drilling by the CNC machine. (K) Drilling the holes of a recipient block made of optimal cutting temperature (OCT) medium for frozen TMAs. The OCT block is mounted on a microtome clamp, which is fixed to the microcompound table on the drill stand. The clamp was cooled in a freezer before drilling. (L) Filled frozen TMA mounted on the clamp of a freezing microtome. (M) Fully automated tissue arrayer using the drilling technique (TMA Grand Master, 3DHistech, Budapest, Hungary). Vogel modified this technique by using a computer numerical control (CNC) drilling machine (Figure 8F–I) for creating up to 2500 holes 0.3 mm in diameter into a standard paraffin block to achieve the highest number of PTCBs per PTMA to this day when using a prefabricated recipient block technique [80,81]. With the CNC drilling machine also special arrangements of the holes (e.g., in a honeycomb pattern, Figure 8J) were easily possible to enhance the number of PTCBs to be installed [82]. The CNC-drilled holes displayed the highest quality in comparison to all other drilling and punching techniques. In the meantime the drilling technique to create the holes in the recipient block is incorporated in an automated PTMA construction machine (e.g., TMA Grand Master) by 3DHISTECH (Budapest, Hungary) (Figure 8M). The drilling technique is also applicable for the construction of frozen TMAs (Figure 8K,L). A tissue arrayer applying the drilling technique is also manufactured by Mr. Mirlacher, University of Hamburg. Apparently, there is only one hint for this arrayer being published in a subordinate clause [83]; the arrayer is only constructed on demand. 2.8. The PTMA of Paraffinized Agar by Song (2005) [42] Independently invented by Mengel et al., Vogel, Yan et al., and Song, Song was apparently the first inventor and consistently got the patent on a method to use stabilization bodies (e.g., of paraffinized agar) as recipient blocks [42,84,85,86]. In brief, Song poured a block of agar, paraffinized it and punched out the holes of the later PTMA by using a Cartesian aligned grid of punches. These holes could be filled with PTCBs of an adequate diameter. Then the filled agar recipient block was put into an ordinary mold and filled with liquid paraffin to create the PTMA whereby the agar block stabilized the PTCBs and prevented them from tumbling. After resolidification, the PTCBs were in perfect contact with the surrounding paraffin (Figure 9N). This system can be purchased from Sakura (Tokyo, Japan [87]; various websites for Japan, Europe and America) which sells the system under the brand of Tissue-Tek Quick-Ray, or from Unitma (Figure 9O, Quick Ray manual tissue microarrayer (full set: 3,500 US$/set, Seoul, South Korea [78]). A disadvantage of the Quick-Ray system may be the limitation to about 170 PTCBs per PTMA. But nonetheless, this is a powerful and simple technique to construct PTMAs especially within the aspect of an incorporated one step full melting procedure. In the meantime, an automated tissue arrayer which punches out the PTCBs of the recipient block and transfers them to the holes of the preformed recipient block is now sold by Unitma (Seoul, South Korea) for about 78,000–98,000 US$ depending on the version (Figure 9Q). Besides, this technique is well suited for a low cost construction of PTMAs because the agar stabilization bodies can be easily fabricated in every laboratory of pathology (Figure 9A–H) [88]. The holes of these agar blocks can be created by drilling, pouring (Figure 9D,E) or even punching with the Beecher tissue arrayer (Figure 9I). Especially with the drilling technique as many holes as needed and every diameter of the holes necessary may be provided if a low cost technique is favored. Figure 9 The Song technique—agar stabilization bodies for a one step fully melting procedure. (A) Boiling agar 2% like in the molecular biologic laboratory for gels or in the kitchen. (B) The liquid agar is poured into the lid of a pipette box (waste material). (C) The solidified agar, which can be simply released from the mold is cut into plates of desired dimension by a scalpel. (D) Pouring the holes into an agar plate: This top pin tissue arrayer is placed into liquid agar. (E) After solidification of the agar the pins are withdrawn from the agar by turning the screws (D). (F) The agar plates (with or without preformed holes) are paraffinized in a standard automatic tissue processor. (G) The agar plate (agar stabilization body) is poured into a paraffin block. (H) Before drilling the holes into such a paraffin block and/or before filling the holes of a cast agar plate (D,E) sectioning of the block is recommended until the agar plate is in contact with the block surface. (I) The holes of the stabilization body can also be punched, e.g., with a manual tissue arrayer. (J) Agar stabilization body cast into a paraffin block with the holes being filled with PTCBs. (K) Agar plates can also be used and filled with PTCBs as a stand alone and may be cast into a paraffin block after melting. Note, this agar plate (asterisk) is thin (see insert) and gives stabilization only for one PTCB per hole. The advantage of this thin plate is the better release of air bubbles at melting. (L) A thick agar plate nearly completely surrounds the PTCBs at the entire length (see also insert (upper right corner) with a thick stabilization body (asterisk)). The holes can be filled with more than one PTCB to ensure an equal length and to prevent the thinning of the PTCBs in deeper sections. PTCBs of different diameters can be installed into the holes (inserts with arrowheads). The holes of the agar plate can be also constructed by punching manually with some more or less precise arrangement of the cores (arrow). Different diameters of the holes are possible. (M) Gap (arrows) between a PTCB and the surrounding agar stabilization body (a) before melting. (N) Gap (arrows) filled with paraffin after melting (a agar stabilization body). (O) Quick Ray manual tissue microarrayer set (Unitma, Seoul, South Korea). (P) Agar stabilization bodies of 1 mm, 1.5 mm, 3 mm, and 5 mm (Unitma, Seoul, South Korea) (Q) Fully automated tissue arrayer of Unitma (Seoul, South Korea) constructing PTMAs with preformed stabilization bodies. 2.9. Ultrahigh Density Microarrays of Solid Samples by LeBaron et al. (2005) [35,36] In 2005, LeBaron published a technique, also called the cutting edge matrix assembly, to construct PTMAs with the highest number of specimens up to day, i.e., up to 10,000 different specimens per block [35,36]. Figure 10 Le Baron technique—PTMA made of thick sections. (A) Routine paraffin tissue block used as a donor block fixed to the clamp of a rotary microtome to cut sections 100 µm thick. (B) 100 µm thick sections of different donor blocks being stacked (primary stack) and glued together by gently warming (30–40 °C). (C) Cut surface of the trimmed primary stack of thick sections. (D) Cross sections of different primary stacks being arranged in a routine steel embedding mold and cast into a paraffin block. (E) Several paraffin blocks with cross sections of a lot of different primary stacks. (F) Secondary stack of 100 µm thick sections of the blocks displayed in E. (G) Trimmed secondary stack before being cast into a paraffin block. (H) Hematoxylin-Eosin stained section of the trimmed secondary stack. The correct arrangement of the sections of the primary and secondary stack does not seem to be so easy to perform to get a Cartesian grid. Furthermore, entrapped air bubbles might cause difficulties. To achieve such a high number of specimens LeBaron et al. avoided the punching technique, which can be applied only to a diameter of the PTCBs equal or greater than 0.3 mm due to the stability of the tissue punches (own unpublished experience). The tissue specimens were cut as plates by knife or microtome with a thickness of about 100 µm (Figure 10A). These tissue plates were melted (Figure 10B) or glued (superglue, i.e., methacrylat) together to receive a primary stack and sectioned again, until 3D plates as secondary stacks (Figure 10F) resulted. Of course, this technique was only useful for tissues with the cells of interest being homogeneously distributed. One spot of tissue in a section reached only 100 µm2. Although this technique is brilliant there may be sometimes some difficulties in gaining such 100 µm thick plates of tissue, e.g., by cracking (unpublished own experience). Furthermore the exact arrangement of the plates and the melting of the plates may not always be easy (Figure 10G,H; own unpublished experience). 2.10. Combined Techniques for PTMA Construction 2.10.1. Punching Technique/Drilling Technique/Pouring Technique Combined with the Double Sided Adhesive Tape Technique of Wilkens/Chen (Vogel, 2007) [47,89] The Kononen/Beecher technique using the manual or automated tissue arrayer may be the mostly used technology for constructing PTMAs word-wide. However, the already mentioned rolling and folding of PTCBs or the cracking of the PTMA at sectioning may be a problem if the paraffin tape transfer system is not used (Figure 11A–C). Of course, these problems also occur when the holes of the recipient block are drilled or poured. The rolling and folding could be prevented by combining the aforementioned techniques with a melting step, especially with the technique of Wilkens/Chen (Figure 11D–K) [27,39,47,89]. Generally three techniques may exist for melting a PTMA: a partial melting procedure (e.g., 18 min, 58 °C in an oven), a one step complete melting procedure and a two step melting procedure as introduced by Mengel [23]. The partial melting technique is a rapid and simple procedure, which may be used very often word-wide. However, extreme care has to be taken for PTMAs with PTCBs of less than 1 mm in diameter that the PTCBs do not tumble during the melting. The larger the diameter of the PTCBs the less the probability of tumbling. The PTCBs have to be stabilized by the still solid paraffin in the upper parts of the PTMA. In contrast to own experience the two-step melting procedure is said to be easy to perform, e.g., by using an in situ hybridization platform and a heating lamp. Firstly, the bottom half of the block is melted by the platform, secondly after cooling the upper half is melted by a heating lamp (see above). Probably the easiest way to get a fully melted PTMA is to cut a filled PTMA (Figure 11D) on a microtome to get a smooth surface, to fix the double sided adhesive tape-x-ray film-sandwich of Wilkens/Chen to the smooth surface of the PTMA (Figure 11E–G) and to melt this sandwich (Figure 11H,I). Independently of the diameter the PTCBs are fixed in place and in an upright position during the melting by the tape. There is no time limit which has to be strictly obeyed at melting like in the partial or two-step melting procedures. The only disadvantage of this kind of melting is that PTCBs may tumble down if more than one PTCB is installed in one hole of the PTMA or if the tissue of the PTCB has no contact to the tape. Figure 11 Combined techniques. (A) PTMA of an external TMA laboratory constructed with the Kononen technique. (B) Large gap between the PTCB and the surrounding paraffin of the recipient block. (C) Section of this PTMA with perfect (asterisk) and rolled PTCBs (arrows) if the paraffin tape transfer system of Instrumedics, Inc., is not used. (D) Predrilled paraffin recipient block filled with 100 PTCBs 0.43 mm in diameter. (E) The PTMA is fixed to the clamp of a rotary microtome and cut to get a smooth surface. A double sided adhesive tape with a brown protective sheet is attached to the surface of the PTMA to get into contact with the PTCBs. (F) The PTMA with the white double sided adhesive tape after removal of the protection sheet. (G) An x-ray film is attached to the double-sided adhesive tape to stabilize the tape at melting. (H) The PTMA-adhesive tape-x-ray film-sandwich is melted in a standard steel embedding mold (note: Do not heat over 65 °C, in order not to shrink the adhesive tape.). (I) Melted PTMA with the PTCBs standing upright and in position due to the adhesive tape. Of course, if a hole is filled with more than one PTCB, the PTCB without contact to the tape will topple down (arrow). (J) After resolidification the double-sided adhesive tape-x-ray-film-sandwich is removed from the surface of the PTMA demonstrating the strong adhesion of the PTCBs to the glue of the adhesive tape. Note the toppled down PTCB (arrow) (K) After resolidification the PTCB displays a strong contact to the surrounding paraffin. (L) Black x-ray film (x) with two strips of the white double sided adhesive tape (t) and an agar stabilization body (a) cast into a standard paraffin block and filled with some PTCBs. (M) After fixing the x-ray film-tape-sandwich to the surface of the agar stabilization body (paraffinized agar binds to the tape) this sandwich is melted in a standard steel embedding mold. The PTCBs are held in position by the stabilization body and not by the adhesive tape. This very small gap between the x-ray film and the stabilization body facilitates the flow of the liquid paraffin into the gaps between the PTCBs and the agar stabilization body. This small gap is ensured by the tape and the x-ray film. Without the x-ray film-tape-sandwich the short and/or small PTCBs may fall out of the holes of the stabilization body at melting. (N) After resolidification the gap (arrow) between the PTCBs and the agar stabilization body (a) is perfectly filled with paraffin; this secures a very low number of rolled PTCBs at sectioning. (O) Predrilled agar stabilization body cast into a paraffin block with a bottomless plastic cassette (Tissue-Tek Paraform Sectionable Cassette System, Sakura, Tokyo, Japan) (Look from above). The surface of the agar stabilization body is fixed to a x-ray film-double sided adhesive tape-sandwich. (P) A paraffinized breast needle biopsy specimen (PNBS), which was punched out of the donor block, melted to remove the adhering paraffin surplus and resolidifed at the tip of a small needle. This PNBS can now be installed into the hole of a stabilization body. (Q) Agar stabilization body filled with PNBSs after melting, resolidification and removal of the x-ray film-tape-sandwich. (R) Higher magnification of the surface of the PTMA (Q) demonstrates a perfect contact between the PNBSs and the agar stabilization body. (S) A Hematoxlin-Eosin stained section of the PTMA filled with PNBSs. This disadvantage may be cured by using an agar stabilization body as recipient block as described by Song. (Figure 11L–N) [42]. 2.10.2. Stabilization Body Technique Combined with the Wilkens/Chen Double Sided Adhesive Tape Technique (Vogel) [27,39] Such a combination of the techniques is especially useful when constructing PTMAs by using paraffinized needle biopsy specimens (PNBSs) (Figure 11O–S). Such PNBSs can be put into punched or drilled holes of an agar stabilization body, which is mounted on a double-sided adhesive tape. After filling the holes, the stabilization body (recipient block) can be melted in a one step procedure without tumbling of the upright PNBSs (Figure 11Q–S). The double-sided adhesive tape-x-ray film-sandwich creates a reversible bottom to the stabilization body and, therefore, prevents the PNBSs to fall out of the holes during filling or melting of the PTMA. This technique may be easier and faster than the checker technique as described by McCarthy et al. [55,60]. 2.10.3. Rod Technique of Miller Combined with the Punched Preformed Recipient Block of Kononen (Shi et al., 2013) [64] The paraffin tissue core biopsies (PTCBs) as described by Wan and Kononen are punched vertical to the surface of the donor paraffin block [6,15]. The disadvantages of these PTCBs are the uneven length as discussed above and the unknown tissue composition in the depth. These obstacles can be overcome by punching or dissecting parallel and not perpendicular to the surface of the donor block. These tissue rods, as they were called by Shi et al., apparently in remembrance of the Miller technique have a definite equal length and the tissue of interest at least near the former surface in all sections to be cut [64]. These tissue rods were prepared using a unique sampling tool for which a Chinese patent exists. The rods were installed in a prewarmed softened paraffin recipient block with holes being drilled with a steel needle and arranged in a Cartesian coordinate system. 2.10.4. The Microarray Group—A Combination of the Tissue Macroarray and the Tissue Microarray (Jiang et al. 2007) [50] In 2007, Jiang et al. presented a technique, which they called the microarray group [50]. They arranged sections of different small PTMAs on one slide thereby greatly enhancing the TMA effectiveness. Up to 2534 PTCBs 0.6 mm in diameter could be examined on one standard glass slide. This technique was also presented by Vogel arguing that smaller PTMAs may also increase the flexibility of the TMA technique. Probably, it would be more reasonable to construct smaller PTMAs, e.g., with subsets of tumors instead of creating one large tumor PTMA [90]. 3. How to Select the Appropriate Technique for Constructing TAs? The selection of the appropriate technique for constructing TAs depends on several factors, which may also interfere with each other (Table 2). microarrays-03-00103-t002_Table 2Table 2 Factors influencing the choice of the tissue array (TA) technique. Factors influencing the choice of the tissue array (TA) technique Intent of the TAs (e.g., as positive control for routine immunohistochemistry, for translational research) Physical property of the tissues/embedding medium (frozen material for frozen TMAs, formalin fixed paraffin embedded tissue for PTMAs, paraformaldehyde fixed material for resin TMAs) Number of TCBs to be installed in the TMAs Dimension of tissues to be evaluated (e.g., needle biopsy specimens, resection specimens, cell blocks) Frequency of the construction of TAs Money to be spent The most important question which has to be answered at first, belongs to the intended use of the TA. It is the purpose, which determines the choice of the material (fresh frozen, formalin fixed paraffin embedded, paraformaldehyde), the number of the TCBs to be installed in a TMA, and the need for a TMA. A compilation of the techniques may be found in Table 3. The simplest form of a TA is the tissue macroarray, i.e., the array only on the slide. For most applications, however, the construction of a TMA is needed. If a TMA should serve as a positive control in immunohistochemistry with less than about 30 specimens per TMA, the construction of PTMAs is advised according to several well functioning zero or low cost techniques (e.g., Miller (tissue rods), Wilkens/Chen (tissue cores)), which don’t need a prefabricated recipient block [10,27,39]. microarrays-03-00103-t003_Table 3Table 3 Selection of the appropriate paraffin TA technique. Specifications Appropriate Techniques Few slides necessary Few specimens (<10) Tissue macroarray (the array on the slide) [1] Many slides necessary Few specimens (20–30) No precise arrangement necessary No preformed recipient block Miller: tissue microarray (TMA): tissue rods [10] Wilkens/Chen technique: TMA: double sided adhesive tape [27,39] Melting step inherent to the technique Many slides necessary Few specimens (30–100) Precise arrangement necessary Preformed recipient block more comfortable Manual transfer of tissue cores Poured, punched or drilled paraffin or paraffinized agar TMAs: commercially available tissue arrayers or ready to use recipient blocks (Mengel, Song, Szekeres, Vogel) [23,42,52,79] Additional melting step with the adhesive tape-x-ray film-sandwich advised [89] Many slides necessary Many specimens (>100) Precise arrangement necessary Preformed recipient block advised Construction of only a few TMAs Commercial services to construct the TMA Cooperation with a TMA facility at university Many slides necessary Many specimens (>100) Precise arrangement necessary Preformed recipient block advised Semiautomatic/automatic transfer of tissue cores Construction of many TMAs Commercially available tissue arrayers: e.g., Beecher Instruments Inc., 3DHistech, Veridiam, Unitma, (see above) Additional melting step with the adhesive tape-x-ray film-sandwich advised if not inherent to the technique itself [89] Although the Wilkens/Chen technique can be used to construct PTMAs with more than 300 PTCBs, it seems more comfortable especially in the setting of translational research to use some preformed poured, punched or drilled recipient blocks when more than 30–50 PTCBs should be installed. These recipient blocks may be pure paraffin blocks or paraffinized agar blocks (stabilization bodies), which are commercially available or can be fabricated by oneself, e.g., with a mold and spacers made of steel or rubber (e.g., Mengel, Song, Szekeres, Vogel) [23,42,52,79]. By applying these systems the PTCBs have to be transferred manually. For more than 100 PTCBs per PTMA the decision should be made whether the PTCBs should be still transferred manually, semi-manually/semi-automatically with the help of a machine or even automatically. This choice of the appropriate device is of course also dependent on the numbers of PTMAs, which should be constructed in the course of time. If only a few PTMAs are intended for construction, it may be inefficient to invest in an expensive automatic tissue arrayer. Commercial services for construction of PTMAs may also be a choice in this scenario. Furthermore, assistance may be provided by colleagues, e.g., of tissue microarray units of university institutes in terms of collaboration. Only if multiple PTMAs with a high number of PTCBs are to be constructed the acquisition of a semi-automatic or automatic tissue arrayer may be favorable and cost-effective. Several suppliers may fulfill the needs for such a device (e.g., Beecher Instruments Inc., 3DHistech, Veridiam, Unitma, see above). Probably the most used machine may be the manual tissue arrayer MTA-1 of Beecher Instruments Inc. for about 10,000 Euros. Of course, possible devices should fit for the embedding medium of choice. Whereas the expensive commercially available tissue arrayers are mostly restricted for paraffin, low cost devices may be also used for frozen TMAs. Furthermore, the equipment of choice should also provide the appropriate diameter of the PTCBs. Depending on the tissue and the scientific question to be answered the diameter of the PTCBs needed may vary from 0.3 mm (e.g., for homogeneous, densely packed tumors, such as endocrine tumors) to 5 mm (e.g., for skin or mucosal preparations where the epithelial-mesenchymal interface is of interest). 4. Low Cost Techniques to Construct TAs of High Quality In this section some low cost techniques may be recommended for those who do not have access to a tissue microarray facility, e.g., at universities, who want to obviate the waiting time at tissue microarray facilities and who do not want to spend money for commercial services, but who want, nonetheless, TAs of high quality. As already discussed at point 3, the technique used may be chosen according to the intent of the TA. If a tissue macroarray, i.e., the array on a slide, is not applicable, a TMA must be constructed. For frozen TMAs the use of a frozen agar stabilization body is strongly recommended with holes being poured or due to low costs preferably drilled [49,79]. Even when the frozen TMA may warm up unexpectedly, the TMA keeps the structure due to the agar. The technique of Tsao et al. using gelatin-based capsules may also be an alternative for low cost frozen TMA construction [57]. microarrays-03-00103-t004_Table 4Table 4 Low cost techniques for the construction of tissue arrays. Specifications Appropriate techniques Few slides necessary Few specimens (<10) Tissue macroarray (the array on the slide) [1] Many slides necessary Few specimens (20–30) No precise arrangement necessary Miller: tissue microarray (TMA): tissue rods [10] Many slides necessary Few specimens (30–50) Precise arrangement necessary Wilkens/Chen technique: TMA: double sided adhesive tape [27,39] Many slides necessary Many specimens (50–300) Precise arrangement necessary Vogel manual drilling: TMA: Preformed recipient blocks [79] Stabilization body advised [88] Melting with the adhesive tape-x-ray film-sandwich [89] Many slides necessary Multiple specimens (>300) Precise arrangement necessary Vogel manual drilling or CNC drilling: TMA: Preformed recipient blocks [79,81] Stabilization body advised [88] Melting with the adhesive tape-x-ray film-sandwich [89] A prerequisite for a high quality paraffin TMA (PTMA) is a complete melting step in the construction process. Only a complete melting step ensures a firm contact between the PTCBs and the surrounding paraffin of the PTMA which is necessary to avoid rolling and folding of the PTCBs at sectioning as described above (Section 2.5). By this step the paraffin tape transfer system mostly becomes needless. Possible low cost techniques are summarized in Table 4. A melting step is always included in the techniques working with tissue rods (e.g., Miller) or with PTCBs without the use of a prefabricated recipient block (e.g., Wilkens/Chen: double sided adhesive tape) [10,27,39]. If PTMAs made of tissue rods or PTMAs with a small number of PTCBs may fulfill the task, these techniques are advised with costs being negligible. If PTMAs with more than 50 PTCBs are needed, a prefabricated recipient block in form of a stabilization body preferably made of agar is favorable to perform a one step complete melting procedure. The holes of these stabilization bodies can be poured before or preferably drilled or punched after the paraffinization process. The most appropriate method to create the holes may be the drilling technique because of the enormous flexibility and the exact arrangement of the holes. By drilling, different diameters of the holes due to the huge amount of commercially available drill bits, different distances of the holes and multiple arrangements of the holes are feasible. Care should be taken to drill through the whole thickness of the stabilization body that the air bubbles can leave the holes during the melting process. Furthermore, the hardware for drilling with costs of less than 300 € is much cheaper than any commercially available pouring system or prefabricated stabilization body. The disadvantage of the pouring systems or the prefabricated stabilization bodies may be the often low number of holes defined by the number of pins of the arrayer. If more than 300 holes/PTMA are needed, the use of a CNC drilling machine is advised although manual drilling may also achieve this task. CNC drilling machines can often be found in the department of fine mechanics of a larger hospital or any neighboring metal-working company. Stabilization bodies may also provide the possibility to use more than one short PTCB in one hole in combination with the melting process. The construction of PTMAs with paraffin needle biopsy specimens is also facilitated by using stabilization bodies because the biopsies are held in an upright position during the melting process. If a stabilization body is not intended to be used the melting can also be done by fixing the PTCBs of a predrilled PTMA to a double sided adhesive tape as described above. However, in such case, the holes of the PTMA can be filled with only one PTCB. To overcome the problem of unevenly long PTCBs the use of a cutting board is advised (Figure 12A–F) [91]. Such a cutting board consists of a plate of polymethylmethacrylate (plexiglas®) for example with a height of 4–5 mm and with at least one hole with a diameter of the correspondent PTCBs (Figure 12A,B). The plate is reversibly fixed or pressed to a firm base. The hole of the plate is filled with as much PTCBs as needed. The protruding part of the last PTCBs can be cut and inserted into another hole of the plate or stored in a vial for further use to spare the often precious tissue. With a stylet the PTCB composite is released (Figure 12C) and injected into the hole of the PTMA (Figure 12F). The cutting board can also be used to store PTCBs for further deployment into PTMAs in the future with the advantage of an already homogeneous length of the PTCBs. The installment of the PTCBs into the cutting board can be also accomplished with a Beecher tissue arrayer (Figure 12K). A very good idea to overcome this problem of unevenly long PTCBs, seems to be the rod technique as proposed by Shi [64], which is similar to the wedge shaped tissue rod technique as described by Sundblad [13]. If the tissue is punched parallel to the donor block surface the length of the rods and the tissue composition is definitely determined. Of interest may be the sampling tool to perform this horizontal tissue acquisition. Figure 12 Vogel technique to construct evenly long PTCBs. (A) A so-called cutting board is a 4 mm thick plate made of e.g., polymethylmethacrylate (plexiglas®) with numerous holes to accommodate PTCBs. The PTCBs can be installed into the plate with a paraffin tissue punch. (B) The holes of the cutting board can be filled with as many PTCBs as necessary to achieve a length of 4 mm. The part of the PTCB which surmounts the surface of the cutting board can be easily cut with a used microtome knife and inserted into another hole to spare the sometimes precious tissue. (C) The composite PTCB (arrows) can be retrieved from the cutting board with the stylet of the paraffin tissue punch. (D) The PTCBs punched out of the donor paraffin blocks are of different length (bottom). After the installment into the cutting board the composite PTCBs are of equals length (top). (E) A 4 mm long composite PTCB composed of several very short PTCBs. (F) The composite PTCBs (arrow) can be transferred into the holes of the recipient blocks separately. (G) A tissue arrayer can also be used to transfer all of the composite cores (arrow) of the cutting board (c) into the holes of the recipient block simultaneously (p plastic cassette of the recipient block. (H) A PTMA filled with composite PTCBs by the tissue arrayer. (I) PTCB library in an Eppendorf vial as already proposed by Wan et al. (J) A commercially available paraffin tissue punch (Beecher Instruments, Inc.) was modified with a countersink to facilitate the installment of stored PTCBs into a punch. (K) The cutting board can also be used in combination with a tissue arrayer (e.g., the Beecher tissue arrayer). In contrast to tissue rods, which can be trimmed with an ordinary knife, PTCBs need tissue punches. The most cost-effective method is to produce tissue punches out of routinely used hypodermic needles by oneself. As described by Wan et al. the tip of the needle is cut with a cutting disk and the new end resharpened with the use of a grinding device [6]. As stylet a smaller hypodermic needle may be used. Due to the widespread use of needles in medicine a large variety of different diameters exist. Care has to be taken to remove the grinding dust before using the needle as a punch. In addition to the low costs these punches are normally very stable because of the relative high wall thickness in comparison to commercially available punches. A special tissue punch with a lateral opening 1 mm away from the tip was manufactured out of conventional hypodermic needles (16 and 18 gauge) by Pires et al. [45]. Commercially available skin biopsy punches with or without a plunger system may also be used as tissue punches (e.g., kai biopsy punches) (Figure 8C). If a skin biopsy punch without a plunger system is applied, a stylet (e.g., of a bone marrow biopsy needle) has to be used to release the PTCBs from the punch. To get the stylet through the punch the transition of the metal punch and the plastic handle has to be widened by an appropriate drill bit. This drilling can be done manually. A minor disadvantage of the skin biopsy punches is the vulnerability of the very sharp tip. The tissue punches may also be purchased from the manufacturers of the tissue arrayers (e.g., Beecher Instruments Inc.) and then used manually. Damaged tissue punches may be obtained from tissue microarray facilities for free and can be resharpened for further manual use. Other construction methods for tissue punches were proposed by Shebl et al., Foda, and Garcia-Garcia et al. (e.g., mechanical pencil tips) [59,65,66]. When working with PTCB libraries, i.e., the storage of PTCBs in vials (Figure 12I), a modification of the paraffin tissue punches with a countersink is helpful (Figure 12J) [92]. This countersink can be fabricated with the same drill grinder, which is used for drilling, and a larger cone-shaped precision cutter (e.g., 3–4 mm in diameter; costs: less than 5 €). 5. Conclusion Since 1965, several efficient techniques to construct TAs have been developed to meet the needs of the user. Especially the intended use of the TAs (e.g., positive controls for immunohistochemistry or translational research) may decide what kind of technique (e.g., tissue rods or PTCBs) or whether a low cost technique or a high priced automated system is chosen. Even with low cost techniques high density and precise TMAs can be constructed. It is a pleasure to see what different kind of techniques colleagues all over the word invented to fulfill only one task: To get tissue core biopsies into holes/blocks. Conflict of Interest The author declares no conflict of interest. ==== Refs References 1. Wang L. Deavers M.T. Malpica A. Silva E.G. Liu J. Tissue macroarray: A simple and cost-effective method for high-throughput studies Appl. Immun. Mol. Morph. 2003 11 174 176 10.1097/00129039-200306000-00015 2. Deng F.M. Zhao Y. Kong X. Lee P. Melamed J. Construction of tissue microarrays using pre-existing slides as source of tissue when paraffin blocks are unavailable J. Clin. Pathol. 2013 66 627 629 10.1136/jclinpath-2012-201171 23476078 3. Lilie R.D. Histopathologic Technic and Practical Histochemistry 3rd ed. McGraw-Hill Book Co. New York, NY, USA 1965 4. Battifora H. The multitumor (sausage) tissue block: Novel method for immunohistochemical antibody testing Lab. Invest. 1986 55 244 248 3525985 5. Kajdacsy-Balla A. Geynisman J.M. Macias V. Setty S. Nanaji N.M. Berman J.J. Dobbin K. Melamed J. Kong X. Bosland M. Practical aspects of planning, building, and interpreting tissue microarrays: The cooperative prostate cancer tissue resource experience J. Mol. Histol. 2007 38 113 121 10.1007/s10735-006-9054-5 17318343 6. Wan W.H. Fortuna M.B. Furmanski P. A rapid and efficient method for testing immunohistochemical reactivity of monoclonal antibodies against multiple tissue samples simultaneously J. Immunol. Methods 1987 103 121 129 10.1016/0022-1759(87)90249-3 3655378 7. Kraaz W. Risberg B. Hussein A. Multiblock: An aid in diagnostic immunohistochemistry J. Clin. Pathol. 1988 41 1337 1339 10.1136/jcp.41.12.1337-a 3225337 8. Rowden G. Fraser R.B. Preparation of “histocomposites” for direct immunohistochemical screening of monoclonal antibodies Stain. Technol. 1988 63 49 52 2451326 9. Battifora H. Mehta P. The checkerboard tissue block. An improved multitissue control block Lab. Invest. 1990 63 722 724 2232717 10. Miller R.T. Groothuis C.L. Multitumor “sausage” blocks in immunohistochemistry. Simplified method of preparation, practical uses, and roles in quality assurance Am. J. Clin. Pathol. 1991 96 228 232 1862778 11. Press M.F. Hung G. Godolphin W. Slamon D.J. Sensitivity of HER-2/neu antibodies in archival tissue samples: Potential source of error in immunohistochemical studies of oncogene expression Canc. Res. 1994 54 2771 2777 12. Rose D.S. Maddox P.H. Brown D.C. Multiblock slides: A useful technique for teaching J. Clin. Pathol. 1994 47 88 89 10.1136/jcp.47.1.88 7510723 13. Sundblad A.S. A simplified multitissue block Am. J. Clin. Pathol. 1994 102 192 193 8042588 14. Petrosyan K. Press M.F. Multispecimen tissue blocks in pathology: An improved technique of preparation Lab. Invest. 1997 77 541 542 9389797 15. Kononen J. Bubendorf L. Kallioniemi A. Bärlund M. Schraml P. Leighton S. Torhorst J. Mihatsch M.J. Sauter G. Kallioniemi O.P. Tissue microarrays for high-throughput molecular profiling of tumor specimens Nat. Med. 1998 4 844 847 10.1038/nm0798-844 9662379 16. Leighton S.B. Instrument for Constructing Tissue Arrays U.S. Patent 6,103,518 2000 17. Gilett C. Springall R. Barnes D. Hanby A.M. Multiple tissue core arrays in histopathology research: A validation study J. Pathol. 2000 192 549 553 10.1002/1096-9896(2000)9999:9999<::AID-PATH721>3.0.CO;2-0 18. Chan J.K. Wong C.S. Ku W.T. Kwan M.Y. Reflections on the use of controls in immunohistochemistry and proposal for application of a multitissue spring-roll control block Ann. Diagn. Pathol. 2000 4 329 336 10.1053/adpa.2000.17892 11073340 19. Hoos A. Cordon-Cardo C. Tissue microarray profiling of cancer specimens and cell lines: Opportunities and limitations Lab. Invest. 2001 81 1331 1338 10.1038/labinvest.3780347 11598146 20. Schoenberg-Fejzo M. Slamon D.J. Frozen tumor tissue microarray technology for analysis of tumor RNA, DNA and proteins Am. J. Pathol. 2001 159 1645 1650 10.1016/S0002-9440(10)63011-8 11696425 21. Packeisen J. Bürger H. Krech R. Boecker W. Tissue microarrays—A new approach for quality control in immunohistochemistry J. Clin. Pathol. 2002 55 613 615 10.1136/jcp.55.8.613 12147657 22. Badve S. Deshpande C. Hua Z. Lögdberg L. Expression of invariant chain (CD 74) and major histocompatibility complex (MHC) class II antigens in the human fetus J. Histochem. Cytochem. 2002 50 473 482 10.1177/002215540205000404 11897800 23. Mengel M. Kreipe H. von Wasielewski R. Rapid and large-scale transition of new tumor biomarkers to clinical biopsy material by innovative tissue microarray systems Appl. Immunohistochem. Mol. Morph. 2003 11 261 268 24. Vogel U. Bültmann B. A simple and cheap method to produce home-made high density multi-tissue arrays Lab. Invest. 2003 83 Suppl. 1 328A 25. Hidalgo A. Pina P. Guerrero G. Lazos M. Salcedo M. A simple method for the construction of small format tissue arrays J. Clin. Pathol. 2003 56 144 146 10.1136/jcp.56.2.144 12560397 26. Matysiak B.E. Brodzeller T. Buck S. French A. Counts C. Boorsma B. Datta M.W. Kajdacsy-Balla A.A. Simple, inexpensive method for automating tissue microarray production provides enhanced microarray reproducibility Appl. Immunohistochem. Mol. Morph. 2003 11 269 273 27. Wilkens L. Verfahren und Vorrichtung zur Präparation von Gewebeproben German Patent, DE 102 03 524 A1 2003 28. Schnetz G. Redl H. Zatloukal K. Method and Device for Manipulating Samples International Patent Application WO 2004/040264 A1 2004 29. Vogel U.F. Bueltmann B.D. Low cost tissue microarrays (TMA) constructed with a common microcompound table Mod. Pathol. 2004 17 Suppl. 1 363A 30. Vogel U.F. Bültmann B.D. Low cost tissue microarrays (TMA) using computer numerical control (CNC) pre-drilled recipient paraffin blocks Path. Res. Pract. 2004 200 254 255 10.1016/S0344-0338(04)80404-7 31. Dan H.L. Zhang Y.L. Zhang Y. Wang Y.D. Lai Z.S. Yang Y.J. Cui H.H. Jian Y.T. Geng J. Ding Y.Q. Guo C.H. Zhou D.Y. A novel method for preparation of tissue microarray World J. Gastroenterol. 2004 10 579 582 14966920 32. Pan C.C. Chen P.C. Chiang H. An easy method for manual construction of high-density tissue arrays Appl. Immunohistochem. Mol. Morphol. 2004 12 370 372 10.1097/00129039-200412000-00015 15536340 33. Vogel U.F. Bültmann B.D. Tissue microarrays with paraffin tissue core biopsies of 0.43 mm in diameter are technically feasible Modern Pathol. 2005 18 Suppl. 1 336A 34. Howat W.J. Warford A. Mitchell J.N. Clarke K.F. Conquer J.S. McCafferty J. Resin tissue microarrays: A universal format for immunohistochemistry J. Histochem. Cytochem. 2005 53 1189 1197 10.1369/jhc.5C6659.2005 15983117 35. LeBaron M.J. Crismon H.R. Utama F.E. Neilson L.M. Sultan A.S. Johnson K.J. Andersson E.C. Rui H. Ultrahigh density microarrays of solid samples Nat. Meth. 2005 2 511 513 10.1038/nmeth772 36. Tran T.H. Lin J. Sjolund A.B. Utama F.E. Rui H. Protocol for constructing tissue arrays by cutting edge matrix assembly Methods Mol. Biol. 2010 664 45 52 10.1007/978-1-60761-806-5_5 20690051 37. Montgomery K. Zhao S. van de Rijn M. Natkunam Y. A novel method for making “tissue” microarrays from small numbers of suspension cells Appl. Immunohistochem. Mol. Morphol. 2005 13 80 84 10.1097/00129039-200503000-00013 15722798 38. Datta M.W. Kahler A. Macias V. Brodzeller T. Kajdacsy-Balla A. A simple inexpensive method for the production of tissue microarrays from needle biopsy specimens: Examples with prostate cancer Appl. Immunohistochem. Mol. Morphol. 2005 13 96 103 10.1097/00129039-200503000-00016 15722801 39. Chen N. Zhou Q. Constructing tissue microarrays without prefabricating recipient blocks A novel approach. Am. J. Clin. Pathol. 2005 124 103 107 10.1309/LHCJRFBUH8Q6QD3N 15923162 40. Meng P.Q. Hou G. Zhou G.Y. Peng J.P. Dong Q. Zheng S. Application of new tissue microarrayer—ZM-1 without recipient paraffin block J. Zhejiang Univ. SCI. 2005 6B 853 858 10.1631/jzus.2005.B0853 41. Mengel M. Hebel K. Kreipe H. von Wasielewski R. Standardized on-slide control for quality assurance in the immunohistochemical assessment of therapeutic target molecules in breast cancer Breast J. 2005 11 34 40 10.1111/j.1075-122X.2005.21445.x 15647076 42. Song Y.-M. Jeong H.-J. Recipient Block and Method for Preparation Thereof U.S. Patent 7,070,950 2006 43. Vogel U.F. Bültmann B.D. Fully melting of tissue microarrays using a stabilization body Lab. Invest. 2006 86 Suppl. 1 336A 337A 44. Vogel U.F. Bültmann B.D. Depositing archived paraffin tissue core biopsies in paraffin tissue microarrays by using a paraffin tissue punch with a countersunk Lab. Invest. 2006 86 Suppl. 1 337A 10.1038/sj.labinvest.3700625 45. Pires A.R. Andreiuolo F.M. de Souza S.R. TMA for all: A new method for the construction of tissue microarrays without recipient paraffin block using custom-built needles Diagn. Pathol. 2006 1 10.1186/1746-1596-1-14 46. Wang S.L. Yang C.H. Chen H.H. Chai C.Y. A simple and economical method for the manual construction of well-aligned tissue arrays Pathol. Res. Pract. 2006 202 485 486 10.1016/j.prp.2006.01.014 16563652 47. Vogel U.F. Bueltmann B.D. Combining different techniques to construct paraffin tissue microarrays of superior quality Lab. Invest. 2007 87 Suppl. 1 1641 48. Vogel U. Bültmann B. Curing cracked or broken paraffin tissue microarrays by using a soldering iron equipped with a copper wire 0.2 mm in diameter or an adhesive tape Pathol. Res. Pract. 2007 203 402 403 49. Zhou L. Hodeib M. Abad J.D. Mendoza L. Kore A.R. Hu Z. New tissue microarray technology for analyses of gene expression in frozen pathological samples BioTechniques 2007 43 101 105 10.2144/000112498 17695259 50. Jiang H.Y. Zhang X.F. Liu L. Li H.L. Zhao T. A novel tissue array technique for high-throughput tissue microarray analysis—Microarray groups In Vitro Cell Dev. Biol. Anim. 2007 43 109 112 10.1007/s11626-007-9019-3 17514512 51. Vogel U.F. Bueltmann B.D. Paraffin tissue microarrays constructed with a cutting board, a board of steel spacers and a soldering iron Modern Pathol. 2008 21 Suppl. 1 373A 52. Szekeres G. Halas Z. Zorn L. Tissue Micro-Array Builder Manual Test U.S. Patent 7,326,560 2008 53. Vogel U.F. A simple and flexible device to pour the holes in paraffin tissue microarrays Mod. Pathol. 2009 22 Suppl. 1 389A 54. Vogel U.F. The production of cast recipient blocks for paraffin tissue microarrays using conventional steel embedding moulds Histopathology 2009 55 470 472 10.1111/j.1365-2559.2009.03384.x 19817900 55. McCarthy F. Fletcher A. Dennis N. Cummings C. O’Donnell H. Clark J. Flohr P. Vergis R. Jhavar S. Parker C. Cooper C.S. An improved method for constructing tissue microarrays from prostate needle biopsy specimens J. Clin. Pathol. 2009 62 694 698 10.1136/jcp.2009.065201 19638540 56. Vogel U.F. Bültmann B. Application of a novel and low cost technique to construct paraffin tissue microarrays out of paraffinized needle biopsy specimens from patients with breast cancer J. Clin. Pathol. 2010 63 640 643 10.1136/jcp.2010.076356 20591915 57. Tsao S.C. Wu C.C. Wen C.H. Huang Y.C. Chai C.Y. A simple and economical method for the manual construction of frozen tissue arrays APMIS 2010 118 739 743 10.1111/j.1600-0463.2010.02652.x 20854467 58. Fridman E. Daya D. Srigley J. Whelan K.F. Lu J.P. Pinthus J.H. Construction of tissue micro array from prostate needle biopsies using the vertical clustering re-arrangement technique Prostate 2011 71 1374 1381 10.1002/pros.21352 21308718 59. Shebl A.M. Zalata K.R. Amin M.M. El-Hawary A.K. An inexpensive method of small paraffin tissue microarrays using mechanical pencil tips Diagn. Pathol. 2011 6 10.1186/1746-1596-6-117 60. McCarthy F. Dennis N. Flohr P. Jhavar S. Parker C. Cooper C.S. High-density tissue microarrays from prostate needle biopsies J. Clin. Pathol. 2011 64 88 90 10.1136/jcp.2010.082339 21081515 61. Pilla D. Bosisio F.M. Marotta R. Faggi S. Forlani P. Falavigna M. Biunno I. Martella E. De Blasio P. Borghesi S. Cattoretti G. Tissue microarray design and construction for scientific, industrial and diagnostic use J. Pathol. Inform. 2012 3 10.4103/2153-3539.104904 62. Yang J. Zhang M. Su B. Chen X. Kang A. A novel tissue microarray instrumentation: The HT-1 tissue microarrayer Indian J. Pathol. Microbiol. 2012 55 314 318 10.4103/0377-4929.101736 23032823 63. Choi C.H. Kim K.H. Song J.Y. Choi S.J. Kim L. Park I.S. Han J.Y. Kim J.M. Chu Y.C. Construction of high-density tissue microarrays at low cost by using self-made manual microarray kits and recipient paraffin blocks Korean J. Pathol. 2012 46 562 568 10.4132/KoreanJPathol.2012.46.6.562 23323107 64. Shi Y. He D. Hou Y. Hu Q. Xu C. Liu Y. Jiang D. Su J. Zeng H. Tan Y. An alternative high output tissue microarray technique Diagn. Pathol. 2013 8 10.1186/1746-1596-8-9 65. Foda A.A. No-cost manual method for preparation of tissue microarrays having high quality comparable to semiautomated methods Appl. Immunohistochem. Mol. Morphol. 2013 21 271 274 23235346 66. García-García R. Rodríguez-Vidales E.P. Adolfo Soto-Domínguez A. Quick and inexpensive method to elaborate tissue punches useful in paraffin tissue microarrays Int. J. Morphol. 2013 31 50 54 10.4067/S0717-95022013000100007 67. Zanini C. Forni M. The cell block technique revisited for cells cultured in adherence and as “spheres” Histochem. Cell. Biol. 2013 140 685 690 10.1007/s00418-013-1139-0 24013649 68. Packeisen J. Korsching E. Herbst H. Boecker W. Buerger H. Demystified ... tissue microarray technology Mol. Pathol. 2003 56 198 204 10.1136/mp.56.4.198 12890740 69. Catchpoole D. Mackie N. McIver S. Chetcuti A. Henwood A. Graf N. Arbuckle S. Tape transfer sectioning of tissue microarrays introduces nonspecific immunohistochemical staining artifacts Biotech. Histochem. 2011 86 421 428 10.3109/10520295.2010.527859 21091080 70. Pinder S.E. Brown J.P. Gilett C. Purdie C.A. Speirs V. Thompson A.M. Shaaban A.M. The manufacture and assessment of tissue microarrays: Suggestions and criteria for analysis, with breast cancer as an example J. Clin. Pathol. 2013 66 169 177 10.1136/jclinpath-2012-201091 23087330 71. Beecher Instruments, Inc Available online:http://www.beecherinstruments.com/ (accessed on 9 January 2014) 72. Estigen Available online:http://www.estigen.com/ (accessed on 9 January 2014) 73. Veridiam Tissue Arrayer Available online:http://www.veridiamtissuearrayer.com/ (accessed on 9 January 2014) 74. Integrated Systems Engineering S.R.L Available online:http://www.isenet.it/ (accessed on 9 January 2014) 75. Lilischkis R. von Wasielewski R. Mengel M. Method for Production of Material Blocks with Multiple Test Samples International Patent Number PCT/DE00/04647 2001 76. Fisher Scientific Available online:http://www.fishersci.com/ (accessed on 9 January 2014) 77. 3DHISTECH Ltd Available online:http://www.3dhistech.com/ (accessed on 9 January 2014) 78. Unitma Available online:http://www.unitma.com/ (accessed on 9 January 2014) 79. Vogel U.F. Bueltmann B. Simple, inexpensive, and precise paraffin tissue microarrays constructed with a conventional microcompound table and a drill grinder Am. J. Clin. Pathol. 2006 126 342 348 10.1309/F2Q38DXN1V1V4GQM 16880136 80. Vogel U.F. The construction of high density tissue microarrays with up to 2500 paraffin tissue core biopsies 0.3 mm in diameter is technically feasible Pathologe 2009 30 Suppl. 1 80 81. Vogel U.F. Inexpensive and precise paraffin tissue microarrays constructed with a computer numerical control (CNC) drilling machine Histopathology 2007 51 136 137 10.1111/j.1365-2559.2007.02713.x 17532771 82. Vogel U.F. Bode J. Bültmann B. Increasing the efficiency of paraffin tissue microarrays by packing the paraffin tissue core biopsies in a honeycomb pattern Appl. Immunohistochem. Mol. Morphol. 2007 15 343 345 10.1097/01.pai.0000213140.47277.f6 17721282 83. Mirlacher M. Simon R. Recipient block TMA technique Methods Mol. Biol. 2010 664 37 44 10.1007/978-1-60761-806-5_4 20690050 84. Lilischkis R. Mengel M. Wasilewski R. Verfahren zur Herstellung von Materialblöcken mit multiplen Untersuchungsproben German Patent Application DE 100 01 136 A1 2001 85. Vogel U. Stablilisationskörper für Gewebemikroarrays German Patent Application DE 10 2005 028 833.2–52 2005 86. Yan P. Seelentag W. Bachmann A. Bosman F.T. An agarose matrix facilitates sectioning of tissue microarray blocks J. Histochem. Cytochem. 2007 55 21 24 16899763 87. Sakura Finetek Japan Available online:http://www.sakura-finetek.com/ (accessed on 9 January 2014) 88. Vogel U.F. One-step complete melting of paraffin tissue microarrays using stabilization bodies Appl. Immunohistochem. Mol. Morphol. 2008 16 382 388 10.1097/PAI.0b013e318158ec68 18528278 89. Vogel U.F. Combining different techniques to construct paraffin tissue microarrays of superior quality Histopathology 2009 54 624 626 10.1111/j.1365-2559.2009.03249.x 19413642 90. Vogel U.F. The construction of high-density paraffin tissue microarrays with 0.43-mm-diameter paraffin tissue core biopsies is technically feasible Virchows Arch. 2008 453 43 46 10.1007/s00428-008-0622-9 18551310 91. Vogel U.F. Paraffin tissue microarrays constructed with a cutting board and cutting board arrayer Appl. Immunohistochem. Mol. Morphol. 2010 18 283 287 10.1097/PAI.0b013e3181c8092b 20048672 92. Vogel U.F. Depositing archived paraffin tissue core biopsy specimens in paraffin tissue microarrays using a paraffin tissue punch modified with a countersink J. Clin. Pathol. 2007 60 206 207 10.1136/jcp.2006.039578 17079355

PMC005xxxxxx/PMC5003445.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays3020137microarrays-03-00137ArticleQualitative and Quantitative Requirements for Assessing Prognostic Markers in Prostate Cancer Burdelski Christoph 1†Matuszewska Aleksandra 1†Kluth Martina 1Koop Christina 1Grupp Katharina 2Steurer Stefan 1Wittmer Corinna 1Minner Sarah 1Tsourlakis Maria Christina 1Sauter Guido 1Schlomm Thorsten 34Simon Ronald 1*1 Institute of Pathology, University Medical Center Hamburg-Eppendorf, Martinistr. 52, 20246, Hamburg, Germany; E-Mails: cburdelski@uke.de (C.B.); matuszewska.aleksandra@yahoo.de (AM); m.kluth@uke.de (M.K.); c.koop@uke.de (C.K.); s.steurer@uke.de (S.S.); c.wittmer@uke.de (C.W.); s.minner@uke.de (S.M.); m.tsourlakis@uke.de (M.C.T.); g.sauter@uke.de (G.S.)2 General, Visceral and Thoracic Surgery Department and Clinic, University Medical Center Hamburg-Eppendorf, Martinistr. 52, 20246, Hamburg, Germany; E-Mail: k.grupp@uke.de3 Martini-Clinic, Prostate Cancer Center, Martinistr. 52, 20246, Hamburg, Germany; E-Mail: tschlomm@uke.de4 Department of Urology, Section for Translational Prostate Cancer Research, University Medical Center Hamburg-Eppendorf, Martinistr. 52, 20246, Hamburg, Germany† These authors contributed equally to this work. * Author to whom correspondence should be addressed; E-Mail: r.simon@uke.de; Tel.: +49-40-7410-57214; Fax: +49-40-7410-55997.17 4 2014 6 2014 3 2 137 158 03 3 2014 28 3 2014 02 4 2014 © 2014 by the authors; licensee MDPI, Basel, Switzerland.2014This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Molecular prognostic markers are urgently needed in order to improve therapy decisions in prostate cancer. To better understand the requirements for biomarker studies, we re-analyzed prostate cancer tissue microarray immunohistochemistry (IHC) data from 39 prognosis markers in subsets of 50 – >10,000 tumors. We found a strong association between the “prognostic power” of individual markers and the number of tissues that should be minimally included in such studies. The prognostic relevance of more than 90% of the 39 IHC markers could be detected if ≥6400 tissue samples were analyzed. Studying markers of tissue quality, including immunohistochemistry of ets-related gene (ERG) and vimentin, and fluorescence in-situ hybridization analysis of human epidermal growth factor receptor 2 (HER2), we found that 18% of tissues in our tissue microarray (TMA) showed signs of reduced tissue preservation and limited immunoreactivity. Comparing the results of Kaplan-Meier survival analyses or associations to ERG immunohistochemistry in subsets of tumors with and without exclusion of these defective tissues did not reveal statistically relevant differences. In summary, our study demonstrates that TMA-based marker validation studies using biochemical recurrence as an endpoint require at least 6400 individual tissue samples for establishing statistically relevant associations between the expression of molecular markers and patient outcome if weak to moderate prognosticators should also be reliably identified. tissue microarrayprostate cancertissue qualitynumber of samplesprognosismarker validation ==== Body 1. Introduction Prostate cancer is the most frequent malignancy in men. The clinical behavior ranges from slowly growing indolent tumors to highly aggressive and metastatic cancers. Based on the results of large autopsy studies demonstrating a high prevalence of prostate cancers also in men who never experienced symptoms of the disease during their lifetime, it is assumed that most prostate cancer patients would be manageable without surgery and its associated side effects [1]. Accordingly, distinguishing between the low malignant and indolent form of the disease that does not require immediate therapy and the aggressive cancers that will eventually progress to life-threatening disease is the clinically most relevant discipline of current prostate cancer research. Only recently, commercial molecular classifiers have become available. These tests are based on mRNA expression profiling of defined gene sets, allow for estimating the biological aggressiveness of a cancer and, therefore, may aid in therapy decisions [2,3,4]. These classifiers underscore the interest of the diagnostic industry in the topic of prostate cancer prognosticators. It can be expected that future generations of such classifiers can be substantially improved if genes are included that on their own already exhibit strong and independent prognostic power. During the last decades, a multitude of studies announced prognostic biomarkers for prostate cancer. However, although more than hundred different prognostic markers have been suggested (reviewed in [5]), none of them has entered clinical routine testing as to yet. This disappointing failure to translate research findings into clinical applications is partly due to the fact that data obtained on virtually all of these markers vary largely between different studies. This is even true for the most established prognostic parameters, such as p53 or phosphatase and tensin homolog (PTEN). More than 50 studies analyzed the impact of p53 alterations on prostate cancer phenotype and prognosis. Although most immunohistochemistry studies reported a link between nuclear p53 accumulation and adverse tumor features, such as high grade, advanced stage, and peripheral zone origin [6], as well as poor prognosis after radical prostatectomy [7] or external beam radiation [8] and unfavorable clinical courses in conservatively managed patients [9], there are also studies that do not confirm these associations [10,11]. Likewise, genomic deletion of PTEN has been unequivocally linked to adverse tumor features in several studies [12,13,14,15,16], other studies again employing immunohistochemistry reported highly variable results on the prognostic value of PTEN expression. For example, an association between loss of PTEN expression and poor patient prognosis was only found in one [13] out of four studies [13,17,18,19], and a link between loss of PTEN expression and high Gleason grade or advanced tumor stage was only reported in two [20,21] out of five studies on this topic [18,19,20,21,22]. It is quite obvious that most of the discrepant results in the literature are due to (i) technical issues, and (ii) relatively small patient cohorts used for most studies. It is obvious that different antibodies, staining protocols, and scoring criteria that are employed in most studies can cause massive experimental variation. Due to an intense dispute with a reviewer of one of our manuscripts on the issue whether our frequency of p53 immunostaining in prostate cancer was lower than the 50% suggested by another group due to protocol issues (in our opinion) or to missed heterogeneity in a tissue microarray (TMA) setting (the reviewer’s opinion), we were forced to experimentally demonstrate that the range of p53 positive prostate cancers could be brought from 2.5% to 98% solely by protocol modifications [7]. However, the example of HER2 immunohistochemistry analysis of breast cancer demonstrates that a considerable (but not a complete) degree of assay standardization can be achieved [23]. However, even in such a highly standardized analysis including various controls, the quality of the tissues samples will impact the results. This is due to the fact that postsurgical tissue fixation cannot be fully standardized. The most frequently used fixative, i.e., formalin, causes proteins to cross-link and makes them impassible for microbial degradation or autolysis. The efficiency of the fixation process depends on the proper penetration of the formalin into the tissue, but obviously, the success of penetration greatly depends on the size and the composition of a given tissue. In case of too much or too little fixation, the tissue may not be suitable for analysis. This is a serious problem particularly in immunohistochemistry studies, where lack of immunoreactivity cannot be distinguished from a true negative result due to biological absence of the protein of interest. The tissue microarray (TMA) technology has proven to be excellently suited for rapid and cost efficient analysis of large numbers of tissue samples [24]. While in studies analyzing conventional large sections the study cohort size was typically limited to less than 100 samples due to the time and costs connected with such “classical” analyses, it is the availability of suitable tissues that first of all limits the size of TMA studies. As a consequence, TMA studies including hundreds of tissue samples are often viewed as “large-scale” analyses. Extent and impact of low-quality tissues that are inevitably included in every large-scale tissue analysis are, however, unknown. In the present study, we took advantage of our very large prostate cancer tissue microarray comprising more than 12,000 tissue spots and molecular data from more than one hundred proteins analyzed by means of immunohistochemistry to better understand the impact of the sample size and the tissue quality on the outcome of TMA studies for marker validation purposes. Biochemical (PSA) recurrence was used as an endpoint in this project dealing with patients having undergone prostatectomy. This reason for this is that PSA recurrence is the “easiest” (most frequent) clinical endpoint to analyze in prostatectomy studies and it is strongly associated with other clinical endpoints such as metastasis or cancer-related death. 2. Experimental Section Patients. Radical prostatectomy specimens were available from 12,427 patients undergoing surgery between 1992 and 2012 at the Department of Urology and the Martini Clinics at the University Medical Center Hamburg-Eppendorf. Follow-up data were available for a total of 12,344 patients with a median follow-up of 36 months (range of 1 to 241 months; Table 1). Prostate specific antigen (PSA) values were measured following surgery and PSA recurrence was defined as a postoperative PSA of 0.2 ng/mL and increasing at first of appearance. All prostate specimens were analyzed according to a standard procedure, including a complete embedding of the entire prostate for histological analysis [7]. The TMA manufacturing process was described earlier in detail [24]. In short, one 0.6 mm core was taken from a representative tissue block from each patient. The tissues were distributed among 27 TMA blocks, each containing 144 to 522 tumor samples. For internal controls, each TMA block also contained various control tissues, including normal prostate tissue. TMA Database. The molecular database attached to this TMA contained results on more than 100 molecular markers. For example, we analyzed expression of therapy target genes like epidermal growth factor receptor (EGFR) [25] and human epidermal growth factor receptor 2 (HER2) [26], putative prognosticators including p53 [7,27], proliferation marker Ki67 [5], mammalian target of rapamycin (mTOR) [28], cluster of differentiation (CD) 10 [29], serine peptidase inhibitor Kazal type 1 (SPINK1) [30], karyopherin alpha 2 (KPNA2) [31], cysteine-rich secretory protein 3 (CRISP3) [32], nibrin (NBS1) [33], RNA binding motif protein 3 (RMB3) [34], and lysophosphatidylcholine acyltransferase 1 (LPCAT1) [35], mitochondrial content [36], prostate-specific markers like prostate specific antigen (PSA), prostate specific membrane antigen (PSMA) [37], alpha-methylacyl-CoA racemase (AMACR), and androgen receptor (AR), microvessel density [38], or immunological target proteins like CD117 [39], CD147 [40], and CD151 [41], and determined gene copy number alterations of important tumor suppressor loci in prostate cancer, including 8p (lipoprotein lipase, LPL) [42], 3p13 (forkhead box P1, FOXP1) [43], 5q21 (chromodomain helicase DNA binding protein 1, CHD1) [44], 6q15 (mitogen-activated protein kinase kinase kinase 7, MAP3K7) [45], 10q23 (PTEN) [46], TMPRSS2:ERG fusion [47] and PTEN breakage [48]. For this study, we selected 39 different protein markers that predicted patient prognosis if analyzed in our current 12,427 samples TMA or in an earlier version of the TMA comprising 11,156 of the 12,427 samples. Based on the results of previous survival analyses using biochemical (PSA) recurrence as a clinical endpoint, our selection included very strong prognostic markers (e.g., the p53 tumor suppressor, Figure 1a [27]), markers with only marginal prognostic relevance (e.g., CD147, Figure 1b [40]), and those with an intermediate prognostic impact (e.g., LPCAT1, Figure 1c [35]). Immunohistochemistry (IHC). Freshly cut TMA sections were used for all experiments. IHC analysis was performed using an ETS-related gene (ERG)-specific antibody as described before [47] and an antibody directed against vimentin to identify tissue samples that might have impaired immunoreactivity. For vimentin detection, freshly cut TMA sections were immunostained on one day and in one experiment. Slides were deparaffinized and exposed to heat-induced antigen retrieval for 5 minutes at 100 °C in pH 6 Tris-EDTA-Citrate buffer. Primary antibody specific for vimentin (mouse monoclonal antibody, DAKO, Glostrup, DK; clone V9; dilution 1:18,000) was applied at 37 °C for 60 minutes. Bound antibody was then visualized using the EnVision Kit (Dako, Glostrup, Denmark) according to the manufacturer’s directions. Presence or absence of ERG and vimentin staining was recorded in all tissue spots. microarrays-03-00137-t001_Table 1Table 1 Composition of the prognosis tissue microarray containing 12,427 prostate cancer specimens. The percentage in the column “Study cohort on tissue microarray (TMA)” refers to the fraction of samples across each category. The percentage in column “Biochemical relapse among categories” refers to the fraction of samples with biochemical relapse within each parameter in the different categories. pT, pathological tumor stage; pN, pathological lymph node stage. Parameter No. of patients (%) Study cohort on TMA Biochemical relapse among categories (n = 12,427) Follow-up (mo) n 11,665 (93.9%) 2769 (23.7%) Mean 48.9 - Median 36.4 - Age (y) ≤50 334 (2.7%) 81 (24.3%) 51–59 3061 (24.8%) 705 (23%) 60–69 7188 (58.2%) 1610 (22.4%) ≥70 1761 (14.3%) 370 (21%) Pretreatment prostate specific antigen (PSA) (ng/mL) <4 1585 (12.9%) 242 (15.3%) 4–10 7480 (60.9%) 1355 (18.1%) 10–20 2412 (19.6%) 737 (30.6%) >20 812 (6.6%) 397 (48.9%) pT category (AJCC 2002) pT2 8187 (66.2%) 1095 (13.4%) pT3a 2660 (21.5%) 817 (30.7%) pT3b 1465 (11.8%) 796 (54.3%) pT4 63 (0.5%) 51 (81%) Gleason grade ≤3 + 3 2983 (24.1%) 368 (12.3%) 3 + 4 6945 (56.2%) 1289 (18.6%) 4 + 3 1848 (15%) 788 (42.6%) ≥4 + 4 584 (4.7%) 311 (53.3%) pN category pN0 6970 (91%) 1636 (23.5%) pN+ 693 (9%) 393 (56.7%) Surgical margin Negative 9990 (81.9%) 1848 (18.5%) Positive 2211 (18.1%) 853 (38.6%) Numbers do not always add up to 12,427 in the different categories because of cases with missing data. Abbreviation: AJCC, American Joint Committee on Cancer. Fluorescence in situ hybridization (FISH). A 4-μm TMA section was used for two-color FISH. For proteolytic slide pretreatment, a commercial kit was used (Paraffin pretreatment reagent kit, Vysis). A Spectrum-Orange–labeled HER2 probe was used together with a Spectrum-Green–labeled centromere 17 probe (PathVysion; Abbott Molecular). Before hybridization, TMA sections were de-paraffinized, air dried, and dehydrated in 70%, 85%, and 100% ethanol followed by denaturation for 5 minutes at 74 °C in 70% formamide-2× SSC solution. Following overnight hybridization at 37 °C in a humidified chamber, slides were washed and counterstained with 0.2 μmol/L 4',6-diamidino-2-phenylindole, an antifade solution. Presence or absence of red and green FISH signals was recorded in all tissue spots. Statistics. Statistical calculations were performed with JPM 9 (JMP®, Version 9. SAS Institute Inc., Cary, NC, USA, 1989–2007) Contingency tables and the chi²-test were performed to search for associations between molecular parameters and tumor phenotype. Survival curves were calculated according to Kaplan‑Meier. The Log-Rank test was applied to detect significant differences between groups. Figure 1 Examples of prognosis markers in prostate cancer. Kaplan Meier plots using biochemical recurrence as a clinical endpoint to demonstrate the clinical impact of (a) p53 as an example of a strong marker, (b) cluster of differentiation 147 (CD147) as an example of a very weak marker, and (c) LPCAT as an example of a moderate marker of prognosis. 3. Results and Discussion 3.1. Impact of the Tissue Quality In order to identify tissues with poor immunoreactivity, we performed ERG and vimentin immunohistochemistry analysis of our TMA. These proteins are expressed in virtually every human tissue. ERG is a member of the E26 transformation-specific (ETS) transcription factor family that is expressed in endothelial cells. ERG had been extensively studied in our TMA before since it is strongly expressed in about 50% of prostate cancers [47], and has been linked to early onset prostate cancer [48]. For the purpose of identifying low quality tissues, we re-analyzed our large 12,427 samples prostate cancer TMA for ERG expression specifically in endothelial cells (Figure 2a). In addition, we stained the TMA for vimentin, a type III intermediate filament that is strongly expressed in mesenchymal cells, which typically accompany prostate cancer cells (Figure 2b). Since blood vessels and mesenchymal cells can be found in virtually every prostate tissue sample, we considered complete absence of vimentin and ERG staining as an indicator of impaired immunoreactivity. In addition, we took advantage of the results from an earlier study, where we demonstrated that low-quality tissues with impaired immunoreactivity also showed a poor performance in fluorescence in-situ hybridization (FISH) analysis of gene copy numbers [49]. In the present study, we performed HER2 FISH analysis on the TMA and considered absence of FISH signals as an additional indicator of poor tissue quality. In summary, all tissue spots that showed simultaneous lack of ERG and vimentin immunostaining and absence of HER2 FISH signals were considered “low-quality”. Figure 2 Examples of immunostainings of markers for tissue quality. (a) ETS-related gene (ERG) expression in endothelial cells and lymphocytes in a prostate cancer tissue spot. Tumor cells are ERG negative. The inset shows magnification of a blood vessel. (b) Vimentin expression in mesenchymal cells in a prostate cancer tissue spot. Tumor cells are vimentin negative. A total of 11,223 tissue spots was included in this analysis. The remaining tissue spots were excluded because they were severely damaged or absent in the TMA slides. Simultaneous lack of ERG, vimentin, and HER2 signals were found in 2056 (18.3%) of the analyzed tissue spots. These “low-quality” tissues were randomly distributed across the TMA, and there was no obvious association between tissue reactivity and tumor phenotype or patient outcome (Figure 3). The marginally significant p-values obtained in these analyses do not reflect true associations but are attributable to slight variations between the groups. To further investigate the performance of these “low-quality” spots in IHC experiments, we next compared them to staining patterns of our 39 IHC markers in subsets of 2000 “low-quality” and 2000 “high-quality” tissues (i.e., samples that were positive for all of ERG, vimentin and HER2). This analysis revealed that, although the “low-quality” tissues can be stained with most of the tested antibodies, there was a average reduction of about 12 percent points in the fraction of positive tissue samples across all of these markers in “low-quality” tissues as compared to “high-quality” tissues (Figure 4). These date demonstrate, that “low-quality” tissues bear a high risk to underestimate the true expression level and may even result in false negative findings. Figure 3 Lack of relevant associations between tissue quality and prostate cancer phenotype. (a) Relationship with the Gleason grade. (b) Relationship with biochemical recurrence. Figure 4 Impact of the tissue quality of the overall fraction of samples yielding a positive result in immunohistochemistry studies. These findings imply that problems may arise if it comes to comparisons between biomarkers analyzed by immunohistochemistry. In such a scenario, false positive associations can potentially occur if the level of immunostaining of the analyzed markers parallels the quality of the tissue in a relevant fraction of samples. In contrast, inverse associations must always be considered valid. Here, the same tissues that are negative for one marker stain positive for the other, thus excluding the possibility of false associations due to reduced immunoreactivity. To assess the potential impact of “low-quality” tissues on the reliability of associations between ERG and other IHC markers, we used our existing ERG IHC data [47], which showed a positive result in about 50% of cancers. We studied the associations of all 39 markers to ERG expression, including markers with strong associations to ERG positivity (e.g., Marker #24, Figure 5a), markers with strong associations to ERG negativity (e.g., Marker #34, Figure 5c), markers with weak associations to ERG positivity (e.g., #12 (MTC02), Figure 5b), and markers lacking such associations (e.g., Marker #32, Figure 5d). Particularly for the latter set of markers, it could be possible that “low-quality” tissues drive such weak associations. All analyses were performed in differently sized subsets of our large TMA, and the significance of associations was compared between tissue sets containing both “low” and “high”‑quality tissues and tissue sets after excluding the 2056 “low-quality” tissues from the data. Since statistical associations will become stronger the more samples are analyzed, we performed the analyses in randomly selected subsets of 1600, 3200, 6400, and 10,000 samples. To compensate for incidental findings that might arise from random subset selection, we repeated each analysis five times. The Log-rank chi2 p-value was recorded from each analysis, and the average Log‑rank chi2 p-value was calculated from the five repeated analyses. All results are shown in Table 2. Following the same analysis strategy, we also questioned whether the reduced immunoreactivity in the “low-quality” tissues impacted the outcome of prognosis associations. For this analysis, we selected five of our 39 prognostic markers set and performed Kaplan-Meier survival plots to compare the impact of these markers (using biochemical (PSA) recurrence as clinical endpoint) before and after exclusion of the 2056 “low-quality” tissues. All results are shown in Table 3, and examples of Kaplan-Meier plots are given in Figure 4. In both sets of calculations, we did not observe changes in the analysis results, regardless if the “low-quality” tissue was excluded from the analysis or not, demonstrating that the ≈20% “low-quality” tissues present in our TMA did not significantly impact the study outcome. Nevertheless, the examples of associations with ERG expression given in Figure 5 confirm that the fraction of entirely negative samples can be slightly overestimated unless the “low-quality” tissues are exclude, as indicated by a difference of five percent points between tumors with a negative result for both Marker #24 and ERG in subsets of cancers before and after exclusion of the “low-quality” tissues. However, the finding that even positive associations resulting from discrete expression differences remained significant after exclusion of the “low-quality” tissues (Figure 5b) clearly demonstrates that associations between different markers can be reliably detected in large-scale TMA studies. Since “low-quality” tissues were randomly distributed across all samples irrespective of the clinical course (Figure 3), it was not surprising that there was no difference in the ability to detect prognostic differences in tissues with or without “low-quality” tissues. Here, the “underestimation” of the true staining intensity resulted in a smooth shift of all survival curves either towards an overall better prognosis (i.e., if strong expression of the marker was linked to poor prognosis), or towards worse prognosis (i.e., if strong expression of the marker was linked to good prognosis), whereas the relative distance between the curves remained largely constant. However, this analysis also suggested that the number of samples included in marker validation analyses might have a much stronger impact on the analysis result than the tissue quality, thus prompting us to analyze the impact of the sample size in more detail below. Figure 5 Examples of associations between expression of ERG and other Immunohistochemistry (IHC) markers in all tissue samples included in the 12,247 prostate cancers TMA (all tissues) and after exclusion of “low-quality” tissues from the analysis. (a) strong positive association, (b) weak positive association, (c) strong inverse association, (d) no association. microarrays-03-00137-t002_Table 2Table 2 Impact of the tissue quality on the association between expression of ERG and other IHC markers. The chi2 p-values are given for survival analyses in subsets of 1600–10,000 tissue spots. “Low quality tissue” indicates whether tissues with impaired immunoreactivity were excluded from analysis or not (included). “Association strength” separates the markers into those with weak, moderate, or strong positive associations (i.e., the marker is more frequently expressed in ERG positive than in ERG negative cancers), those with inverse associations (i.e., the marker is more frequently expressed in ERG negative than in ERG positive cancers), and those that are unrelated to ERG (no association). Marker Low quality tissue Number of analyzed tissue spots Association strength 1600 3200 6400 10,000 Marker #32 included 0.1039 0.5243 0.1879 0.0952 No association excluded 0.9431 0.6612 0.7653 0.6515 Marker #13 included 0.0112 <0.0001 <0.0001 <0.0001 Weak excluded 0.012 0.0004 <0.0001 <0.0001 Marker #12 (MTC02) included <0.0001 <0.0001 <0.0001 <0.0001 Weak excluded 0.0238 <0.0001 <0.0001 <0.0001 Marker #31 included 0.1307 0.0091 <0.0001 <0.0001 Weak excluded 0.2414 0.0031 <0.0001 <0.0001 Marker #37 included 0.0017 <0.0001 <0.0001 <0.0001 Weak excluded 0.1159 0.0012 0.0008 <0.0001 Marker #7 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #10 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #2 (CD10) included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #21 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #27 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #39 (p53) included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #4 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker#3 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #5 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #33 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #16 (NBS1) included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #18 (AR) included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #22 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #24 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #23 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #36 (KPNA2) included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #35 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #30 included <0.0001 <0.0001 <0.0001 <0.0001 Moderate excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #6 (FOXP2) included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #8 included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #9 included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #11 included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #14 included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #15 (LPCAT) included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #19 (RBM3) included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #26 included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #26 included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #25 included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #29 included <0.0001 <0.0001 <0.0001 <0.0001 Strong excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #1 (CD147) included <0.0001 <0.0001 <0.0001 <0.0001 Inverse excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #17 included <0.0001 <0.0001 <0.0001 <0.0001 Inverse excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #20 included <0.0001 <0.0001 <0.0001 <0.0001 Inverse excluded 0.0004 <0.0001 <0.0001 <0.0001 Marker #34 included <0.0001 <0.0001 <0.0001 <0.0001 Inverse excluded <0.0001 <0.0001 <0.0001 <0.0001 Marker #38 included <0.0001 <0.0001 <0.0001 <0.0001 Inverse excluded <0.0001 <0.0001 <0.0001 <0.0001 microarrays-03-00137-t003_Table 3Table 3 Impact of the tissue quality on the outcome of Kaplan-Meier survival analysis. The chi2p-values are given for survival analyses in subsets of 1600–10,000 tissue spots. “Low quality tissue” indicates whether tissues with impaired immunoreactivity were excluded from analysis or not (included). Marker Low quality tissue Number of analyzed tissue spots 1600 3200 6400 10,000 #2 (CD10) included 0.0937 0.0006 <0.0001 <0.0001 excluded 0.1215 0.0005 <0.0001 <0.0001 #3 included 0.0761 0.0946 0.0595 0.0037 excluded 0.1146 0.0761 0.0385 0.0043 #4 included 0.0810 0.1082 0.0151 <0.0001 excluded 0.1059 0.0810 0.0060 <0.0001 #18 (AR) included 0.0082 0.0206 0.0006 <0.0001 excluded 0.0197 0.0082 0.0003 <0.0001 #35 included <0.0001 <0.0001 <0.0001 <0.0001 excluded 0.0230 <0.0001 <0.0001 <0.0001 3.2. Impact of the Sample Size In order to estimate the minimal sample size that is required to yield statistically stable results in prostate cancer prognosis marker validation studies, we carried out serial analyses in randomly selected subsets of 50, 100, 200, 400, 800, 1600, 3200, 6400 and all (12,427) samples included in our TMA. We performed Kaplan-Meier survival plots and Log-rank chi2 tests including a total of 39 protein markers with confirmed prognostic relevance from our molecular database. The smallest sample set that revealed a Log-rank p-value of 0.001 or less was considered to be sufficient for reliable marker analysis provided that this significance level held also true in the analysis of all larger sample sets. In addition, in order to rank the “prognostic power” of our 39 markers, we summarized the Log-rank values emerging from all subset analyses of each marker. This strategy was selected because chi2 values can be easily extracted from all tests and thus provide a simple, however objective, index of the power of individual markers. We grouped our markers according to the accumulated chi2 values into markers with “weak” (sum of all chi2 values <100), “moderate” (sum chi2 101–299), and “strong” prognostic power (sum chi2 ≥300). The Log-rank p-values for all markers in each sample subset and the accumulated Log-rank chi2 values per marker are shown in Table 4, and exemplary Kaplan-Meier plots are given in Figure 6. The results of this analysis first of all demonstrate a close relationship between the “prognostic power” of a marker and the numbers of samples that need to be analyzed in order to reliably evaluate the marker’s prognostic potential (Figure 7 and Table 4). Given that the power of a marker of interest is typically not known before the analysis is performed (particularly in case of novel and uncharacterized candidate markers), and that four markers revealed prognostic relevance only if the entire sample set was analyzed (Table 4), our findings imply that as many samples as possible should be included in such marker validation experiments in order to also reliably detect minor associations between prostate cancer genotype and clinical behavior. However, from a more practical point of view, our data also demonstrates that a cohort size of 6400 prostate cancers is sufficient to reproduce the prognostic value of the vast majority (i.e., 35 out of 39, 90%) of the markers included in our study. microarrays-03-00137-t004_Table 4Table 4 Impact of the sample size on the outcome of Kaplan-Meier survival analyses. The chi2p-values are given for survival analyses in subsets of 50–12,427 tissue spots. “n analyzable” gives the number of interpretable tissue spots if the entire TMA was analyzed. “Marker power” indicates the relative prognostic power as described in the results section. Bold face indicates the minimal sample set that was considered sufficient to evaluate the respective marker. Grey color indicates sample sizes that yielded strong prognostic relevance. Marker n analy-zable Number of analyzed tissue spots Marker Power 50 100 200 400 800 1600 3200 6400 12,427 Marker #1 (CD147) 7605 0.2417 0.4218 0.4550 0.4984 0.5396 0.3539 0.3428 0.0379 0.0019 weak Marker #4 8025 0.0863 0.0998 0.9252 0.4330 0.4648 0.5781 0.0571 0.0666 0.0002 weak Marker #3 7574 0.2276 0.0720 0.0673 0.7038 0.6825 0.0934 0.3195 0.0351 0.0016 weak Marker #5 9473 0.8553 0.2279 0.9171 0.5532 0.1026 0.3192 0.0797 0.0097 0.0016 weak Marker #6 (FOXP2) 8284 0.7963 0.4436 0.0082 0.4058 0.4309 0.0776 0.0308 0.0011 <0.0001 weak Marker #7 9485 <0.0001 0.4999 0.0056 0.2196 0.0492 0.0464 0.0796 0.0007 <0.0001 moderate Marker #8 8158 0.1175 0.5066 0.7638 0.1192 0.3055 0.0020 0.0548 <0.0001 <0.0001 moderate Marker #9 6494 0.0494 0.9606 0.5709 0.1780 0.0603 0.0421 0.0564 <0.0001 <0.0001 moderate Marker #33 9262 0.1088 0.2166 0.5932 0.6114 0.0017 0.0013 0.0105 <0.0001 <0.0001 moderate Marker #10 9516 0.6617 0.8651 0.0836 0.1074 0.1415 0.0489 0.0041 0.0034 <0.0001 weak Marker #2 (CD10) 8488 0.0722 0.5977 0.6268 0.2465 0.3409 0.5137 0.0001 0.0062 0.0012 weak Marker #11 9627 0.5518 0.1662 0.0652 0.4009 0.5480 0.1911 0.0009 0.0046 <0.0001 weak Marker #12 (MTC02) 8407 0.4313 0.9545 0.7656 0.9315 0.3693 0.2517 0.0013 0.0004 <0.0001 weak Marker #14 8654 0.7139 0.0146 0.6741 0.4165 0.1473 0.4119 0.0094 0.0002 <0.0001 weak Marker #16 (NBS1) 8026 0.6506 0.0525 0.2071 0.7379 0.8382 0.2079 0.0004 0.0026 <0.0001 weak Marker #13 9875 0.8336 0.3172 0.2557 0.6717 0.0790 0.0962 0.0026 0.0004 <0.0001 weak Marker #15 (LPCAT) 8762 0.7141 0.1713 0.5933 0.6978 0.2428 0.0449 0.0020 0.0003 <0.0001 weak Marker #17 7275 0.7259 0.2301 0.2169 0.3562 0.3492 0.0661 0.0094 <0.0001 <0.0001 moderate Marker #18 (AR) 7856 0.4576 0.0246 0.9451 0.6439 0.7573 0.3058 <0.0001 0.0034 <0.0001 moderate Marker #20 6638 0.7489 0.7507 0.2524 0.4930 0.9583 0.0863 0.0001 <0.0001 <0.0001 moderate Marker #19 (RBM3) 8303 0.4712 0.1035 0.4893 0.2494 0.0134 0.0658 0.0006 <0.0001 <0.0001 moderate Marker #21 9643 0.8921 0.6673 0.2718 0.0004 0.0086 0.0170 <0.0001 <0.0001 <0.0001 moderate Marker #28 6824 0.5465 0.0252 0.8696 0.6919 0.2038 0.0075 0.0026 <0.0001 <0.0001 moderate Marker #22 7117 0.2824 0.8367 0.5302 0.9432 0.0591 0.0095 <0.0001 <0.0001 <0.0001 moderate Marker #24 9756 0.5724 0.3132 0.8763 0.6290 0.0686 0.0016 <0.0001 0.0002 <0.0001 moderate Marker #23 9744 0.3087 0.0266 0.0008 0.1771 0.3495 0.0001 0.0004 <0.0001 <0.0001 moderate Marker #29 9403 0.8852 0.0612 0.1260 0.1830 0.8574 0.0028 <0.0001 <0.0001 <0.0001 moderate Marker #27 7588 0.0607 0.2364 0.2661 0.0050 0.4899 0.0001 <0.0001 <0.0001 <0.0001 moderate Marker #26 9633 0.0169 0.5727 0.1760 0.7661 0.0491 0.0003 <0.0001 <0.0001 <0.0001 moderate Marker #25 7677 0.8371 0.3814 0.7065 0.0233 0.9297 <0.0001 <0.0001 <0.0001 <0.0001 moderate Marker #30 8174 0.3570 0.5619 0.3006 0.0028 0.1186 0.0002 <0.0001 <0.0001 <0.0001 strong Marker #31 10215 0.4072 0.4062 0.2342 0.3143 0.0040 0.0023 0.0002 <0.0001 <0.0001 moderate Marker #35 10216 0.8828 0.4813 0.1523 0.1304 0.0018 <0.0001 <0.0001 <0.0001 <0.0001 strong Marker #34 7670 0.0833 0.3010 0.2137 0.0703 0.0001 0.0002 <0.0001 <0.0001 <0.0001 strong Marker #37 7822 0.0044 0.0106 0.1209 0.0247 0.0002 <0.0001 <0.0001 <0.0001 <0.0001 strong Marker #32 9467 0.8994 0.3867 0.0006 0.0640 <0.0001 0.0004 <0.0001 <0.0001 <0.0001 strong Marker #36 (KPNA2) 7943 0.1490 0.3803 0.0483 0.0044 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 strong Marker #39 (p53) 10946 0.2686 0.1962 0.0192 <0.0001 0.0055 <0.0001 <0.0001 <0.0001 <0.0001 strong Marker #38 9576 0.7878 0.3611 0.0091 0.0063 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 strong Figure 6 Examples of Kaplan-Meier plots obtained from Marker #4 in subsets of 1600–10,000 samples. “All tissues” indicates that the samples contain low-quality and high-quality samples. Figure 7 Association between the “prognostic power” of different immunohistochemistry markers and the minimal number of samples that is required for statistically sound marker validation studies. “Marker Power” is given as the sum of Log-rank chi2 values per marker from the analysis in subsets of 50, 100, 200, 400, 800, 1600, 3200, 6400, and 12,427 samples. Some markers are annotated as examples. Importantly, only two (5%) of the 39 markers in our study, including p53 as a prime example of a very strong prognostic marker [27], had sufficient prognostic power to allow for conclusive results also in small cohorts including less than 500 prostate cancers. This finding is of particular interest since the majority of prostate cancer marker studies still analyze less than 500 cancer samples [5]. Our findings provide a simple explanation for the highly discrepant results on most potential prognostic biomarkers. We also found significant associations in less than 500 samples that, however, did not hold true in the next larger subsets and must, therefore, be considered incidental. For example, Marker #7 revealed significant p-values in subsets of 50 and 200 samples, but not in subsets of 100, 400, or even 3200 samples, demonstrating that analysis of small subsets can occasionally lead to incidental statistical results. It is further of note that there is always a considerable fraction of samples that does not yield interpretable results. This is due to typical TMA-related issues, including exhausted tissue cores resulting in empty spots in the TMA section or lack of tumor cells in the tissue spot. In our study, the fraction of non-interpretable tissue cores was about 35%, independent from the size of the subset selected for analysis (Figure 8a). As a consequence, the number of interpretable samples varied between 6494 and 10,946 (average 8592) spots for the different markers in the entire dataset (n = 12,427) and averaged for example 33.4 cancers in the 50 samples subset, 1019.7 cancers in the 1600 samples subset, or 4065 cancers in the 6400 samples subset (Figure 8b). Therefore, a certain dropout rate of TMA spots should always be taken into account if a TMA is built. The fraction of interpretable samples can potentially be increased if multiple samples of the same cancer specimen are included in the TMA. We do not recommend this procedure, however. For example, building a 6000‑samples TMA from 2000 cancers with three spots from each cancer can be expected to result in about 1800 interpretable cancers (which still is too small a number for reliable statistical analysis according to our data), but is connected with the same costs, analysis time, and tissue consumption as compared to a 6000 samples TMA built from one punch per tumor, which will, however, result in about 3900 interpretable cancers. In addition, analysis of multiple cores always introduces a statistical bias into the analysis. This is because not all of the multiple tissue spots per tumor will be analyzable, and tumors with three to four interpretable spots might have a higher likelihood to detect positive staining as compared to tumors with only one to two interpretable tissue spots. Figure 8 Association between the number of tissue samples analyzed in TMA studies and (a) the fraction or (b) total numbers of interpretable samples. 4. Conclusions The availability of a very large prostate cancer prognosis TMA with an extensive molecular database, including samples from more than 12,000 individual prostate cancers as well as molecular data from 39 prognostic relevant protein markers enabled us to evaluate the impact of qualitative and quantitative factors for prostate cancer biomarker studies. The results of our analyses suggest that such studies should aim at the analysis of at least 6000 individual prostate cancer samples to obtain reliable statistical findings allowing for a concluding judgment of a potential prognostic value of a marker of interest. Only for particularly strong markers, reliable results can also be obtained from substantially smaller cohorts. However, very strong prognostic markers appear to be rare, and the power of a marker is often not known before the analysis is made. Our data further suggest that almost 20% of the tissues included in a prostate cancer TMA may have limited tissue reactivity, potentially compromising the results of some analyses. While there is no impact of tissue reactivity on the results of prognostic studies, this issue is more relevant if it comes to comparisons between biomarkers analyzed by immunohistochemistry. Our data suggest, however, that even such associations that result from only discrete expression differences can be reliably identified in large-scale TMA analyses. Acknowledgments We thank Janett Lüttgens, Sünje Seekamp, Inge Brandt, Silvia Schnöger and Sascha Eghtessadi for excellent technical assistance. Author Contributions CB, KG, SS, CW, SM, MCT analyzed all immunostainings. MK analyzed FISH. AM and CB performed statistical analyses. GS, TS, and RS wrote the manuscript. Conflicts of Interest The authors declare no conflict of interest. ==== Refs References 1. Graefen M. Ahyai S. Heuer R. Salomon G. Schlomm T. Isbarn H. Budaus L. Heinzer H. Huland H. Active surveillance for prostate cancer Urologe A 2008 47 261 269 10.1007/s00120-008-1638-0 18273597 2. Cooperberg M.R. Simko J.P. Cowan J.E. Reid J.E. Djalilvand A. Bhatnagar S. Gutin A. Lanchbury J.S. Swanson G.P. Stone S. Validation of a cell-cycle progression gene panel to improve risk stratification in a contemporary prostatectomy cohort J. Clin. Oncol. 2013 31 1428 1434 10.1200/JCO.2012.46.4396 23460710 3. Cuzick J. Swanson G.P. Fisher G. Brothman A.R. Berney D.M. Reid J.E. Mesher D. Speights V.O. Stankiewicz E. Foster C.S. Prognostic value of an RNA expression signature derived from cell cycle proliferation genes in patients with prostate cancer: A retrospective study Lancet Oncol. 2011 12 245 255 10.1016/S1470-2045(10)70295-3 21310658 4. Badani K. Thompson D.J. Buerki C. Davicioni E. Garrison J. Ghadessi M. Mitra A.P. Wood P.J. Hornberger J. Impact of a genomic classifier of metastatic risk on postoperative treatment recommendations for prostate cancer patients: A report from the decide study group Oncotarget 2013 4 600 609 23592338 5. Schlomm T. Erbersdobler A. Mirlacher M. Sauter G. Molecular staging of prostate cancer in the year 2007 World J. Urol. 2007 25 19 30 10.1007/s00345-007-0153-z 17334767 6. Zellweger T. Ninck C. Bloch M. Mirlacher M. Koivisto P.A. Helin H.J. Mihatsch M.J. Gasser T.C. Bubendorf L. Expression patterns of potential therapeutic targets in prostate cancer Int. J. Canc. 2005 113 619 628 10.1002/ijc.20615 7. Schlomm T. Iwers L. Kirstein P. Jessen B. Kollermann J. Minner S. Passow-Drolet A. Mirlacher M. Milde-Langosch K. Graefen M. Clinical significance of p53 alterations in surgically treated prostate cancers Mod. Pathol. 2008 21 1371 1379 10.1038/modpathol.2008.104 18552821 8. Vergis R. Corbishley C.M. Thomas K. Horwich A. Huddart R. Khoo V. Eeles R. Sydes M.R. Cooper C.S. Dearnaley D. Parker C. Expression of Bcl-2, p53, and MDM2 in localized prostate cancer with respect to the outcome of radical radiotherapy dose escalation Int. J. Radiat. Oncol. Biol. Phys. 2010 78 35 41 10.1016/j.ijrobp.2009.07.1728 20092961 9. Kudahetti S. Fisher G. Ambroisine L. Foster C. Reuter V. Eastham J. Moller H. Kattan M.W. Cooper C.S. Scardino P. Cuzick J. Berney D.M. P53 immunochemistry is an independent prognostic marker for outcome in conservatively treated prostate cancer BJU Int. 2009 104 20 24 10.1111/j.1464-410X.2009.08407.x 19239456 10. Uzoaru I. Rubenstein M. Mirochnik Y. Slobodskoy L. Shaw M. Guinan P. An evaluation of the markers p53 and Ki-67 for their predictive value in prostate cancer J. Surg. Oncol. 1998 67 33 37 10.1002/(SICI)1096-9098(199801)67:1<33::AID-JSO7>3.0.CO;2-N 9457254 11. Incognito L.S. Cazares L.H. Schellhammer P.F. Kuban D.A. van Dyk E.O. Moriarty R.P. Wright G.L. Jr. Somers K.D. Overexpression of p53 in prostate carcinoma is associated with improved overall survival but not predictive of response to radiotherapy Int. J. Oncol. 2000 17 761 769 10995889 12. Han B. Mehra R. Lonigro R.J. Wang L. Suleman K. Menon A. Palanisamy N. Tomlins S.A. Chinnaiyan A.M. Shah R.B. Fluorescence in situ hybridization study shows association of PTEN deletion with erg rearrangement during prostate cancer progression Mod. Pathol. 2009 22 1083 1093 10.1038/modpathol.2009.69 19407851 13. McCall P. Witton C.J. Grimsley S. Nielsen K.V. Edwards J. Is pten loss associated with clinical outcome measures in human prostate cancer? Br. J. Canc. 2008 99 1296 1301 10.1038/sj.bjc.6604680 14. Sircar K. Yoshimoto M. Monzon F.A. Koumakpayi I.H. Katz R.L. Khanna A. Alvarez K. Chen G. Darnel A.D. Aprikian A.G. TEN genomic deletion is associated with p-akt and ar signalling in poorer outcome, hormone refractory prostate cancer J. Pathol. 2009 218 505 513 10.1002/path.2559 19402094 15. Yoshimoto M. Cunha I.W. Coudry R.A. Fonseca F.P. Torres C.H. Soares F.A. Squire J.A. Fish analysis of 107 prostate cancers shows that PTEN genomic deletion is associated with poor clinical outcome Br. J. Canc. 2007 97 678 685 10.1038/sj.bjc.6603924 16. Reid A.H. Attard G. Ambroisine L. Fisher G. Kovacs G. Brewer D. Clark J. Flohr P. Edwards S. Berney D.M. Molecular characterisation of ERG, ETV1 and PTEN gene loci identifies patients at low and high risk of death from prostate cancer Br. J. Canc. 2010 102 678 684 10.1038/sj.bjc.6605554 17. Koumakpayi I.H. Le Page C. Mes-Masson A.M. Saad F. Hierarchical clustering of immunohistochemical analysis of the activated ErbB/PI3K/Akt/NF-kappaB signalling pathway and prognostic significance in prostate cancer Br. J. Canc. 102 1163 1173 18. Bedolla R. Prihoda T.J. Kreisberg J.I. Malik S.N. Krishnegowda N.K. Troyer D.A. Ghosh P.M. Determining risk of biochemical recurrence in prostate cancer by immunohistochemical detection of PTEN expression and akt activation Clin. Canc. Res. 2007 13 3860 3867 10.1158/1078-0432.CCR-07-0091 19. Osman I. Dai J. Mikhail M. Navarro D. Taneja S.S. Lee P. Christos P. Shen R. Nanus D.M. Loss of neutral endopeptidase and activation of protein kinase b (AKT) is associated with prostate cancer progression Cancer 2006 107 2628 2636 10.1002/cncr.22312 17083125 20. McMenamin M.E. Soung P. Perera S. Kaplan I. Loda M. Sellers W.R. Loss of PTEN expression in paraffin-embedded primary prostate cancer correlates with high gleason score and advanced stage Canc. Res. 1999 59 4291 4296 21. Bertram J. Peacock J.W. Fazli L. Mui A.L. Chung S.W. Cox M.E. Monia B. Gleave M.E. Ong C.J. Loss of PTEN is associated with progression to androgen independence Prostate 2006 66 895 902 10.1002/pros.20411 16496415 22. Thomas G.V. Horvath S. Smith B.L. Crosby K. Lebel L.A. Schrage M. Said J. de Kernion J. Reiter R.E. Sawyers C.L. Antibody-based profiling of the phosphoinositide 3-kinase pathway in clinical prostate cancer Clin. Canc. Res. 2004 10 8351 8356 10.1158/1078-0432.CCR-04-0130 23. Sauter G. Lee J. Bartlett J.M. Slamon D.J. Press M.F. Guidelines for human epidermal growth factor receptor 2 testing: Biologic and methodologic considerations J. Clin. Oncol. 2009 27 1323 1333 10.1200/JCO.2007.14.8197 19204209 24. Kononen J. Bubendorf L. Kallioniemi A. Barlund M. Schraml P. Leighton S. Torhorst J. Mihatsch M.J. Sauter G. Kallioniemi O.P. Tissue microarrays for high-throughput molecular profiling of tumor specimens Nat. Med. 1998 4 844 847 10.1038/nm0798-844 9662379 25. Schlomm T. Kirstein P. Lwers L. Daniel B. Steuber T. Walz J. Chun F.H.K. Haese A. Kollermann J. Graefen M. Clinical significance of epidermal growth factor receptor protein overexpression and gene copy number gains in prostate cancer Clin. Canc. Res. 2007 13 6579 6584 10.1158/1078-0432.CCR-07-1257 26. Minner S. Jessen B. Stiedenroth L. Burandt E. Kollermann J. Mirlacher M. Erbersdobler A. Eichelberg C. Fisch M. Brummendorf T.H. Low level HER2 overexpression is associated with rapid tumor cell proliferation and poor prognosis in prostate cancer Clin. Canc. Res. 2010 16 1553 1560 10.1158/1078-0432.CCR-09-2546 27. Kluth M. Harasimowicz S. Burkhardt L. Grupp K. Krohn A. Prien K. Gjoni J. Hass T. Galal R. Graefen M. Clinical significance of different types of p53 gene alteration in surgically treated prostate cancer Int. J. Canc. 2014 10.1002/ijc.28784 28. Muller J. Ehlers A. Burkhardt L. Sirma H. Steuber T. Graefen M. Sauter G. Minner S. Simon R. Schlomm T. Michl U. Loss of p(ser2448)-mTOR expression is linked to adverse prognosis and tumor progression in erg-fusion-positive cancers Int. J. Canc. 2013 132 1333 1340 10.1002/ijc.27768 29. Fleischmann A. Schlomm T. Huland H. Kollermann J. Simon P. Mirlacher M. Salomon G. Chun F.H. Steuber T. Simon R. Distinct subcellular expression patterns of neutral endopeptidase (CD10) in prostate cancer predict diverging clinical courses in surgically treated patients Clin. Canc. Res. 2008 14 7838 7842 10.1158/1078-0432.CCR-08-1432 30. Grupp K. Diebel F. Sirma H. Simon R. Breitmeyer K. Steurer S. Hube-Magg C. Prien K. Pham T. Weigand P. SPINK1 expression is tightly linked to 6q15- and 5q21-deleted erg-fusion negative prostate cancers but unrelated to psa recurrence Prostate 2013 73 1690 1698 23843146 31. Grupp K. Habermann M. Sirma H. Simon R. Steurer S. Hube-Magg C. Prien K. Burkhardt L. Jedrzejewska K. Salomon G. High nuclear karyopherin alpha 2 expression is a strong and independent predictor of biochemical recurrence in prostate cancer patients treated by radical prostatectomy Mod. Pathol. 2014 27 96 106 10.1038/modpathol.2013.127 23887301 32. Grupp K. Kohl S. Sirma H. Simon R. Steurer S. Becker A. Adam M. Izbicki J. Sauter G. Minner S. Schlomm T. Tsourlakis M.C. Cysteine-rich secretory protein 3 overexpression is linked to a subset of PTEN-deleted ERG fusion-positive prostate cancers with early biochemical recurrence Mod. Pathol. 2013 26 733 742 10.1038/modpathol.2012.206 23196798 33. Grupp K. Boumesli R. Tsourlakis M.C. Koop C. Wilczak W. Adam M. Sauter G. Simon R. Izbicki J.R. Graefen M. The prognostic impact of high nijmegen breakage syndrome (NBS1) gene expression in erg negative prostate cancers lacking PTEN deletion is driven by kpna2 expression Int. J. Canc. 2014 10.1002/ijc.28778 34. Grupp K. Wilking J. Prien K. Hube-Magg C. Sirma H. Simon R. Steurer S. Budaus L. Haese A. Izbicki J. High RNA-binding motif protein 3 expression is an independent prognostic marker in operated prostate cancer and tightly linked to erg activation and pten deletions Eur. J. Canc. 2014 50 852 861 10.1016/j.ejca.2013.12.003 35. Grupp K. Sanader S. Sirma H. Simon R. Koop C. Prien K. Hube-Magg C. Salomon G. Graefen M. Heinzer H. High lysophosphatidylcholine acyltransferase 1 expression independently predicts high risk for biochemical recurrence in prostate cancers Mol. Oncol. 2013 7 1001 1011 10.1016/j.molonc.2013.07.009 23941784 36. Grupp K. Jedrzejewska K. Tsourlakis M.C. Koop C. Wilczak W. Adam M. Quaas A. Sauter G. Simon R. Izbicki J.R. High mitochondria content is associated with prostate cancer disease progression Mol. Canc. 2013 12 145 10.1186/1476-4598-12-145 37. Minner S. Wittmer C. Graefen M. Salomon G. Steuber T. Haese A. Huland H. Bokemeyer C. Yekebas E. Dierlamm J. High level PSMA expression is associated with early psa recurrence in surgically treated prostate cancer Prostate 2011 71 281 288 10.1002/pros.21241 20809553 38. Erbersdobler A. Isbarn H. Dix K. Steiner I. Schlomm T. Mirlacher M. Sauter G. Haese A. Prognostic value of microvessel density in prostate cancer: A tissue microarray study World J. Urol. 2010 28 687 692 10.1007/s00345-009-0471-4 19714336 39. Fleischmann A. Schlomm T. Kollermann J. Sekulic N. Huland H. Mirlacher M. Sauter G. Simon R. Erbersdobler A. Immunological microenvironment in prostate cancer: High mast cell densities are associated with favorable tumor characteristics and good prognosis Prostate 2009 69 976 981 10.1002/pros.20948 19274666 40. Grupp K. Hohne T.S. Prien K. Hube-Magg C. Tsourlakis M.C. Sirma H. Pham T. Heinzer H. Graefen M. Michl U. Reduced CD147 expression is linked to ERG fusion-positive prostate cancers but lacks substantial impact on psa recurrence in patients treated by radical prostatectomy Exp. Mol. Pathol. 2013 95 227 234 10.1016/j.yexmp.2013.08.002 23948277 41. Minner S. de Silva C. Rink M. Dahlem R. Chun F. Fisch M. Hoppner W. Wagner W. Bokemeyer C. Terracciano L. Reduced CD151 expression is related to advanced tumour stage in urothelial bladder cancer Pathology 2012 44 448 452 10.1097/PAT.0b013e32835576ee 22772340 42. El Gammal A.T. Bruchmann M. Zustin J. Isbarn H. Hellwinkel O.J. Kollermann J. Sauter G. Simon R. Wilczak W. Schwarz J. Chromosome 8p deletions and 8q gains are associated with tumor progression and poor prognosis in prostate cancer Clin. Canc. Res. 2010 16 56 64 10.1158/1078-0432.CCR-09-1423 43. Krohn A. Seidel A. Burkhardt L. Bachmann F. Grupp K. Becker A. Adam M. Graefen M. Huland H. Steurer S. Recurrent deletion of 3p13 targets multiple tumor suppressor genes and defines a distinct subgroup of aggressive erg fusion positive prostate cancers J. Pathol. 2013 231 130 141 10.1002/path.4223 23794398 44. Burkhardt L. Fuchs S. Krohn A. Masser S. Mader M. Kluth M. Bachmann F. Huland H. Steuber T. Graefen M. CHD1 is a 5q21 tumor suppressor required for erg rearrangement in prostate cancer Canc. Res. 2013 73 2795 2805 45. Kluth M. Hesse J. Heinl A. Krohn A. Steurer S. Sirma H. Simon R. Schumacher U. Grupp K. Izbicki J. Genomic deletion of MAP3K7 at 6q12–22 is associated with early PSA recurrence in prostate cancer and absence of TMPRSS2:ERG fusions Mod. Pathol. 2013 26 975 983 10.1038/modpathol.2012.236 23370768 46. Krohn A. Diedler T. Burkhardt L. Mayer P.S. De Silva C. Meyer-Kornblum M. Kotschau D. Tennstedt P. Huang J. Gerhauser C. Genomic deletion of PTEN is associated with tumor progression and early PSA recurrence in erg fusion-positive and fusion-negative prostate cancer Am. J. Pathol. 2012 181 401 412 10.1016/j.ajpath.2012.04.026 22705054 47. Minner S. Enodien M. Sirma H. Luebke A.M. Krohn A. Mayer P.S. Simon R. Tennstedt P. Muller J. Scholz L. ERG status is unrelated to PSA recurrence in radically operated prostate cancer in the absence of antihormonal therapy Clin. Canc. Res. 2011 17 5878 5888 10.1158/1078-0432.CCR-11-1251 48. Weischenfeldt J. Simon R. Feuerbach L. Schlangen K. Weichenhan D. Minner S. Wuttig D. Warnatz H.J. Stehr H. Rausch T. Integrative genomic analyses reveal androgen-driven somatic alteration landscape in early-onset prostate cancer Canc. Cell 2013 23 159 170 10.1016/j.ccr.2013.01.002 49. Tapia C. Schraml P. Simon R. Al-Kuraya K.S. Maurer R. Mirlacher M. Novotny H. Spichtin H. Mihatsch M.J. Sauter G. HER2 analysis in breast cancer: Reduced immunoreactivity in fish non-informative cancer biopsies Int. J. Oncol. 2004 25 1551 1557 15547690

PMC005xxxxxx/PMC5003446.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5010001microarrays-05-00001ReviewSNP Array in Hematopoietic Neoplasms: A Review Song Jinming Shao Haipeng *Bortoluzzi Stefania Academic EditorDepartment of Hematopathology and Laboratory Medicine, H. Lee Moffitt Cancer Center and Research Institute, 12902 Magnolia Drive, Tampa, FL 33612, USA; Jinming.Song@moffitt.org* Correspondence: Haipeng.Shao@moffitt.org; Tel.: +1-813-745-2672; Fax: +1-813-745-170822 12 2015 3 2016 5 1 131 7 2015 14 12 2015 © 2015 by the authors; licensee MDPI, Basel, Switzerland.2015This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Cytogenetic analysis is essential for the diagnosis and prognosis of hematopoietic neoplasms in current clinical practice. Many hematopoietic malignancies are characterized by structural chromosomal abnormalities such as specific translocations, inversions, deletions and/or numerical abnormalities that can be identified by karyotype analysis or fluorescence in situ hybridization (FISH) studies. Single nucleotide polymorphism (SNP) arrays offer high-resolution identification of copy number variants (CNVs) and acquired copy-neutral loss of heterozygosity (LOH)/uniparental disomy (UPD) that are usually not identifiable by conventional cytogenetic analysis and FISH studies. As a result, SNP arrays have been increasingly applied to hematopoietic neoplasms to search for clinically-significant genetic abnormalities. A large numbers of CNVs and UPDs have been identified in a variety of hematopoietic neoplasms. CNVs detected by SNP array in some hematopoietic neoplasms are of prognostic significance. A few specific genes in the affected regions have been implicated in the pathogenesis and may be the targets for specific therapeutic agents in the future. In this review, we summarize the current findings of application of SNP arrays in a variety of hematopoietic malignancies with an emphasis on the clinically significant genetic variants. SNP arrayhematopoieticmyelodysplastic syndromeleukemialymphoma ==== Body 1. Introduction The progressive accumulation of genetic changes plays an essential role in the tumorigenesis and evolution of human cancers. The genetic changes commonly seen in human cancers include chromosomal translocations, amplifications, allelic loss, loss of heterozygosity, deletions, mutations, and epigenetic changes/DNA methylation affecting oncogenes and tumor suppressor genes [1]. The resolution of genetic alterations identified in clinical specimens has been pushed to the single nucleotide level over the decades with advancements in genetic technologies. Conventional cytogenetic analysis with G-banding karyotyping, a routine clinical analysis in cytogenetic labs, allows differentiation of approximately 400–500 bands per haploid genome [2]. At this level of resolution, chromosomal change over 10 Mb can be detected. Fluorescence in situ hybridization (FISH) offers high sensitivity and specificity of detecting genetic abnormalities such as translocations, aneuploidy, deletions, inversions, or amplifications by using DNA probes targeted to known DNA sequences [3]. FISH can identify genetic changes at a resolution up to a few kilobases (kb), but is not suited for identification of unknown genetic changes or global chromosomal abnormalities. Array-based comparative genomic hybridization (aCGH) developed in the early 1990s offers efficient high-throughput analysis of the entire genome for identification of copy number variations/aberrations (CNVs/CNAs) that are usually not detectable by conventional karyotyping or targeted FISH studies, and has an improved resolution down to 100 kb [4,5,6,7]. Single nucleotide polymorphism (SNP) arrays, manufactured by Affymetrix and Illumina, were initially designed for high-throughput SNP genotyping, but were quickly applied to cancer genomics [8,9,10,11]. In contrast to aCGH, SNP arrays are able to detect both CNVs/CNAs and loss of heterozygosity (LOH) or copy-neutral LOH/uniparental disomy (UPD), which are frequently involved in the development of cancers. With the advance in technology and marked improvements in resolution, the new SNP array offers over 90% coverage of known copy number variants by using more than 946,000 probes and an average inter-marker distance of 680 base pairs. This high level of resolution of cytogenetic changes has only recently been surpassed by next generation sequencing (NGS) technology developed in the last decade [12,13]. Ever since the invention of SNP arrays, they have been extensively applied to various hematologic malignancies. While currently there are no clinical guidelines on the use of SNP array in hematopoietic malignancies, SNP array will certainly be useful in difficult cases, especially in myelodysplastic syndrome (MDS) diagnosis, when other methodologies fail to identify cytogenetic abnormalities. A proposed flow chart for the application of SNP array in hematopoietic malignancies is presented in Figure 1. In this review, we summarize the important findings of chromosomal changes in hematopoietic malignancies identified by SNP array analysis. Figure 1 Proposed application of SNP array in hematopoietic malignancies. 2. Acute Lymphoblastic Leukemia/Lymphoma Acute lymphoblastic leukemia/lymphoma is the most frequent pediatric malignancy, affecting 20–40 patients per million children in developed countries [14], and accounts for 20% of all acute leukemias in adults. B-lymphoblastic leukemia/lymphoma (B-ALL) is the most common type of acute lymphoblastic leukemia, and comprises genetically distinct subtypes including B-ALL with Philadelphia chromosome t(9;22)(q34;q11.2) (BCR-ABL1), t(v;11q23) (MLL rearranged), t(12;21)(p13;q22) (TEL-AML1), t(5;14)(q31;q32) (IL3-IGH), t(1;19)(q23;p13.3) (E2A-PBX1), hyperdiploidy, hypodiploidy, and about 25% cases without defined cytogenetic abnormalities [15]. Pediatric B-ALL has a favorable prognosis with approximately 80% rate of cure, while adult B-ALL has an inferior prognosis with about 40% rate of cure [16,17,18,19,20,21,22,23,24]. B-ALL with t(9;22)(q34;q11.2) (BCR-ABL1) is associated with the worst prognosis in both children and adults. Studies have shown that the currently identified chromosomal translocations are early initiating genetic events but are not sufficient to induce ALL [25]. To identify additional genetic lesions important for leukaemogenesis, a SNP array is well-suited for global genomic mapping of B-ALL. Irving et al. [26] first applied The Affymetrix 10K SNP array with resolution of 100 to 200 kb was used in 10 cases of pediatric B-ALL and demonstrated the usefulness of this technique in studying B-ALL. Of the 10 cases, LOH was detected in eight cases with the most frequent abnormality (50%) in chromosome 9p harboring the CDKN2A/B (INK4) gene locus. The loss of INK4 gene locus was only observed at relapse in three of the four cases, suggesting its association with treatment failure. Subsequently, Mullighan et al. [27] performed the first large-scale study of 242 cases of paediatric ALL, including 192 B-ALL and 50 T-ALL, by using Affymetrix SNP arrays that examine over 350,000 loci with an average resolution of less than 5 kb. Matched remission samples allowed the identification of somatic CNAs and LOH in leukemic blasts. The SNP arrays showed a low number of somatic copy number alterations (mean of 6.46) per case in ALL, with deletions outnumbering amplifications almost 2:1. The frequency of CNAs varied significantly between different cytogenetically defined ALL subtypes, with deletions more frequent than gains of DNA. Chromosomal deletions occurred more frequently in B-ALL with ETV6–RUNX1 and hypodiploidy with average of six deletions per case, up to 21 deletions, and only one deletion in MLL rearranged B-ALL. Gains of DNA occurred most frequently in hyperdiploid B-ALL (average of 10 gains), and uncommon in other types of ALL. The study identified 54 recurring regions of deletion that were mostly focal with the minimal deletion less than 1 Mb, and 24 deletions harboring only one single gene. The most important finding was that genes regulating normal B-cell development were deleted or mutated in approximately 40% cases of B-ALL. Copy number changes of PAX5, which is essential for B-cell differentiation, occurred in about 30% cases of B-ALL, making PAX5 the most frequently altered gene in B-ALL. These changes resulted in either reduced level or hypomorphic alleles of PAX5. Sequencing studies also identified a variety of somatic mutations in PAX5 resulting in either lost or altered DNA-binding or transcriptional functions. Other important genes deleted in ALL included EBF1, TCF3, LEF1, IKZF1 (IKAROS), and IKZF3 (AIOLOS). These findings suggest that ALL is not a neoplasm characterized by chromosomal instability, and genetic alterations in genes controlling B-cell development (PAX5, EBF1 and IKZF1) are common and play important roles in B-ALL leukaemogenesis. In a separate study, Kawamata et al. [28] studied 399 pediatric ALL samples with matched remission marrow using Affymetrix 50K SNP arrays, and identified three most common genetic alterations: deletion of ETV6, deletion of CDKN2A/p16INK4A, and hyperdiploidy. This study also confirmed the deletions of PAX5 (9p13), EBF (5q33), IKAROS (7p12.2), AIOLOS (17q12), LEF1 (4q25), RAG1 (11p12), and RAG2 (11p12) in pediatric ALL, albeit with a lower frequency except PAX5. Uniparental disomy (UPD) was frequently identified, especially in chromosome 9. In addition, hyperdiploid ALL without gains of chromosomes 17 and 18 was found to have poor prognosis. The common deletions of CDKN2A at 9p21 (29%) and ETV6 (TEL) at 12p13 (3/24, 12%) were also confirmed by Bungaro et al. [29] in a separate study. T-ALL comprises about 25% cases of adult ALL and approximately 15% cases of childhood ALL and is most commonly present in adolescents as mediastinal lymphoma. The most frequent recurrent cytogenetic abnormalities involve translocation of T-cell receptor gene locus (14q11.2, 7q35, and 7p14-15) with a variety of partner genes such as HOX11, MYC, TAL1, RBTN1/2, and LYL1 [30,31] and, thus, T-ALL is genetically more heterogeneous. In their SNP array analysis of 50 cases of T-ALL, Mullighan et al. [32] identified multiple new genomic changes in T-ALL, including deletions of TAL1, RB1, and PTEN, and duplications of protooncogene MYB. Most recently, Karrman et al. [33] investigated 47 cases of T-ALL with an Illumina HumanOmni1-Quad BeadChip containing >1 million markers and a median marker interval of 1.5 kb. Copy number changes and UPD were identified in the majority of cases (92%), with a median of three changes/per case. This study identified recurring region of deletion harboring genes CDKN2A, CDKN2B, LEF1, PTEN, RBI, and STIL. In terms of uniparental disomy (UPD), the T-ALL lacked whole chromosome UPD, but showed segmental UPDs (sUPDs) in 42% of cases, with a high proportion of sUPD 9p (30% of the cases) associated with homozygous CDKN2A deletion. Therefore, disruption of the p53 and RB1 pathways through deletion of INK4/ARF on CDKN2A gene locus appears to be important in the pathogenesis of T-ALL. The genomic changes that may explain the difference in survival between pediatric and adult ALL were addressed by Okamoto et al. [34] with Affymetrix 50K or 250K SNP arrays. The authors studied 75 cases of adult ALL and 399 cases of pediatric ALL. This study showed 572 genomic alterations with a mean of 7.6 genomic changes per case in adult ALL. The genomic changes in adult ALL were comparable to those identified in pediatric ALL, including deletions of 3p14.2 (FHIT), 5q33.3 (EBF), 6q, 9p21.3 (CDKN2A/B), 9p13.2 (PAX5), 13q14.2 (RB1), and 17q11.2 (NF1). The recurrent genomic alterations had similar rate of occurrence in pediatric and adult ALL. As the adult ALL cases were all non-hyperdiploid, the pediatric ALL cases were divided into hyperdiploid (HD) and non-hyperdiploid groups. There was no significant difference between adult and non-HD pediatric ALL in terms of deletions of 3p14.2 (FHIT), 9p21.3 (CDKN2A/B), 9p13.2 (PAX5), 13q14.2 (RB1), and 17q11.2 (NF1), and adult ALL showed more frequent deletion of 17p (TP53) and duplication of 17q than non-HD pediatric ALL (11% vs. 2%, and 9% vs. 1%, respectively). Overall, there were no unequivocal CNAs identified by SNP array that can account for the differences of prognosis between adult and pediatric ALL. The characteristic deletions of CDKN2A, PAX5, IKZF1, ETV6, RB1, and EBF1 genes were also identified by Safavi et al. [35] in adult ALL using SNP array covering 5 million markers and a resolution of 10 kb. A number of novel recurrent cryptic genetic changes involving BCAT1, SERP2, RAB30, SRPR, ST3GAL4, ASS1, RASSF3, FUBP3, BCL11A, GAB1, LINGO2, TOX, and CXCR4 genes, and partial and whole-chromosome UPDs were also discovered in adult ALL. Their significance in pediatric and adult ALL remains to be determined. SNP arrays were designed for genome-wide association (GWA) study, and it was naturally applied in ALL for germline SNPs that may have an association with ALL. Trevino et al. [36] studied 317 cases of pediatric ALL along with 17,958 control cases with an Affymetrix 500K Mapping array, and found two SNPs at chromosome 10q21 (rs10821936 and rs10994982) located in intron 3 of the ARID5B gene to be associated pediatric ALL. In addition, both SNPs discriminated hyperdiploid B-ALL from other major ALL subtypes. A genome-wide association study by Papaemmanuil et al. [37] on two case-control series with 907 ALL cases and 2398 controls also identified an association between a SNP at 10q21.2 in the ARID5B gene (rs7089424) and pediatric ALL. The 10q21.2 (ARID5B) risk association was selective for hyperdiploid B-ALL. In addition, they also found two additional risk loci for ALL at 7p12.2 (IKZF1, rs4132601), and 14q11.2 (CEBPE, rs2239633). The association between genetic variations at 7p12.2 (IKZF1), 10q21.2 (ARIDB5), and 14q11.2 (CEBPE) with pediatric ALL was replicated by Prasad et al. [38] in genotyping 1384 cases of pediatric B-ALL and 1877 controls. These findings indicate that common germline variants contribute to the risk of development of pediatric ALL. As both ARID5B and IKZF1 play important roles in B-lymphocyte growth and differentiation, the possibility of SNPs in these genes predispose the patients to the development of B-ALL is high. 3. Acute Myeloid Leukemia Acute myeloid leukemia (AML) is genetically heterogeneous, ranging from cases with recurrent cytogenetic abnormalities to approximately 40% cases with normal karyotype, some of which have prognostically significant somatic mutations. AMLs with t(15;17), inv(16) and t(8;21) respond well to chemotherapy and have good prognosis, while AMLs with Philadelphia chromosome t(9;22), complex karyotype, −5/−5q, −7/−7q, and 11q23 translocations have poor prognosis. In AMLs with normal karyotype, FLT3 internal tandem duplications (ITDs) are associated with poor prognosis [39], while mutations of NPM1 and CEBPA are associated with favorable prognosis [40]. In 2005, Raghavan et al. [41] first applied a 10K SNP array in 64 cases of AML and identified partial uniparental disomy (pUPD) in approximately 20% cases. A similar study with 10K SNP array demonstrated homozygous mutations involving genes FLT3, CEBPA, and RUNX1 in approximately 50% patients with UPD involving corresponding chromosomal regions [42]. Subsequently, Gupta et al. [43] expanded the study with 10K SNP array on 454 cases of AML from young adults, and found nonrandom acquired UPD (aUPD) in 17% cases, preferentially affecting chromosomes 13q, 11p, and 11q, similar to the findings by Raghavan et al. [43], and additional recurrent aUPDs at 2p, 17p, 2q, 17q, 1p, and Xq. AMLs with FLT3-ITD had aUPD 13q involving the FLT3 gene, while AMLs with FLT3-ITD-/FLT3-TKD+ mutation did not have aUPD 13q. These early studies suffered from the low resolution of the SNP arrays available at that time. With advances in SNP array technology, Bullinger et al. [44] were able to use 50 and 500K Affymetrix SNP arrays on 157 cases of cytogenetically normal AML. The cohort showed 12% aUPDs with chromosomal regions 6p, 11p, and 13q most commonly affected and all aUPDs were >29 Mb. The aUPDs were associated with mutations in NPM1 or CEBPA, which suggested that aUPDs may affect genes critical for hematopoiesis. In terms of aCNAs, as expected, the 500K SNP array was much more sensitive than 50K SNP array, which missed approximately 60%–70% of CNAs detected by the 500K SNP array. aCNAs were identified in 49% cases, with genomic losses (1.21/case) more frequent than gains (0.25/case). The recurrent genomic deletions included bands 3p14.1-p13, 6q27, 8q23.3, 10q11.21, 11q25, 12p13.2, and 15q21.3, that harbor genes FOXP1 and RYBP (3p14.1-p13), RPS6KA2 (6q27), TRPS1 (8q23.3), HNRPF (10q11.21), ETV6 (12p13.2) and RFXDC2 (15q21.3). In a separate study, Walter et al. [45] applied a 1.85 million SNP array from Affymetrix in 86 adult patients with de novo AML, and identified a total of 201 aCNAs in 44% of the cases with a mean of 2.34 CNAs per genome. Acute erythroid leukemia and acute megakaryocytic leukemia had more CNAs (10–29 CNAs per genome) than other morphologic variants of AML. CNAs were detected in 24% AMLs with normal cytogenetics and in 40% AMLs with abnormal karyotype. 12 chromosomal regions (eight deletions and four amplifications) containing at least one gene implicated in AML or MDS (deletions of 3p14.1: FHIT, 5q31.1: CTNNA1, 12p12.3: ETV6, 16q22.1: CBFB, 17p13.1: TP53, 17q11.2: NF1, and amplifications of 8q23.2: MYC, 11q23.3: MLL, and 21q22.2: ETS2) were identified in multiple AML cases. CNAs in chromosomal regions 17q11.2 and 21q22.2 (CNAs spanning NF1 and ETS2) were found in at least five cases, most of which had complex karyotypes, and associated with worse overall survival. aUPD was infrequent occurring in eight of 86 genomes, with most in AML with normal cytogenetics. 50% of the AML studied had no CNA or UPD at this resolution. The identified CNAs did not predict overall or event-free survival independent of cytogenetics. Radtke et al [46] showed similarly very low burden of genomic alterations in pediatric de novo AML, with a mean of only 2.38 somatic copy-number alterations per case. These studies indicate that AMLs are not genomically unstable and the genes implicated in these CNAs likely play important role in the leukemogenesis of AML. While Walter’s study did not identify any prognostically significant CNAs or UPDs, three groups clearly showed prognostic significance of SNP array lesions in AML, likely due to utilization of more stringent SNP lesion detection algorithms. Parkin et al. [47] examined 114 previously untreated prospectively enrolled AML patients with Affymetrix SNP 6.0 arrays, and showed that ≥2 genomic lesions detected by SNP 6.0 array almost doubled the risk of death after controlling for age- and karyotype-based risk by multivariate analyses. P53 mutations, or P53 mutations coupled with 17p-LOH conferred an independent negative prognosis. Tiu et al. [48] performed 250 K and 6.0 SNP arrays on 140 cases of primary (p) and secondary (s) AML, and demonstrated that patients with genomic lesions including acquired somatic UPD identified by SNP array had worse overall survival (OS) and event-free survival (EFS) in pAML with normal cytogenetics and in pAML/sAML with abnormal cytogenetics. The SNP array lesions, AML type, and metaphase cytogenetics had independent predictive value for OS by multivariate analyses. In a study of 133 cases of AML with normal cytogenetics by Genome-Wide Human SNP 6.0 Array, Yi et al. [49] found at least one abnormal SNP lesion in 32.3% cases. Detection of abnormal SNP lesions by SNP-A karyotyping conferred an unfavorable prognosis for overall survival by multivariate analyses. All these studies confirmed the clinical relevance as well as prognostic significance of SNP lesions in AML, especially AML with normal cytogenetics, which would allow a better prognostic stratification of patients with AML for appropriate treatment. 4. Myelodysplastic Syndrome Myelodysplastic syndrome (MDS) is a group of clonal hematopoietic neoplasm characterized by ineffective hematopoiesis, cytopenia, morphologic dysplasia, and potential progression to acute myeloid leukemia [50]. Most patients with MDS die of bone marrow failure rather than transformation to acute myeloid leukemia. MDS is diagnosed based on the World Health Organization (WHO) classification, but the prognosis for survival is stratified based on the international prognostic scoring system (IPSS) and revised IPSS (IPSS-R) that incorporate cytogenetic abnormalities, percentage of bone marrow myeloblasts, and number of cytopenias [51,52]. As part of the IPSS-R system, cytogenetic abnormalities are stratified from very good with single del(11q) and −Y to very poor with complex karyotype (>3 abnormalities). Approximately half of patients with MDS have normal karyotype that is associated with good prognosis in the IPSS-R system, but MDS patients with normal karyotype are still heterogeneous genetically. SNP arrays were therefore performed on MDS to identify additional occult genetic abnormalities, especially in patients with normal karyotype. Gondek et al. [53,54] was the first group to study MDS with SNP array. They applied an Affymetrix 50K SNP Assay to 66 and 72 patients with MDS, and found chromosomal defects in 82% of MDS patients, including 68% patients with normal karyotype, 81% patients with abnormal karyotypes, with chromosomes 8, 7, 5, and 11 most frequently involved. Segmental uniparental disomy (sUPD) was found in 33% of patients with MDS, and usually in regions frequently affected by deletions detected by metaphase cytogenetic analysis, including 7q and 11q. In a similar study, Mohamedali et al. [55] studied 119 patients with low-risk MDS with 50 K, 250 K and 500 K SNP arrays and identified deletions in 10%, amplifications in 8%, and UPD in 46% of cases. Nowak et al. [56] showed that CNAs and LOH can be identified in the CD34 positive blasts. These early studies suffered from absence of paired normal tissue for each case, which made distinction of inherited CNAs and LOH from somatic acquired ones difficult. Heinrichs et al. [57] performed a prospective study of matched pairs of bone marrow and buccal cell (normal) DNA from 51 patients with MDS by 250K SNP array, and identified somatically acquired genomic abnormalities in 41% patients, including 15% in MDS with normal karyotypes. UPDs affecting chromosome 7q was associated with rapidly progressive clinical course despite a low-risk IPSS score [57]. Similarly, Tiu et al. [58] analyzed 250 cases of MDS by 250K and Affymetrix SNP Array 6.0 with paired bone marrow and CD3+ lymphocytes to distinguish germline lesions. In this study, they showed that combined metaphase cytogenetics and SNP array had a higher diagnostic yield of chromosomal defects (74% vs. 44%), compared with conventional karyotyping. The genetic abnormalities detected by SNP arrays were deletions and aUPD involving chromosomes 1, 5, 7, 11, 17, and 21. While Mohamedali’s study failed to show independent prognostic significance of the genetic lesions identified by SNP array on multivariate analysis, Tiu showed that the presence of new genetic lesions detected by SNP array was predictive of poor prognosis in MDS by univariate and multivariate analyses [58]. These studies proved the utility of SNP array in detecting submicroscopic genetic lesions in MDS as a complement to metaphase cytogenetics, and the lesions identified by SNP arrays can further help prognostic stratification of MDS patients. MDS can be difficult to diagnose clinically due to many mimickers of the disease and the lack of significant morphologic dysplasia in a small subset of cases. For example, severe aplastic anemia (AA) may be difficult to distinguish from hypoplastic MDS morphologically and cytogenetically. Afable et al. [59] demonstrated the utility of SNP analysis in AA to complement metaphase cytogenetics for the detection of clonal chromosomal lesions. Combined metaphase cytogenetics and SNP array identified chromosomal lesions in 19% of AA and 54% of hypoplastic MDS. Persistent detection of chromosomal lesions by SNP array would be highly suspicious for hypoplastic MDS and less response to immunotherapy (ATG/cyclosporine) for AA. Therefore, in diagnostically-challenging and equivocal cases of MDS, SNP array can be used to establish the presence of clonal hematopoiesis in patients with normal karyotype and allow appropriate management of the patients. With the advent of next generation sequencing, SNP arrays are not likely to be used for the identification of genes involved in disease. In the last decade, SNP arrays have played an important role in the identification of individual genes important for the pathogenesis of MDS. TET2 gene was identified by SNP array genomic profiling and genomic sequencing in 102 patients with MDS, and acquired deletions, missense and nonsense mutations in the TET2 gene were found in 26% cases of MDS [60]. Recurrent aUPD and microdeletion of chromosome 7q led to identification of EZH2 gene and mutations in MDS [61,62]. The TET2 gene mutations have been found to be associated with better response to hypomethylating agents [63], while EZH2 mutations are poor prognostic marker for MDS. By combining SNP-array and gene expression profiling, Merkerova et al. [64] identified BMP2 and TRIB3 genes located in 20p UPD as potential candidate genes for the pathogenesis of MDS. 5. Chronic Myelogenous Leukemia Chronic myelogenous leukemia (CML) is characterized by the presence of Philadelphia chromosome t(9;22)(q34;q11) (BCR-ABL1), for which targeted therapy with tyrosine kinase inhibitors (TKI) have revolutionized the treatment of CML. In an effort to define genetic lesions that cooperate with BCR-ABL1 to transform to Philadelphia chromosome positive acute leukemia, Mullighan et al. [65] studied 304 cases of ALL, including 23 CML cases with a 250K Affymetrix SNP array, and found only 0.47 copy number alterations per case in chronic phase CML (range 0–8), which suggested that chronic phase CML is genomically stable and BCR–ABL1 is sufficient to induce CML. However, a separate study by Khorashad et al. [66] on 10 chronic phase CML patients with high resolution 2.1 million oligonucleotide array comparative genomic hybridization (CGH) showed an average of 53 CNAs per patient (range: 4–166) with majority being amplifications. The difference in CNAs detected by the two techniques is most likely due to the marked difference in resolutions of the arrays. In Mullighan’s study, IKZF1 was deleted in CML lymphoid blast phase, but not in CML chronic phase [65]. The IKZF1 encodes IKAROS which is an essential transcription factor for normal lymphoid development. Deletion of IKZF1 results in monoallelic, expression of dominant-negative form, or loss of expression of IKAROS. Deletion of PAX5 and CDKN2A/B, together with loss of IKFZ1 in lymphoid but not myeloid blast phase of CML indicates that these genes play important role in the transformation of CML to lymphoid blast phase of CML [32]. A subset of the CML patients is resistant to TKI therapy, mainly due to BCR-ABL1 mutations or amplification. In a study to identify genetic alterations in 45 TKI-resistant CML by 250K SNP array, Nowak et al. [67] found recurrent submicroscopic alterations, including aUPD in chromosomes 1, 8, 9, 17, 19, and 22. Recurrent deletions of IGLC1 locus on chromosome 22 were identified in three patients with previous blast crisis, suggesting dedifferentiation into immature progenitors as a possible mechanism of TKI resistance. 6. Polycythemia Vera, Essential Thrombocythemia and Primary Myelofibrosis Polycythemia (Rubra) vera (PV) is a clonal hematopoietic neoplasm characterized by increased red cell mass and JAK2 V617F mutation in 95% cases, and potential to progress to myelofibrosis or transformation to acute myeloid leukemia. Essential thrombocythemia (ET) is a chronic myeloproliferative neoplasm affecting the megakaryocytic lineage and characterized by abnormally large megakaryocytes in the marrow and persistent thrombocytosis. Primary myelofibrosis (PMF) is characterized by megakaryocytic and myeloid proliferation and progressive fibrosis in the marrow. JAK2 V617F mutation is identified in approximately 40%–50% cases of ET and approximately 50% of PMF. Of note, JAK2 V617F mutation can be detected in healthy individuals by high sensitivity methods. CNAs were rare in PV and ET and approximately one third cases of PMF showed small genomic losses (<5 Mb) with 250K SNP array [68]. In terms of copy-neutral aberrations (UPD), recurrent changes were only observed on chromosome 9p. Rice et al. [69] confirmed that chromosome 9 abnormalities including 9p LOH, trisomy 9, amplifications of 9p13.3–23.3, 9q33.1–34.13, and 9q34.13 were most frequent in myeloproliferative neoplasms when analyzing 87 myeloproliferative neoplasmas (MPN) cases with an Affymetrix 250K SNP array. Other less frequent recurrent genetic alterations included gains of 1p36.31–36.33, 17q21.2–q21.31, and 17q25.1–25.3, and deletions affecting 18p11.31–11.32. In the study of genetic profiles of myeloproliferative neoplasms by 50K SNP array, Kawamata et al. [70] found rare genomic abnormalities in ET, and deletion of chromosomal regions harboring RB (13q14) or NF1 (17q11) in 25% PMF cases. aUPD involving JAK2 was found in five PV cases with homozygous JAK2 V617F [70]. A subpopulation with 9p aUPD was detected in 30% of PV and approximately 50% PMF cases, and UPD at 1p was identified in one case of PV. The relationship between JAK2 V617F mutation and 9p aUPD in PV was further addressed by Wang et al. [71]. They investigated 31 PV patients with SNP array and whole genome sequencing and validated the findings in 59 additional PV patients [71]. They defined four PV subgroups based on the quantitative relationship between JAK2 V617F and 9p aUPD: 42% of patients with heterozygous JAK2 V617F and no detectable 9p aUPD (subgroup I); 45% of patients with homozygous JAK2 V617F and an allelic fraction directly proportional to the level of 9p aUPD (subgroup II); 10% with 9p aUPD at approximately twice the level of heterozygous JAK2 V617F allelic burden (subgroup III) and 3% with trisomy 9p and two copies of JAK2 V617F allele (subgroup IV), which likely suggest different pathways leading to PV phenotype. These genomic profiling studies indicated that ET and PV are genomically stable and JAK2 on chromosome 9p is critical for the pathogenesis of MPN. A genome-wide associate study seems to support this, with the identification of a SNP in the JAK2 locus (rs10974944), which predisposed to the development of JAK2 V617F-positive myeloproliferative neoplasm [72]. A subset of the patients with MPNs eventually progresses to acute myeloid leukemia. Identification of acquired genetic alterations facilitating this transformation would be valuable for patient stratification. In a study comparing genome profiles of 88 cases of MPN and 71 cases of MPN-blasts phase with 50 and 250 K SNP arrays, leukemic transformation of MPN was accompanied by up to three-fold more genomic alterations per case than chronic phase [73]. The genomic regions commonly affected during leukemic transformation harbored established genes such as ETV6, TP53, and RUNX1, and also new candidate genes on 7q, 16q, 19p, and 21q. Trisomy 8 or amplification of 8q24 (MYC) was identified exclusively in JAK2 V617F(−) MPN-blast phase. A poor prognosis after leukemic transformation was associated with copy number-neutral loss of heterozygosity (CNN-LOH) on either 7q or 9p including homozygous JAK2 V617F. With higher resolution SNP 6.0 arrays, Rumi et al. [74] showed a close relationship between UPD and/or gain of chromosome 9p with progression from PV to post-PV myelofibosis, and genetic aberrations of chromosome 5, 7, or 17p associated with progression to AML and overall survival. 7. Myelodysplastic/Myeloproliferative Neoplasms Myelodysplastic/myeloproliferative neoplasms (MDS/MPN) are characterized by the presence of features of both MDS and MPN. Chronic myelomonocytic leukemia (CMML) is the most common form of MDS/MPN and characterized by persistent monocytosis, dysplasia, and transformation to acute myeloid leukemia. SNP Array 6.0 identified genomic alterations in 60% of patients with CMML with cryptic CN-LOH in 71% and microdeletions in 45% cases [75]. CN-LOH was frequent on 7q harboring EZH2, 11q harboring CBL, and 4q harboring TET2. The presence of multiple chromosomal defects detected by SNP array was associated with a worse overall survival by univariate analysis [75]. In a separate study, UPD occurred in 48% of CMML by 250K SNP array analysis, with the most frequently affected chromosomal region in 11q harboring proto-oncogene c-CBL [76]. All patients with UPD 17q and UPD 4q were found to have CMML or M5 primary AML. These studies indicated that CMML is genetically heterogeneous with different pathways to a common disease phenotype, and CBL mutations may activate the RAS pathway and aberrant pSTAT5 activation in CMML. A recent study suggested that abnormal SNP array lesions were associated with an inferior complete and partial remission rate, and worse overall survival when compared with patients without SNP lesions after decitabine therapy [77]. This finding, if confirmed, would certainly help better prognostic stratification and treatment of CMML patients. Juvenile myelomonocytic leukemia (JMML) is a rare clonal hematopoietic disorder of childhood characterized by monocytosis and loss of function of neurofibromatosis 1 (NF1) or somatic mutations of genes in RAS/MAPK pathway. Flotho et al. [78] first applied SNP array on 16 cases of JMML with normal karyotype and identified large regions of UPD on chromosome 17 spanning approximately 55 Mb, which contained the locus of the NF1 tumor suppressor gene on 17q11.2, in four of five patients with JMML and NF1, but not in other cases without NF1. Inactivating NF1 lesion on both alleles was found by mutational analysis in each case. This study indicates that 17q UPD with homozygous loss of normal NF1 plays a critical role for the pathogenesis of JMML in NF1 patients. 8. Classical Hodgkin Lymphoma Classical Hodgkin lymphoma (cHL) is characterized by presence of small numbers of neoplastic Reed-Sternberg/Hodgkin cells admixed with mixed inflammatory cells. Metaphase cytogenetic study is typically unsuccessful due to low numbers of neoplastic cells. For the same reason, SNP array is not expected to yield useful information on patient samples. Instead, SNP array 6.0 was performed on cHL cell lines and showed a UPD of chromosome 14q, which was associated with biallelic deletion of TRAF3 in one cell line, and a gain of copy number for MAP3K14 in three other cell lines [79]. With primary cHL tissues, interphase cytogenetic analyses confirmed monoallelic deletion of TRAF3 in 3/20 cases and gains of MAP3K14 in 5/16 cases. Both TRAF3 and MAP3K14 are regulators of the NF-κB pathway, which is constitutively activated in cHL. The study suggested that genetic alterations of the components of the NF-κB pathways contributed to the pathogenesis of cHL, at least in a subset of cases. A genome-wide association study identified five SNPs on chromosome 6p21.32 associated with nodular sclerosis cHL (NSHL), which is a common subtype of cHL [80]. Two of the SNPs, rs6903608 and rs2858870, were significantly associated with NSHL. The extended haplotype containing these five SNPs was the strongest overall predictor of risk for NSHL [80]. The haplotype with all five risk alleles for the SNPs (Hap3: AGGCT) was associated with a 70% increased risk of NSHL; while the haplotype with all five protective alleles (Hap6: GAATC) was associated with a 60% decreased risk. The DRB1*07:01 allele, which was carried by all individuals with haplotype 6 (GAATC), was associated with a 50% decreased risk of NSHL, suggesting HLA-DRB1 polymorphisms likely implicated in NSHL susceptibility. 9. Mature B-Cell Lymphoproliferative Disorder Mature B-cell neoplasms account for over 90% lymphomas, with the most common types being diffuse large B-cell lymphoma and follicular lymphoma. Diffuse large B-cell lymphoma (DLBCL) is composed of large neoplastic B-cells and genetically heterogeneous. Based on gene expression profiles, DLBCL can be further classified into germinal center B-cell like (GCB) type with good prognosis and activated B-cell like (ABC) type with poor prognosis. Scholtysik et al. [81] applied 250K SNP array on 148 cases of DLBCL and found recurrent genomic gains in 24 regions and recurrent genomic losses in 38 regions, with a median of 19 imbalances per case in GCB-DLBCL and 25 per case in ABC-DLBCL. A number of genetic alterations showed different frequencies in GCB and ABC-DLBCL, such as gains of HDAC7A on chromosome 12 predominantly in GCB-DLBCL (38% of cases) and losses of BACH2 and CASP8AP2 on chromosome 6 predominantly in ABC-DLBCL (35%), suggesting different pathways to lymphomagenesis in the two subtypes of DLBCL. Two new potential tumor suppressor genes CASP3 and IL5RA were identified in the analysis and showed no somatic mutations, suggesting a haploinsufficiency effect of the genes. Another study showed high frequency of LOH over chromosomal region 11p11.2 harboring the gene encoding the protein tyrosine phosphatase receptor type J (PTPRJ), which regulates a number of survival pathways [82]. The combination of SNP array and whole exome/whole transcriptome sequencing was especially productive and identified ARID1B, ROBO2, and MRS1 as potential tumor suppressor genes and KLHL6, IL31, and LRP1 as oncogenes in DLBCL [83]. The impact of genomic alterations on clinical course was studied by 250 SNP array in 124 patients treated with R-CHOP [84]. 20 recurrent genetic lesions were shown to have an impact on the clinical course, in which loss of 8p23.1 had the strongest statistical significance. In this study, five clusters of DLBCL showed distinct genetic profiles, clinical features, and outcomes. Follicular lymphoma (FL) is a mature B-cell lymphoma composed of malignant germinal center B-cells with some cases eventually transforming to diffuse large B-cell lymphoma. Early study with 10K SNP array on 26 cases of FL revealed recurrent aUDP on 6p, 9p, 12q, and 17p [85]. Homozygosity of 9p and 17p were found predominantly in transformed FL with homozygosity of pre-existing mutation of either CDKN2A or TP53 identified in a subset of cases. Interestingly, 10 cases showed chromosomal regions of homozygosity in FL that were absent in the subsequent transformed FL, suggesting the transformed FL derived from a common malignant precursor but not from step-wise clonal evolution of the preceding FL, at least in some cases [85]. The prognostic significance of aUPD was investigated in 185 cases of FL by 10K SNP array [86]. This study found genetic abnormalities in 65% cases, and more than three abnormalities were associated with inferior overall survival. Recurrent aUPD were detected on 6p, 16p, 12q, 1p36, 10q, and 6q, which were confirmed by other studies [87,88]. O’Shea et al. [86] showed that aUPD on 1p36 was correlated with shorter overall survival on multivariate analysis, and aUPD on 16p predicted transformation and poorer progression free survival. Somatic mutations of TNFRSF14 gene was identified by exon sequencing of the minimum region of deletion of ∼97 kb within 1p36.32, and TNFRSF14 mutations and 1p36 deletions were associated with inferior overall survival and disease specific survival after adjustment for the International Prognostic Index [89]. Individual genes including CDKN2A, CDKN2B, FHIT, KIT, PEX14, and PTPRD, which were associated with canonical pathways, were implicated by SNP array analysis in FL [87]. FL is characterized by t(14;18)(q32;q21) with IGH/BCL2 fusion, but a small subset of cases of FL is negative for t(14;18). The genetic profiles of t(14;18) positive and negative FL were compared by SNP array, and showed that gains/amplifications of the BCL2 gene locus at 18q were only present in the t(14;18)-positive FL [90]. Chronic lymphocytic leukemia is the most common leukemia in adults, and characterized by proliferation of small mature lymphoid cells in the peripheral blood, bone marrow, spleen and lymph nodes. Risk stratification based on FISH findings are routinely performed clinically. Early studies with 10 K and 50 K Affymetrix SNP arrays showed chromosomal imbalances in 65.6% and 82% cases, respectively, including UPD in 20% cases [91,92]. The cytogenetic changes commonly identified by FISH studies including trisomy 12, deletions of TP53 (17p13), ATM (11q22), and 13q14 were readily identified by SNP arrays. SNP arrays found a total of 45 CNAs in 45% cases excluding the four common cytogenetic changes identifiable by FISH [92]. High resolution Affymetrix SNP array 6.0 study on 353 samples of chronic lymphocytic leukemia (CLL) showed similar findings [93], with CN-LOH in 6% of CLL cases, most frequently on 13q, 17p, and 11q. Minimally-deleted regions were identified on 13q14 to the DLEU1 and DLEU2 genes, 11q22.3 to ATM, 2p16.1-2p15 to a 1.9-Mb fragment containing nine genes, and 8q24.21 to a 486 kb region proximal to the MYC locus. Breakpoint cluster regions flanking 13q deletions were found. A 3.5-Mb gain at 2p16 harboring REL and BCL11A oncogenes and deletion at 6q21 that involved the AIM1 gene were identified in a subset of cases [91,92], suggesting involvement of these genes in the pathogenesis of CLL. UPD was detected in 7% cases, with 50% involving whole chromosome 13 resulting in homozygous deletion of micro-RNA-15a (miR-15a)/miR-16-1 [92], which was confirmed by another study [94]. The genomic complexity identified by SNP array was an independent risk factor for aggressive CLL and short survival on multivariate analysis [94,95,96,97]. Large genomic aberrations identified by SNP array but not covered by the standard FISH panel was found to be an independent prognosticator of a shorter time to first treatment in CLL, by multivariate analysis [98]. Clonal evolution was shown to developed in 33% patients with unmutated IGHV, and in 16% treated patients with mutated IGHV, and included known recurrent aberrations such as del(13q) [94]. Analysis of clonal diversity by SNP array for genomic alterations and mosaic distribution of clones was also shown to be predictive of disease progression [99]. Del13q14 is the most frequent cytogenetic abnormality in CLL and associated with good prognosis. Ouillette et al. [100] showed that del13q14 was heterogeneous and composed of multiple subtypes, with deletion of RB or the miR15a/miR16 loci as anatomic landmarks. Large (type II) 13q14 deletions spanning the RB gene were associated with elevated genomic complexity, accelerated clinical course and short survival [101]. Similarly a separate study was able to classify CLL with del13q14 into two separate clusters characterized by short/biallelic deletion with loss of the miR-15a/16-1 versus wide/monoallelic 13q14 deletions [102]. Therefore, despite the established good prognosis of del13q14 by FISH, SNP arrays are clinically useful to identify a subset of cases with del13q14 that has poor prognosis. SNP arrays were shown to be able to supplement FISH studies with unusual signal patterns [103]. Mantle cell lymphoma is an aggressive B-cell lymphoma characterized by IGH/CCND1 translocation resulting in overexpression of cyclin D1. A large numbers of genomic abnormalities have been identified through metaphase cytogenetics and comparative genomic hybridization in addition to the characteristic t(11;14)(q13;q32). With 250K SNP array, Kawamata et al. [104] confirmed the presence of known genetic alterations, including deletion of INK4A/ARF, duplication/amplification of MYC, deletion of ATM, and deletion of TP53 in 33 samples of MCL. A duplication/amplification at 13q involving oncogenic microRNA, miR17-92, other genomic abnormalities, including duplication/amplification of cyclin D1, del(1p), del(6q), dup(3q) and dup(18q), and a number of aUPD sites, including whole chromosome 9 aUPD and 9p aUPD were identified by the SNP array. To identify the target genes in the genomic lesions, Beà et al. [105] combined SNP array and gene expression profiling and detected high number of partial UPDs with UPD 17p one of the most common associated with TP53 gene inactivation. 4 known tumor suppressor genes (CDKN2C, BCL2L11, CDKN2A, and RB1) and six new genes (FAF1, MAP2, SP100, MOBKL2B, ZNF280A, and PRAME) were identified in homozygous deletions. The most recurrent amplifications were at 11q13.3–q13.5, 13q31.3, and 18q21.33, targeting CCND1, C13orf25, and BCL2, which may be important for the lymphomogenesis of MCL. Marginal zone lymphoma is a diverse group of B-cell lymphoma derived from post-germinal center B-cells. SNP array studies in this group of lymphoma are limited. Flossbach et al. [106] examined a series of marginal zone B-cell lymphomas of the gastrointestinal tract including ones with large cell transformation by SNP array. They found increase of genomic complexity with lymphoma progression to large cell lymphoma. Gains of protooncogenes such as REL, BCL11A, ETS1, PTPN1, PTEN, and KRAS were identified exclusively in the large cell variants. Copy numbers of ADAM3A, SCAPER and SIRPB1 were also associated with progression from small to large cell lymphoma. The gene for tumor necrosis factor alpha-induced protein 3 (TNFAIP3, A20), a negative regulator of NF-κB was found to be deleted at chromosomal region 6q23 in a small subset of marginal zone lymphomas by SNP array analysis [107]. In ocular adnexal MALT lymphoma, one study showed CNAs in approximately 70% cases and UPD detected on 6q (14%) and 3q (10%) [108]. The UPD on 6q likely involves the A20 gene. In this study, chromosomal gains were most commonly trisomy 3 (31%), trisomy 18 (17%), and 6p and 21q (14%), and the most frequent copy neutral (CN) loss regions were 6q and 9p (7%). Importantly, CNAs were not detected in reactive lymphoid hyperplasia, suggesting that SNP may be useful in difficult cases for diagnostic purpose. Burkitt lymphoma (BL) is an aggressive B-cell lymphoma mainly in pediatric patients and characterized cytogenetically by t(8;14)(q24;q32) involving the MYC gene. Illumina 1M SNP array was applied to 20 cases of BL to identify additional genomic lesions in addition to t(8;14)(q24;q32) [109]. Genomic imbalances were found in 95% cases by SNP array, including recurrent losses of 6q14.1–q22.33, 9p21.3, and 13q14.2–q14.3, gains of 1q23.3–q31.3, chromosome 7, 13q31.3, and partial UPD for 6p12.2-pter, 9p23-pter, and 17p11.2-pter. These genetic alterations resulted in deletion of CDKN2A and TP53 genes, and gains/losses of other genes including MIR17HG and E2F2K that are involved in the MYC pathway, suggesting dysregulation of the MYC pathway by 8q24/MYC translocation or secondary genomic alterations are essential for development of Burkitt lymphoma. Hairy cell leukemia (HCL) is an indolent B-cell neoplasm with “hairy cells” in the peripheral blood and bone marrow aspirates, and characteristic BRAF V600E mutation. With current therapy, patients with HCL have an excellent prognosis and near normal life span. HCL was found to have a remarkably stable genome by a 250K SNP array analysis [110]. With high resolution SNP Array 6.0, Rinaldi et al. [111] confirmed this finding. In this study, the 19 cases of HCL showed an extremely low numbers of CNAs with only two heterozygous losses detected (11%). These studies indicate very limited genetic damages in HCL, which may partly explain the excellent treatment response in current therapy. Multiple myeloma (MM) is a neoplasm of terminally differentiation B-cells (plasma cells) characterized by multisystem damage and presence of M-Spike in the peripheral blood. In a study of 30 samples of patients with newly diagnosed MM by a 50K SNP array, genomic alterations at 1p, 1q, 6q, 8p, 13, and 16q were most frequent [112]. Multiple regions of UPD were identified for the first time and were found to be interspersed throughout the genome, with a median of three UPD per sample (range, 0–19). These findings were largely confirmed by Agnelli et al. [113] in analyzing 41 cases of MM and four cases of plasma cell leukemia. Using unsupervised clustering methods five main groups of genetic imbalances were identified and showed strict correlation with transcriptional expression: cluster I with hyperdiploidy, particularly trisomy 11; cluster II with no or limited alterations; cluster III with 1q gain and chromosome 13 deletion; cluster IV with deletions of 1p, 13, 14, plus deletions of 8p and 22; and cluster V with near-tetraploidy. In an effort to address the prognostic significance of genetic lesions detected by high resolution SNP array, Avet-Loiseau et al. [114] showed deletions and amplifications in 98% of patients with MM. Amplifications in 1q and deletions in 1p, 12p, 14q, 16q, and 22q were frequently associated with adverse prognosis, and recurrent amplifications of chromosomes 5, 9, 11, 15, and 19 was associated with a favorable prognosis. Amp(1q23.3), amp(5q31.3), and del(12p13.31) retained independent prognostic value in multivariate analysis. Del(12p13.31) alone, or amp(5q31.3) and del(12p13.31), and high Sβ2M predicted a very poor prognosis. The prognostic significance of 1q amplification was confirmed by two other groups even after removing cases with the most adverse cytogenetic factors such as translocations involving FGFR3/MMSET, MAF, and MAFB, and del17p [115,116]. Walker et al. [115] identified UPD on 1q (8%), 16q (9%), and X (20%), that was associated with regions of gain and loss. Kamada et al. [116] showed accumulation of deletions and UPD at 22q12.1 associated with poor prognosis in hyperdiploid MM. With a 500K SNP array, Jenner et al. [117] identified LOH at 16q involving CYLD, a negative regulator of the NF-κB pathway, and WWOX, a tumor suppressor involved in apoptosis, that were independently associated with poor prognosis in MM. Multiple myeloma nowadays is typically treated with hypomethylating agent, bortezomib plus melphalan and prednisone. Kim et al. [118] showed that increasing genomic complexity identified by SNP arrays correlated with the outcome of the bortezomib plus melphalan and prednisone therapy. Patients with deletion of 1p and gain of 3q did not achieve very good partial response, while complex karyotype and gain of 3q were associated with progressive disease after therapy. Finally, López-Corral et al. [119] showed progressive increase in the incidence of CNAs from precursor monoclonal gammopathy of undetermined significance (MGUS) to MM. Gains on 1q, 3p, 6p, 9p, 11q, 19p, 19q, and 21q along with deletions of 1p, 16q, and 22q were significantly less frequent in MGUS than in MM. The frequency of UPD was higher in active MM than in the asymptomatic MM. As expected, the increasing genomic complexity from MGUS to MM is consistent with acquisition of additional genomic lesions as essential pathway to the progression from MGUS to MM. 10. Mature T/NK-Cell Lymphoproliferative Disorders The T/NK-cell lymphoproliferative disorders comprise a diverse group of mature lymphomas or leukemias with variable etiology and clinical course. SNP array studies in this group of neoplasms are limited to a few selected types, largely due to the much lower frequency of occurrence of mature T/NK-cell leukemia/lymphoma overall. Peripheral T-cell lymphoma, not otherwise specified (PTCL, NOS) is the most common T-cell lymphoma and morphologically, phenotypically and cytogenetically heterogeneous. Hartmann et al studied 47 cases of PTCL, NOS with a 250K SNP array, and found genomic alterations in 47% of cases, including recurrent gains of chromosome regions 1q32–43, 2p15–16, 7, 8q24, 11q14–25, 17q11–21 and 21q11–21 and losses of chromosome regions 1p35–36, 5q33, 6p22, 6q16, 6q21–22, 8p21–23, 9p21, 10p11–12, 10q11–22, 10q25–26, 13q14, 15q24, 16q22, 16q24, 17p11, 17p13, and Xp22 [120]. Genomic gains of REL gene locus at 2p15–16 and nuclear expression of the REL protein by immunohistochemistry were identified in approximately 25% cases of PTCL, NOS, suggesting pathogenetic relevance of REL in a subset of PTCL, NOS cases. Angioimmunoblastic T-cell lymphoma (AILT) is the second most common mature T-cell lymphoma and characterized by proliferation of malignant follicular T-helper cells associated with Epstein-Barr virus infection. In a study comparing the genomic profiles of 40 cases of AILT and 33 cases of PTCL, NOS by 50 K SNP array, three quarters of the cases had relatively stable genomes, while the remaining one quarter had CNAs of various sizes [121]. The presence of CNAs was associated with poor prognosis. Highly-recurrent chromosomal gains in both AILT and PTCL, NOS were clustered at three distinct regions of 8q, 9p, and 19q, and genomic losses at two distinct regions of 3q and 9p. The most common region of LOH was identified in a 440-kb region at 2q32.3. AILT- or PTCL NOS-specific CNAs or LOH were present at 21 regions. Furthermore, overexpression of CARMA1 at 7p22 and MYCBP2 at frequently amplified 13q22 predicted poor prognosis. A novel isoform of IKZF2 was identified in the LOH region at 2q34, which likely acted as a dominant negative form to participate in the transformation to AILT or PTCL, NOS. Adult T-cell leukemia/lymphoma is a well-defined malignant T-cell neoplasm caused by human T-cell leukemia virus type 1 (HTLV-1). To understand the genetic events occurring after HTLV-1 infection, spectral karyotyping and SNP array of 61 ATLL cases revealed a 2-Mb deletion region breakpoint in 10p11.2 in 35% cases [122]. Transcription Factor 8 (TCF8) was identified within this region by gene expression studies as a possible tumor suppressor for ATLL. Loss of TCF8 resulted in resistance to transforming growth factor β1 (TGF-β1) mediated growth inhibition in ATLL cells, which likely contributed to the pathogenesis of ATLL. The same group in a subsequent study showed localization of the breakpoints at 10p11.2 within the EPC1 locus by SNP array in two cases [123]. EPC1 is a member of the polycomb group gene family and participate in chromatin formation and gene regulation. EPC1/ASXL2 and truncated EPC1 were identified in the two cases respectively, and both were able to induce cellular proliferation in in vitro studies, implicating EPC1 in the pathogenesis of ATLL in some cases. T-cell prolymphocytic leukemia (T-PLL) is an aggressive T-cell leukemia characterized by proliferation of prolymphocytes in the peripheral blood, bone marrow, liver, spleen and lymph nodes. The most frequent genetic change is inversion of chromosome 14. Gains in 6p (3/12), 8q (10/12), and of losses in 6q (5/12), 8p (7/12), 10p (4/12), 11q (3/12), and 18p (3/12) were identified by a 50K SNP array when analyzing 11 T-PLL with inv(14)(q11q32) or t(14;14)(q11;q32) and one T-PLL case without inv(14)/t(14;14) [124]. Recurrent UPD in 3q, and non-recurrent partial UPD on chromosomes 3, 6, 9, 11, and 13 were identified. In a subsequent study of 18 cases of T-PLL by 250K SNP array, Nowak et al. [125] identified abundant copy number alterations, and confirmed the characteristic genetic lesions described before. Recurrent microdeletions targeting microRNA 34b/c, ETS1, and FLI1 were implicated in losses in chromosome 11, and PLEKHA2, NBS1, NOV and MYST3 genes were found to be involved in the breakpoints of chromosome 8. New recurrent lesions were identified on chromosomes 5p, 12p, 13q, 17, and 22 including aUPD on chromosome 17q, with genes DNAH5, ETV6, miR-15a, and miR-16-1, p53, BIRC5, and SOCS3 implicated in the regions. Future studies of the implicated genes identified by SNP array are likely going to further our understanding of the pathogenesis and provide potential targets for therapy. Sézary syndrome (SS) is rare but aggressive neoplasm characterized by erythroderma, generalized adenopathy, and infiltrate of cerebriform Sézary cells in the peripheral blood, skin and lymph nodes. A low resolution 10K SNP array of eight patients with SS identified frequent SNP copy number changes and LOH involving 1, 2p, 3, 4q, 5q, 6, 7p, 8, 9, 10, 11, 12q, 13, 14, 16q, 17, and 20 [126]. SNP copy number loss was most frequent at FAT gene at 4q35 (75%), followed by VEGFC at 4q34.1q34.3 (50%), NFIB at chromosome 12 (38%), and TRIM16 at 17p11.2 (38%). SNP LOH gene clusters at chromosome regions of 9q31q34, 10p11q26, and 13q11q12 were only present in SS but not in normal controls, suggesting their involvement in SS pathology. 11. Conclusions In summary, SNP array studies have contributed significantly in our understanding of the genomics of various hematopoietic malignancies. A summary of the common genomic abnormalities identified by SNP arrays is presented in Table 1. Several oncogenes have been mapped through application of SNP array and were shown to be important in clinical applications. As the cost of next generation sequencing (NGS) continues to drop, NGS will be increasingly applied in clinical labs. As a result, SNP array will gradually phase out. However, as a mature technology with fully developed data analysis and relatively low cost, SNP array will continue to play a role in the clinical lab, especially in situations where diagnostic and prognostic significance of SNP lesions were well established. microarrays-05-00001-t001_Table 1Table 1 Common genetic abnormalities detected by SNP array in hematopoietic malignancies. Disease CNVs/CNAs and/or Associated Genes LOH/UPD and/or Associated Genes Prognostic Association Ref. B-ALL Deletion of PAX5, EBF1, TCF3, LEF1, IKZF1 (IKAROS), IKZF3 (AIOLOS), ETV6, and CDKN2A/p16INK4A 9p (CDKN2A/B) - [26,27,28,29,35] T-ALL Deletion of TAL1, RB1, PTEN, CDKN2A, CDKN2B, LEF1, and STIL; Gains of MYB 9p (CDKN2A) - [32] AML Deletion of 3p14.1–p13 (FOXP1, RYBP, FHIT), 6q27 (RPS6KA2), 8q23.3 (TRPS1), 10q11.21 (HNRPF), 11q25, 12p13.2 (ETV6), 15q21.3 (RFXDC2), 5q31.1 (CTNNA1), 16q22.1 (CBFB), 17p13.1 (TP53), 17q11.2 (NF1) 13q (FLT3), 11p (WT1, PU1) and 11q (MLL), 19q (CEBPA), 6p and 21q (RUNX1) Worse prognosis: ≥2 genomic lesions detected by SNP array [42,43,44,45,47,48,49] Amplifications of 8q23.2 (MYC), 11q23.3 (MLL), and 21q22.2 (ETS2) P53 mutations, or P53 mutations coupled with 17p-LOH; Genomic lesions including aUPD identified by SNP array MDS Deletion and aUPD of chromosomes 1, 5q, 7, 11, 17, and 21 Worse prognosis: UPDs of 7q; New genetic lesions detected by SNP array; EZH2 mutations [57,58,60,61,62,63,64] Deletion of EZH2 and TET2 UPD 20p (BMP2 and TRIB3) Favorable prognosis: TET2 mutations (better response to hypomethylating agents) CML Frequent amplifications in chronic phase 1, 8, 9, 17, 19, and 22 in TKI-resistant CML - [66,67] Deletion of IKFZ1 in lymphoid blast phase MPN Rare in ET and PV 9p (JAK2) Worse prognosis: CNN-LOH on 7q or 9p (JAK2 V617F); Genetic aberrations of chromosome 5, 7, or 17p [68,69,70,73,74] Deletion of 13q14 (RB) or 17q11 (NF1) in PMF CMML Frequent microdeletions 7q (EZH2), 11q (CBL), and 4q (TET2) Worse prognosis: Multiple chromosomal defects detected by SNP array [75,76] cHL Gain of MAP3K14 14q (TRAF3) - [79] DLBCL Frequent gains and deletions; gains HDAC7A on chromosome 12 predominantly in GCB-DLBCL, losses of BACH2 and CASP8AP2 on chromosome 6 predominantly in ABC-DLBCL; Potential tumor suppressor genes: CASP3, IL5RA ARID1B, ROBO2 and MRS1; Potential oncogenes: KLHL6, IL31 and LRP1 11p11.2 (PTPRJ), Worse prognosis: Loss of 8p23.1 [81,82,83,84] FL CDKN2A, CDKN2B, FHIT, KIT, PEX14, and PTPRD 1p36 (TNFRSF14), 6p, 6q, 9p (CDKN2A), 10q, 12q, 16p, and 17p (TP53) Worse prognosis: >3 SNP abnormalities; aUPD and deletion of 1p36, aUPD of 16p [85,86] CLL Deletion of 17p13 (TP53), 11q22 (ATM) and 13q14 (DLEU1 and DLEU2), 2p16.1–2p15, 8q24.21, 6q21 (AIM1) 13q, 13 (miR-15a/miR-16-1), 17p, and 11q Worse prognosis: Genomic complexity; large genomic aberrations; large (type II) 13q14 deletions [91,92,93,101] Gain of 12, 2p16 (REL, BCL11A) MCL Deletion of INK4A/ARF, ATM, TP53, 1p, 6q, CDKN2C, BCL2L11, CDKN2A, and RB1, FAF1, MAP2, SP100, MOBKL2B, ZNF280A, and PRAME. 9p, 9, 17p (TP53) - [104,105] Amplification of MYC, 11q13(cyclin D1), 13q (miR17-92, C13 or f25), dup(3q), 18q (BCL2) MZL Deletion of 6q23 (TNFAIP3, A20), 9p 6q (A20), 3q - [106,107,108] Gain of 3, 18, 6p and 21q Gains of REL, BCL11A, ETS1, PTPN1, PTEN and KRAS in transformation to DLBCL BL Losses of 6q14.1–q22.33, 9p21.3 (CDKN2A), and 13q14.2–q14.3 6p12.2-pter, 9p23-pter, and 17p11.2-pter (TP53). - [109] Gains of 1q23.3–q31.3, 7, 13q31.3, MM Genomic alterations at 1p, 1q, 6q, 8p, 13, and 16q 1q, 16q (CYLD), and X Worse prognosis: Amplifications in 1q and deletions in 1p, 12p, 14q, 16q, and 22q [111,112,115,116,117] UPD of 16q (CYLD) Favorable prognosis: Amplifications of 5, 9, 11, 15, and 19 PTCL, NOS Losses of 1p35-36, 3q, 5q33, 6p22, 6q16, 6q21–22, 8p21–23, 9p21, 10p11–12, 10q11-22, 10q25–26, 13q14, 15q24, 16q22, 16q24, 17p11, 17p13 and Xp22 2q32.3 - [120,121] Gains of 1q32–43, 2p15–16 (REL), 7, 8q24, 11q14–25, 17q11–21 and 21q11–21, 9p and 19q AILT Loss of 3q and 9p 2q32.3 Worse prognosis: The presence of CNAs; overexpression of CARMA1 at 7p22 and MYCBP2 at 13q22 [121] Gains of 8q, 9p and 19q ATLL Deletion of 10p11.2 ( TCF8) - - [122] T-PLL Loss in 6q, 8p, 10p, 11q (microRNA 34b/c, ETS1 and FLI1), and 18p and Gains of 6p, 8q 3q, 17q - [124,125] Aberrations in 5p, 12p, 13q, 17 and 22 (DNAH5, ETV6, miR-15a and miR-16-1, p53, BIRC5, and SOCS3) SS Loss at 4q35 (FAT), 4q34 (VEGFC), 12 (NFIB), and 17p11.2 (TRIM16) 9q31q34, 10p11q26, and 13q11q12 - [126] CNV, copy number variations; CAN, copy number aberrations; LOH, loss of heterozygosity; UPD, uniparental disomy; Ref, reference; B-ALL, B lymphoblastic leukemia; T-ALL, T lymphoblastic leukemia; AML, acute myeloid leukemia; MDS, myelodysplastic syndrome; CML, chronic myelogenous leukemia; MPN, myeloproliferative neoplasm; PV, polycythemia (Rubra) vera; ET, essential thrombocythemia; PMF, primary myelofibrosis; CMML, chronic myelomonocytic leukemia; cHL, classical Hodgkin lymphoma; DLBCL, diffuse large B-cell lymphoma; FL, follicular lymphoma; CLL, chronic lymphocytic leukemia; MCL, mantle cell lymphoma; MZL, marginal zone lymphoma; BL, Burkitt lymphoma; MM, multiple myeloma; PTCL, NOS, peripheral T-cell lymphoma, NOS; AILT, angioimmunoblastic T-cell lymphoma; ATLL, adult T-cell leukemia/lymphoma; T-PLL, T-cell prolymphocytic leukemia; SS, Sezary syndrome. Author Contributions All authors performed literature searches and wrote this review. Conflicts of Interest The authors declare no conflict of interest. ==== Refs References 1. Hahn W.C. Weinberg R.A. Rules for making human tumor cells N. Engl. J. Med. 2002 347 1593 1603 10.1056/NEJMra021902 12432047 2. Bayani J.S.J. Traditional banding of chromosomes for cytogenetic analysis Curr. Protoc. Cell Biol. 2004 S23 10.1002/0471143030.cb2203s23 3. Levsky J.M. Singer R.H. Fluorescence in situ hybridization: Past, present and future J. Cell Sci. 2003 116 2833 2838 10.1242/jcs.00633 12808017 4. Kallioniemi A. Kallioniemi O.P. Sudar D. Rutovitz D. Gray J.W. Waldman F. Pinkel D. Comparative genomic hybridization for molecular cytogenetic analysis of solid tumors Science 1992 258 818 821 10.1126/science.1359641 1359641 5. Du Manoir S. Speicher M.R. Joos S. Schrock E. Popp S. Dohner H. Kovacs G. Robert-Nicoud M. Lichter P. Cremer T. Detection of complete and partial chromosome gains and losses by comparative genomic in situ hybridization Hum. Genet. 1993 90 590 610 10.1007/BF00202476 8444465 6. Pinkel D. Albertson D.G. Comparative genomic hybridization Annu Rev. Genomics Hum. Genet. 2005 6 331 354 10.1146/annurev.genom.6.080604.162140 16124865 7. De Ravel T.J. Devriendt K. Fryns J.P. Vermeesch J.R. What’s new in karyotyping? The move towards array comparative genomic hybridisation (CGH) Eur. J. Pediatr. 2007 166 637 643 10.1007/s00431-007-0463-6 17372759 8. Kennedy G.C. Matsuzaki H. Dong S. Liu W.M. Huang J. Liu G. Su X. Cao M. Chen W. Zhang J. Large-scale genotyping of complex DNA Nat. Biotechnol. 2003 21 1233 1237 10.1038/nbt869 12960966 9. Lindblad-Toh K. Tanenbaum D.M. Daly M.J. Winchester E. Lui W.O. Villapakkam A. Stanton S.E. Larsson C. Hudson T.J. Johnson B.E. Loss-of-heterozygosity analysis of small-cell lung carcinomas using single-nucleotide polymorphism arrays Nat. Biotechnol. 2000 18 1001 1005 10973224 10. Wong K.K. Tsang Y.T. Shen J. Cheng R.S. Chang Y.M. Man T.K. Lau C.C. Allelic imbalance analysis by high-density single-nucleotide polymorphic allele (SNP) array with whole genome amplified DNA Nucleic Acids Res. 2004 32 10.1093/nar/gnh072 15148342 11. La Framboise T. Single nucleotide polymorphism arrays: A decade of biological, computational and technological advances Nucleic Acids Res. 2009 37 4181 4193 10.1093/nar/gkp552 19570852 12. Mardis E.R. Ding L. Dooling D.J. Larson D.E. McLellan M.D. Chen K. Koboldt D.C. Fulton R.S. Delehaunty K.D. McGrath S.D. Recurring mutations found by sequencing an acute myeloid leukemia genome N. Engl. J. Med. 2009 361 1058 1066 10.1056/NEJMoa0903840 19657110 13. Ley T.J. Mardis E.R. Ding L. Fulton B. McLellan M.D. Chen K. Dooling D. Dunford-Shore B.H. McGrath S. Hickenbotham M. DNA sequencing of a cytogenetically normal acute myeloid leukaemia genome Nature 2008 456 66 72 10.1038/nature07485 18987736 14. Stiller C.A. Parkin D.M. Geographic and ethnic variations in the incidence of childhood cancer Br. Med. Bull. 1996 52 682 703 10.1093/oxfordjournals.bmb.a011577 9039726 15. Swerdlow S.H. Campo E. Harris N.L. Jaffe E.S. Pileri S.A. Stein H. Thiele J. Vardiman J.W. Who Classification of Tumours of Haematopoietic and Lymphoid Tissues International Agency for Research on Cancer Lyon, France 2008 167 178 16. Schrappe M. Reiter A. Ludwig W.D. Harbott J. Zimmermann M. Hiddemann W. Niemeyer C. Henze G. Feldges A. Zintl F. Improved outcome in childhood acute lymphoblastic leukemia despite reduced use of anthracyclines and cranial radiotherapy: Results of trial ALL-BFM 90. German-Austrian-Swiss ALL-BFM study group Blood 2000 95 3310 3322 10828010 17. Gaynon P.S. Trigg M.E. Heerema N.A. Sensel M.G. Sather H.N. Hammond G.D. Bleyer W.A. Children’s cancer group trials in childhood acute lymphoblastic leukemia: 1983–1995 Leukemia 2000 14 2223 2233 10.1038/sj.leu.2401939 11187913 18. Harms D.O. Janka-Schaub G.E. Co-operative study group for childhood acute lymphoblastic leukemia (COALL): Long-term follow-up of trials 82, 85, 89 and 92 Leukemia 2000 14 2234 2239 10.1038/sj.leu.2401974 11187914 19. Silverman L.B. Gelber R.D. Dalton V.K. Asselin B.L. Barr R.D. Clavell L.A. Hurwitz C.A. Moghrabi A. Samson Y. Schorin M.A. Improved outcome for children with acute lymphoblastic leukemia: Results of dana-farber consortium protocol 91–01 Blood 2001 97 1211 1218 10.1182/blood.V97.5.1211 11222362 20. Gustafsson G. Schmiegelow K. Forestier E. Clausen N. Glomstein A. Jonmundsson G. Mellander L. Makipernaa A. Nygaard R. Saarinen-Pihkala U.M. Improving outcome through two decades in childhood all in the nordic countries: The impact of high-dose methotrexate in the reduction of CNS irradiation. Nordic society of pediatric haematology and oncology (NOPHO) Leukemia 2000 14 2267 2275 10.1038/sj.leu.2401961 11187918 21. Pui C.H. Sandlund J.T. Pei D. Rivera G.K. Howard S.C. Ribeiro R.C. Rubnitz J.E. Razzouk B.I. Hudson M.M. Cheng C. Results of therapy for acute lymphoblastic leukemia in black and white children JAMA 2003 290 2001 2007 10.1001/jama.290.15.2001 14559953 22. Gokbuget N. Hoelzer D. Recent approaches in acute lymphoblastic leukemia in adults Rev. Clin. Exp. Hematol. 2002 6 114 141 10.1046/j.1468-0734.2002.00068.x 12196212 23. Kantarjian H.M. O’Brien S. Smith T.L. Cortes J. Giles F.J. Beran M. Pierce S. Huh Y. Andreeff M. Koller C. Results of treatment with hyper-CVAD, a dose-intensive regimen, in adult acute lymphocytic leukemia J. Clin. Oncol. 2000 18 547 561 10653870 24. Linker C. Damon L. Ries C. Navarro W. Intensified and shortened cyclical chemotherapy for adult acute lymphoblastic leukemia J. Clin. Oncol. 2002 20 2464 2471 10.1200/JCO.2002.07.116 12011123 25. Greaves M.F. Wiemels J. Origins of chromosome translocations in childhood leukaemia Nat. Rev. Cancer 2003 3 639 649 10.1038/nrc1164 12951583 26. Irving J.A. Bloodworth L. Bown N.P. Case M.C. Hogarth L.A. Hall A.G. Loss of heterozygosity in childhood acute lymphoblastic leukemia detected by genome-wide microarray single nucleotide polymorphism analysis Cancer Res. 2005 65 3053 3058 15833833 27. Mullighan C.G. Goorha S. Radtke I. Miller C.B. Coustan-Smith E. Dalton J.D. Girtman K. Mathew S. Ma J. Pounds S.B. Genome-wide analysis of genetic alterations in acute lymphoblastic leukaemia Nature 2007 446 758 764 10.1038/nature05690 17344859 28. Kawamata N. Ogawa S. Zimmermann M. Kato M. Sanada M. Hemminki K. Yamatomo G. Nannya Y. Koehler R. Flohr T. Molecular allelokaryotyping of pediatric acute lymphoblastic leukemias by high-resolution single nucleotide polymorphism oligonucleotide genomic microarray Blood 2008 111 776 784 10.1182/blood-2007-05-088310 17890455 29. Bungaro S. Dell’Orto M.C. Zangrando A. Basso D. Gorletta T. Lo Nigro L. Leszl A. Young B.D. Basso G. Bicciato S. Integration of genomic and gene expression data of childhood all without known aberrations identifies subgroups with specific genetic hallmarks Genes Chromosomes Cancer 2009 48 22 38 10.1002/gcc.20616 18803328 30. Harrison C.J. Foroni L. Cytogenetics and molecular genetics of acute lymphoblastic leukemia Rev. Clin. Exp. Hematol. 2002 6 91 113 10.1046/j.1468-0734.2002.00069.x 12196211 31. Graux C. Cools J. Michaux L. Vandenberghe P. Hagemeijer A. Cytogenetics and molecular genetics of T-cell acute lymphoblastic leukemia: From thymocyte to lymphoblast Leukemia 2006 20 1496 1510 10.1038/sj.leu.2404302 16826225 32. Mullighan C.G. Williams R.T. Downing J.R. Sherr C.J. Failure of CDKN2A/b (INK4A/b-ARF)-mediated tumor suppression and resistance to targeted therapy in acute lymphoblastic leukemia induced by BCR-ABL Genes Dev. 2008 22 1411 1415 10.1101/gad.1673908 18519632 33. Karrman K. Castor A. Behrendtz M. Forestier E. Olsson L. Ehinger M. Biloglav A. Fioretos T. Paulsson K. Johansson B. Deep sequencing and SNP array analyses of pediatric T-cell acute lymphoblastic leukemia reveal NOTCH1 mutations in minor subclones and a high incidence of uniparental isodisomies affecting CDKN2A J. Hematol. Oncol. 2015 8 10.1186/s13045-015-0138-0 25903014 34. Okamoto R. Ogawa S. Nowak D. Kawamata N. Akagi T. Kato M. Sanada M. Weiss T. Haferlach C. Dugas M. Genomic profiling of adult acute lymphoblastic leukemia by single nucleotide polymorphism oligonucleotide microarray and comparison to pediatric acute lymphoblastic leukemia Haematologica 2010 95 1481 1488 10.3324/haematol.2009.011114 20435627 35. Safavi S. Hansson M. Karlsson K. Biloglav A. Johansson B. Paulsson K. Novel gene targets detected by genomic profiling in a consecutive series of 126 adults with acute lymphoblastic leukemia Haematologica 2015 100 55 61 10.3324/haematol.2014.112912 25261097 36. Trevino L.R. Yang W. French D. Hunger S.P. Carroll W.L. Devidas M. Willman C. Neale G. Downing J. Raimondi S.C. Germline genomic variants associated with childhood acute lymphoblastic leukemia Nat. Genet. 2009 41 1001 1005 10.1038/ng.432 19684603 37. Papaemmanuil E. Hosking F.J. Vijayakrishnan J. Price A. Olver B. Sheridan E. Kinsey S.E. Lightfoot T. Roman E. Irving J.A. Loci on 7p12.2, 10q21.2 and 14q11.2 are associated with risk of childhood acute lymphoblastic leukemia Nat. Genet. 2009 41 1006 1010 10.1038/ng.430 19684604 38. Prasad R.B. Hosking F.J. Vijayakrishnan J. Papaemmanuil E. Koehler R. Greaves M. Sheridan E. Gast A. Kinsey S.E. Lightfoot T. Verification of the susceptibility loci on 7p12.2, 10q21.2, and 14q11.2 in precursor B-cell acute lymphoblastic leukemia of childhood Blood 2010 115 1765 1767 10.1182/blood-2009-09-241513 20042726 39. Thiede C. Steudel C. Mohr B. Schaich M. Schakel U. Platzbecker U. Wermke M. Bornhauser M. Ritter M. Neubauer A. Analysis of FLT3-activating mutations in 979 patients with acute myelogenous leukemia: Association with FAB subtypes and identification of subgroups with poor prognosis Blood 2002 99 4326 4335 10.1182/blood.V99.12.4326 12036858 40. Frohling S. Schlenk R.F. Stolze I. Bihlmayr J. Benner A. Kreitmeier S. Tobis K. Dohner H. Dohner K. CEBPA mutations in younger adults with acute myeloid leukemia and normal cytogenetics: Prognostic relevance and analysis of cooperating mutations J. Clin. Oncol. 2004 22 624 633 10.1200/JCO.2004.06.060 14726504 41. Raghavan M. Lillington D.M. Skoulakis S. Debernardi S. Chaplin T. Foot N.J. Lister T.A. Young B.D. Genome-wide single nucleotide polymorphism analysis reveals frequent partial uniparental disomy due to somatic recombination in acute myeloid leukemias Cancer Res. 2005 65 375 378 15695375 42. Fitzgibbon J. Smith L.L. Raghavan M. Smith M.L. Debernardi S. Skoulakis S. Lillington D. Lister T.A. Young B.D. Association between acquired uniparental disomy and homozygous gene mutation in acute myeloid leukemias Cancer Res. 2005 65 9152 9154 10.1158/0008-5472.CAN-05-2017 16230371 43. Gupta M. Raghavan M. Gale R.E. Chelala C. Allen C. Molloy G. Chaplin T. Linch D.C. Cazier J.B. Young B.D. Novel regions of acquired uniparental disomy discovered in acute myeloid leukemia Genes Chromosomes Cancer 2008 47 729 739 10.1002/gcc.20573 18506749 44. Bullinger L. Kronke J. Schon C. Radtke I. Urlbauer K. Botzenhardt U. Gaidzik V. Cario A. Senger C. Schlenk R.F. Identification of acquired copy number alterations and uniparental disomies in cytogenetically normal acute myeloid leukemia using high-resolution single-nucleotide polymorphism analysis Leukemia 2009 24 438 449 10.1038/leu.2009.263 20016533 45. Walter M.J. Payton J.E. Ries R.E. Shannon W.D. Deshmukh H. Zhao Y. Baty J. Heath S. Westervelt P. Watson M.A. Acquired copy number alterations in adult acute myeloid leukemia genomes Proc. Natl. Acad. Sci. USA 2009 106 12950 12955 10.1073/pnas.0903091106 19651600 46. Radtke I. Mullighan C.G. Ishii M. Su X. Cheng J. Ma J. Ganti R. Cai Z. Goorha S. Pounds S.B. Genomic analysis reveals few genetic alterations in pediatric acute myeloid leukemia Proc. Natl. Acad. Sci. USA 2009 106 12944 12949 10.1073/pnas.0903142106 19651601 47. Parkin B. Erba H. Ouillette P. Roulston D. Purkayastha A. Karp J. Talpaz M. Kujawski L. Shakhan S. Li C. Acquired genomic copy number aberrations and survival in adult acute myelogenous leukemia Blood 2010 116 4958 4967 10.1182/blood-2010-01-266999 20729466 48. Tiu R.V. Gondek L.P. O’Keefe C.L. Huh J. Sekeres M.A. Elson P. McDevitt M.A. Wang X.F. Levis M.J. Karp J.E. New lesions detected by single nucleotide polymorphism array-based chromosomal analysis have important clinical impact in acute myeloid leukemia J. Clin. Oncol. 2009 27 5219 5226 10.1200/JCO.2009.21.9840 19770377 49. Yi J.H. Huh J. Kim H.J. Kim S.H. Kim Y.K. Sohn S.K. Moon J.H. Kim K.H. Won J.H. Mun Y.C. Adverse prognostic impact of abnormal lesions detected by genome-wide single nucleotide polymorphism array-based karyotyping analysis in acute myeloid leukemia with normal karyotype J. Clin. Oncol. 2011 29 4702 4708 10.1200/JCO.2011.35.5719 22084373 50. Tefferi A. Vardiman J.W. Myelodysplastic syndromes N. Engl. J. Med. 2009 361 1872 1885 10.1056/NEJMra0902908 19890130 51. Greenberg P. Cox C. LeBeau M.M. Fenaux P. Morel P. Sanz G. Sanz M. Vallespi T. Hamblin T. Oscier D. International scoring system for evaluating prognosis in myelodysplastic syndromes Blood 1997 89 2079 2088 9058730 52. Greenberg P.L. Tuechler H. Schanz J. Sanz G. Garcia-Manero G. Sole F. Bennett J.M. Bowen D. Fenaux P. Dreyfus F. Revised international prognostic scoring system for myelodysplastic syndromes Blood 2012 120 2454 2465 10.1182/blood-2012-03-420489 22740453 53. Gondek L.P. Tiu R. Haddad A.S. O’Keefe C.L. Sekeres M.A. Theil K.S. Maciejewski J.P. Single nucleotide polymorphism arrays complement metaphase cytogenetics in detection of new chromosomal lesions in mds Leukemia 2007 21 2058 2061 10.1038/sj.leu.2404745 17525728 54. Gondek L.P. Haddad A.S. O’Keefe C.L. Tiu R. Wlodarski M.W. Sekeres M.A. Theil K.S. Maciejewski J.P. Detection of cryptic chromosomal lesions including acquired segmental uniparental disomy in advanced and low-risk myelodysplastic syndromes Exp. Hematol. 2007 35 1728 1738 10.1016/j.exphem.2007.08.009 17920760 55. Mohamedali A. Gaken J. Twine N.A. Ingram W. Westwood N. Lea N.C. Hayden J. Donaldson N. Aul C. Gattermann N. Prevalence and prognostic significance of allelic imbalance by single-nucleotide polymorphism analysis in low-risk myelodysplastic syndromes Blood 2007 110 3365 3373 10.1182/blood-2007-03-079673 17634407 56. Nowak D. Nolte F. Mossner M. Nowak V. Baldus C.D. Hopfer O. Noll S. Thiel E. Wagner F. Hofmann W.K. Genome-wide DNA-mapping of CD34+ cells from patients with myelodysplastic syndrome using 500k SNP arrays identifies significant regions of deletion and uniparental disomy Exp. Hematol. 2009 37 215 224 10.1016/j.exphem.2008.10.012 19135900 57. Heinrichs S. Kulkarni R.V. Bueso-Ramos C.E. Levine R.L. Loh M.L. Li C. Neuberg D. Kornblau S.M. Issa J.P. Gilliland D.G. Accurate detection of uniparental disomy and microdeletions by SNP array analysis in myelodysplastic syndromes with normal cytogenetics Leukemia 2009 23 1605 1613 10.1038/leu.2009.82 19387468 58. Tiu R.V. Gondek L.P. O’Keefe C.L. Elson P. Huh J. Mohamedali A. Kulasekararaj A. Advani A.S. Paquette R. List A.F. Prognostic impact of SNP array karyotyping in myelodysplastic syndromes and related myeloid malignancies Blood 2011 117 4552 4560 10.1182/blood-2010-07-295857 21285439 59. Afable M.G. II Wlodarski M. Makishima H. Shaik M. Sekeres M.A. Tiu R.V. Kalaycio M. O’Keefe C.L. Maciejewski J.P. SNP array-based karyotyping: Differences and similarities between aplastic anemia and hypocellular myelodysplastic syndromes Blood 2011 117 6876 6884 10.1182/blood-2010-11-314393 21527527 60. Langemeijer S.M. Kuiper R.P. Berends M. Knops R. Aslanyan M.G. Massop M. Stevens-Linders E. van Hoogen P. van Kessel A.G. Raymakers R.A. Acquired mutations in TET2 are common in myelodysplastic syndromes Nat. Genet. 2009 41 838 842 10.1038/ng.391 19483684 61. Ernst T. Chase A.J. Score J. Hidalgo-Curtis C.E. Bryant C. Jones A.V. Waghorn K. Zoi K. Ross F.M. Reiter A. Inactivating mutations of the histone methyltransferase gene EZH2 in myeloid disorders Nat. Genet. 2010 42 722 726 10.1038/ng.621 20601953 62. Nikoloski G. Langemeijer S.M. Kuiper R.P. Knops R. Massop M. Tonnissen E.R. van der Heijden A. Scheele T.N. Vandenberghe P. de Witte T. Somatic mutations of the histone methyltransferase gene EZH2 in myelodysplastic syndromes Nat. Genet. 2010 42 665 667 10.1038/ng.620 20601954 63. Bejar R. Lord A. Stevenson K. Bar-Natan M. Perez-Ladaga A. Zaneveld J. Wang H. Caughey B. Stojanov P. Getz G. TET2 mutations predict response to hypomethylating agents in myelodysplastic syndrome patients Blood 2014 124 2705 2712 10.1182/blood-2014-06-582809 25224413 64. Merkerova M.D. Bystricka D. Belickova M. Krejcik Z. Zemanova Z. Polak J. Hajkova H. Brezinova J. Michalova K. Cermak J. From cryptic chromosomal lesions to pathologically relevant genes: Integration of SNP-array with gene expression profiling in myelodysplastic syndrome with normal karyotype Genes Chromosomes Cancer 2012 51 419 428 10.1002/gcc.21927 22250017 65. Mullighan C.G. Miller C.B. Radtke I. Phillips L.A. Dalton J. Ma J. White D. Hughes T.P. le Beau M.M. Pui C.H. BCR-ABl1 lymphoblastic leukaemia is characterized by the deletion of Ikaros Nature 2008 453 110 114 10.1038/nature06866 18408710 66. Khorashad J.S. de Melo V.A. Fiegler H. Gerrard G. Marin D. Apperley J.F. Goldman J.M. Foroni L. Reid A.G. Multiple sub-microscopic genomic lesions are a universal feature of chronic myeloid leukaemia at diagnosis Leukemia 2008 22 1806 1807 10.1038/leu.2008.210 18668129 67. Nowak D. Ogawa S. Muschen M. Kato M. Kawamata N. Meixel A. Nowak V. Kim H.S. Kang S. Paquette R. SNP array analysis of tyrosine kinase inhibitor-resistant chronic myeloid leukemia identifies heterogeneous secondary genomic alterations Blood 2010 115 1049 1053 10.1182/blood-2009-03-210377 19965645 68. Stegelmann F. Bullinger L. Griesshammer M. Holzmann K. Habdank M. Kuhn S. Maile C. Schauer S. Dohner H. Dohner K. High-resolution single-nucleotide polymorphism array-profiling in myeloproliferative neoplasms identifies novel genomic aberrations Haematologica 2010 95 666 669 10.3324/haematol.2009.013623 20015882 69. Rice K.L. Lin X. Wolniak K. Ebert B.L. Berkofsky-Fessler W. Buzzai M. Sun Y. Xi C. Elkin P. Levine R. Analysis of genomic aberrations and gene expression profiling identifies novel lesions and pathways in myeloproliferative neoplasms Blood Cancer J. 2011 1 10.1038/bcj.2011.39 22829077 70. Kawamata N. Ogawa S. Yamamoto G. Lehmann S. Levine R.L. Pikman Y. Nannya Y. Sanada M. Miller C.W. Gilliland D.G. Genetic profiling of myeloproliferative disorders by single-nucleotide polymorphism oligonucleotide microarray Exp. Hematol. 2008 36 1471 1479 10.1016/j.exphem.2008.06.006 18723266 71. Wang L. Swierczek S.I. Lanikova L. Kim S.J. Hickman K. Walker K. Wang K. Drummond J. Doddapaneni H. Reid J.G. The relationship of JAK2(V617F) and acquired upd at chromosome 9p in polycythemia vera Leukemia 2014 28 938 941 10.1038/leu.2014.20 24463469 72. Kilpivaara O. Mukherjee S. Schram A.M. Wadleigh M. Mullally A. Ebert B.L. Bass A. Marubayashi S. Heguy A. Garcia-Manero G. A germline JAK2 SNP is associated with predisposition to the development of JAK2(V617F)-positive myeloproliferative neoplasms Nat. Genet. 2009 41 455 459 10.1038/ng.342 19287384 73. Thoennissen N.H. Krug U.O. Lee D.H. Kawamata N. Iwanski G.B. Lasho T. Weiss T. Nowak D. Koren-Michowitz M. Kato M. Prevalence and prognostic impact of allelic imbalances associated with leukemic transformation of philadelphia chromosome-negative myeloproliferative neoplasms Blood 2010 115 2882 2890 10.1182/blood-2009-07-235119 20068225 74. Rumi E. Harutyunyan A. Elena C. Pietra D. Klampfl T. Bagienski K. Berg T. Casetti I. Pascutto C. Passamonti F. Identification of genomic aberrations associated with disease transformation by means of high-resolution SNP array analysis in patients with myeloproliferative neoplasm Am. J. Hematol. 2011 86 974 979 10.1002/ajh.22166 21953568 75. Jankowska A.M. Makishima H. Tiu R.V. Szpurka H. Huang Y. Traina F. Visconte V. Sugimoto Y. Prince C. O’Keefe C. Mutational spectrum analysis of chronic myelomonocytic leukemia includes genes associated with epigenetic regulation: UTX, EZH2, AND DNMT3A Blood 2011 118 3932 3941 10.1182/blood-2010-10-311019 21828135 76. Dunbar A.J. Gondek L.P. O’Keefe C.L. Makishima H. Rataul M.S. Szpurka H. Sekeres M.A. Wang X.F. McDevitt M.A. Maciejewski J.P. 250k single nucleotide polymorphism array karyotyping identifies acquired uniparental disomy and homozygous mutations, including novel missense substitutions of c-Cbl , in myeloid malignancies Cancer Res. 2008 68 10349 10357 10.1158/0008-5472.CAN-08-2754 19074904 77. Yi J.H. Huh J. Kim H.J. Kim S.H. Kim K.H. Do Y.R. Mun Y.C. Kim H. Kim M.K. Kim T. Genome-wide single-nucleotide polymorphism array-based karyotyping in myelodysplastic syndrome and chronic myelomonocytic leukemia and its impact on treatment outcomes following decitabine treatment Ann. Hematol. 2013 92 459 469 10.1007/s00277-012-1635-7 23262795 78. Flotho C. Steinemann D. Mullighan C.G. Neale G. Mayer K. Kratz C.P. Schlegelberger B. Downing J.R. Niemeyer C.M. Genome-wide single-nucleotide polymorphism analysis in juvenile myelomonocytic leukemia identifies uniparental disomy surrounding the NF1 locus in cases associated with neurofibromatosis but not in cases with mutant RAS or PTPN11 Oncogene 2007 26 5816 5821 10.1038/sj.onc.1210361 17353900 79. Otto C. Giefing M. Massow A. Vater I. Gesk S. Schlesner M. Richter J. Klapper W. Hansmann M.L. Siebert R. Genetic lesions of the TRAF3 and MAP3K14 genes in classical hodgkin lymphoma Br. J. Haematol. 2012 157 702 708 10.1111/j.1365-2141.2012.09113.x 22469134 80. Cozen W. Li D. Best T. Van Den Berg D.J. Gourraud P.A. Cortessis V.K. Skol A.D. Mack T.M. Glaser S.L. Weiss L.M. A genome-wide meta-analysis of nodular sclerosing hodgkin lymphoma identifies risk loci at 6p21.32 Blood 2012 119 469 475 10.1182/blood-2011-03-343921 22086417 81. Scholtysik R. Kreuz M. Hummel M. Rosolowski M. Szczepanowski M. Klapper W. Loeffler M. Trumper L. Siebert R. Kuppers R. Characterization of genomic imbalances in diffuse large B-cell lymphoma by detailed SNP-chip analysis Int. J. Cancer 2015 136 1033 1042 10.1002/ijc.29072 25042405 82. Aya-Bonilla C. Green M.R. Camilleri E. Benton M. Keane C. Marlton P. Lea R. Gandhi M.K. Griffiths L.R. High-resolution loss of heterozygosity screening implicates PTPRJ as a potential tumor suppressor gene that affects susceptibility to non-hodgkin’s lymphoma Genes Chromosomes Cancer 2013 52 467 479 10.1002/gcc.22044 23341091 83. Trifonov V. Pasqualucci L. Dalla Favera R. Rabadan R. Mutcomfocal: An integrative approach to identifying recurrent and focal genomic alterations in tumor samples BMC Syst. Biol. 2013 7 10.1186/1752-0509-7-25 23531283 84. Scandurra M. Mian M. Greiner T.C. Rancoita P.M. De Campos C.P. Chan W.C. Vose J.M. Chigrinova E. Inghirami G. Chiappella A. Genomic lesions associated with a different clinical outcome in diffuse large B-cell lymphoma treated with R-CHOP-21 Br. J. Haematol. 2010 151 221 231 10.1111/j.1365-2141.2010.08326.x 20813005 85. Fitzgibbon J. Iqbal S. Davies A. O’Shea D. Carlotti E. Chaplin T. Matthews J. Raghavan M. Norton A. Lister T.A. Genome-wide detection of recurring sites of uniparental disomy in follicular and transformed follicular lymphoma Leukemia 2007 21 1514 1520 10.1038/sj.leu.2404696 17495976 86. O’Shea D. O’Riain C. Gupta M. Waters R. Yang Y. Wrench D. Gribben J. Rosenwald A. Ott G. Rimsza L.M. Regions of acquired uniparental disomy at diagnosis of follicular lymphoma are associated with both overall survival and risk of transformation Blood 2009 113 2298 2301 10.1182/blood-2008-08-174953 19141865 87. Cheung K.J. Delaney A. Ben-Neriah S. Schein J. Lee T. Shah S.P. Cheung D. Johnson N.A. Mungall A.J. Telenius A. High resolution analysis of follicular lymphoma genomes reveals somatic recurrent sites of copy-neutral loss of heterozygosity and copy number alterations that target single genes Genes Chromosomes Cancer 2010 49 669 681 10.1002/gcc.20780 20544841 88. Cheung K.J. Rogic S. Ben-Neriah S. Boyle M. Connors J.M. Gascoyne R.D. Horsman D.E. SNP analysis of minimally evolved t(14;18)(q32;q21)-positive follicular lymphomas reveals a common copy-neutral loss of heterozygosity pattern Cytogenet. Genome Res. 2012 136 38 43 10.1159/000334265 22104078 89. Cheung K.J. Johnson N.A. Affleck J.G. Severson T. Steidl C. Ben-Neriah S. Schein J. Morin R.D. Moore R. Shah S.P. Acquired TNFRSF14 mutations in follicular lymphoma are associated with worse prognosis Cancer Res. 2010 70 9166 9174 10.1158/0008-5472.CAN-10-2460 20884631 90. Leich E. Salaverria I. Bea S. Zettl A. Wright G. Moreno V. Gascoyne R.D. Chan W.C. Braziel R.M. Rimsza L.M. Follicular lymphomas with and without translocation t(14;18) differ in gene expression profiles and genetic alterations Blood 2009 114 826 834 10.1182/blood-2009-01-198580 19471018 91. Pfeifer D. Pantic M. Skatulla I. Rawluk J. Kreutz C. Martens U.M. Fisch P. Timmer J. Veelken H. Genome-wide analysis of DNA copy number changes and LOH in CLL using high-density SNP arrays Blood 2007 109 1202 1210 10.1182/blood-2006-07-034256 17053054 92. Lehmann S. Ogawa S. Raynaud S.D. Sanada M. Nannya Y. Ticchioni M. Bastard C. Kawamata N. Koeffler H.P. Molecular allelokaryotyping of early-stage, untreated chronic lymphocytic leukemia Cancer 2008 112 1296 1305 10.1002/cncr.23270 18246537 93. Edelmann J. Holzmann K. Miller F. Winkler D. Buhler A. Zenz T. Bullinger L. Kuhn M.W. Gerhardinger A. Bloehdorn J. High-resolution genomic profiling of chronic lymphocytic leukemia reveals new recurrent genomic alterations Blood 2012 120 4783 4794 10.1182/blood-2012-04-423517 23047824 94. Gunnarsson R. Mansouri L. Isaksson A. Goransson H. Cahill N. Jansson M. Rasmussen M. Lundin J. Norin S. Buhl A.M. Array-based genomic screening at diagnosis and during follow-up in chronic lymphocytic leukemia Haematologica 2011 96 1161 1169 10.3324/haematol.2010.039768 21546498 95. Kujawski L. Ouillette P. Erba H. Saddler C. Jakubowiak A. Kaminski M. Shedden K. Malek S.N. Genomic complexity identifies patients with aggressive chronic lymphocytic leukemia Blood 2008 112 1993 2003 10.1182/blood-2007-07-099432 18436738 96. Ouillette P. Collins R. Shakhan S. Li J. Peres E. Kujawski L. Talpaz M. Kaminski M. Li C. Shedden K. Acquired genomic copy number aberrations and survival in chronic lymphocytic leukemia Blood 2011 118 3051 3061 10.1182/blood-2010-12-327858 21795749 97. Schweighofer C.D. Coombes K.R. Majewski T. Barron L.L. Lerner S. Sargent R.L. O’Brien S. Ferrajoli A. Wierda W.G. Czerniak B.A. Genomic variation by whole-genome SNP mapping arrays predicts time-to-event outcome in patients with chronic lymphocytic leukemia: A comparison of CLL and hapmap genotypes J. Mol. Diagn. 2013 15 196 209 10.1016/j.jmoldx.2012.09.006 23273604 98. Mian M. Rinaldi A. Mensah A.A. Rossi D. Ladetto M. Forconi F. Marasca R. Uhr M. Stussi G. Kwee I. Large genomic aberrations detected by SNP array are independent prognosticators of a shorter time to first treatment in chronic lymphocytic leukemia patients with normal fish Ann. Oncol. 2013 24 1378 1384 10.1093/annonc/mds646 23372049 99. Zhang L. Znoyko I. Costa L.J. Conlin L.K. Daber R.D. Self S.E. Wolff D.J. Clonal diversity analysis using SNP microarray: A new prognostic tool for chronic lymphocytic leukemia Cancer Genet. 2011 204 654 665 10.1016/j.cancergen.2011.10.012 22285017 100. Ouillette P. Erba H. Kujawski L. Kaminski M. Shedden K. Malek S.N. Integrated genomic profiling of chronic lymphocytic leukemia identifies subtypes of deletion 13q14 Cancer Res. 2008 68 1012 1021 10.1158/0008-5472.CAN-07-3105 18281475 101. Ouillette P. Collins R. Shakhan S. Li J. Li C. Shedden K. Malek S.N. The prognostic significance of various 13q14 deletions in chronic lymphocytic leukemia Clin. Cancer Res. 2011 17 6778 6790 10.1158/1078-0432.CCR-11-0785 21890456 102. Mosca L. Fabris S. Lionetti M. Todoerti K. Agnelli L. Morabito F. Cutrona G. Andronache A. Matis S. Ferrari F. Integrative genomics analyses reveal molecularly distinct subgroups of B-cell chronic lymphocytic leukemia patients with 13q14 deletion Clin. Cancer Res. 2010 16 5641 5653 10.1158/1078-0432.CCR-10-0151 20947517 103. Gardiner A. Parker H. Glide S. Mould S. Robinson H. Tracy I. Stankovic T. Oscier D. Strefford J. A new minimal deleted region at 11q22.3 reveals the importance of interpretation of diminished fish signals and the choice of probe for ATM deletion screening in chronic lymphocytic leukemia Leuk. Res. 2012 36 307 310 10.1016/j.leukres.2011.08.002 21955805 104. Kawamata N. Ogawa S. Gueller S. Ross S.H. Huynh T. Chen J. Chang A. Nabavi-Nouis S. Megrabian N. Siebert R. Identified hidden genomic changes in mantle cell lymphoma using high-resolution single nucleotide polymorphism genomic array Exp. Hematol. 2009 37 937 946 10.1016/j.exphem.2009.04.012 19477219 105. Bea S. Salaverria I. Armengol L. Pinyol M. Fernandez V. Hartmann E.M. Jares P. Amador V. Hernandez L. Navarro A. Uniparental disomies, homozygous deletions, amplifications, and target genes in mantle cell lymphoma revealed by integrative high-resolution whole-genome profiling Blood 2009 113 3059 3069 10.1182/blood-2008-07-170183 18984860 106. Flossbach L. Holzmann K. Mattfeldt T. Buck M. Lanz K. Held M. Moller P. Barth T.F. High-resolution genomic profiling reveals clonal evolution and competition in gastrointestinal marginal zone B-cell lymphoma and its large cell variant Int. J. Cancer 2013 132 E116 E127 10.1002/ijc.27774 22890838 107. Novak U. Rinaldi A. Kwee I. Nandula S.V. Rancoita P.M. Compagno M. Cerri M. Rossi D. Murty V.V. Zucca E. The NF-κB negative regulator TNFAIP3 (A20) is inactivated by somatic mutations and genomic deletions in marginal zone lymphomas Blood 2009 113 4918 4921 10.1182/blood-2008-08-174110 19258598 108. Takahashi H. Usui Y. Ueda S. Yamakawa N. Sato-Otsubo A. Sato Y. Ogawa S. Goto H. Genome-wide analysis of ocular adnexal lymphoproliferative disorders using high-resolution single nucleotide polymorphism array Investig. Ophthalmol. Vis. Sci. 2015 56 4156 4165 10.1167/iovs.15-16382 26120819 109. Lundin C. Hjorth L. Behrendtz M. Ehinger M. Biloglav A. Johansson B. Submicroscopic genomic imbalances in burkitt lymphomas/leukemias: Association with age and further evidence that 8q24/MYC translocations are not sufficient for leukemogenesis Genes Chromosomes Cancer 2013 52 370 377 10.1002/gcc.22034 23225516 110. Forconi F. Poretti G. Kwee I. Sozzi E. Rossi D. Rancoita P.M. Capello D. Rinaldi A. Zucca E. Raspadori D. High density genome-wide DNA profiling reveals a remarkably stable profile in hairy cell leukaemia Br. J. Haematol. 2008 141 622 630 10.1111/j.1365-2141.2008.07106.x 18397341 111. Rinaldi A. Kwee I. Young K.H. Zucca E. Gaidano G. Forconi F. Bertoni F. Genome-wide high resolution DNA profiling of hairy cell leukaemia Br. J. Haematol. 2013 162 566 569 10.1111/bjh.12393 23692203 112. Walker B.A. Leone P.E. Jenner M.W. Li C. Gonzalez D. Johnson D.C. Ross F.M. Davies F.E. Morgan G.J. Integration of global SNP-based mapping and expression arrays reveals key regions, mechanisms, and genes important in the pathogenesis of multiple myeloma Blood 2006 108 1733 1743 10.1182/blood-2006-02-005496 16705090 113. Agnelli L. Mosca L. Fabris S. Lionetti M. Andronache A. Kwee I. Todoerti K. Verdelli D. Battaglia C. Bertoni F. A SNP microarray and fish-based procedure to detect allelic imbalances in multiple myeloma: An integrated genomics approach reveals a wide gene dosage effect Genes Chromosomes Cancer 2009 48 603 614 10.1002/gcc.20668 19396863 114. Avet-Loiseau H. Li C. Magrangeas F. Gouraud W. Charbonnel C. Harousseau J.L. Attal M. Marit G. Mathiot C. Facon T. Prognostic significance of copy-number alterations in multiple myeloma J. Clin. Oncol. 2009 27 4585 4590 10.1200/JCO.2008.20.6136 19687334 115. Walker B.A. Leone P.E. Chiecchio L. Dickens N.J. Jenner M.W. Boyd K.D. Johnson D.C. Gonzalez D. Dagrada G.P. Protheroe R.K. A compendium of myeloma-associated chromosomal copy number abnormalities and their prognostic value Blood 2010 116 e56 e65 10.1182/blood-2010-04-279596 20616218 116. Kamada Y. Sakata-Yanagimoto M. Sanada M. Sato-Otsubo A. Enami T. Suzukawa K. Kurita N. Nishikii H. Yokoyama Y. Okoshi Y. Identification of unbalanced genome copy number abnormalities in patients with multiple myeloma by single-nucleotide polymorphism genotyping microarray analysis Int. J. Hematol. 2012 96 492 500 10.1007/s12185-012-1171-1 22972171 117. Jenner M.W. Leone P.E. Walker B.A. Ross F.M. Johnson D.C. Gonzalez D. Chiecchio L. Dachs Cabanas E. Dagrada G.P. Nightingale M. Gene mapping and expression analysis of 16q loss of heterozygosity identifies WWOX and CYLD as being important in determining clinical outcome in multiple myeloma Blood 2007 110 3291 3300 10.1182/blood-2007-02-075069 17609426 118. Kim M. Lee S.H. Kim J. Lee S.E. Kim Y.J. Min C.K. Copy number variations could predict the outcome of bortezomib plus melphalan and prednisone for initial treatment of multiple myeloma Genes Chromosomes Cancer 2015 54 20 27 10.1002/gcc.22213 25145975 119. Lopez-Corral L. Sarasquete M.E. Bea S. Garcia-Sanz R. Mateos M.V. Corchete L.A. Sayagues J.M. Garcia E.M. Blade J. Oriol A. SNP-based mapping arrays reveal high genomic complexity in monoclonal gammopathies, from MGUS to myeloma status Leukemia 2012 26 2521 2529 10.1038/leu.2012.128 22565645 120. Hartmann S. Gesk S. Scholtysik R. Kreuz M. Bug S. Vater I. Doring C. Cogliatti S. Parrens M. Merlio J.P. High resolution SNP array genomic profiling of peripheral T cell lymphomas, not otherwise specified, identifies a subgroup with chromosomal aberrations affecting the REL locus Br. J. Haematol. 2010 148 402 412 10.1111/j.1365-2141.2009.07956.x 19863542 121. Fujiwara S.I. Yamashita Y. Nakamura N. Choi Y.L. Ueno T. Watanabe H. Kurashina K. Soda M. Enomoto M. Hatanaka H. High-resolution analysis of chromosome copy number alterations in angioimmunoblastic T-cell lymphoma and peripheral T-cell lymphoma, unspecified, with single nucleotide polymorphism-typing microarrays Leukemia 2008 22 1891 1898 10.1038/leu.2008.191 18633432 122. Hidaka T. Nakahata S. Hatakeyama K. Hamasaki M. Yamashita K. Kohno T. Arai Y. Taki T. Nishida K. Okayama A. Down-regulation of TCF8 is involved in the leukemogenesis of adult T-cell leukemia/lymphoma Blood 2008 112 383 393 10.1182/blood-2008-01-131185 18467597 123. Nakahata S. Saito Y. Hamasaki M. Hidaka T. Arai Y. Taki T. Taniwaki M. Morishita K. Alteration of enhancer of polycomb 1 at 10p11.2 is one of the genetic events leading to development of adult T-cell leukemia/lymphoma Genes Chromosomes Cancer 2009 48 768 776 10.1002/gcc.20681 19484761 124. Durig J. Bug S. Klein-Hitpass L. Boes T. Jons T. Martin-Subero J.I. Harder L. Baudis M. Duhrsen U. Siebert R. Combined single nucleotide polymorphism-based genomic mapping and global gene expression profiling identifies novel chromosomal imbalances, mechanisms and candidate genes important in the pathogenesis of T-cell prolymphocytic leukemia with INV(14)(q11q32) Leukemia 2007 21 2153 2163 10.1038/sj.leu.2404877 17713554 125. Nowak D. le Toriellec E. Stern M.H. Kawamata N. Akagi T. Dyer M.J. Hofmann W.K. Ogawa S. Koeffler H.P. Molecular allelokaryotyping of T-cell prolymphocytic leukemia cells with high density single nucleotide polymorphism arrays identifies novel common genomic lesions and acquired uniparental disomy Haematologica 2009 94 518 527 10.3324/haematol.2008.001347 19278963 126. Mao X. Chaplin T. Young B.D. Integrated genomic analysis of sezary syndrome Genet. Res. Int. 2011 2011 10.4061/2011/980150 22567373

PMC005xxxxxx/PMC5003447.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5010002microarrays-05-00002ArticleIdentification of Critical Region Responsible for Split Hand/Foot Malformation Type 3 (SHFM3) Phenotype through Systematic Review of Literature and Mapping of Breakpoints Using Microarray Data Li Catherine F. Angione Katie Milunsky Jeff M. *Louhelainen Jari Academic EditorCenter for Human Genetics, Cambridge, MA 02139, USA; cli@chginc.org (C.F.L.); kangione@chginc.org (K.A.)* Correspondence: jmilunsky@chginc.org; Tel.: +1-617-492-7083; Fax: +1-617-492-709224 12 2015 3 2016 5 1 208 9 2015 16 12 2015 © 2015 by the authors; licensee MDPI, Basel, Switzerland.2015This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Split hand/foot malformation (SHFM) is a limb malformation with underdeveloped or absent central digital rays, clefts of hands and feet, and variable syndactyly of the remaining digits. There are six types of SHFM. Here, we report a boy with SHFM type 3 having normal 4th and 5th digits, absent 2nd and 3rd digits, and a 4th finger flexion deformity, as well as absent 2nd, 3rd and 4th toes bilaterally. His father, two paternal uncles, and two paternal first cousins have similar phenotype. Chromosome analysis showed a normal male karyotype. A 514 kb gain at 10q24.31–q24.32 (chr10:102,962,134–103,476,346, hg19) was identified using 6.0 Single nucleotide polymorphism (SNP) microarray, resulting in the duplication of nine genes, including BTRC and FBXW4. A detailed systematic review of literature and mapping of breakpoints using microarray data from all reported cases in PubMed and DECIPHER were conducted, and exon 1 of BTRC gene was identified as the critical region responsible for the SHFM3 phenotype. The potential mechanism and future studies of this critical region causing the SHFM3 phenotype are discussed. Split hand/foot Malformation (SHFM3)10q24.31–q24.32duplicationFBXW4BTRC ==== Body 1. Introduction Split-hand/foot malformation (SHFM) is a limb malformation with underdeveloped or absent central digital rays, clefts of hands and feet, and variable syndactyly of the remaining digits. It is a rare condition, affecting 1 in 8500–25,000 newborns. So far, six genetic loci have been identified for non-syndromic SHFM, and classified as SHFM 1 to SHFM6. SHFM1 is at 7q21 (OMIM 183600), SHFM2 at Xq26 (OMIM 313350), SHFM3 at 10q24 due to tandem duplication (OMIM 246560), SHFM4 at 3q27 due to mutations in TP63 gene (OMIM 605289), SHFM5 at 2q31 (OMIM 606708), and SHFM6 at 12q13 due to mutations in WNT10B gene. Here, we report a boy with SHFM type 3, having 514 kb duplication at 10q24.31–q24.32 identified using 6.0 SNP microarray. A literature review of reported cases, mapping of breakpoints using microarray data to identify the critical region responsible for the SHFM3 phenotype, and discussion of the potential mechanism and future studies of this critical region causing the SHFM3 phenotype are presented below. 2. Results and Discussion 2.1. Chromosome Karyotyping and SNP Microarray Results for the Proband High resolution chromosome and 6.0 SNP microarray analyses were performed. The result showed a normal 46,XY karyotype and a 514 kb duplication at 10q24.31-10q24.32 (chr10:102,962,134–103,476,346, hg19) (Figure 1), containing LBX1, FLJ41350, AX747408, BTRC, POLL, DPCD, MIR3158-1, MIR3158-2, and FBXW4 from centromeric to telomeric direction (Figure S1a, and Table S1, Patient 1). Fluorescence in situ hybridization (FISH) confirmation of the microarray aberration was not performed. Microarray and FISH results on parental and other affected family members were not available. Figure 1 SNP microarray result in the proband: arr[hg19] 10q24.31q24.32(102,962,134-103,476,346)×3. Tracks were arranged from top to bottom showing Log2 ratio (blue dots), copy number state (blue dot line), segments with copy number gain (blue), genomic variants (grey) and genes in the region (green), as well as chromosome band and genomic coordinates. Gray line is the center of the region with copy number gain. 2.2. Breakpoint Mapping for SHFM3 Cases 2.2.1. SHFM3 Cases Reported in PubMed PubMed literature review on reported cases was conducted up to 23 July 2015, and there were 14 publications regarding duplication at 10q24 in SHFM3, including seven publications using microarray to define the boundary of the duplication. The genes and exons in the duplicated region are presented from centromeric to telomeric direction (Figure S1a, and Table S1). Tandem duplication at 10q24 in patients with SHFM3 was first reported by de Mollerat et al. [1] in 2003 using Southern, pulsed field gel electrophoresis and dosage analyses. This duplicated region contained a disrupted extra copy of the FBXW4 gene and the entire LBX1 and BTRC genes, known to be involved in limb development. The smallest duplicated region (440 kb) defines the minimal duplicated region common to all patients, which narrows the number of genes included in the duplicated region to genes from LBX1 to a portion of the FBXW4 gene involving a region of about 440–570 kb (Table S1, Patient 26–32). The proximal and distal breakpoints were within a 130 and 80 kb region, respectively. The proximal breakpoints are located at the intergenic region centromeric to LOC159673 (two cases), centromeric to HUG1 (TLX1NB1) (one case), telomeric to TBX1 (one case), telomeric to LOC203696 (centromeric to LBX1) (three cases), respectively. Therefore, the majority of the proximal breakpoints are located at the intergenic region centromeric to LBX1. The distal breakpoints located at 5′ untranslated region (UTR) (two cases), intron 2 (one case) and intron 5 (four cases) of FBXW4 gene, respectively. Therefore, the majority of the distal breakpoints are located at intron 5 of FBXW4 gene. To refine the minimum duplicated region and to further characterize the SHFM3 locus, Kano et al. [2] in 2005 screened 28 non-syndromic SHFM families for tandem genomic duplication of 10q24 by Southern blot and sequence analysis of the FBXW4 gene. Of 28 families, only two showed genomic rearrangements. Representative patients from the two families exhibit typical SHFM phenotypes with symmetrically affected hands and feet. One patient is a familial case with a 512 kb tandem duplication containing genes from LBX1 to FBXW4 (Table S1, Patient 33), which is similar to the proband in this study. The other patient is a sporadic case arising from a de novo, 447 kb duplication of maternal origin, containing genes from LBX1 to a portion of FBXW4 (exon 9–6) (Table S1, Patient 34). Therefore, the proximal breakpoints in both cases are located at the intergenic region centromeric to LBX1, while the distal breakpoints are located at intron 5 of FBXW4 gene in the smaller duplication case, and at 5′UTR of FBXW4 gene in the longer duplication case. Using FISH and genomic DNA quantitative PCR, Lyle et al. [3] in 2006 showed duplication at 10q24 locus in 12 patients with SHFM3, in which seven patients have one duplicated segment, while five patients have two discontinuous duplicated segments. The proximal and distal breakpoint clusters are maximally about 160 kb and about 60 kb, respectively. In patients with only one duplicated segment, the distal breakpoint is between exon 3 and 7 of FBXW4 in all patients, while the proximal breakpoint is between KAZALD1 exon 5 and TLX1NB1 (HUG1) intron 2 in one patient, between TLX1NB1 (HUG1) intron 2 and LBX1 exon 1 in four patients including the daughter in the RK040 family, and from LBX1 exon 1 to the intergenic region between AX747408 and BTRC in two patients including the father in the RK040 family. Therefore, the proximal breakpoint changes when it transmits from the father to the daughter. The duplicated segment contains genes from TLX1NB1 to a portion of FBXW4 (exon 9–7) in one patient; from LBX1 to a portion of FBXW4 (exon 9–7) in four patients including the daughter in the RK040 family; from BTRC to a portion of FBXW4 (exon 9–7) in two patients including the father in the RK040 family (Table S1, Patient 28, and 35 to 40). In patients with two discontinuous duplicated segments, the centromeric duplicated segment contains TLX1NB1 and the telomeric duplicated segment contains genes from LBX1 to a portion of FBXW4 (exon 9–7) in two patients; the centromeric duplicated segment contains TLX1NB1 and TLX1 and the telomeric duplicated segment contains genes from BTRC to a portion of FBXW4 (exon 9–7) in two patients; the centromeric duplicated segment contains genes from LBX1 to a portion of FBXW4 (exon 9–3) and telomeric duplicated segment contains FGF8 in one patient. For the centromeric duplicated region, the proximal breakpoint is between KAZALD1 exon 5 and TLX1NB1 intron 2, and the distal breakpoint is between TLX1 exon 3 and LBX1 exon 1 in four patients; the proximal breakpoint is between TLX1 exon 3 and LBX1 exon 1, and the distal breakpoint is between FBXW4 exon 3 and FGF8 exon 2 in one patient. For the telomeric duplicated region, the proximal breakpoint is between TLX1 exon 3 and LBX1 exon 1, and the distal breakpoint is between exon 3 and 7 of FBXW4 in two patients; the proximal breakpoint is from LBX1 exon 1 to the intergenic region between AX747408 and BTRC, and the distal breakpoint is between exon 3 and 7 of FBXW4 in two patients; the proximal breakpoint is between FBXW4 exon 3 and FGF8 exon 2, and the distal breakpoint is between FGF8 exon 2 and MGEA5 exon 2 in one patient (Table S1, Patient 26, and 29–32). In addition, using RNA quantitative PCR, expression analysis of 13 candidate genes within and flanking the duplicated region showed that BTRC (present in three copies) and SUFU (present in two copies) were overexpressed in SHFM3 patients compared to controls, suggesting that SHFM3 may be caused by overexpression of BTRC and SUFU, both of which are involved in beta-catenin signaling [3]. Abnormal bands were observed using pulsed-field gel electrophoresis in eight cases, all but one being non-syndromic reported by Everman et al. in 2006 [4]. In all but one case, abnormal bands were observed with probes to the genes FBXW4 and BTRC and to the centromeric region approximately 400-kb upstream of FBXW4. In the remaining case, abnormal bands were observed with probes to FBXW4 and BTRC, but not with a probe to the centromeric region. Array comparative genomic hybridization was first performed by Dimitrov et al. [5] in 2010 on seven individuals from four families with SHFM. In the first family, two siblings have distal limb deficiency with micrognathia syndrome (DLDMS). A 532.77 kb triplication at 10q24 was found in the more severely affected brother (Table S1, Patient 2), while duplication of the same region was present in the mildly affected sister (Table S1, Patient 3). This region contains genes from LBX1 to FBXW4. The patient in the second family with a de novo DLDMS has 528.72 kb duplication at 10q24. There are genes from LBX1 to FBXW4 in the duplicated region (Table S1, Patient 4). In the third family, a 597.29 kb duplication at 10q24 was found in the sister with SHFM, the brother with DLDMS, and shown as somatic mosaicism for this duplication in the phenotypically normal mother. The duplicated region contains genes from LINC01514 to FBXW4 (Table S1, Patient 5–7). The patient in the fourth family has syndromic SHFM, who has 658.43 kb duplication at 10q24, which contains genes from a portion of TLX1NB (exon 2–1, breakpoint at intron 2) to a portion of FGF8 (exon 6–2, breakpoint at intron 1) (Table S1, Patient 8). FISH was performed in Patients 2, 3 and 5, while qPCR was conducted in all cases to confirm the array comparative genomic hybridization results. Two affected brothers were found to have a small duplication of approximately 539 kb at 10q24.31-q24.32 through array-based comparative genomic hybridization by Filho et al. [6] in 2011. The duplicated region contains genes from a portion of LINC01514 (exon 6, breakpoint at intron 5) to FBXW4. The patients’ sister and father do not have the duplication, but qPCR showed that the mother’s DNA carries the duplication in 20% of blood lymphocytes (Table S1, Patient 9–11). The microarray aberrations found in both affected brothers were confirmed by qPCR. Dai et al. [7] in 2013 found the duplication at 10q24.31-q24.32 containing two discontinuous DNA fragments using Affymetrix cytogenetic 2.7M array in a SHFM family with four-generation-span. The proband (IV:3) was a three-year-old boy with severe distal deficiency affecting all four limbs. The other three patients exhibited similar phenotypes, although phenotypic variations were observed between affected family members. Triphalangeal thumb was only identified in the boy’s paternal grandfather (II:5), and complete 1/2 toe syndactyly in the boy (IV:3). Similar hypoplasia/agenesis of 1st ray existed in the boy, his father (III:9) and paternal aunt (III:10). No non-limb malformations were identified in any patient. The centromeric duplicated segment is 247 kb in the boy, 257 kb in the boy’s paternal grandfather and paternal aunt, and 259 kb in the boy’s father. The centromeric duplicated segment involves genes from LINC01514 to a portion of BTRC (exon 1, breakpoint at intron 1) in all patients. The proximal breakpoint is at the same location for all patients, while the distal breakpoint is different among the boy, his father, and his paternal grandfather, although it is the same in the paternal grandfather and paternal aunt. Therefore, the distal breakpoint for the centromeric duplicated segment changes when it transmits from the father to the son, however, it stays the same when it transmits from the father to the daughter. The telomeric duplication is 114 kb in the boy and his paternal grandfather, 116 kb in his paternal aunt, and 125 kb in his father. The telomeric duplicated segment encompasses genes from POLL to a portion of FBXW4 (exon 9–2, breakpoint at intron 1) in all patients. The distal breakpoint is at the same location for all patients, while the proximal breakpoint is different among his paternal grandfather, his father and his paternal aunt, but it is the same between the boy and his paternal grandfather. Therefore, the proximal breakpoint for the telomeric duplicated segment changes when it transmits from the father to both his son and daughter (Table S1, Patient 12–15). The microarray aberrations found in all cases were confirmed by qPCR. A two-year-old boy having SHFM of all four limbs with the hands more severely affected than the feet was reported by Ockeloen et al. [8] in 2013. A 250K Affymetrix SNP array analysis showed a 600 kb gain at 10q24.31–q24.32. The duplicated region contains genes from a portion of TLX1NB (part of exon 3–1, breakpoint at exon 3) to a portion of FBXW4 (exon 9–2, breakpoint at intron 1) (Table S1, Patient 16). Affymetrix SNP 6.0 array was used to perform a genome-wide copy number variation scan, and quantitative real-time PCR (qPCR) was applied to validate the identified genomic duplication by Liu et al. [9] in 2014 in order to identify the potential pathogenic mutation in a Chinese family with SHFM. A microduplication of about 560 kb on the chromosome 10q24 was identified, and the qPCR assay confirmed the presence of this microduplication in all the available affected family members (Table S1, Patient 47). To identify genomic aberrations underlying pathogenesis of SHFM in two Chinese families and to provide genetic counseling and prenatal diagnosis for them, Wang et al. [10] used array-based comparative genomic hybridization to analyze both blood and amniotic fluid samples from one of the families and showed a 662.3 kb duplication at 10q24.31–q24.32 (Table S1, Patient 48). The most recent publication was from Chen et al. [11] in 2014 for a patient with SHFM using genome-wide copy number variation SNP microarray, and the tiny copy number variations were verified by real-time fluorescent quantitative PCR. The results of SNP microarray has revealed that the patient has carried a 394 kb duplication at 10q24.31–q24.32, which contains genes from LBX1 to a portion of DPCD (exon 1, breakpoint at intron 1). By real-time fluorescent quantitative PCR, the duplicate area encompassing the pathogenic genes was verified, and duplication in exon 9 of the nearby FBXW4 gene was detected (Table S1, Patient 17). 2.2.2. SHFM3 Cases Reported in DECIPHER Database Searching of DECIPHER database using SHFM3 as a keyword was performed on 23 July 2015, and there were 16 entries. Clinical information was available in 11 cases, and limb anomalies were present in nine cases. Seven patients had duplication at 10q24.31-q24.32. Clinical information was available in five patients, all having limb anomalies with classical SHFM phenotype, except one patient having short palm, small feet and duplication of genes from a portion of BTRC (exon 2–14, breakpoint at intron 1) to a portion of FBXW4 (exon 9–2, breakpoint at intron 1) (Table S1, Patient 18). The inheritance was unknown. In four cases with classical SHFM phenotype, the duplicated segment contains genes from a portion of LINC01514 (exon 3–6, breakpoint at intron 2) to a portion of FBXW4 (exon 9–5, breakpoint at intron 4) having SHFM in the hands only (Table S1, Patient 19), while from LBX1 to a portion of FBXW4 (exon 9–6, breakpoint at intron 5) (Table S1, Patient 20), from a portion of LINC01514 (exon 4 to 6, breakpoint at intron 3) to a portion of FBXW4 (exon 9–5, breakpoint at intron 4) (Table S1, Patient 21) and from LBX1 to a portion of FBXW4 (exon 9 to part of exon 1, breakpoint at exon 1) (Table S1, Patient 22) having SHFM in both hands and feet. One patient with SHFM phenotype had a 1687 kb duplication at 10q23.33-q24.31, which contains 43 genes from a portion of DNMBP (exon 4 to 1, breakpoint at intron 4) to a portion of FBXW4 (exon 9–6, breakpoint at intron 5) (Table S1, Patient 23). The inheritance was unknown. Four patients had the same 3112 kb duplicated region at 10q23.33-q25.1, which contains 82 genes from a portion of DNMBP (exon 4–1, breakpoint at intron 4) to a portion of CNNM2 (exon 1–4, breakpoint at intron 4). SHFM phenotype was found in two patients (Table S1, Patient 24 and 25), but it was unknown in the other two cases. The inheritance was unknown for all four cases. 2.3. Breakpoint Mapping for Cases Without SHFM3 Phenotypes, but Gaining at the SHFM3 Locus 2.3.1. PubMed Case Using array comparative genomic hybridization, Fernández-Jaén et al. [12] in 2014 reported the first clinical case of a 126 kb microduplication at 10q24.31, affecting LINC01514, LBX1, FLJ41350 and AX747408, in a 12 year-old girl with attention problems, dyspraxia, idiopathic congenital scoliosis, and marked atrophy of paravertebral muscles, and her paternal aunt with a severe and progressive myopathy (Table S1, Patient 41). LBX1 plays a cardinal role in neuronal and muscular development in animal models, and has been reported as a candidate gene for idiopathic scoliosis, although its function in humans is unknown. 2.3.2. DECIPHER Cases One patient without classical SHFM phenotype, but having behavioural/psychiatric abnormality, constipation, deep-set eyes, delayed speech and language development, intellectual disability, macrocephaly, pectus excavatum, plagiocephaly, short stature and ventricular septal defect, had a 17,188 kb gain at 10q23.1–q26.11, which contains 243 genes from a portion of IDE (exon 14–1, breakpoint at intron 14) to RNU6-53P (Table S1, Patient 42). The inheritance was unknown. 2.4. Breakpoint Mapping for Cases Having Loss or Mutation at the SHFM3 Locus 2.4.1. PubMed Cases Vergult et al. [13] in 2013 screened a cohort of 54 patients with radial ray deficiencies (RRDs) for genomic aberrations using array comparative genomic hybridization. In eight of 54 cases, an aberration was detected, including an 80.2 kb microdeletion at 10q24.32 in a female with absence of right radius and thumb, shorter right ulna, left thenar hypoplasia and small apical ventricular septal defect at birth (Table S1, Patient 43). This deletion region contains a portion of DPCD (exon 2–6, breakpoint at intron 1), the entire MIR3158-1, MIR3158-2, and a portion of FBXW4 (exon 9–4, breakpoint at intron 3). Molecular analysis of the parents revealed that the deletion was inherited from the unaffected mother. Since reduced penetrance, as also seen with the 10q24.3 duplications, may be possible in this case, the author postulated that the 10q24.3 deletion may result in RRDs by deletion of one or more specific regulatory regions. To test this hypothesis, six conserved noncoding elements (≥350 bp and sharing ≥90% identity with the mouse and/or ≥100 bp and sharing ≥70% identity with the frog and/or zebrafish) were selected in the deleted region on 10q24.3 based on literature review and on the use of the ECR (available online: http://ecrbrowser.dcode.org/), Ancora (available online: http://ancora.genereg.net/) and UCSC Genome Browser (available online: http://genome.ucsc.edu/), and were sequenced in the entire patient cohort. In a male patient with bilateral radius dysplasia, an A to G transition at genomic position 103380009 (GRCh 37, hg 19) (g.103380009A>G) was detected that was neither a single-nucleotide polymorphism nor present in a control cohort of 96 samples matched to the ethnicity of this patient or in the 1000 Genomes Project data (available online: http://browser.1000genomes.org). For this patient, no causal aberrations were detected using array CGH (Table S1, Patient 44). Unfortunately, parental DNA was unavailable. This A→G substitution could account for the RRDs seen in this patient, but functional studies need to be performed. Sequence analysis of the exons of FBXW4 did not reveal causal mutations in the patient with the heterozygous FBXW4 deletion or in any other patient of this cohort. This substitution is located at intron 6 of FBXW4. Although duplications in the 10q24.3 region result in split hand-foot malformations, this publication indicates that deletions may cause radial ray defects. However, the question of whether the deletion of the FBXW4 gene itself or the presence of mutations in the flanking conserved noncoding regions are responsible for the limb defect remains open. 2.4.2. DECIPHER Cases Two male patients had loss or deletion at 10q24.31–q24.32. The first patient with short phalanx of finger, abnormality of the kidney, cleft palate, intellectual disability, microphthalmos, non-midline cleft lip, pulmonic stenosis, sensorineural hearing impairment, short stature, soft skin, truncal obesity and vesicoureteral reflux had a 546.42 kb loss at 10q24.31–q24.32, which contains 13 genes from LINC01514 to a portion of FBXW4 (exon 9–2, breakpoint at intron 1) (Table S1, Patient 45). The inheritance was unknown. The second patient with aortic dilatation, median cleft palate, Pierre-Robin sequence, renal hypoplasia and secundum atrial septal defect had a de novo 389.02 kb deletion at 10q24.31–q24.32, which contains genes from BTRC to a portion of FBXW4 (exon 9–6, breakpoint at intron 5) (Table S1, Patient 46). 2.5. Discussion Among the 48 cases presented above, 25 cases with microarray data have SHFM phenotypes (Table S1, Patients 1–25), 15 cases without microarray data have SHFM phenotypes (Table S1, Patients 26–40), two cases without SHFM phenotypes have gain at the SHFM3 locus (Table S1, Patients 41–42), four cases have loss or mutation at the SHFM3 locus (Table S1, Patients 43–46), and two cases with SHFM3 phenotypes and microarray data with duplication at the SHFM3 locus do not have breakpoint information (Table S1, Patient 47–48). 2.5.1. Genotype and Phenotype Correlation To explore the genotype and phenotype correlation, clinical presentation and genomic aberration in all cases were compiled in Table S1. In cases having microarray data, the shortest duplication is 126 kb at 10q24.31 resulting in the duplication of LINC01514, LBX1, FLJ41350, and AX747408 from centromeric to telomeric direction in Patient 41 who does not have the SHFM phenotype, but has attention problems, dyspraxia, idiopathic congenital scoliosis, and marked atrophy of paravertebral muscles, and her paternal aunt has a severe and progressive myopathy. A 241.59 kb duplication at 10q24.31-q24.32 containing a portion of BTRC (exon 2–14, breakpoint at intron 1), the entire POLL, DPCD, MIR3158-1, MIR3158-2, and a portion of FBXW4 (exon 9–2, breakpoint at intron 1) was found in Patient 18 who does not have classical SHFM phenotype, but has short palm and small feet, hypertelorism, intellectual disability, and obesity. The shortest duplication resulting in classical SHFM phenotype is 394 kb at 10q24.31–q24.32 containing LBX1, FLJ41350, AX747408, BTRC, POLL, and a portion of DPCD (exon 1, breakpoint at intron 1) from centromeric to telomeric direction in Patient 17. By real-time fluorescent quantitative PCR, the duplicate area encompassing the pathogenic genes was verified, and duplication in exon 9 of the nearby FBXW4 gene was detected. In the majority of cases with SHFM phenotype, the duplication contains LBX1, FLJ41350, AX747408, BTRC, POLL, DPCD, MIR3158-1, MIR3158-2, and FBXW4 or a portion of FBXW4 from centromeric to telomeric direction. Some cases have a longer duplication region, such as a 1687 kb duplication at 10q23.33-q24.31 resulting in the duplication of 43 genes from a portion of DNMBP (exon 4–1, breakpoint at intron 4) to a portion of FBXW4 (exon 9–6, breakpoint at intron 5) in Patient 23 who has the SHFM phenotype, and a 3112 kb duplication at 10q23.33-q25.1 resulting in the duplication of 82 genes from a portion of DNMBP (exon 4–1, breakpoint at intron 4) to a portion of CNNM2 (exon 1–4, breakpoint at intron 4) in Patient 24 and 25, who have the SHFM phenotype. However, a 17,188 kb duplication at 10q23.1-q26.11 containing 243 genes from a portion of IDE (part of exon 15 to exon 1, breakpoint at exon 15) to RNU6-53P was found in Patient 42 who does not have the SHFM phenotype, but has behavioural/psychiatric abnormality, constipation, deeply set eyes, delayed speech and language development, intellectual disability, macrocephaly, pectus excavatum, plagiocephaly, short stature, and a ventricular septal defect. In SHFM cases with two discontinuous duplications at 10q24.31–q24.32 (Patient 12–15), the centromeric duplicated segment contains LINC01514, LBX1, FLJ41350, AX747408, and a portion of BTRC (exon 1, breakpoint at intron 1) from centromeric to telomeric direction, and the telomeric duplicated segment contains POLL, DPCD, MIR3158-1, MIR3158-2, and a portion of FBXW4 (exon 9–2, breakpoint at intron 1) from centromeric to telomeric direction. In SHFM cases without microarray data, the shortest duplication contains BTRC, POLL, DPCD, MIR3158-1, MIR3158-2, and a portion of FBXW4 (exon 9–7), which was found in Patient 35 and 40. The majority of SHFM cases have duplication of LBX1, BTRC, POLL, DPCD, MIR3158-1, MIR3158-2, and FBXW4 or a portion of FBXW4, which is the same as what was found in the cases having microarray data. In SHFM cases with two discontinuous duplications at 10q24, the centromeric duplicated segment containing TLX1NB, and a telomeric duplicated segment containing LBX1, BTRC, POLL, and a portion of FBXW4 (exon 9–7) was found in Patient 26 and 32. Patient 29 has a centromeric duplicated segment containing LBX1, BTRC, POLL, and a portion of FBXW4 (exon 9–3), and a telomeric duplicated segment containing FGF8, who has normal hands, but cleft and syndactyly of feet. Patients 30 and 31 have a centromeric duplicated segment containing TLX1NB and TLX1 and a telomeric duplicated segment containing BTRC, POLL, DPCD, MIR3158-1, MIR3158-2, and a portion of FBXW4 (exon 9–7). The phenotype of Patient 30 is unknown, but cleft hands and feet, oligodactyly and syndactyly were found in the affected family members. Patient 31 has duplication of a hand digit, triphalangeal thumb, and cleft, syndactyly and oligodactyly of feet. The majority of the SHFM3 patients have anomalies in all four limbs, but Patient 19 has anomalies in the hands only, while Patients 29, 37 and 38 have anomalies in the feet only. In most cases with the SHFM phenotypes, the duplicated segment is at 10q24.31-q24.32 from the intergenic region centromeric to LBX1 to the intergenic region telomeric to FBXW4 (Figure S1), while the duplication in Patient 19 is from LINC0154 (exon 3–6, breakpoint at intron 2) to FBXW4 (exon 9–5, breakpoint at intron 4), Patient 29 has two duplicated segments, one containing genes from LBX1 to FBXW4 (exon 9–3) and the other containing FGF8, Patients 37 and 38 are from the same family having duplication from LBX1 to FBXW4 (exon 9–7). Although there was no clear genotype and phenotype correlation found, the critical region responsible for the SHFM phenotype was identified and presented below. 2.5.2. Critical Region Responsible for the SHFM Phenotype To define the critical region resulting in the SHFM phenotype, the breakpoints of all cases having gain at the SHFM3 locus (Patient 1–42) were mapped, and the sizes of aberrations were compared, aligned and presented from centromeric to telomeric direction in Figure S1b. Patient 41 without the SHFM phenotype has duplication from LINC01514 to AX747408, and Patient 18 without classical SHFM phenotype has duplication from a portion of BTRC (exon 2–14, breakpoint at intron 1) to a portion FBXW4 (exon 9–2, breakpoint at intron 1); while Patients 12–15 with SHFM phenotypes have two duplicated segments, one from LINC01514 to a portion of BTRC (exon 1, breakpoint at intron 1), and the other from POLL to a portion FBXW4 (exon 9–2, breakpoint at intron 1). Therefore, the critical region responsible for the SHFM phenotypes is most likely located at exon 1 of BTRC gene, which is duplicated in all cases with SHFM phenotypes. Patient 42 has a 17188 kb duplication at 10q23.1–q26.11 containing 243 genes, including a portion of IDE (part of exon 15–1, breakpoint at exon 15) at the most centromeric region and RNU6-53P at the most telomeric region, which includes exon 1 of BTRC. However, Patient 42 does not have the SHFM phenotype. The reason is that large duplications may be associated with other phenotypes due to other pathomechanisms. There is only one copy number variation (CNV) overlapping BTRC exon 1 in the normal population (Figure S2), and it was observed as duplication in 4 out of 771 samples analyzed, located at 10q24.32 (chr10:103054982–103452645, hg 19) and containing genes from BTRC to a portion of FBXW4 (exon 9–2) [14]. Therefore, the frequency of this duplication in the normal population is about 0.5%, which is very low. In addition, the four normal individuals having this duplication may be mosaic or have a more subtle SHFM phenotype, which remains unknown as detailed clinical information is not available. 3. Materials and Methods 3.1. Patient’s Phenotypes Patient FH (Figure 2, III:4), a five year old male, presented to us for an initial syndromic evaluation. He was born by normal spontaneous vaginal delivery at term to non-consanguineous parents of Hispanic descent following an uncomplicated antenatal history. He had mild fine motor delay and otherwise normal growth and development. He was generally healthy, aside from exercise-induced asthma and recurrent reflux and constipation. On his left hand, FH was noted to have normal 4th and 5th digits, absent 2nd and 3rd digits, and camptodactyly of his thumb. On his right hand, he had a normal 5th digit, a 4th finger flexion deformity, absent 2nd and 3rd digits, and thumb camptodactyly (Figure 3a,b). He had pes planus, laterally deviated halluces, and absent 2nd, 3rd, and 4th toes bilaterally (Figure 3c). Other minor dysmorphic features included epicanthal folds, a bulbous nasal tip, anteverted nares, posteriorly rotated ears, a high arched palate, micrognathia, mild cubitus valgus, and a mild pectus excavatum deformity. His father (Figure 2, II:5) was observed to have similar hand and foot malformations. On his right hand, he had 5th finger campotodactyly, 4th finger camptodactyly and metacarpal aplasia, and rudimentary 2nd and 3rd digits. His right thumb was not fully formed, with anonychia and camptodacyly. On his left hand, he was noted to have mild 4th and 5th finger camptodactyly, rudimentary 2nd and 3rd digits, and thumb camptodactyly (Figure 4a,b). On his right foot, his hallux was curved inward, his 2nd and 3rd toes were missing with a cleft, and his 4th and 5th toes were fused together and curved inward. On his left foot, his hallux was curved inward perpendicular to his foot, his 2nd and 3rd toes were missing with a cleft, and his 4th and 5th toes were normal with a common base (Figure 4c). He was also noted to have retrognathia and several dental fillings. Detailed pedigree analysis revealed that similar hand and foot abnormalities were also present in two of FH’s three paternal uncles (Figure 2, II:2 and II:3), as well as two paternal first cousins (Figure 2, III:1 and III:2). Figure 2 Pedigree of FH family: II:5 was examined clinically, and III:4 was examined clinically and genetically. Hollow box = normal male; black box = affected male; hollow circle = normal female; black circle = affected female; arrow = proband. Figure 3 (a) FH palms; (b) FH hands; (c) FH feet. Figure 4 (a) Father’s palms; (b) Father’s hands; (c) Father’s feet. 3.2. High Resolution Chromosome Analysis Peripheral blood in heparin was obtained after informed consent, and high resolution chromosome analysis (band level > 650) was performed according to standard protocols. 3.3. Single Nucleotide Polymorphism (SNP) Microarray Analysis A peripheral blood specimen was obtained after informed consent, and genomic DNA was extracted according to standard protocols. Single nucleotide polymorphism (SNP) microarray analysis was conducted using 6.0 SNP microarray from Affymetrix, which contains approximately 900,000 SNPs and 940,000 copy number probes throughout the human genome. The sample was analyzed at a resolution of 25 probes per 50 kb for the known microdeletion or microduplication syndromes and subtelomeric regions. The remaining genome was analyzed at a resolution of 50 probes per 200 kb. Affymetrix Genotyping Console was used to analyze the data. The practical resolution was set at ≥500 kb for duplications and ≥200 kb for deletions. 3.4. Database Search of Reported Cases PubMed search for all available publications on SHFM3 was performed, as well as search on reported cases in the DECIPHER database using SHFM3 as the keyword. All PubMed cases after discovery of a tandem duplication at 10q24 in patients with SHFM3 in 2003 and all DECIPHER cases having clinical information were reviewed and presented below. 3.5. Mapping of Breakpoints Breakpoints were mapped to the genomic location using available genomic information from the publications, and inputting them into UCSC Genome Browser (available online: http://genome.ucsc.edu/) with hg 18 or 19 assembly according to the assembly used in the publication to find the genetic coordinate for all cases reported in PubMed and DECIPHER databases. 4. Conclusions Duplication of sequence in the exon 1 of BTRC is responsible for the formation of SHFM3 phenotype, which may be via cis-acting or trans-acting effects on genes or regulatory sequences involved in the limb development pathway. The etiology of SHFM3 is due to disruption of the limb apical ectodermal ridge (AER). The SHFM3 locus at 10q24 is conserved in vertebrates from zebrafish to human, especially the region from TLX1 to FGF8. AER is characterized by an unexpected regulatory complexity, with at least five distinct enhancers in the intragenic region of FBXW4 (CE58, CE59, CE61 and CE66) and between FBXW4 and FGF8 (CE80) being autonomously active in the mouse embryo. In the human genome, the highest density of conserved noncoding elements is found in BTRC. Conserved noncoding elements can act as developmental enhancers with varying degrees of reproducibility. Extensive study of the regulatory architecture of this region showed that this region behaves as an integrated unit, a holo-enhancer: the internal organization of this intricate 200 kb interval has in itself an important role in integrating and filtering the activities of the multiple regulatory modules present within this region into a restricted tissue- and gene-specific output. This regulatory system may contribute to evolution of gene expression and account for the SHFM3 phenotypic consequences of genomic structural variants found in humans [15,16]. However, the above results were drawn from examination of the conserved noncoding elements. Future studies with an animal model containing sequence from the coding region, such as the sequence from exon 1 of BTRC, may explore its effect on AER and confirm our finding from mapping of the clinical SHFM3 cases. In addition, application of next generation sequencing, such as single molecular real-time (SMRT) sequencing, may identify the exact breakpoints at the nucleotide level to further refine the mapping of the critical region involved and the mechanism of rearrangements found in the SHFM3 cases, especially in those cases with two discontinuous duplications, as well as to verify the change of breakpoints if transmitted by the father. Furthermore, simultaneous epigenetic characterization during SMRT sequencing in cases with reduced penetrance may verify whether epigenetics is also involved in the pathogenesis of SHFM3, as has been found in the mouse model of this disorder [17]. Acknowledgments We acknowledge the family for the permission to reproduce photographs for publication, and medical laboratory technologists at the Center for Human Genetics for performing the SNP microarray and karyotyping of the proband. Supplementary Materials The following are available online at http://www.mdpi.com/2076-3905/5/1/2/s1, Figure S1: Genetic landscape at SHFM3 locus, Figure S2: CNVs overlapping BTRC in normal population: Only variant esv2750847 (Pinto2007) shows as duplication overlapping exon 1 of BTRC. Table S1: Genotype and phenotype correlation. Click here for additional data file. Author Contributions Catherine F. Li has analyzed the SNP microarray data, mapped the breakpoints on all reported cases, and drafted the manuscript. Katie Angione has drafted the clinical information of the proband and his father. Jeff M. Milunsky has evaluated the proband and his father, performed physical examination and taken the photographs of the anomalies, and reviewed the manuscript. Conflicts of Interest The authors declare no conflict of interest. ==== Refs References 1. De Mollerat X.J. Gurrieri F. Morgan C.T. Sangiorgi E. Everman D.B. Gaspari P. Amiel J. Bamshad M.J. Lyle R. Blouin J.L. A genomic rearrangement resulting in a tandem duplication is associated with split hand-split foot malformation 3 (SHFM3) at 10q24 Hum. Mol. Genet. 2003 12 1959 1971 10.1093/hmg/ddg212 12913067 2. Kano H. Kurosawa K. Horii E. Ikegawa S. Yoshikawa H. Kurahashi H. Toda T. Genomic rearrangement at 10q24 in non-syndromic split-hand/split-foot malformation Hum. Genet. 2005 118 477 483 10.1007/s00439-005-0074-0 16235095 3. Lyle R. Radhakrishna U. Blouin J.L. Gagos S. Everman D.B. Gehrig C. Delozier-Blanchet C. Solanki J.V. Patel U.C. Nath S.K. Split-hand/split-foot malformation 3 (SHFM3) at 10q24, development of rapid diagnostic methods and gene expression from the region Am. J. Med. Genet. A 2006 140 1384 1395 10.1002/ajmg.a.31247 16691619 4. Everman D.B. Morgan C.T. Lyle R. Laughridge M.E. Bamshad M.J. Clarkson K.B. Colby R. Gurrieri F. Innes A.M. Roberson J. Frequency of genomic rearrangements involving the SHFM3 locus at chromosome 10q24 in syndromic and non-syndromic split-hand/foot malformation Am. J. Med. Genet. A 2006 140 1375 1383 10.1002/ajmg.a.31246 16761290 5. Dimitrov B.I. de Ravel T. van Driessche J. de Die-Smulders C. Toutain A. Vermeesch J.R. Fryns J.P. Devriendt K. Debeer P. Distal limb deficiencies, micrognathia syndrome, and syndromic forms of split hand foot malformation (SHFM) are caused by chromosome 10q genomic rearrangements J. Med. Genet. 2010 47 103 111 10.1136/jmg.2008.065888 19584065 6. Filho A.B. Souza J. Faucz F.R. Sotomaior V.S. Dupont B. Bartel F. Rodriguez R. Schwartz C.E. Skinner C. Alliman S. Somatic/gonadal mosaicism in a syndromic form of ectrodactyly, including eye abnormalities, documented through array-based comparative genomic hybridization Am. J. Med. Genet. A 2011 155A 1152 1156 21485001 7. Dai L. Deng Y. Li N. Xie L. Mao M. Zhu J. Discontinuous microduplications at chromosome 10q24.31 identified in a Chinese family with split hand and foot malformation BMC Med. Genet. 2013 14 10.1186/1471-2350-14-45 23596994 8. Ockeloen C.W. Cobben J.M. Marcelis C.L. Koolen D.A. A rare complex malformation of the hand in split hand foot malformation type 3 (SHFM3) Clin. Dysmorphol. 2013 22 106 108 10.1097/MCD.0b013e328363025c 23722700 9. Liu Y. Huang Y. Yang W. Zhang X. Identification of a pathogenic microduplication in a Chinese split-hand/split-foot malformation family Chinese J. Med. Genet. 2014 31 276 279 (In Chinese) 10. Wang H. Xie J. Chen W. Geng Q. Xu X. Genetic analysis and prenatal diagnosis of two Chinese families with split hand foot malformation Chinese J. Med. Genet. 2014 31 280 284 (In Chinese) 11. Chen Y. Li H. Tang S. Hu T. Du J. Analysis of genomic copy number variation for a Chinese patient with split hand/split foot malformation Chinese J. Med. Genet. 2014 31 774 777 (In Chinese) 12. Fernández-Jaén A. Suela J. Fernández-Mayoralas D.M. Fernández-Perrone A.L. Wotton K.R. Dietrich S. Castellanos Mdel C. Cigudosa J.C. Calleja-Pérez B. López-Martín S. Microduplication 10q24.31 in a Spanish girl with scoliosis and myopathy: The critical role of LBX Am. J. Med. Genet. A 2014 164A 2074 2078 10.1002/ajmg.a.36589 24782348 13. Vergult S. Hoogeboom A.J. Bijlsma E.K. Sante T. Klopocki E. de Wilde B. Jongmans M. Thiel C. Verheij J.B. Perez-Aytes A. Complex genetics of radial ray deficiencies: Screening of a cohort of 54 patients Genet. Med. 2013 15 195 202 10.1038/gim.2012.120 22995989 14. Pinto D. Marshall C. Feuk L. Scherer S.W. Copy-number variation in control population cohorts Hum. Mol. Genet. 2007 16 R168 R173 10.1093/hmg/ddm241 17911159 15. Komisarczuk A.Z. Kawakami K. Becker T.S. Cis -regulation and chromosomal rearrangement of the Fgf8 locus after the teleost/tetrapod split Dev. Biol. 2009 336 301 312 10.1016/j.ydbio.2009.09.029 19782672 16. Marinić M. Aktas T. Ruf S. Spitz F. An integrated holo-enhancer unit defines tissue and gene specificity of the Fgf8 regulatory landscape Dev. Cell 2013 24 530 542 10.1016/j.devcel.2013.01.025 23453598 17. Kono H. Kurahashi H. Toda T. Genetically regulated epigenetic transcriptional activation of retrotransposon insertion confers mouse dactylaplasia phenotype Proc. Natl. Acad. Sci. USA 2007 104 19034 19039 10.1073/pnas.0705483104 17984064

PMC005xxxxxx/PMC5003448.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5010003microarrays-05-00003ReviewGlycoarray Technologies: Deciphering Interactions from Proteins to Live Cell Responses Puvirajesinghe Tania M. 1234Turnbull Jeremy. E. 5*Negrini Massimo Academic Editor1 CRCM (Centre de Recherche en Cancérologie de Marseille), Cell Polarity, Cell Signalling and Cancer “Equipe labellisée Ligue Contre le Cancer”, Inserm, U1068, Marseille F-13009, France; taniap@liverpool.ac.uk2 Institut Paoli-Calmettes, Marseille F-13009, France3 Aix-Marseille Université, Marseille F-13284, France4 CNRS (Centre National de la Recherche Scientifique), UMR7258, Marseille F-13009, France5 Centre for Glycobiology, Department of Biochemistry, Institute of Integrative Biology, University of Liverpool, Liverpool L69 7ZB, UK* Correspondence: j.turnbull@liverpool.ac.uk; Tel.: +44-(0)151-795-442704 1 2016 3 2016 5 1 322 10 2015 14 12 2015 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/4.0/).Microarray technologies inspired the development of carbohydrate arrays. Initially, carbohydrate array technology was hindered by the complex structures of glycans and their structural variability. The first designs of glycoarrays focused on the HTP (high throughput) study of protein–glycan binding events, and subsequently more in-depth kinetic analysis of carbohydrate–protein interactions. However, the applications have rapidly expanded and now achieve successful discrimination of selective interactions between carbohydrates and, not only proteins, but also viruses, bacteria and eukaryotic cells, and most recently even live cell responses to immobilized glycans. Combining array technology with other HTP technologies such as mass spectrometry is expected to allow even more accurate and sensitive analysis. This review provides a broad overview of established glycoarray technologies (with a special focus on glycosaminoglycan applications) and their emerging applications to the study of complex interactions between glycans and whole living cells. glycomicsglycobioarraysglycoconjugatessaccharide librariesheparan sulfate ==== Body 1. Introduction Carbohydrates are a major group of biomolecules, which can be subdivided into smaller families of molecules categorized by their structures. In natural states, carbohydrates are found conjugated to other biomolecules including proteins and lipids. Glycolipids are carbohydrates covalently attached to lipids and consist of important constituents including glycosphingolipids (GSLs), diacylglycolipids (DAGs) and lipopolysaccharides (LPs). Glycoproteins are formed when carbohydrates are conjugated to proteins via a serine or threonine residue (O-linked) or an asparagine residue (N-linked) depending on their distinct biosynthetic pathways. Proteoglycans represent a class of glycoconjugates richly dense in carbohydrates structures, which are generally long, unbranched molecules including glycosaminoglycans (GAGs), attached to a serine residue within the core protein via a xylose residue [1]. Free oligosaccharides also exhibit significant biological roles. One important example is milk glycans, whose expression pattern is highly regulated and controlled [2]. Bacterial polysaccharides and viral polysaccharides are another important form of biologically important glycan structures. Transient glycan variation has been documented in key physiological events including roles in pregnancy, lactation, infection, or acute phase response, and tissue and cell development [3]. The glycome describes the total collection of glycans synthesized by a cell, tissue, or organism under specified conditions of time, space, and environment [4]. Glycans vary in structure due to the high number of possible structural modifications, variable chain length of glycan chains and the fact that the biosynthetic process is not template driven and does not go to completion [5]. However, only a small proportion of each type of glycan structure has been found in mammalian systems. Therefore, the assumption is that nature does not require all possible glycan structures to function [6]. This has important consequences for the rationale design of glycan structures for glycan array platforms. Other important factors to consider with glycans are the flexibility of glycans due to anomericity; the ability of glycans to tightly cluster together to form three-dimensional internal structures (as documented from X-ray crystal structures) [7]. Introduction of array-based technology has significantly advanced the field of biology. In 1998, DNA microarrays heralded the introduction of an array platform, offering the simultaneous and high-throughput analysis of immobilized DNA molecules. Glycomic analyses seek to understand how a collection of glycans fulfil a range of particular biological functions. Therefore array-based technologies are particularly suited to the study of carbohydrate structures, providing the two main benefits, the need for low quantities of samples and achieving high-throughput parallel screening. Indeed this provided the rationale for the creation of the Consortium for Functional Glycomics (CFG) in 2001, which is now a widely used web-resource aimed at providing researchers with data, tools, resources, and information about community activities in the growing field of functional glycomics. The use of microarrays for the study of carbohydrate interactions has lagged behind other biological molecules, with scientific literature appearing only in 2002 [8,9]. This is due to the wide structural variability of carbohydrate structures compared to other macromolecules and the difficulties in obtaining a large number of highly purified compounds. However, advances have been made in various areas including purification from natural sources, full chemical synthesis, immobilization and detection methods and automatization of sample handling. This has meant that carbohydrate arrays can now be used in screening molecular interactions as well as functional screening of whole organisms including eukaryotic cells, bacteria and viruses, and even living cell responses such as activation of cell signaling (Figure 1). Figure 1 Schematic diagram of the various applications of glycan arrays relating to applications including screening protein and antibody interactions with various glycans, virus and bacteria interactions with glycoproteins; and interactions of mammalian cells with glycosaminoglycan structures which lead to live cell responses relating to phosphorylation of cell signaling cascades. Symbols for glycan structures use nomenclature from [7]. Yellow circles: galactose; yellow squares: N-acetylgalactosamine; blue squares: N-acetylglucosamine; green circles: mannose; blue/white diamonds: glucuronic acid; brown/white diamond: iduronic acid; sulphation shown by 2S, 6S and NS; extra lines on the diamond represent unsaturated bonds. 2. Sources of Glycan Structures One major consideration for the construction of glycan arrays is the source of the glycoconjugate used, which in itself determines the aim of the experimental strategy. Sources of glycoconjugates can be broadly divided into three main subgroups: (1) Naturally occurring structures, which are extracted and purified from biological sources. Drawbacks of this source are that purification of glycans are hampered with factors such as multiple experimental steps associated with sample losses at every stage and reagents and conditions that can change the original glycan structures [10]. Methods adapted at using smaller sample sizes and providing more sensitive sample analysis circumvent these disadvantages [11]. For these reasons, carbohydrate arrays commonly use the latter two options of glycan libraries. (2) Libraries generated from natural glycans. Saccharide libraries can be generated by partial digestion of tissue-derived glycan chains and chromatographic fractionation of the resulting saccharide mixtures. Fractionation is initially based on the hydrodynamic volume, using size exclusion chromatography. Further fractionation, on the basis of charge or hydrophobicity can be used to separate the different glycan entities [12]. (3) Chemically-synthesized libraries. Traditionally, these libraries suffered from processes which were lengthy and needed highly specialized processes [13] and drawbacks included the fact that as the size of oligosaccharides increased, the yield of the coupling step decreased [14]. However, more rapid advances have been made in this type of technology with the use of enzymatic or chemoenzymatic de novo synthetic approaches. One example of the fundamental challenges of generating chemical libraries is the capacity to modify one specific hydroxyl group in the presence of many others. Strategies of glycan synthesis use steps to protect and mask chemical moieties in order to preferentially react the chemical groups of interest [15]. The second important consideration for glycan synthesis is the synthesis of the glycosidic bond. One strategy employed to generate glycosidic bonds is the use of recombinant glycosyltransferases [16]. The use of recombinant enzymes also means that the introduction of chemical groups such as sialic acid can be more easily accomplished as compared to chemical modification techniques [17]. The use of solid-support synthesis for glycans was inspired by the great advances previously made in peptide synthesis [18]. Translation of glycan synthesis to solid-phase platforms frequently employs the strategic placement of an amine-linker, which can in turn be covalently immobilized to glass surfaces [19]. Automation of the glycan chemical synthesis is now possible for several oligosaccharides on a solid-phase synthesizer [20], which is important in obtaining substances of high purity, a critical factor for the study of structure-activity relationships [21]. 3. Choice of Solid Supports and Immobilization of Glycans onto Microarrays Important factors to consider in the initial choice of solid supports for glycan arrays are twofold: whether derivatization of glycans is necessary and which type of chemistry is needed for immobilization of sugars onto the surface. However, this depends on the final experimental technique or the versatility of the resulting glycan array. Certain supports such as gold-based substrates provide advantages in that not only are they compatible for traditional fluorescence measurements using microarray scanners, but they can also be used as a platform for Matrix-assisted laser desorption ionization mass spectrometry (MALDI-MS), surface plasmon resonance (SPR), and quartz crystal microbalances [22]. The use of linkers, including polyethylene glycol also function as spacers and may have advantages in creating a separation from the matrix/plate surface [23]. There are different types of immobilization methods for glycans which exploit the large structural variety of glycoconjugates (see Table 1). One of the key components of a glycan array is the surface. Typical surfaces are microtiter plates, functionalized glass slides, nitrocellulose coated slides and gold slides [24,25,26,27]. The immobilization of the glycan structures onto a surface is the second consideration. Chemistries for the attachment of carbohydrates to surfaces can be broadly divided into four main categories, which are listed below: One criticism of glycan arrays is that the structure of oligosaccharides may affect the efficiency of their immobilization onto a surface, which may in turn affect the end-point signal measured [14]; in some platforms this has been largely addressed through production of tagged conjugates and their purification prior to immobilization. Examples include fluorescent labels such as 2-aminobenzamide (AB) or 2-aminobenzoid acid (AA) [28] and derivatives containing alkyl amines [29] or lipid tags [9]. 3.1. Affinity Adsorption One relatively straightforward method of immobilization of glycans onto a surface is adsorption. Glycans can be non-covalently and non-specifically immobilized onto nitrocellulose or oxidized black polystyrene surfaces [30,31]. Neoglyolipids can also be efficiently adsorbed onto nitrocellulaose slides [9]. Electrostatic interactions can be used to mediate immobilization between negatively charged glycans and surfaces coated with positively charged proteins such as poly-lysine [32]. 3.2. Covalent Immobilization of Glycans Modification of glycan structures can be time-consuming and costly; therefore, methods of immobilization that can use unmodified glycans are sometimes preferable. This type of immobilization can be achieved with photoactivatable supports which contain photolabile groups such as aryl(trifluoromethyl)diazirine [33]. Upon irradiation, photoproducts form intermediates of singlet carbene structure which can rapidly react with free glycan groups [34]. Immobilization of glycoconjugates and lectins has been achieved using this non-discriminatory type of immobilization [35]. The attachment of glycans via covalent bonds has been particularly useful for custom-based arrays such as heparin and heparan sulfate glycosaminoglycans, whereby the reducing-end aldehyde can be linked to amino and hydrazide surfaces groups on a surface [22,36,37,38,39]. microarrays-05-00003-t001_Table 1Table 1 Different types of covalent attachment methods used for glycan arrays. Type of Interaction Type of Reaction Details of Immobilization Reference Covalent Condensation Unmodified carbohydrates onto hydrazide surfaces. [22,40,41,42] Covalent Michael Addition Malemide-linked carbohydrates and thiol-coated glass slides. [27,43] Covalent Epoxide opening Covalently attach carbohydrates, glycoproteins and neoglycoconjugates to glass slides. [44,45,46] Covalent Amide coupling – – Covalent Diels-Alder reaction Covalent immobilization of glycans by cycloaddition. [31] Covalent Carbene Covalent immobilization of glycans by insertion. [35] Covalent Radical coupling Covalent immobilization of unmodified glycans by insertion. [47] 3.3. Site-Specific Covalent Immobilization The localization of glycan immobilization can be assured by using methods that result in covalent site-specific attachment of glycans. Immobilization methods based on the reaction between thiol and malemide groups are an example of covalent site-specific attachment of glycans. Studies have used malemide-conjugated sugars immobilized onto thiol-derivatized surfaces [27,43]. Conversely, thiol-linked sugars can be attached to malemide-coated surfaces [42,48]. Other types of covalent site-specific attachment include conjugation involving cyclopentadiene-linked sugars covalently attached to benzoquinone-coated surfaces [31]. In the field of GAG arrays, the ligation reaction between the aldehyde group and aminoxy or amino group has been crucial in the conjugation of heparin, heparin sulfate (HS) and chondroitin sulfate. 3.4. Non-Covalent Immobilization One of the main disadvantages of covalent attachment to a monolayer is that high ligand concentrations are required which is somewhat quixotic when working with complex oligosaccharides [38]. Glycan structures with lipid tags allow the non-covalent immobilization onto MALDI plate by insertion into a self-assembled alkylthiolate monolayer. This simple and efficient procedure results in the orientated immobilization of glycans, and avoids the use of fluorescent tags [49]. 4. Techniques for Detection of Protein Binding to Glycoarrays The choice of surface is important as it determines the type of detection that can be used, as the detection method depends on the fundamental properties of the surface. The most common type of detection method of carbohydrate arrays relies on fluorescence detection and the use of fluorescently labeled proteins binding directly or indirectly to the glycan structures. The standard mode of detection in this case is a fluorescent microarray scanner. Fluorescence detection can be achieved using different methodologies: (1) using proteins labeled with fluorophores, (2) using fluorophores as a secondary reagent or (3) using fluorescent proteins that bind tags. The latter two sandwich assay options can be used to increase signal amplification as the number of fluorophores available for detection is increased [50]. Disadvantages in fluorescence detection are based on the drawbacks of fluorophores themselves, which include their sensitivity to light and the fragility to oxidative degradation. Carbohydrate arrays, which involve applications with cells, can utilize phase-contrast and fluorescence microscopy for detection steps. Acquisition of live-cell images is made possible by motorized microscope incubator stages which are designed to maintain physiological conditions for cell cultures during the image acquisition process. Other improvements in microscopy include increased automatization in autofocusing image stability and speed of image analysis [51]. Relative straightforward replacements such as replacing the light source with a programmable light emitting diode (LED) array of modules in standard microscope equipment can turn low-power microscopes into high-resolution imagers [52]. This can increase signal intensity, which can be an important factor in discriminating between different samples. 5. Use of Glycan Arrays for the High-Throughput Analysis of Glycan–Protein Interactions As the cell membrane is a fluidic lipid bilayer environment, it is potentially important that the generation of glycan arrays mimic the lateral movement of glycans in their natural state at the cell membrane. Another important factor in protein–carbohydrate interactions is multivalency. Protein–carbohydrate interactions tend to be of low affinity but high specificity and so use multivalency to generate the affinity required for biologically relevant binding. In this way glycans can be organized to form clustered saccharide patches (CSPs) [53]. Applications that incorporate these important aspects include fluidic microarrays and CSP recognition in glycan microarrays. Generic glycan arrays are available and allow the attachment of glycans from sources including mammals and microorganisms. One example of such a platform is described by Fukui et al. [9], which have exploited the attachment of oligosaccharides to lipid tails (neoglyolipids, NGLs) that can then be spotted on nitrocellulose membranes and consequently probed using proteins or peroxidase-conjugated lectins. Other formats have taken advantage of non-covalent yet strong interactions between biotin and streptavidin in order to conjugate biotinylated glycosides [54]. In comparison, custom-made arrays are also designed to answer specific questions. One example is the inclusion of the “designer microarray” with other combinatorial approaches to define the carbohydrate sequence of the Prostate Cancer-associated Antigen F77. The generation of designer NGLs probes from O-glycans was achieved by alkaline reductive release from a source of epithelial mucin, porcine stomach mucin (PSM). This method surprising found the F77 antigen to be expressed in blood group H on a 6-linked branch of a poly-N-acetyllactosamine backbone [55]. Another example of the application of the designer microarray approach pertains to glucanpolysaccharides, which are d-glucose polymers with differing linkages in linear or branched sequences and function as secreted virulence factors in bacteria. By using combinatorial approaches involving negative-ion electrospray tandem MS, information was obtained on linkage sequence and chain length requirement of glucan-recognition proteins such as Dendritic Cell-Specific Intercellular Adhesion molecule-3-Grabbing Non Integrin (DC-SIGN) [56]. Other examples of combinatorial arrays focus on the design of novel autoantibody targets formed from glycolipid and lipid complexes, formed from two or more individual species, can interact to create molecular shapes capable of being recognized by these autoantibodies [53,57,58]. A further example of custom-based arrays is based on the structure of heparan sulfates (see below). 6. Use of Glycan Arrays for Studying Heparin/Heparan Sulfate Interactions with Proteins Heparan sulfate proteoglycans (HSPGs) contain protein core proteins that are covalently attached with HS chains. HS is a ubiquitous linear polysaccharide molecule, belonging to the GAG family of macromolecules. Attachment of HS to different core proteins results in HS having the ability to alter its location and topography as core proteins can occur at the cell surface and the extracellular matrix [59]. HSPGs are responsible for a multitude of different types of molecular interactions with different families of proteins and in different cellular contexts. HS is functionally important in many stages of the tumor process such as cellular transformation, tumor growth, invasion and metastasis. Factors which are important in modulating growth and metastasis are the charge density of HSPGs, level of expression of core proteins and the structure of the HS [60] Syndecan-1 is responsible for the maintenance of morphological differentiation and localization of epithelial cells. Additionally, a direct correlation between heparanases expression and the invasiveness of tumor cells has been shown [30]. Due to the interaction of HS with structural proteins such as laminin, HS can also provide physical barriers to tumor cells. In neurological contexts, HS also has important functions in Alzheimer’s disease [53,59] and roles in HIV attachment [61,62]. When located at the cell membrane, HSPGs interact with various growth factors, including members of fibroblast growth factors (FGFs), which are a large family of molecules (over 20 members), with similarities in sequence and functional properties [63]. The discovery that heparin binds acidic (FGF-1) and basic FGF (FGF-2) was made in 1989 by Thornton and coworkers who showed heparin potentiated the biological activity of crude preparations of acidic FGF [58]. The minimum sequence needed for HS binding to FGF2 contains a relatively common disaccharide structures in a pentasaccharide unit and consists of an N-sulfated glucosamine (GlcNS) and one Ido2S containing disaccharide [64]. Other examples show that more specificity is needed in the sequence of HS regulating its interactions. This is shown by the fact that a specific pentasaccharide sequence containing relatively rare modifications of 3-O-sulfate and the acetyl residue on an otherwise N-sulfated saccharide is responsible for heparin binding to increase the activity of antithrombin [65]. The minimal binding sequences are not always sufficient to increase biological activity. For FGF-2, twice the size of the minimum sequence is needed to produce a mitogenic response and a dodecasaccharide or longer sequence containing both IdoA (2-OSO3) and GlcNSO3(6-OSO3) residues is required [66,67]. Not all heparin binding proteins show this level of specificity. For example, hepatocyte growth factor/scatter factor (HGF/SF) is a 90 kDa paracrine factor synthesized by mesenchymal cells and involved in embryonic organ development and adult organ regeneration [68]. It has been shown that both HS and dermatan sulfate (DS) are both able to bind with a high affinity to a single common GAG-binding site and so have parallel abilities to modulate the HGF/SF induced functions [68]. Therefore it is unlikely that there is only one specific structure for binding and activity for any given protein and some structures can function with multiple proteins. HS disaccharide units form long chain polysaccharide structures. There are at least 8 different common disaccharide structures that make up HS. For longer oligosaccharides, the number of variant structures increases exponentially [69]. However, the constraints of the biosynthetic process limit the number of possible structural variations. HS structures also have specific domain type structure that occurs in the full length HS polysaccharide chains, which typically varies between 50–200 disaccharide units long (that is equivalent to 25–100 kDa in size [14]. This manifests as regions of high sulfation, consisting of mainly IdoA residues sulfated at the 2 position (IdoA-2-S) and N-sulfated residues that are known as “S” or “NS” domain. One of the main first applications of glycan arrays was to study how glycan structure would affect protein binding properties. This is being studied using arrays of heparin structures. Libraries containing heparin structures varying in their degree of sulfation as well as the length of their saccharide chains have been generated using heparin depolymerization [22] techniques in addition to chemical synthesis techniques [70]. The advantages of these assays are that, not only can quantitative data be generated on binding affinity but also the discovery of uncharacterized interactions can allow the investigation of novel therapeutic interventions. One example of the data generated using this type of approach is the measurement of relative binding affinities of different structures of heparin to FGF-1 and FGF-2 growth factors. This has shown that there is strong agreement between previously reported data [63]. Glycan arrays can then be used to calculate and determine the Kd values between proteins and immobilized glycans. The fluorescent intensities of bound proteins on glycan microarrays can be used to calculate the apparent dissociation constants [71]. Studies using the gold-standard approach for measuring Kd have shown that Kd values are similar to those determined using SPR experiments [72]. The measurement of proteins and glycans was originally used for the calculation of IC50 (half maximal inhibition) values for soluble inhibitors of proteins binding to glycans immobilized to surfaces. An example of the experimental setup includes a glycan array to which fluorescently labeled proteins are added, followed by soluble inhibitors. Following washing steps, the fluorescent intensities of the bound proteins are measured, which determines the IC50 values of the soluble inhibitor [73]. 7. Glycoarrays for Measuring Glycan–Cell Interactions Lectins are carbohydrate-binding proteins and macromolecules that are highly specific for sugar moieties. Lectins are present in plants, microbes and animals [4]. Lectins use specific carbohydrate-binding domains (CRD) to bind to carbohydrates. Glycan–lectin interactions mediate various processes notably in innate immunity and pathogenesis of viruses [74]. Carbohydrate microarrays have been used to identify and compare the binding preferences of different lectins, including C-type lectins. DC-SIGN, which is a receptor that plays a dual role in interaction of dendritic cells with pathogen surfaces as well as with T cells [75,76], was one of the first receptors tested against glycan arrays. Glycan arrays showed that DC-SIGN was able to bind high-mannose structures, in addition to fucose-terminated sugars including Lewis A and Lewis B structures [77]. Cell adhesion has been quantitatively assessed using glycan arrays. This has been shown using glycan arrays with lectin structures on hepatocytes. It has also been shown using glass slides with covalently attached monosaccharides and oligosaccharides of non-reducing terminal N-acetylglucosamine (GlcNAc) residues, galactose (Gal) and N-acetylgalactosamine residues. Primary chicken hepatocytes express a well-defined C-type lectin that binds to non-reducing terminal N-acetylglucosamine residues, and was labeled with a fluorescent dye. A specific chamber was used to remove non-adherent cells (GlycoChip® Centrifugation Chamber, Agilent Technology, Santa Clara, California, USA) and adherent cells were measured using fluorescence detection. Chicken hepatocytes bound selectively to lectins derivatized with GlcNAc structures rather than spots of lectin with Gal or no modifications [78]. 8. Glycoarrays for Measuring Virus and Bacteria–Glycan Interactions Glycan arrays have been very important in the understanding of surface interaction of various other microorganisms including interactions with viruses. The interaction of envelope glycoproteins with protein binding partners has been the basis of the development of the design of vaccines, which has been extensively exploited to interfere with the interaction involving the entry of the HIV virus. Glyco-protein microarrays have been used to analyze the glycan interactions dictating the interaction between the two HIV-1 (human immunodeficiency virus) glycoproteins, gp-120 and gp-41. The structures decorating the viral-surface envelope glycoproteins of HIV include high-mannose oligosaccharides including triantennary mannoside-1 [79]. Synthesized glycan structures contain a thiol-terminated ethylene glycol linker which allow attachment of the sugars onto a malemide-functionalized glass slide using the stable covalent bond [80]. This has made possible the identification of HIV vaccine candidate antigens [81]. Carbohydrate–protein interactions also dictate interactions between human cells and other pathogens including various bacteria such as Helicobacter pylori and Escherichia coli. Initial interactions determine adherence, which is an important factor of pathogenicity in microorganisms and separate human microbiota from pathogenic bacteria [82]. Carbohydrate microarray studies on glycan–bacteria interactions can also be exploited for screening for attachment inhibitors that can be used as novel antibiotics, as well as serving as detection platforms which can detect as well as purify certain bacterial species which are capable of differential expression of glycan-binding proteins [71]. Immobilization techniques employ ethanolamine linkers at the reducing end of synthesized glycans in combination with CodeLink slides. These slides are coated with a hydrophilic polymer containing N-hydroxylsuccinimide (NHS) ester reactive group and so enable the covalent attachment of sugars [83]. After hybridization of bacteria, cells can be stained with cell-permeant fluorescent nuclei staining dyes such as SYTO 62. These carbohydrate arrays have shown the ability to successfully identify species of bacteria which specifically express mannose-receptors and bind mannose sugars immobilized to surfaces [84]. Therefore the large scale screening of bacteria can be used to isolate and identify species of bacteria which are potentially pathogenic, in order to delineate a subspecies of bacteria which can be further analyzed [85]. Carbohydrate arrays can also be used to characterize aminoglycoside antibiotics. Aminoglycosides contain aminocyclitol rings with one or more amino sugars. Aminoglycosides bind to various sites including the A site of 16S rRNA of bacteria as well as other RNA species [86,87,88,89] and lead to the alteration of translation at diverse steps including initiation, elongation and termination and therefore inhibit bacterial protein synthesis. One of the major drawbacks of aminoglycosides is the increasing resistance in bacteria [90]. Aminoglycoside arrays have been useful in screening different RNA structures from bacteria including Candida albicans, which show the ability to bind aminoglycoside structures. In addition, differential binding can be detected by varying amounts of fluorescence. 9. Glycoarrays for Reporting Live Cell Responses Including Cellular Signaling Pathways The interactions of specific glycan structures with proteins may produce various functional consequences such as activation or inhibition of cellular processes. This means that functional assays are very important in addition to the protein interaction measurements; recent work has shown that glycoarrays can also be used to assess living cell responses to immobilized glycans. For example in the HS structure-function paradigm, many studies have analyzed the effects of HS structure on long-term cellular outcomes, such as cell proliferation, using standard cell biology techniques. Recently however, the direct investigation of the effect of different HS structures on cellular signaling in a glycoarray format has been described for the first time [39]. Previous studies have shown that chain length of HS influence the interaction with FGF-2 and affect ERK1/2 signaling events and eventual outcomes of FGF-2 signaling such as cell proliferation. ERK1/2 can be activated or switched on without the presence of HSPG, however, this effect is transient [91,92]. This is typical of the behavior of ERK1/2 which has been described to act as a continuously variable switch that controls transcription [93] in diverse cellular programs including embryogenesis, proliferation, differentiation and apoptosis [94]. The activation of ERK1/2 is critical in G0-G1-S phase progression due to regulation of cyclin D1. Mitogens such as FGF2 and particular HS saccharides structures in chlorate- treated Swiss 3T3 cells evoke a biphasic increase in ERK1/2 activity involved a large initial increase in activity followed by a degree of activity sustained at a lower level. This pathway could be studied in-depth by the development of a high-throughput (HTP) assay which allowed study of the functional effect of many HS structures simultaneously. 96-well plate assays have been used effectively in the past [95]. Puvirajesinghe et al. [39] have described a cell-responsive glycoarray format—termed “glycobioarrays”—which allow analysis of cell signaling responses of cells overlaid on spotted glycans on glass slides (Figure 2). Glycobioarrays provide an innovative platform to analyze the consequences of stimulation using different structures of glycans in terms of the activation of different signaling cascades. The data obtained in proof-of-concept studies on HS activation of FGF signaling corresponded to that reported in literature, where dp10 and dp12 heparin and HS oligosaccharides are sufficiently large to induce a functional response in cells [96,97,98]. This platform has potential for wider application with glycoarrays for measurement of cell responses requiring extracellular interactions of cell surface proteins with their regulatory glycan targets. Figure 2 Schematic diagram of a glycobioarray platform for screening live cell fibroblast growth factor signaling responses to immobilized heparin saccharides. See Puvirajesinghe et al. [39] for details. Saccharides immobilized onto an aminosilane glass surface via a Schiff’s base linkage with their reducing ends, is shown. Cells (shown using 40× magnification) can be overlaid onto the slide surface and cultured for a specified period, followed by fixation and immunostaining to detected specific epitopes for phosphorylation (green fluorescence for phosphorylated ERK and red fluorescence for total ERK) events using a microarray slide scanner. Symbols for glycan structures use nomenclature from [7]. Blue squares: N-acetylglucosamine; blue/white diamonds: glucuronic acid; brown/white diamond: iduronic acid; sulphation shown by 2S, 6S and NS; extra lines on the diamond represent unsaturated bonds. More recently, the study of molecular interactions dictating cellular processes such as cellular proliferation has now been made possible using 3D platforms and the use of 3D block printing. This has resulted in the development of high-throughput miniaturized 3D-chip platforms that can examine the importance of specific structures in HS and CS (chondroitin sulfate) in ternary structures of growth factor and growth factor receptor signaling complexes [99]. 3D block printers have been used to print immortalized bone marrow (BaF3) cells that have no HSPGs on the cellular surface and express a single FGFR [63,100], which makes them an ideal cellular model for the use with exogenous addition of FGF growth factor and HS structures. In order to translate 96-well cell proliferation assays to an array-based platform, the following experimental setup was used. Acid-washed glass slides were first modified with polystyrene co-malic anhydride and dried. Onto the slide, spots of BaCl2 and polylysine were arrayed onto the surface and then BaF3 cells in a viscous solution of alginate were then spotted onto the polylysine spots. Soluble growth factors and exogenous structures of HS and chondroitin sulfate (CS) were added to the media. The mechanism of FGF-FGFR-GAG signal transduction and how it relates to cellular activity can be examined by assessing cell proliferation using reagents which distinguish live cells from dead cells and assess intracellular esterase activity and plasma membrane integrity. Calcein AM and ethidium homodimer (EthD-1) dyes were used for this application [101,102,103]. This type of assay means that numerous combinations of FGF growth factors and GAG structures can be performed in parallel and in replicates [99]. A second important factor is that glycan microarray platforms require the use of lower quantities of reagent [99]. 10. Interrogation of Glycoarrays Using Mass Spectrometry Coupling mass spectrometry (MS) with glycan arrays provides a promising application for the discovery of new glycan-binding ligands and binding proteins. To date, the main achievements in this field have been made with respect to lectin microarrays. Applications include the systematic identification of carbohydrate-binding proteins in proteomes [104]. One such example of a lectin glycan array is based on the covalent immobilization of lectins onto a molded silicone polymer, polydimethylsiloxane (PDMS) by the use of oxidation of PDMS, silanization with aminopropyltrimethoxysilane and cross-linking with glutaraldehyde [41]. The high degree of flexibility of PDMS enables the substrate to be easily attached to a MALDI plate for MALDI-MS measurement, following incubation of lectin arrays with patient sera [41]. Mass spectrometry is also a powerful method for defining the saccharides arrayed in “designer” glycome arrays produced from biological sources [56]. 11. Conclusions and Future Perspectives Key advances in glycan arrays over the past two decades have been in screening carbohydrate-binding proteins in proteomes, calculating protein binding affinities and automatization of solid-support synthesis for glycans. Integration and examination of this wealth of information is beginning to become more standardized with the use of public data repositories within consortia such as the CFG. Aspects of immobilization, choice of support and detection have now been studied for different types of glycans depending on the type of application. Technological developments mean that increased sensitivity as well as combinatorial approaches exploiting mass spectrometry techniques for glycan array interrogation will permit new methodological advances to continue. Further exploitation of glycobioarrays may also provide higher throughput functional level screening of glycan activities. Collectively, these will improve the quality and amount of data, plus improve quantitation of data. Ultimately, this will yield a wealth of insights into the functional diversity and functional specificity of glycans and will underpin new routes to exploit this knowledge in biomedical applications. Acknowledgments The authors acknowledge funding from the Medical Research Council and the Biotechnology and Biological Sciences Research Council in the UK, and an A*MIDEX project (No. ANR-11-IDEX-0001-02) funded by the Investissements d’Avenir French Government program, managed by the French National Research Agency (ANR) with Aix Marseille Université (TMP, A_M-AAP-ID-14-15-140314-09.45-GUENNEAU-PUVIRAJESINGHE-HLS_SAT). Author Contributions Tania M. Puvirajesinghe and Jeremy. E. Turnbull wrote this manuscript. Conflicts of Interest The authors declare no conflict of interest. ==== Refs References 1. Prydz K. Dalen K.T. Synthesis and sorting of proteoglycans J. Cell Sci. 2000 113 193 205 10633071 2. Boehm G. Stahl B. Oligosaccharides from milk J. Nutr. 2007 137 Suppl. S2 847S 849S 17311985 3. Haltiwanger R.S. Lowe J.B. Role of glycosylation in development Annu. Rev. Biochem. 2004 73 491 537 10.1146/annurev.biochem.73.011303.074043 15189151 4. Bertozzi C.R. Sasisekharan R. Glycomics Essentials of Glycobiology 2nd ed. Ajit V. Richard D.C. Jeffrey D.E. Hudson H.F. Pamela S. Carolyn R.B. Gerald W.H. Marilynn E.E. Cold Spring Harbor Laboratory Press Cold Spring Harbor, NY, USA 2009 5. Turnbull J.E. Field R.A. Emerging glycomics technologies Nat. Chem. Biol. 2007 3 74 77 10.1038/nchembio0207-74 17235338 6. Aoki-Kinoshita K.F. Glycome Informatics- Methods and Applications CRC Press New York, NY, USA 2010 7. Varki A. Cummings R.D. Esko J.D. Freeze H.H. Stanley P. Marth J.D. Bertozzi C.R. Hart G.W. Etzler M.E. Symbol nomenclature for glycan representation Proteomics 2009 9 5398 5399 10.1002/pmic.200900708 19902428 8. Wang D. Carbohydrate microarrays Proteomics 2003 3 2167 2175 10.1002/pmic.200300601 14595816 9. Fukui S. Feizi T. Galustian C. Lawson A.M. Chai W. Oligosaccharide microarrays for high-throughput detection and specificity assignments of carbohydrate-protein interactions Nat. Biotechnol. 2002 20 1011 1017 10.1038/nbt735 12219077 10. Inoue Y. Nagasawa K. Selective N -desulfation of heparin with dimethyl sulfoxide containing water or methanol Carbohydr. Res. 1976 46 87 95 10.1016/S0008-6215(00)83533-8 1248016 11. Guimond S.E. Puvirajesinghe T.M. Skidmore M.A. Kalus I. Dierks T. Yates E.A. Turnbull J.E. Rapid purification and high sensitivity analysis of heparan sulfate from cells and tissues: Toward glycomics profiling J. Biol. Chem. 2009 284 25714 25722 10.1074/jbc.M109.032755 19596853 12. Powell A.K. Ahmed Y.A. Yates E.A. Turnbull J.E. Generating heparan sulfate saccharide libraries for glycomics applications Nat. Protoc. 2010 5 821 833 10.1038/nprot.2010.17 20379137 13. Grootenhuis P.D. Westerduin P. Meuleman D. Petitou M. van Boeckel C.A. Rational design of synthetic heparin analogues with tailor-made coagulation factor inhibitory activity Nat. Struct. Biol. 1995 2 736 739 10.1038/nsb0995-736 7552742 14. Powell A.K. Yates E.A. Fernig D.G. Turnbull J.E. Interactions of heparin/heparan sulfate with proteins: Appraisal of structural factors and experimental approaches Glycobiology 2004 14 17R 30R 10.1093/glycob/cwh051 14718374 15. Seeberger P.H. Finney N. Rabuka D. Bertozzi C.R. Chemical and Enzymatic Synthesis of Glycans and Glycoconjugates Cold Spring Harbor Laboratory Press Cold Spring Harbor, NY, USA 2009 16. Palcic M.M. Glycosyltransferases as biocatalysts Curr. Opin. Chem. Biol. 2011 15 226 233 10.1016/j.cbpa.2010.11.022 21334964 17. Blixt O. Collins B.E. van den Nieuwenhof I.M. Crocker P.R. Paulson J.C. Sialoside specificity of the siglec family assessed using novel multivalent probes: Identification of potent inhibitors of myelin-associated glycoprotein J. Biol. Chem. 2003 278 31007 31019 10.1074/jbc.M304331200 12773526 18. Merrifield R.B. Solid Phase Peptide Synthesis. I. The Synthesis of a Tetrapeptide J. Am. Chem. Soc. 1963 85 2149 2154 10.1021/ja00897a025 19. De Paz J.L. Noti C. Seeberger P.H. Microarrays of synthetic heparin oligosaccharides J. Am. Chem. Soc. 2006 128 2766 2767 10.1021/ja057584v 16506732 20. Plante O.J. Palmacci E.R. Seeberger P.H. Automated solid-phase synthesis of oligosaccharides Science 2001 291 1523 1527 10.1126/science.1057324 11222853 21. Cai C. Li L. Harvey C. Liu J. Linhardt R.J. Towards the chemoenzymatic synthesis of heparan sulfate oligosaccharides: Oxidative cleavage of -nitrophenyl group with ceric ammonium salts Tetrahedron Lett. 2013 54 4471 4474 10.1016/j.tetlet.2013.06.044 23929984 22. Zhi Z.L. Powell A.K. Turnbull J.E. Fabrication of carbohydrate microarrays on gold surfaces: Direct attachment of nonderivatized oligosaccharides to hydrazide-modified self-assembled monolayers Anal. Chem. 2006 78 4786 4793 10.1021/ac060084f 16841896 23. Wehner J.W. Weissenborn M.J. Hartmann M. Gray C.J. Sardzik R. Eyers C.E. Flitsch S.L. Lindhorst T.K. Dual purpose S-trityl-linkers for glycoarray fabrication on both polystyrene and gold Org. Biomol. Chem. 2012 10 8919 8926 10.1039/c2ob26118a 23059912 24. Ratner D.M. Carbohydrate microarrays: Advancing the burgeoning field of glycomics Biol. Tech. Int. 2005 17 8 11 25. Bryan M.C. Plettenburg O. Sears P. Rabuka D. Wacowich-Sgarbi S. Wong C.H. Saccharide display on microtiter plates Chem. Biol. 2002 9 713 720 10.1016/S1074-5521(02)00155-2 12079783 26. Fazio F. Bryan M.C. Blixt O. Paulson J.C. Wong C.H. Synthesis of sugar arrays in microtiter plate J. Am. Chem. Soc. 2002 124 14397 14402 10.1021/ja020887u 12452714 27. Park S. Shin I. Fabrication of carbohydrate chips for studying protein-carbohydrate interactions Angew. Chem. Int. Ed. Engl. 2002 41 3180 3182 10.1002/1521-3773(20020902)41:17<3180::AID-ANIE3180>3.0.CO;2-S 12207382 28. De Boer A.R. Hokke C.H. Deelder A.M. Wuhrer M. General microarray technique for immobilization and screening of natural glycans Anal. Chem. 2007 79 8107 8113 10.1021/ac071187g 17922555 29. Song X. Xia B. Stowell S.R. Lasanajak Y. Smith D.F. Cummings R.D. Novel fluorescent glycan microarray strategy reveals ligands for galectins Chem. Biol. 2009 16 36 47 10.1016/j.chembiol.2008.11.004 19171304 30. Wang D. Liu S. Trummer B.J. Deng C. Wang A. Carbohydrate microarrays for the recognition of cross-reactive molecular markers of microbes and host cells Nat. Biotechnol. 2002 20 275 281 10.1038/nbt0302-275 11875429 31. Houseman B.T. Mrksich M. Carbohydrate arrays for the evaluation of protein binding and enzymatic modification Chem. Biol. 2002 9 443 454 10.1016/S1074-5521(02)00124-2 11983333 32. Ma X. Mohammad S.F. Kim S.W. Heparin removal from blood using poly(l -lysine) immobilized hollow fiber Biotechnol. Bioeng. 1992 40 530 536 10.1002/bit.260400412 18601148 33. Barie N. Rapp M. Sigrist H. Ache H.J. Covalent photolinker-mediated immobilization of an intermediate dextran layer to polymer-coated surfaces for biosensing applications Biosens. Bioelectron. 1998 13 855 860 10.1016/S0956-5663(98)00052-9 9828382 34. Platz M. Admasu A.S. Kwiatkowski S. Crocker P.J. Imai N. Watt D.S. Photolysis of 3-aryl-3-(trifluoromethyl)diazirines: A caveat regarding their use in photoaffinity probes Bioconjug. Chem. 1991 2 337 341 10.1021/bc00011a008 1790173 35. Angeloni S. Ridet J.L. Kusy N. Gao H. Crevoisier F. Guinchard S. Kochhar S. Sigrist H. Sprenger N. Glycoprofiling with micro-arrays of glycoconjugates and lectins Glycobiology 2005 15 31 41 10.1093/glycob/cwh143 15342550 36. Yates E.A. Jones M.O. Clarke C.E. Powell A.K. Johnson S.R. Porch A. Edwards P.P. Turnbull J.E. Microwave enhanced reaction of carbohydrates with amino-derivatised labels and glass surfaces J. Mater. Chem. 2003 13 2061 2063 10.1039/b305248f 37. Powell A.K. Zhi Z.L. Turnbull J.E. Saccharide microarrays for high-throughput interrogation of glycan-protein binding interactions Methods Mol. Biol. 2009 534 313 329 19277542 38. Zhi Z.L. Laurent N. Powell A.K. Karamanska R. Fais M. Voglmeir J. Wright A. Blackburn J.M. Crocker P.R. Russell D.A. A versatile gold surface approach for fabrication and interrogation of glycoarrays ChemBioChem 2008 9 1568 1575 10.1002/cbic.200700788 18561346 39. Puvirajesinghe T.M. Ahmed Y.A. Powell A.K. Fernig D.G. Guimond S.E. Turnbull J.E. Array-based functional screening of heparin glycans Chem. Biol. 2012 19 553 558 10.1016/j.chembiol.2012.03.011 22633407 40. Lee M.R. Shin I. Facile preparation of carbohydrate microarrays by site-specific, covalent immobilization of unmodified carbohydrates on hydrazide-coated glass slides Org. Lett. 2005 7 4269 4272 10.1021/ol051753z 16146404 41. Hu S. Wong D.T. Lectin microarray Proteom. Clin. Appl. 2009 3 148 154 10.1002/prca.200800153 21132067 42. Houseman B.T. Gawalt E.S. Mrksich M. Maleimide-functionalized self-assembled monolayers for the preparation of peptide and carbohydrate biochips Langmuir 2003 19 1522 1531 10.1021/la0262304 43. Park S. Lee M.R. Pyo S.J. Shin I. Carbohydrate chips for studying high-throughput carbohydrate-protein interactions J. Am. Chem. Soc. 2004 126 4812 4819 10.1021/ja0391661 15080685 44. Oyelaran O. Li Q. Farnsworth D. Gildersleeve J.C. Microarrays with varying carbohydrate density reveal distinct subpopulations of serum antibodies J. Proteome Res. 2009 8 3529 3538 10.1021/pr9002245 19366269 45. Oyelaran O. McShane L.M. Dodd L. Gildersleeve J.C. Profiling human serum antibodies with a carbohydrate antigen microarray J. Proteome Res. 2009 8 4301 4310 10.1021/pr900515y 19624168 46. Zhang Y. Campbell C. Li Q. Gildersleeve J.C. Multidimensional glycan arrays for enhanced antibody profiling Mol. Biosyst. 2010 6 1583 1591 10.1039/c002259d 20711537 47. Carroll G.T. Wang D. Turro N.J. Koberstein J.T. Photochemical micropatterning of carbohydrates on a surface Langmuir 2006 22 2899 2905 10.1021/la0531042 16519501 48. Seo J.H. Adachi K. Lee B.K. Kang D.G. Kim Y.K. Kim K.R. Lee H.Y. Kawai T. Cha H.J. Facile and rapid direct gold surface immobilization with controlled orientation for carbohydrates Bioconjug. Chem. 2007 18 2197 2201 10.1021/bc700288z 17915957 49. Sanchez-Ruiz A. Serna S. Ruiz N. Martin-Lomas M. Reichardt N.C. MALDI-TOF mass spectrometric analysis of enzyme activity and lectin trapping on an array of N -glycans Angew. Chem. Int. Ed. Engl. 2011 50 1801 1804 10.1002/anie.201006304 21328643 50. Vora G.J. Meador C.E. Anderson G.P. Taitt C.R. Comparison of detection and signal amplification methods for DNA microarrays Mol. Cell Probes 2008 22 294 300 10.1016/j.mcp.2008.07.002 18675897 51. Fero M. Pogliano K. Automated quantitative live cell fluorescence microscopy Cold Spring Harb. Perspect. Biol. 2010 2 10.1101/cshperspect.a000455 20591990 52. Zheng G. Horstmeyer R. Yang C. Wide-field, high-resolution Fourier ptychographic microscopy Nat. Photonics 2013 7 739 745 10.1038/nphoton.2013.187 25243016 53. Cohen M. Varki A. Modulation of glycan recognition by clustered saccharide patches Int. Rev. Cell Mol. Biol. 2014 308 75 125 24411170 54. Leppanen A. Penttila L. Renkonen O. McEver R.P. Cummings R.D. Glycosulfopeptides with O -glycans containing sialylated and polyfucosylated polylactosamine bind with low affinity to P-selectin J. Biol. Chem. 2002 277 39749 39759 10.1074/jbc.M206281200 12145302 55. Gao C. Liu Y. Zhang H. Zhang Y. Fukuda M.N. Palma A.S. Kozak R.P. Childs R.A. Nonaka M. Li Z. Carbohydrate sequence of the prostate cancer-associated antigen F77 assigned by a mucin O -glycome designer array J. Biol. Chem. 2014 289 16462 16477 10.1074/jbc.M114.558932 24753245 56. Palma A.S. Liu Y. Zhang H. Zhang Y. McCleary B.V. Yu G. Huang Q. Guidolin L.S. Ciocchini A.E. Torosantucci A. Unravelling glucan recognition systems by glycome microarrays using the designer approach and mass spectrometry Mol. Cell Proteom. 2015 14 974 988 10.1074/mcp.M115.048272 25670804 57. Rinaldi S. Brennan K.M. Willison H.J. Heteromeric glycolipid complexes as modulators of autoantibody and lectin binding Prog. Lipid Res. 2010 49 87 95 10.1016/j.plipres.2009.09.001 19735674 58. Galban-Horcajo F. Halstead S.K. McGonigal R. Willison H.J. The application of glycosphingolipid arrays to autoantibody detection in neuroimmunological disorders Curr. Opin. Chem. Biol. 2014 18 78 86 10.1016/j.cbpa.2014.01.008 24495749 59. Van Horssen J. Wesseling P. van den Heuvel L.P. de Waal R.M. Verbeek M.M. Heparan sulphate proteoglycans in Alzheimer’s disease and amyloid-related disorders Lancet Neurol. 2003 2 482 492 10.1016/S1474-4422(03)00484-8 12878436 60. Wang Z. Xu H. Jiang L. Zhou X. Lu C. Zhang X. Positive association of heparanase expression with tumor invasion and lymphatic metastasis in gastric carcinoma Mod. Pathol. 2004 18 205 211 10.1038/modpathol.3800282 15475937 61. Sasisekharan C.R. Essentials of Glycobiology Cold Spring Harbor Laboratory Press Cold Spring Harbor, NY, USA 2009 62. McFeters A.G. Yu F.P. Pyle B.H. Stewart P.S. Physiological assessment of bacteria using fluorochromes J. Microbiol. Methods 1995 21 1 13 10.1016/0167-7012(94)00027-5 11538412 63. De Paz J.L. Angulo J. Lassaletta J.M. Nieto P.M. Redondo-Horcajo M. Lozano R.M. Gimenez-Gallego G. Martin-Lomas M. The activation of fibroblast growth factors by heparin: synthesis, structure, and biological activity of heparin-like oligosaccharides ChemBioChem 2001 2 673 685 10.1002/1439-7633(20010903)2:9<673::AID-CBIC673>3.0.CO;2-7 11828504 64. Esko J.D. Lindahl U. Molecular diversity of heparan sulfate J. Clin. Investig. 2001 108 169 173 10.1172/JCI200113530 11457867 65. Lindahl U. Thunberg L. Backstrom G. Riesenfeld J. Nordling K. Bjork I. Extension and structural variability of the antithrombin-binding sequence in heparin J. Biol. Chem. 1984 259 12368 12376 6490618 66. Guimond S. Maccarana M. Olwin B.B. Lindahl U. Rapraeger A.C. Activating and inhibitory heparin sequences for FGF-2 (basic FGF). Distinct requirements for FGF-1, FGF-2, and FGF-4 J. Biol. Chem. 1993 268 23906 23914 7693696 67. Pye D.A. Vives R.R. Turnbull J.E. Hyde P. Gallagher J.T. Heparan sulfate oligosaccharides require 6-O -sulfation for promotion of basic fibroblast growth factor mitogenic activity J. Biol. Chem. 1998 273 22936 22942 10.1074/jbc.273.36.22936 9722514 68. Deakin J.A. Lyon M. Differential regulation of hepatocyte growth factor/scatter factor by cell surface proteoglycans and free glycosaminoglycan chains J. Cell Sci. 1999 112 1999 2009 10341217 69. Yates E.A. Guimond S.E. Turnbull J.E. Highly diverse heparan sulfate analogue libraries: Providing access to expanded areas of sequence space for bioactivity screening J. Med. Chem. 2004 47 277 280 10.1021/jm0309755 14695842 70. Orgueira H.A. Bartolozzi A. Schell P. Litjens R.E. Palmacci E.R. Seeberger P.H. Modular synthesis of heparin oligosaccharides Chemistry 2003 9 140 169 10.1002/chem.200390009 12506372 71. Park S. Gildersleeve J.C. Blixt O. Shin I. Carbohydrate microarrays Chem. Soc. Rev. 2013 42 4310 4326 10.1039/C2CS35401B 23192235 72. Park S. Shin I. Carbohydrate microarrays for assaying galactosyltransferase activity Org. Lett. 2007 9 1675 1678 10.1021/ol070250l 17394347 73. Liang P.H. Wang S.K. Wong C.H. Quantitative analysis of carbohydrate-protein interactions using glycan microarrays: Determination of surface and solution dissociation constants J. Am. Chem. Soc. 2007 129 11177 11184 10.1021/ja072931h 17705486 74. Vigerust D.J. Shepherd V.L. Virus glycosylation: Role in virulence and immune interactions Trends Microbiol. 2007 15 211 218 10.1016/j.tim.2007.03.003 17398101 75. Geijtenbeek T.B. Kwon D.S. Torensma R. van Vliet S.J. van Duijnhoven G.C. Middel J. Cornelissen I.L. Nottet H.S. KewalRamani V.N. Littman D.R. DC-SIGN, a dendritic cell-specific HIV-1-binding protein that enhances trans-infection of T cells Cell 2000 100 587 597 10.1016/S0092-8674(00)80694-7 10721995 76. Engering A. Geijtenbeek T.B. van Vliet S.J. Wijers M. van Liempt E. Demaurex N. Lanzavecchia A. Fransen J. Figdor C.G. Piguet V. The dendritic cell-specific adhesion receptor DC-SIGN internalizes antigen for presentation to T cells J. Immunol. 2002 168 2118 2126 10.4049/jimmunol.168.5.2118 11859097 77. Powlesland A.S. Ward E.M. Sadhu S.K. Guo Y. Taylor M.E. Drickamer K. Widely divergent biochemical properties of the complete set of mouse DC-SIGN-related proteins J. Biol. Chem. 2006 281 20440 20449 10.1074/jbc.M601925200 16682406 78. Nimrichter L. Gargir A. Gortler M. Altstock R.T. Shtevi A. Weisshaus O. Fire E. Dotan N. Schnaar R.L. Intact cell adhesion to glycan microarrays Glycobiology 2004 14 197 203 10.1093/glycob/cwh022 14638630 79. Bewley C.A. Otero-Quintero S. The potent anti-HIV protein cyanovirin-N contains two novel carbohydrate binding sites that selectively bind to Man8 D1D3 and Man9 with nanomolar affinity: Implications for binding to the HIV envelope protein gp120 J. Am. Chem. Soc. 2001 123 3892 3902 10.1021/ja004040e 11457139 80. Ratner D.M. Adams E.W. Su J. O’Keefe B.R. Mrksich M. Seeberger P.H. Probing protein-carbohydrate interactions with microarrays of synthetic oligosaccharides ChemBioChem 2004 5 379 382 10.1002/cbic.200300804 14997532 81. Adams W.E. Ratner D.M. Bokesch H.R. McMahon J.B. O’Keefe B.R. Seeberger P.H. Oligosaccharide and glycoprotein microarrays as tools in HIV glycobiology; glycan-dependent gp120/protein interactions Chem. Biol. 2004 11 875 881 10.1016/j.chembiol.2004.04.010 15217620 82. Hooper L.V. Gordon J.I. Glycans as legislators of host-microbial interactions: Spanning the spectrum from symbiosis to pathogenicity Glycobiology 2001 11 1R 10R 10.1093/glycob/11.2.1R 11181556 83. CodeLink Activated slides Available online: http://www.surmodics.com/assets/uploads/documents/CodeLink_User_Guide.pdf (accessed on 24 December 2015) 84. De Paz J.L. Horlacher T. Seeberger P.H. Oligosaccharide Microarrays to Map Interactions of Carbohydrates in Biological Systems Elsevier Inc. Amsterdam, The Netherlands 2006 85. Mahal L.K. Catching bacteria with sugar Chem. Biol. 2004 11 1602 1604 10.1016/j.chembiol.2004.11.017 15610842 86. Morris J.C. Ping-Sheng L. Zhai H.X. Shen T.Y. Mensa-Wilmot K. Phosphatidylinositol phospholipase C is activated allosterically by the aminoglycoside G418. 2-deoxy-2-fluoro-scyllo-inositol-1-O -dodecylphosphonate and its analogs inhibit glycosylphosphatidylinositol phospholipase C J. Biol. Chem. 1996 271 15468 15477 8663028 87. Ren Y.G. Martinez J. Kirsebom L.A. Virtanen A. Inhibition of Klenow DNA polymerase and poly(A)-specific ribonuclease by aminoglycosides RNA 2002 8 1393 1400 10.1017/S1355838202021015 12458793 88. Moazed D. Noller H.F. Interaction of antibiotics with functional sites in 16S ribosomal RNA Nature 1987 327 389 394 10.1038/327389a0 2953976 89. Walter F. Vicens Q. Westhof E. Aminoglycoside-RNA interactions Curr. Opin. Chem. Biol. 1999 3 694 704 10.1016/S1367-5931(99)00028-9 10600721 90. Jana S. Deb J.K. Molecular understanding of aminoglycoside action and resistance Appl. Microbiol. Biotechnol. 2006 70 140 150 10.1007/s00253-005-0279-0 16391922 91. Delehedde M. Sergeant N. Lyon M. Rudland P.S. Fernig D.G. Hepatocyte growth factor/scatter factor stimulates migration of rat mammary fibroblasts through both mitogen-activated protein kinase and phosphatidylinositol 3-kinase/Akt pathways Eur. J. Biochem. 2001 268 4423 4429 10.1046/j.1432-1327.2001.02363.x 11502202 92. Lundin L. Larsson H. Kreuger J. Kanda S. Lindahl U. Salmivirta M. Claesson-Welsh L. Selectively desulfated heparin inhibits fibroblast growth factor-induced mitogenicity and angiogenesis J. Biol. Chem. 2000 275 24653 24660 10.1074/jbc.M908930199 10816596 93. Hazzalin C.A. Mahadevan L.C. MAPK-regulated transcription: A continuously variable gene switch? Nat. Rev. Mol. Cell Biol. 2002 3 30 40 10.1038/nrm715 11823796 94. Raman M. Chen W. Cobb M.H. Differential regulation and properties of MAPKs Oncogene 2007 26 3100 3112 10.1038/sj.onc.1210392 17496909 95. Guimond S.E. Turnbull J.E. Fibroblast growth factor receptor signalling is dictated by specific heparan sulphate saccharides Curr. Biol. 1999 9 1343 1346 10.1016/S0960-9822(00)80060-3 10574766 96. Delehedde M. Lyon M. Gallagher J.T. Rudland P.S. Fernig D.G. Fibroblast growth factor-2 binds to small heparin-derived oligosaccharides and stimulates a sustained phosphorylation of p42/44 mitogen-activated protein kinase and proliferation of rat mammary fibroblasts Biochem. J. 2002 366 235 244 10.1042/bj20011718 12000311 97. Turnbull J.E. Fernig D.G. Ke Y. Wilkinson M.C. Gallagher J.T. Identification of the basic fibroblast growth factor binding sequence in fibroblast heparan sulfate J. Biol. Chem. 1992 267 10337 10341 1587820 98. Ishihara M. Shaklee P.N. Yang Z. Liang W. Wei Z. Stack R.J. Holme K. Structural features in heparin which modulate specific biological activities mediated by basic fibroblast growth factor Glycobiology 1994 4 451 458 10.1093/glycob/4.4.451 7827407 99. Sterner E. Meli L. Kwon S.J. Dordick J.S. Linhardt R.J. FGF-FGFR signaling mediated through glycosaminoglycans in microtiter plate and cell-based microarray platforms Biochemistry 2013 52 9009 9019 10.1021/bi401284r 24289246 100. Ornitz D.M. Xu J. Colvin J.S. McEwen D.G. MacArthur C.A. Coulier F. Gao G. Goldfarb M. Receptor specificity of the fibroblast growth factor family J. Biol. Chem. 1996 271 15292 15297 8663044 101. Yu F.P. Pyle B.H. McFeters G.A. A direct viable count method for the enumeration of attached bacteria and assessment of biofilm disinfection J. Microbiol. Methods 1993 17 167 180 10.1016/0167-7012(93)90044-I 11537721 102. Papadopoulos N.G. Dedoussis G.V. Spanakos G. Gritzapis A.D. Baxevanis C.N. Papamichail M. An improved fluorescence assay for the determination of lymphocyte-mediated cytotoxicity using flow cytometry J. Immunol. Methods 1994 177 101 111 10.1016/0022-1759(94)90147-3 7822816 103. Hayes A.W. Principles and Methods of Toxicology 5th ed. Hayes A.W. CRC Press Boca Raton, FL, USA 2008 104. Feizi T. Chai W. Oligosaccharide microarrays to decipher the glyco code Nat. Rev. Mol. Cell Biol. 2004 5 582 588 10.1038/nrm1428 15232576

PMC005xxxxxx/PMC5003449.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5010004microarrays-05-00004EditorialAcknowledgement to Reviewers of Microarrays in 2015 Microarrays Editorial Office MDPI AG, Klybeckstrasse 64, CH-4057 Basel, Switzerland; microarrays@mdpi.com25 1 2016 3 2016 5 1 4© 2016 by the author; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/). ==== Body 1. Introduction The editors of Microarrays would like to express their sincere gratitude to the following reviewers for assessing manuscripts in 2015. We greatly appreciate the contribution of expert reviewers, which is crucial to the journal’s editorial decision-making process. Several steps have been taken in 2015 to thank and acknowledge reviewers. Good, timely reviews are rewarded with a discount off their next MDPI publication. By creating an account on the submission system, reviewers can access details of their past reviews, see the comments of other reviewers, and download a letter of acknowledgement for their records. In addition, MDPI has launched a collaboration with Publons, a website that seeks to publicly acknowledge reviewers on a per journal basis. This is all done, of course, within the constraints of reviewer confidentiality. Feedback from reviewers shows that most see their task as a voluntary and mostly unseen work in service to the scientific community. We are grateful to our reviewers for the contribution they make. Andreotti, Gabriella Heller, Gerwin Seidel, Michael Bidarra, Silvia Ingelman-Sundberg, Magnus Seitz, Harald Bouckaert, Julie Iqbal, M. Anwar Signore, Michele Brait, Mariana Jayavelu, Naresh Doni Smedby, Karin Ekström Brander, Christian Jonoska, Natasha Solé, Francesc Braun, Frank K. Kato, Takamitsu A. Somoza, Mark M. Brunham, Liam R. Kelly, Christopher V. Strillacci, Maria Giuseppina Burnside, Rachel Kibriya, Muhammad G. Sugimura, Haruhiko Canevari, Silvana Kidd, Mark Sumoy, Lauro Castiglioni, Bianca Kim, J. Sykacek, Peter Castresana, Javier Klopocki, Eva Terbrueggen, Robert Chandra, Divya Korf, Ulrike Ting, Angela Chen, Xiangning Kschischo, Maik Tombelli, Sara Chu, Chi-Ming Kudithipudi, Srikanth Tong, Tiejun Chung, Brian H.Y. Kunkalla, Kranthi Torisawa, Yu-suke Corfield, Anthony P. Kurth, Ina Townley, Helen E. Craw, Pascal Labaer, Joshua Usui, Kenji Cretich, Marina Labarge, Mark Vaudry, David Damaraju, Sambasivarao Lampe, Paul Vazza, Giovanni Darcy, Isabel Laurenceau, Emmanuelle Vogel, Ulrich De Crescenzo, Gregory Lee, Sheng-An Wang, Yufeng De Paz, Jose Luis Louhelainen, Jari Watts, Lora Talley Dixit, Chandra K. Lu, Huibin Wei, Hairong Dodd, Janice Manzano, Marisa Weinhäusel, Andreas Duverger, Eric McDermott, Jason E. Whelan, Rebecca J. Eder, Iris E. Min, Junhong Wild, Peter J. Efthimiadis, Georgios K Moreno, Miguel Willison, Hugh Emmert-Streib, Frank Nagatani, Naoki Winegarden, Neil Eroles, Pilar Nga, Min En Winssinger, Nicolas Esfandyarpour, Rahim Nicolau, Monica Woodbury, Neal Fernández, Ana Nicolini, Claudio Wright, Gavin J. Filiatrault, Melanie Olsen, Catharina Wu, Wesson Fogel, Brent L. Pasquer, Frédérique Yamada, Tesshi Fukuoka, Junya Pena, Romi N Yamagata, Takanori Gach, Johannes S. Peppelenbosch, Maikel Yao, Jun Ghosh, Gargi Pereira, Julia Santucci Zanella, Fabian Giordano, Antonio Phipson, Belinda Zheng, Bo Gonzalo Claros, M Potashkin, Judith Zhu, Xiangdong Granade, Timothy C. Quagliata, Luca Zhu, Heng Griffin, Robert J. Roupioz, Yoann Ziemssen, Focke Guinney, Justin Schmitt, Joachim Zubarev, Eugene Heemstra, Jennifer M. Segura, Víctor

PMC005xxxxxx/PMC5003450.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5010005microarrays-05-00005ArticleA Double-Hybridization Approach for the Transcription- and Amplification-Free Detection of Specific mRNA on a Microarray Haider Michaela 1Haselgrübler Thomas 1Sonnleitner Alois 1Aberger Fritz 2Hesse Jan 1*Negrini Massimo Academic Editor1 Center for Advanced Bioanalysis GmbH, Gruberstrasse 40-42, 4020 Linz, Austria; michaela.haider@cbl.at (M.H.); thomas.haselgruebler@cbl.at (T.H.); alois.sonnleitner@cbl.at (A.S.)2 Department of Molecular Biology, University of Salzburg, 5020 Salzburg, Austria; fritz.aberger@sbg.ac.at* Correspondence: jan.hesse@cbl.at; Tel.: +43-732-2468-7509; Fax: +43-732-2468-753023 2 2016 3 2016 5 1 517 12 2015 15 2 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).A double-hybridization approach was developed for the enzyme-free detection of specific mRNA of a housekeeping gene. Targeted mRNA was immobilized by hybridization to complementary DNA capture probes spotted onto a microarray. A second hybridization step of Cy5-conjugated label DNA to another section of the mRNA enabled specific labeling of the target. Thus, enzymatic artifacts could be avoided by omitting transcription and amplification steps. This manuscript describes the development of capture probe molecules used in the transcription- and amplification-free analysis of RPLP0 mRNA in isolated total RNA. An increase in specific signal was found with increasing length of the target-specific section of capture probes. Unspecific signal comprising spot autofluorescence and unspecific label binding did not correlate with the capture length. An additional spacer between the specific part of the capture probe and the substrate attachment site increased the signal significantly only on a short capture probe of approximately 30 nt length. microarraymRNA detectionenzyme-freegene expressionfluorescence microscopy ==== Body 1. Introduction The majority of methods for mRNA expression analysis, such as quantitative polymerase chain reaction (qPCR) [1], RNA-sequencing [2,3] and microarray analysis [3], start with the reverse transcription (RT) of mRNA into complementary DNA (cDNA). The RT step, however, can introduce quantification bias due to variable efficacy and fidelity of the RT enzyme. For instance, Ståhlberg et al. [4] found RT yields varying up to 100-fold dependent on the RT enzyme and the target gene. Moreover, with oligo(dT) primers, which are often used for priming the RT reaction, only polyadenylated RNA molecules are reverse-transcribed, while RNA species without poly(A) tail [5] are not available for analysis [6]. Furthermore, cDNAs often do not represent full-length mRNA molecules, which can limit the subsequent analysis to certain mRNA sections [6]. Reverse transcription can also introduce artifacts due to template switching [7,8], primer-independent cDNA synthesis [9] and DNA-dependent DNA polymerase activity [10]. Being the starting point for exponential quantification, the RT step is therefore one of the main contributors to technical variation in RT-qPCR analysis [11,12]. Exponential amplification of nucleic acids with PCR is moreover inherently error-prone, for instance, by introduction of Taq DNA polymerase errors and the formation of chimeric and heteroduplex molecules [13]. Both PCR and linear amplification based on in vitro transcription with T7 RNA polymerase can moreover introduce amplification bias [14,15]. Single-molecule fluorescence in situ hybridization (smFISH) is a widely used technology for direct detection of mRNA without prior reverse transcription or target amplification [16]. While being a highly specific and sensitive method [16], smFISH assays are generally limited to multiplexing up to a dozen transcripts by virtual color barcoding [17,18]. In contrast, microarray technology allows for the parallel analysis of tens of thousands of genes due to spatial separation of individual target genes. The NanoString nCounter gene expression assay is another hybridization-based system which enables the detection of target mRNA without prior reverse transcription. The protocol has a multiplexing capability of several hundred target genes per sample [19]. Analysis of single cells, however, still requires reverse transcription and linear PCR amplification prior to hybridization [20,21]. Our group has recently shown an enzyme-free double-hybridization assay for the specific detection of short DNA and RNA oligonucleotides on a microarray with single-molecule sensitivity [22]. In the present report, we optimized this approach to enable the analysis of endogenous mRNA of a housekeeping gene contained in isolated total RNA. This report presents design improvements for capture probes used in the transcription- and amplification-free analysis of RPLP0 mRNA on a microarray, which may be also applied for the development of probes for other genes. 2. Materials and Methods If not otherwise stated, chemical reagents were purchased from Sigma-Aldrich (Vienna, Austria). Denatured ethanol (EtOH) was purchased from Carl Roth (Vienna, Austria), deionized formamide was purchased from PanReac AppliChem (Darmstadt, Germany), EtOH absolute was purchased from AustrAlco (Spillern, Austria), and UltraPure salmon sperm DNA solution was purchased from Thermo Fisher Scientific (Vienna, Austria). Cell culture media, media supplements and antibiotics were purchased from Thermo Fisher Scientific. All buffers were prepared using ultrapure water (Milli-Q, 18.2 MΩ·cm at 25 °C, Merck Millipore, Vienna, Austria) and filtered through sterile 0.22 µm polyvinylidene fluoride (PVDF) syringe filters (Carl Roth, Austria). Surfaces and instruments were wiped with the surface decontaminant RNase Away (Carl Roth, Austria) prior to work. 2.1. Panc-1 Cell Culture As described elsewhere [23], the human pancreas epithelioid carcinoma cell line Panc-1 (ATCC CRL-1469) was maintained in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 100 units/mL penicillin and 100 μg/mL streptomycin (PenStrep) in a humidified atmosphere at 37 °C and 5% CO2. The cells were routinely passaged at 80%–90% confluence twice a week. 2.2. Double-Hybridization Principle As illustrated in Figure 1b, the double-hybridization approach was based on specific immobilization of target mRNA molecules by hybridization to spotted complementary DNA capture probes on a microarray. Fluorescent labeling of target mRNA was achieved by hybridizing 5′‑Cyanine 5 (Cy5)‑labeled complementary DNA probes (subsequently denoted as labels) to a different section of the mRNA. This approach has been shown before for the ultra-sensitive detection of in vitro synthesized short DNA and RNA molecules in a dynamic microfluidic chip [22]. 2.3. Probe Design All capture and label probes were designed in silico based on GenBank reference sequences [24]. Probe sequences were chosen such that cross-hybridization events and secondary structures were minimized. Moreover, a GC content between 40% and 60% should be obtained. Lyophilized probes were purchased from Microsynth (Balgach, Switzerland). Label probes were dissolved in Tris-EDTA (TE) buffer (1 mM ethylenediaminetetraacetic acid (EDTA), 10 mM Tris; pH 7.0) and kept in the dark. 3′‑Amino‑modified capture probes were dissolved in TE buffer (pH 8.0) according to the manufacturer’s instructions. All oligonucleotides were stored as single-use aliquots at −80 °C. The sequences of all probes are stated in Table S1. 2.4. Isolation of Total RNA for Microarray Analysis Total RNA for microarray analysis was isolated from Panc-1 cells with an RNeasy Mini Kit (Qiagen, Vienna, Austria) according to the manufacturer’s instructions. In brief, 3 × 106 Panc-1 cells were lysed by adding 350 µL lysis buffer (RLT buffer) and thoroughly resuspending. After mixing the homogenized lysate with 350 µL of 70% EtOH absolute, the solution was transferred to an RNeasy spin column placed in a collection tube. Following centrifugation (15 s, 10,000× g), the flow-through was discarded and the column was washed sequentially with 700 µL RW1 buffer and 2 × 500 µL RPE buffer. Purified total RNA was subsequently eluted in 30 µL nuclease-free water. 2.5. DNA Microarray Fabrication A Microgrid II contact-printer and SMP2 stealth pins (for a spot diameter of approximately 62.5 µm) (ArrayIt, Sunnyvale, CA, USA) were used to print 3′-amino-(CH2)7-modified probes (subsequently denoted as captures) on epoxy-functionalized substrates (NEXTERION Slide E, 50 × 24 × 0.175 mm3, Schott Technical Glass Solution GmbH, Jena, Germany). Therefore, the capture probes were diluted in spotting buffer (1.5 M betaine monohydrate, 0.001% (w/v) CHAPS, 0.005% (w/v) sodium dodecyl sulfate (SDS), 4× saline sodium citrate buffer (SSC)) to a working concentration of 15 µM. All substrates were washed with denatured EtOH and nuclease-free water prior to spotting. Following spotting, the substrates were allowed to rest at room temperature overnight for covalent attachment of the amine-modified capture probes to the epoxy-functionalized substrate. Excess capture probes were washed away by dipping the substrate into three beakers of washing buffer (0.1% (w/v) SDS, 1× SSC) and three beakers of nuclease-free water. 2.6. Double-Hybridization of RPLP0 mRNA As depicted in Figure 1, spotted substrates were incubated with 50 µL 1% (w/v) SDS, 50 µL blocking buffer (100 mM ethanolamine, 0.1% (w/v) SDS, 100 mM Tris; adjusted to pH 9.0), and 50 µL washing buffer at 40 °C for 10 min each prior to hybridization. During incubation, all solutions were covered with plastic coverslips (22 × 22 mm2, Cole-Parmer). All incubation steps were performed in a hybridization cassette (ArrayIt, Sunnyvale, CA, USA) that contained water-filled reservoirs to prevent evaporation of buffers. Afterwards, the substrates were dipped into a beaker of nuclease-free water to get a clean and dry surface. A LifterSlip™ (Science Services, Munich, Germany) with a volume of 7.6 µL (18 × 18 mm2) was placed atop of the microarray. After filling the LifterSlip™ with hybridization mix (HM), the hybridization cassette was closed and heated to 70 °C for 5 min to denature mRNA secondary structures. The HM comprised either 100 ng/μL isolated total Panc-1 RNA or the equal volume of nuclease-free water for mock hybridization experiments, as well as 1% (w/v) SDS, 3.5 × SSC, 1 mM EDTA, 50% (v/v) deionized formamide, 100 μg/mL UltraPure salmon sperm DNA solution, 100 pM Control(+) label and 100 pM Control(+) RNA. The RPLP0-specific labels that were used in the respective experiments are listed in Table 1. Overnight incubation at 40 °C in the dark allowed for double-hybridization of RPLP0 mRNA to complementary capture and label probes. 2.7. Fluorescence Microscopy Imaging was performed on an Axiovert 200 inverted microscope (Zeiss, Vienna, Austria) equipped with a motorized scanning stage (Scan IM 120 × 100, Märzhäuser, Wetzlar, Germany) and a CCD camera (CoolSnap HQ, Photometrics, Tucson, AZ, USA). Fluorescence imaging was performed in time delay and integration (TDI)‑mode [25] using a 642 nm diode laser (iBeam, Toptica Photonics AG, Gräfelfing, Germany) and 111 ms effective illumination time. Samples were illuminated through a long-pass filter (OG-550, Schott, Mainz, Germany) and a high-NA 100 × objective (α-Plan FLUAR 100 × 1.45 oil, Zeiss, Vienna, Austria). The obtained fluorescence signals were separated from excitation light by a dichroic beamsplitter (Q660LP, Chroma, Bellows Falls, VT, USA) and an emission filter (HQ700/75M, Chroma) before being imaged on the CCD camera. The microscope setup was further equipped with a focus hold system, which kept the distance from objective to sample constant [26,27]. Prior to imaging, the LifterSlip™ was removed from the microarray and unbound components were washed away by shaking the hybridized coverslip in 30 mL washing buffer (preheated to 37 °C) for 5 min. To prevent drying of the microarray surface during fluorescence measurement, the substrates were covered with measurement buffer (1% (w/v) SDS, 3.5 × SSC, 1 mM EDTA) filled into LifterSlip™ coverslips. All microarrays were heated to 42 °C using an objective heater (TempControl 37-2 digital, PeCon, Erbach, Germany) during measurement to avoid precipitation of the measurement buffer. 2.8. Data Analysis Analysis of the average spot brightness was performed with MATLAB (MathWorksTM Inc., Natick, MA, USA) as previously described [22]. In brief, for each spot, a sub-image was extracted from the raw data for further processing. To determine the average net intensity of the microarray spots, the local background adjacent to the respective spot was subtracted from the mean fluorescence intensity of the whole spot sub-image. Statistical analysis was conducted in SigmaPlot 12.0 (Systat Software, Inc., Washington, WA, USA). A one-way ANOVA was applied to test for statistically significant differences of net spot intensities. A p value of < 0.05 was considered to be statistically significant. 3. Results 3.1. Label Probe Optimization The assay optimization was based on detection of the large ribosomal protein P0 (RPLP0). The lengths of the initial label and capture probes were based on a recent publication of our group dealing with double-hybridization of nucleic acids in a microfluidic chip [22]. First, the label probe concentration for double-hybridization analysis of RPLP0 mRNA in 100 ng/µL isolated total Panc-1 RNA was optimized. Therefore, a RPLP0 label dilution series was conducted with 100 pM–100 nM Label_33nt probe concentration while keeping the RNA concentration constant. As shown in Figure S1, the maximum specific RPLP0 signal was obtained with a label concentration of 10 nM, wherefore this label concentration was used in all subsequent experiments. Then the effect of label probe elongation was tested. Signal intensities obtained using a 29 nt long RPLP0-specific capture probe as well as three label probes with different lengths and/or RPLP0 mRNA target sections were compared. No increase in signal intensity was found with two approximately 50 nt long labels when comparing them to a label probe of approximately 30 nt length (Figure S2). To keep a maximum of specificity, the short label probe (Label_33nt) was used in the following experiments. However, identification of a possible trend for well-performing label probes would need the systematic analysis of a larger amount of different labels. 3.2. Evaluation of Capture Probe Length Next, the effect of specific capture probe elongation was tested by comparing capture probes with different lengths. A comparison of the specific signal obtained on 29 nt (Capture_29nt_I) and 47 nt long capture probes (Capture_47nt) revealed a higher signal on the longer capture (Figure 2a). Similar signal distributions (approximately 70:30 ratio for Capture_47nt) were found with three different labels (Figure S3). To investigate if the higher signal on the 47 nt long capture was due to a better accessibility of the mRNA target section, the modification of the probes used in the initial experiment (Figure 2a) was reversed, i.e., initially 3′-amino-modified probes (captures) were 5′‑Cy5‑modified (labels) and vice versa (Figure 2b). In detail, the label used in the initial experimental setup (Label_33nt) was transformed into a capture (Capture_33nt) and spotted onto a microarray. Furthermore, the former captures (Capture_29nt_I and Capture_47nt) were transformed into label probes (Label_29nt and Label_47nt). As can be seen in Figure 2b, the longer label (Label_47nt) did not deliver a higher signal compared to the shorter label (Label_29nt), which indicates different optimal lengths of label and capture probes, respectively. 3.3. Longer Specific Capture Probes Increase Specific Signal For a further comparison of different capture probe lengths, the 47 nt long capture probe (Capture_47nt) was truncated by 18 nt at its 3′-end to generate the capture probe Capture_29nt. Moreover, Capture_47nt was elongated by 18 nt at its 3′-end to generate Capture_65nt, which was in turn elongated by 28 nt to generate Capture_93nt (Table 2 and S1). As shown in Table 2, all specific elongation steps increased the specific signal. The most prominent signal increase was obtained for the elongation step of the specific capture sequence from 29 nt to 47 nt (7.0 ± 2.6-fold change of the net signal). When comparing fully complementary captures to equally long captures with an unspecific 18 nt or 28 nt long sequence at the 3′ end, the fully complementary captures always delivered a higher signal (Table 2). It was therefore concluded that captures with a longer specific sequence increase the specific signal. Importantly, no trend was found for the unspecific signal on captures with different lengths (Figure S4). 3.4. Effect of Capture Probe Spacers It was investigated whether additional spacers between the gene-specific part of the spotted capture probe and the substrate attachment site could further enhance the hybridization efficiency. Therefore, capture probes were tagged at their 3′-ends with either hexaethylene glycol spacers (HEG) or elongated by adding 18 or 28 non-specific nucleotides (NONS), respectively (Figure 3a, Table S1). As shown in Figure 3b–d, the positive effect of spacers decreased with increasing length of the gene-specific capture sequence. Thus, spacers were especially beneficial for the shortest investigated capture probe. 4. Discussion This report shows for the first time the enzyme-free and gene-specific detection of cellular mRNA of a housekeeping gene on a microarray. This was realized by optimizing an innovative double‑hybridization approach previously characterized for detection of synthetic DNA and RNA [22]. RPLP0 mRNA molecules were labeled by hybridization to complementary DNA label probes with a conjugated 5′-Cy5 fluorophore. Target mRNA was furthermore immobilized on a microarray by hybridization to spotted complementary DNA capture probes. Scanning with an ultra-sensitive fluorescence microscope enabled the highly sensitive detection of mRNA hybridization [22,25,28]. Similar to smFISH techniques, the absence of reverse transcription and amplification steps enabled direct counting of mRNA molecules without introducing enzymatic bias [4,13]. In this publication, we present the design optimization of capture probes used in the direct double‑hybridization analysis of endogenous RPLP0 mRNA. Tests with RPLP0-specific capture probes of different lengths revealed that a target-specific elongation could increase the obtained net signal. Moreover, we found that unspecific spacers located between the gene-specific part of a capture and the substrate attachment site were especially beneficial for the shortest investigated capture probe. This is most likely due to surface effects including electrostatic and steric hindrance, which can reduce the hybridization rate [29,30]. In future experiments, the capture probe optimization presented in this report may also be useful for direct microarray analysis of RNA specimens labeled with alternative methods. Acknowledgments The authors wish to thank Simone Schweiggl and Michael Schobesberger for excellent technical assistance. This work was supported by the Austrian Science Fund (project L422-N20), by the Austrian Research Promotion Agency (FFG) under the scope of the RSA program (contract 844738), by the State of Upper Austria and by the European Fund for Regional Development. Supplementary Materials The following are available online at www.mdpi.com/2076-3905/5/1/5/s1. Click here for additional data file. Author Contributions Jan Hesse, Fritz Aberger and Michaela Haider conceived and designed the experiments; Michaela Haider performed the experiments; Michaela Haider and Jan Hesse analyzed the data; Alois Sonnleitner and Fritz Aberger contributed reagents/materials/analysis tools; and Michaela Haider, Jan Hesse and Thomas Haselgrübler wrote the paper. Conflicts of Interest The authors declare no conflict of interest. Figure 1 Assay principle: (a) Amino-modified microarray capture probes were spotted on epoxy-coated glass substrates. After washing away excess capture probes, the reactive surface of the substrates was subsequently blocked with 1% (w/v) sodium dodecyl sulfate and blocking buffer (grey). An 18 × 18 mm2 glass coverslip with raised edges (LifterSlip™) was placed atop the microarray. Then hybridization mix (red) was injected, mRNA secondary structures were denatured and overnight double-hybridization was performed. The next day, unbound compounds were washed away and the microarray was imaged using a fluorescence microscope; (b) Double-hybridization was based on immobilization of target mRNA on a spotted complementary DNA probe (capture). The mRNA was labeled by hybridizing a complementary Cy5-tagged DNA probe (label) to another section of the target mRNA. Figure 2 Capture/Label swap: (a) Using Label_33nt, the probe Capture_47nt delivered more signal compared to Capture_29nt_I. Bars represent the mean net spot intensities obtained on the RPLP0-specific captures depicted on the x-axis; (b) Upon transforming the former label into a capture (Capture_33nt) and the former captures into labels (Label_29nt, Label_47nt), hybridizations utilizing the elongated Label_47nt did not deliver a higher signal compared to hybridizations performed with Label_29nt. Bars represent the mean net spot intensities obtained on the RPLP0-specific capture Capture_33nt; (a,b) Error bars represent the standard deviation of the mean of three technical replicates. Asterisks indicate the significance level in a one-way ANOVA (* p < 0.05). Figure 3 Comparison of capture probes with and without unspecific spacers. (a) RPLP0‑specific capture probes of different lengths were compared with probes of equal specific length with an additional unspecific 3′ spacer attached. HEG (hexaethylene glycol)-modified capture probes comprised four hexaethylene glycol chains separated by two nucleotides each at their 3′ ends, NONS (non-specific) referred to additional 18–28 non-complementary nucleotides at the 3′ end of the respective capture probe; (b) An additional HEG spacer significantly increased the signal on Capture_29nt; (c,d) The signal on captures with 47 nt or 65 nt specific length could not be increased significantly with additional spacers; (b–d) Bars represent the mean net spot intensities normalized to the respective RPLP0-specific capture without additional spacer, i.e., to Capture_29nt (b); Capture_47nt (c); or Capture_65nt (d). Error bars depict the standard deviation of the mean of five (b,c) or four (d) technical replicates. Asterisks indicate the significance level in a one-way ANOVA (** p < 0.01). microarrays-05-00005-t001_Table 1Table 1 RPLP0-specific label probes used in double-hybridization experiments. Experiment RPLP0 a-Specific Label Probe b Label_29nt Label_33nt Label_47nt Label_51nt Label_52nt Capture/Label swap (Figure 2) 10 nM 10 nM 10 nM – – Effect of capture length and spacers (Table 2) – 10 nM – – – Spacer effects (Figure 3) – 10 nM – – – Label dilution series (Figure S1) – 100 pM, 1 nM, 10 nM, 100 nM – – – Label comparison (Figure S2) – 10 nM – 10 nM 10 nM Capture length comparison (Figure S3) – 10 nM – 10 nM 10 nM Unspecific signal with increasing capture length (Figure S4) – 10 nM – – – a Large ribosomal protein P0 (RPLP0); b Hybridization mixes comprised additionally one of the listed RPLP0-specific label probes in the concentration indicated in the table. microarrays-05-00005-t002_Table 2Table 2 Pair-wise fold-change on RPLP0-specific capture probes. Compared Capture Probes Fold-Change ± SD a n b Original Probe → Modified Probe Capture_29nt → Capture_47nt 7.0 ± 2.6 5 Capture_47nt → Capture_65nt 1.9 ± 0.8 9 Capture_65nt → Capture_93nt 2.0 ± 0.3 4 Capture_29nt+18 → Capture_47nt 3.2 ± 1.9 5 Capture_47nt+18 → Capture_65nt 2.3 ± 1.4 5 Capture_65nt+28 → Capture_93nt 2.0 ± 0.8 4 target-specific capture sequence; target-specific 3′-elongation; non-specific (NONS) 3′‑elongation; → indicates compared capture probe-pairs; a For all capture probe-pairs, the fold-change was calculated by dividing the modified probe value by the original probe value; The mean fold-change was obtained by averaging the fold-change values across the technical replicates; SD = Standard deviation of the mean fold-change; b n = Number of technical replicates. ==== Refs References 1. Nolan T. Hands R.E. Bustin S.A. Quantification of mRNA using real-time RT-PCR Nat. Protoc. 2006 1 1559 1582 10.1038/nprot.2006.236 17406449 2. Mutz K.-O. Heilkenbrinker A. Lönne M. Walter J.-G. Stahl F. Transcriptome analysis using next-generation sequencing Curr. Opin. Biotechnol. 2013 24 22 30 10.1016/j.copbio.2012.09.004 23020966 3. Zhao S. Fung-Leung W.-P. Bittner A. Ngo K. Liu X. Comparison of RNA-Seq and microarray in transcriptome profiling of activated T cells PLoS ONE 2014 9 10.1371/journal.pone.0078644 24454679 4. Ståhlberg A. Kubista M. Pfaffl M. Comparison of reverse transcriptases in gene expression analysis Clin. Chem. 2004 50 1678 1680 10.1373/clinchem.2004.035469 15331507 5. Yang L. Duff M.O. Graveley B.R. Carmichael G.G. Chen L.-L. Genomewide characterization of non-polyadenylated RNAs Genome Biol. 2011 12 10.1186/gb-2011-12-2-r16 21324177 6. Tang F. Lao K. Surani M.A. Development and applications of single-cell transcriptome analysis Nat. Methods 2011 8 S6 S11 10.1038/nmeth.1557 21451510 7. Roy S.W. Irimia M. When good transcripts go bad: Artifactual RT-PCR “splicing” and Genome Analysis BioEssays News Rev. Mol. Cell. Dev. Biol. 2008 30 601 605 10.1002/bies.20749 18478540 8. Cocquet J. Chong A. Zhang G. Veitia R.A. Reverse transcriptase template switching and false alternative transcripts Genomics 2006 88 127 131 10.1016/j.ygeno.2005.12.013 16457984 9. Haddad F. Qin A.X. Giger J.M. Guo H. Baldwin K.M. Potential pitfalls in the accuracy of analysis of natural sense-antisense RNA pairs by reverse transcription-PCR BMC Biotechnol. 2007 7 10.1186/1472-6750-7-21 17480233 10. Perocchi F. Xu Z. Clauder-Münster S. Steinmetz L.M. Antisense artifacts in transcriptome microarray experiments are resolved by actinomycin D Nucleic Acids Res. 2007 35 10.1093/nar/gkm683 17897965 11. Bengtsson M. Hemberg M. Rorsman P. Ståhlberg A. Quantification of mRNA in single cells and modelling of RT-qPCR induced noise BMC Mol. Biol. 2008 9 10.1186/1471-2199-9-63 18631407 12. Ståhlberg A. Håkansson J. Xian X. Semb H. Kubista M. Properties of the reverse transcription reaction in mRNA quantification Clin. Chem. 2004 50 509 515 10.1373/clinchem.2003.026161 14726469 13. Acinas S.G. Sarma-Rupavtarm R. Klepac-Ceraj V. Polz M.F. PCR-induced sequence artifacts and bias: Insights from comparison of two 16S rRNA clone libraries constructed from the same sample Appl. Environ. Microbiol. 2005 71 8966 8969 10.1128/AEM.71.12.8966-8969.2005 16332901 14. Degrelle S.A. Hennequet-Antier C. Chiapello H. Piot-Kaminski K. Piumi F. Robin S. Renard J.-P. Hue I. Amplification biases: Possible differences among deviating gene expressions BMC Genomics 2008 9 10.1186/1471-2164-9-46 18226214 15. Subkhankulova T. Livesey F.J. Comparative evaluation of linear and exponential amplification techniques for expression profiling at the single-cell level Genome Biol. 2006 7 10.1186/gb-2006-7-3-r18 16542485 16. Itzkovitz S. van Oudenaarden A. Validating transcripts with probes and imaging technology Nat. Methods 2011 8 S12 S19 10.1038/nmeth.1573 21451512 17. Kosman D. Mizutani C.M. Lemons D. Cox W.G. McGinnis W. Bier E. Multiplex detection of RNA expression in Drosophila embryos Science 2004 305 10.1126/science.1099247 15297669 18. Levsky J.M. Shenoy S.M. Pezo R.C. Singer R.H. Single-cell gene expression profiling Science 2002 297 836 840 10.1126/science.1072241 12161654 19. Geiss G.K. Bumgarner R.E. Birditt B. Dahl T. Dowidar N. Dunaway D.L. Fell H.P. Ferree S. George R.D. Grogan T. Direct multiplexed measurement of gene expression with color-coded probe pairs Nat. Biotechnol. 2008 26 317 325 10.1038/nbt1385 18278033 20. Guo G. Luc S. Marco E. Lin T.-W. Peng C. Kerenyi M.A. Beyaz S. Kim W. Xu J. Das P.P. Mapping cellular hierarchy by single-cell analysis of the cell surface repertoire Cell. Stem Cell. 2013 13 492 505 10.1016/j.stem.2013.07.017 24035353 21. McDavid A. Dennis L. Danaher P. Finak G. Krouse M. Wang A. Webster P. Beechem J. Gottardo R. Modeling bi-modality improves characterization of cell cycle on gene expression in single cells PLoS Comput. Biol. 2014 10 10.1371/journal.pcbi.1003696 25032992 22. Mayr R. Haider M. Thünauer R. Haselgrübler T. Schütz G.J. Sonnleitner A. Hesse J. A microfluidic platform for transcription- and amplification-free detection of zepto-mole amounts of nucleic acid molecules Biosens. Bioelectron. 2016 78 1 6 10.1016/j.bios.2015.11.013 26580983 23. Eberl M. Klingler S. Mangelberger D. Loipetzberger A. Damhofer H. Zoidl K. Schnidar H. Hache H. Bauer H.-C. Solca F. Hedgehog-EGFR cooperation response genes determine the oncogenic phenotype of basal cell carcinoma and tumour-initiating pancreatic cancer cells EMBO Mol. Med. 2012 4 218 233 10.1002/emmm.201100201 22294553 24. GenBank Available online: http://www.ncbi.nlm.nih.gov (accessed on 13 Feburary 2014) 25. Hesse J. Sonnleitner M. Sonnleitner A. Freudenthaler G. Jacak J. Höglinger O. Schindler H. Schütz G.J. Single-molecule reader for high-throughput bioanalysis Anal. Chem. 2004 76 5960 5964 10.1021/ac049300f 15456321 26. Freudenthaler G. Sonnleitner M. Sonnleitner A. Device for the Microscopic Examination of Samples WO 2006/066286 A1 29 6 2006 27. Hesch C. Hesse J. Jacak J. Schütz G.J. Two-stage focus-hold system for rapid ultra-sensitive read-out of large-area biochips J. Microsc. 2009 234 251 254 10.1111/j.1365-2818.2009.03165.x 19493102 28. Hesse J. Jacak J. Kasper M. Regl G. Eichberger T. Winklmayr M. Aberger F. Sonnleitner M. Schlapak R. Howorka S. Muresan L. Frischauf A.-M. Schütz G.J. RNA expression profiling at the single molecule level Genome Res. 2006 16 1041 1045 10.1101/gr.4999906 16809670 29. Gao Y. Wolf L.K. Georgiadis R.M. Secondary structure effects on DNA hybridization kinetics: A solution versus surface comparison Nucleic Acids Res. 2006 34 3370 3377 10.1093/nar/gkl422 16822858 30. Shchepinov M.S. Case-Green S.C. Southern E.M. Steric factors influencing hybridisation of nucleic acids to oligonucleotide arrays Nucleic Acids Res. 1997 25 1155 1161 10.1093/nar/25.6.1155 9092624

PMC005xxxxxx/PMC5003452.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays3010001microarrays-03-00001ArticlePigeons: A Novel GUI Software for Analysing and Parsing High Density Heterologous Oligonucleotide Microarray Probe Level Data Lai Hung-Ming 12†May Sean T. 13Mayes Sean 14*1 School of Biosciences, University of Nottingham, Sutton Bonington, Loughborough LE12 5RD, UK; E-Mails: hung-ming.lai@kcl.ac.uk (H.-M.L.); sean@arabidopsis.org.uk (S.T.M.)2 Department of Informatics, King’s College London, Strand, London WC2R 2LS, UK3 Nottingham Arabidopsis Stock Centre (NASC), University of Nottingham, Sutton Bonington, Loughborough LE12 5RD, UK4 Crops for the Future Research Centre, University of Nottingham Malaysia Campus (UNMC), Jalan Broga, Semenyih 43500, Malaysia† The first author carried out the research at the University of Nottingham and is now at King’s College London. * Author to whom correspondence should be addressed; E-Mail: sean.mayes@nottingham.ac.uk; Tel.: +44-115-951-6234; Fax: +44-115-951-6060.03 1 2014 3 2014 3 1 1 23 11 11 2013 17 12 2013 19 12 2013 © 2014 by the authors; licensee MDPI, Basel, Switzerland.2014This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Genomic DNA-based probe selection by using high density oligonucleotide arrays has recently been applied to heterologous species (Xspecies). With the advent of this new approach, researchers are able to study the genome and transcriptome of a non-model or an underutilised crop species through current state-of-the-art microarray platforms. However, a software package with a graphical user interface (GUI) to analyse and parse the oligonucleotide probe pair level data is still lacking when an experiment is designed on the basis of this cross species approach. A novel computer program called Pigeons has been developed for customised array data analysis to allow the user to import and analyse Affymetrix GeneChip® probe level data through XSpecies. One can determine empirical boundaries for removing poor probes based on genomic hybridisation of the test species to the Xspecies array, followed by making a species-specific Chip Description File (CDF) file for transcriptomics in the heterologous species, or Pigeons can be used to examine an experimental design to identify potential Single-Feature Polymorphisms (SFPs) at the DNA or RNA level. Pigeons is also focused around visualization and interactive analysis of the datasets. The software with its manual (the current release number version 1.2.1) is freely available at the website of the Nottingham Arabidopsis Stock Centre (NASC). affymetrixheterologous microarrayoligonucleotide probe selectionPigeonsprobe pair data analysisSFPsXspecies ==== Body 1. Introduction Microarrays have become a powerful and widely exploited tool when studying the complete gene expression profiles of a multitude of cells and complex tissues in many different organisms. The major technical advance was the hybridisation of reverse transcribed RNA from tissues or cells to either cDNA or oligonucleotides fixed on glass slides or on a nylon membrane [1]. High-density oligonucleotide gene expression arrays have recently been applied to many areas of biomedical research to assess the abundance of mRNA transcripts for many genes at the same time [2]. Affymetrix (Santa Clara, CA, USA) generated GeneChip® arrays and dominated the market of high-density microarray for many years. Although significant quantities of informative, reproducible, and high quality data is generated by the use of a GeneChip® for expression profiling, the Affymetrix chips are only available for a limited number of species of eukaryotes and a small number of model/commercial plant species, including Arabidopsis thaliana, barley, rice, maize, tomato, soybean, sugar cane, grape and wheat [3,4]. A genomic DNA-based probe selection technology, known as the Xspecies approach, has been developed to investigate the transcriptomes of heterologous plants and to allow the sensitivity of high-density oligonucleotide microarrays to be applied to species where chips have not yet been designed [3,4]. The approach begins with a genomic DNA/DNA hybridisation, hybridising DNA from species X onto an appropriate Affymetrix GeneChip® of a heterologous species. The next step uses a Script to parse an Affymetrix CDF file of the selected chip. The parser uses the CDF file of the chip and the CEL file of the hybridisation to identify and remove “bad” probe-pairs whose perfect match probe intensities are below a cut-off value defined by the user, eventually making a “new” CDF file for Species X [5]. The new probe–masked file, namely the species X.CDF, can be used for Xspecies transcriptomic analysis of RNA hybridisation. Hammond et al. [3] showed that the Xspecies approach had been successfully applied to analyzing the transcriptome of Brassica oleracea L. by labeling gDNA from B. oleracea and hybridising it to the ATH1-121501 (ATH1) GeneChip® array. The approach with heterologous oligonucleotide microarrays was also utilised to profile and to compare the transcriptional levels of Thlaspi caerulescens and Thlaspi arvense, both being species where no GeneChip® is available [4]. A further application of this novel approach was to examine the evidence for neutral transcriptome evolution in plants by quantifying more than 18,000 genes transcripts at the level of 14 taxa from the Brassica family [6]. However, the original script parser has a specific limitation in choosing the cut-off - the selection of the value is essentially arbitrary, although a more recent iteration does allow a degree of sub-sampling to suggest thresholds. One method to improve on this approach is to generate many custom CDF files according to different cut-offs, from low to high. Then, a range of good probes pairs and probe-sets with respect to the chosen specific cut-offs are obtained. The researcher, using a spreadsheet, plots these data as background information and uses them to finally decide the optimal value of the cut-off and the corresponding CDF file [7]. The approach is valid but is still human-dependent since people choose the threshold based on their observations and experience when looking at the plot. Recently, oligonucleotide arrays have been used to recognise allelic variation, the variants being termed single-feature polymorphisms (SFPs). The polymorphism is often detected by a single probe in an oligonucleotide array—the so-called “features”. There is no a priori understanding of the DNA nature of the polymorphism, simply that it is a reproducible polymorphism. With this cross-species approach using Affymetrix GeneChip®s, researchers have the ability to screen hybridisation datasets for potential SFP markers that exist in minor species. Thus, it is essential to design biological and algorithmic approaches for heterologous oligonucleotide microarray analysis, to help facilitate the genomic investigation of minor plants and animals. Here, we have developed an innovative software package “Pigeons”, abbreviated from “Photographically InteGrated En-suite for the OligoNucleotide Screening”, to work towards a solution to the issues mentioned above. Pigeons allows the user to input and analyse microarray data from the Xspecies microarray approach. This can be DNA hybridisations across species, to determine the empirical boundaries for custom CDF files for Xspecies transcriptomics or to examine an experimental design to identify SFPs at single oligonucleotides within the probe-sets, either at the DNA or RNA level. To allow intuitive interaction and final selection of features of interest, we have also developed a specific visualization interface to facilitate navigation through the hundreds of thousands of Affymetrix oligonucleotides. 2. Methods and Algorithms In this paper, there are three algorithms (automated threshold mapping (ATM), dual fold-change (DFC), probe-wise one-sample statistical test (POST)) presented to fulfill the needs of analysing and parsing the Xspecies microarray data at the probe level. We aim to improve on current Xspecies parser scripts by using several traditional and modern computing techniques including interpolation, projection and clustering [8,9]. Meanwhile, recent microarray gene selection approaches, such as a fold-change (FC) analysis and a variety of statistical tests [10,11,12], have also been extended and modified to address the issue of searching for the single oligonucleotides containing the feature of interest. The experimental material used for this paper is derived from the underutilized African legume species Bambara groundnut (Vigna subterranea (L) Verdc.) which is grown as part of subsistence and small-scale agriculture in many of the sub-Saharan countries of Africa [13,14]. A controlled cross between a genotype derived from a wild non-domesticated landrace (VSSP11; Parent 1; P1; “1”) and a genotype derived from a domesticated landrace (DipC; Parent 2; P2; “2”) was made and a single hybrid seed (F1) allowed to grow and produce an F2 population of seed. This population was planted and recorded at the Tropical Crops Research Unit at the University of Nottingham in 2003. Individual plants were recorded for numerous traits, including “number of stems per plant”. The extremes of the “number of stems per plant” distribution were identified and 10 plants from each extreme had DNA extracted by standard techniques and mixed in equal amounts to produce a bulked sample of “low stem number” (“3”) and a bulked sample of “high stem number” (“4”), respectively. 2.1. ATM As a mixed model of numerical analysis and a soft computing technique, a heuristic method called Automated Threshold Mapping (ATM) was proposed to improve on the human-dependent cut-off selection of poorly hybridising oligonucleotide probes. One of the requirements to be able to exploit an Xspecies array is to select a threshold to generate a custom CDF file for further analysis. Therefore, it is necessary to understand the relationships between a particular threshold value, the probe-pairs and the probe-sets retained at this threshold, i.e., three two-ways and one three-way comparisons. Because this problem involves one input (threshold level) and two outputs (probe-pairs and probe-sets), an idea was drawn from vector calculus to assess the relationships among the three variables and to generalize a solution to this problem. Through the generation of a plane curve (Figure 1), we have found that the retained probe-sets and the retained probe-pairs decline when the threshold value is increased and that the relationship between the two retained variables is a monotonic function. This relates to the point that a probe-set is removed only if there are no retained probe-pairs in that probe-set, so that the number of retained probe-pairs declines more sharply than the number of retained probe-sets does, when the threshold value rises. We also find that the plane curve is like a learning curve with a plateau. Thus, an appropriate selection of threshold values could come from the portion of the curve (circle in Figure 1) between the end of the plateau and the beginning of the linear-like drop. Considering the greyness of the position, we want to provide a suggested threshold value, together with an interval of feasible thresholds available for selection using projection, fuzzy clustering and interpolation techniques. From the observation of the plane curve, given a series of vectors that consist of a threshold and its retained units, the vectors are initially projected onto the retained probe-set space, where fuzzy clustering is performed. Since the section of the curve targeted is a limited bridge between the plateau and the linear-like drop, a good fuzzy clustering approach would lead the bridge to a refined overlap of the first two clusters. Based on this, a suggested threshold value could be produced by an interpolation technique. Figure 1 Plane Curve. A vector valued function traced out by retention units with respect to the cut-off of poorly hybridising oligonucleotides using the heterologous GeneChip® platform, with ATH1-121501 used as the basis to generate the image. To define the methods and principles mathematically, first of all, a vector-valued function is introduced to perform an in-depth analysis of the problem. Let X be a scalar variable and Y be a vector variable with two dimensions. A vector function F: → 2 is defined as follows: (1) where X is a set of cut-offs and the component functions f1 and f2 are real-valued functions of the parameter x. The two components of Y, Y1 and Y2, are therefore viewed as sets of retained probe-pairs and probe-sets, respectively, when a defined cut-off is given. Using the vector-valued retention function F, we can easily trace the graph of a curve to know the relationships among cut-off and the retention units of probe-pairs and probe-sets. The point of the position vector F(x) coincides with the point (y1, y2) on the plane curve given by the component equations, as shown in Figure 1. The arrowhead on the curve represents the curve’s orientation by pointing in the direction of increasing values of x, namely x3 > x2 > x1. Due to the nature of the problem, the retention function F monotonically decreases in the direction of the point (0, 0). This characteristic means that mapping from y1 to y2 is also a monotone function, and moreover, it is actually like a learning curve with a stagnant occurrence. A tangent vector-based numerical analysis could be applied to the evaluation and the differentiation of the function at a given point. For example, a turning point F(xtp) can be defined as the intersection between a tangent to the stagnant phase of the curve and the tangent to the linear-like decreasing portion of the curve. The inverse of this point F−(F(xtp)) could be selected as the threshold value. However, the cut-off decision problem is not deterministic, and it usually needs to take biological sense into account, so requires more tolerance in the selection of the threshold. The ATM offers a turning portion (TP) covering the turning point and derived from a closed interval I from which realistic thresholds can be retrieved. Let I be the surrounding area of xtp ϵ X such that F(xtp) is the turning point, and then we construct the turning portion by TP={F(x): x ϵ I }. Construction requires careful definition of a lower boundary (xlb) and an upper boundary (xub) of I, with the aim of developing the idea of selecting a flexible region, rather than a single turning point. Since F is a one-one function well-defined in the interval I, which decreases monotonically; in theory, we can define xlb and xub such that F(xlb) would be in the terminating phase of the plateau and F(xub) would be in the earliest phase of sharp decline, respectively. The ATM is a data-driven mapping method using a two-stage unsupervised learning process for I determination. The first stage involves orthogonal projection in order to highlight the turning portion. To achieve this, we consider an inner-product vector space = 3, let be an r-dimensional subspace of and ┴ be the orthogonal complement of . Given a matrix B3×r such that the column space of B is , and then for there exists a projector P to project v onto along ┴, i.e., Pv = u, u ϵ . The unique linear operator P can be acquired by P = B(BTB)-1BT, in particular, if B constitutes orthonormal bases, then P = BBT. During simplification of the system, the goal at this stage is to minimize the loss of information relevant to the problem of concern. As a consequence, given B (e.g., [0,0,1]T) and n numbers of vectors of thresholds with their retention units, and after linear transformation of each vector vi ϵ , i = 1,…, n we will then gain a learning data set D = {ui: Pvi = ui ϵ that ideally has the most informative features for turning portion discovery. Suppose that all the data vectors in TP have been projected onto a particular area, we define the area as a hotspot D′ such that (2) I = [xinf(J), xsup(J)] (3) where J is an index set to collect and distinguish the elements of the hotspot, and inf(J) & sup(J) denote infimum and supremum of J respectively. Obviously, J is a subset of {1, …, n}, and both D' and J are well-ordered closed sets. In other words, the second task is to identify the hotspot to discover a range of feasible cut-off thresholds. Grouping methods would be appropriate for this task as the object of clustering is to group a set of data vectors to include only those vectors which are similar to each other. Although there are similarities between data points within a group to a certain extent, it is also believed that some of the similarities might also occur between groups. This is due to the intrinsic design element aiming to develop a flexible choice of realistic thresholds. Some elements within the turning portion of the curve are closer to the end of plateau, others are near to the beginning of linear-like decline, and still others will be around the turning point of the curve. Part of the problem with depicting the hotspot is to capture the “grayness” of the cross-cluster similarities so it is essential to allow some degree of uncertainty in its description. The ATM applies Fuzzy c-Means (FCM) clustering to this issue since the FCM allows us to build clusters with vague boundaries, where some overlapping clusters include the same object, to a certain degree [15]. Based on an objective function or performance index , the weighted within-class sum of squares, to quantify how good the quality of clustering models is, the FCM attempts to find the best allocation of data to clusters with a gradual membership matrix M. Given a number of clusters c (1 < c < n), then the learning data set is dominated by fuzzy sets and the fuzzy partition matrix M =[mij]c×n, where and . For the individual entries in M, mij are the membership degree of element uj ϵ D to cluster i, i.e., . Let be a set of cluster prototypes so that each cluster is represented with a cluster centre vector ωi, and the objective function with two constraints can then be defined as below: (4) (5) (6) Here, q ϵ >1 is termed the “fuzzifier” or weighting exponent, and dij is the distance between object uj and cluster centre ωi, within ATM, the Euclidean inner product norm denoted by ‖∙‖ is taken, i.e., dij = ‖uj - ωi‖. The purpose of the clustering algorithm is to obtain the solution M and Ω minimizing the cost function , and this can be carried out by: (7) (8) namely the FCM proceeds with two events: the computation of cluster centroids and the allocation of data elements to these centroids. In practice, the cost function is minimized by an alternating optimization (AO) scheme, i.e., the membership degrees are first optimized given recently fixed cluster parameters, followed by optimizing the cluster prototypes given currently fixed membership degrees. This reiterative procedure will be repeated until the cluster centres have reached equilibrium which is equivalent in mathematics to the optimal objective function . After the grouping scheme is accomplished, the hotspot D' can be deciphered by defining the greatest lower bound and the least upper bound of its index set. The first two clusters (, ) are concentrated for the purpose of deciphering since most of elements of are very likely to be projected from vectors in stagnant phase while data points near the beginning of the sharp drop have mostly fallen into . Thus, we let D′ be the subset of the union of the two clusters and set the infimum and the supremum of J according to the objects whose membership values are the maximum of and , i.e., (9) Not only do the above equations define the index set J, but also they reveal that the tolerance interval I has been established. Besides the selection of feasible cut-offs, the ATM also provides an automated threshold value xATM and a target interval I' for the selection of candidate cut-offs. Both xATM and I' are evaluated through the fuzzy boundary between the first two fuzzy sets. The elements in the boundary imply that and have them in common with various membership values. Owing to the grayness characteristic and the continuity of the learning-like curve, we believe that a good threshold value for parsing the Affymetrix chip description files would come from a projected object that simultaneously belongs to the two clusters with remarkable membership degrees. As a result, the fuzzy boundary can enable us to offer a more reasonable selection of threshold boundary cut-offs. Two indices, l and k, are utilised to determine the highly likely threshold boundary cut-offs and the automated threshold value, determined by (10) Here ϵ is a small number to assess the possibility of the overlap between the two clusters. By this definition, the fuzzy boundary is then portrayed as the set of and another closed interval [xl,xk] is constructed as the target interval I'. Let be u the arithmetic mean of the elements of , and xATM can also be calculated by linear interpolation or by the Lagrange polynomial, as shown in the following formulae: (11) In summary, the ATM returns a 3-tuple (xATM,I',I) to resolve the issue of the threshold cut-off choices. The suggested cut-off given by the ATM, xATM, can directly be exploited to remove the weak intensity signals while any values within a target interval, I' = [xl,xk], can be taken as the potential threshold boundary cut-offs. The design of the target interval gives users a chance of picking a scientifically reasonable value on their own. Those values in a tolerance interval, i.e., x ϵ I= [xinf(J),xsup(J)], can be used as feasible thresholds and values outside the interval are viewed as less feasible choices. 2.2. DFC Dual fold-change analysis (DFC) is an approach to seek potential single-feature polymorphism markers through screening all of the 25-mer oligonucleotide probes of the heterologous microarray. Initially, there are two groups (G1 and G2) under the design of the single trait experiment. While two distinct parental genotype gDNAs are involved in generating G1, G2 is composed of two different phenotypically based F2 bulk segregant pools, derived from a hybrid between the two parental genotypes. We then label the four Xspecies chips with 1 & 2 for the two parent samples and with 3 & 4 for the two F2 bulks. In practice, these F2 bulks are constructed from the pooled DNA of F2 individuals. These are derived from the controlled cross between the parental genotypes with allocation to the contrasting bulk based upon a specific trait of interest. The phenotype classification is a necessary prerequisite for the numerical analysis of potential SFP markers. 1 and 3 are classified into one type under a single trait experiment whereas 2 and 4 belong to the other trait version—the prerequisite can be denoted as . Let N be the number of genes and #() be the cardinal number of a probe-set then each chip can be represented as follows: (12) (13) where bijm denotes the j-th signal intensity of the i-th probe-set on the m-th chip. Let Qij1 = bij1/bij2 and Qij2 = bij3/bij4 be the intensity ratio of G1 and G2, respectively, thus the ratio value of one for this feature represents unchanged hybridisation signal in this experiment and less than or greater than one is for differentially hybridised oligonucleotides. To generate a symmetric distribution of intensity ratios, the fold-change ratio is defined by (14) where FCij1 is used to assess the differential probe hybridisation of the parental group. For the evaluation of the offspring group, FCij2 is calculated in the same way as FCij1 simply replacing Qij1 with Qij2. Given the threshold of weak signals xATM, the cut-off of a fold-change between the parents and that between the offspring , a number of logical criteria are applied to globally screen and search Affymetrix’s single oligoprobes for SFP markers. For , let the first condition be bijm > xATM since any signals whose intensities are below the threshold should not be used for good probes in the analysis of heterologous data—this satisfies the demand of the XSpecies technology. When the first criterion holds, the DFC enables the procedure to run the second condition with the two fold-change indicators FCij1 and FCij2, FCij1 ≥ ϵ1 and FCij2 ≥ ϵ2, to measure whether still holds at the genomic level. The FC approach is commonly used in microarray data analysis to identify differentially expressed genes (DEGs) between a treatment and a control. Calculated as the ratio of two conditions/samples, the FC gives the absolute ratio of normalized intensities in a non-log scale. We extend the same concept in our approach by introducing an additional FC—one ratio assesses the differential hybridisation within G1 and the other assesses the differential hybridisation within G2. The extra FC tests whether the difference in phenotype could result from a difference in genotype at a single locus. Therefore, when there are any differentially hybridised oligonucleotides for the feature of interest between the two parental genotypes, the inherited attribute of would imply that we could expect those differentially hybridised oligonucleotides to have also been transmitted into the F2 individuals. In a word, the corresponding fold-change of the F2 is introduced as a cross-check mechanism for identifying SFPs which are consistent between parental genotype/trait and bulk genotype/trait. The mixture of F2 genotypes (which are bulked according to the trait difference which segregates within the cross) should mean that the attribute difference is only detected when the location of the parental SFP is close to the gene controlling the trait difference. The accuracy of this approach is dependent upon bulk size used. Smaller bulk sizes will lead to the identification of SFPs which are located distantly from (and probably on different chromosomes to) the target trait associated SFPs. Oligo-probes that satisfy the second criterion above are potential SFP markers distinguishing the two phenotypes and could be further tested and used for genetic mapping of the gene controlling the phenotypic difference. 2.3. POST The FC is typically viewed to be significant if there is at least a two-fold difference [10]. In addition, the FC threshold is selected arbitrarily and does not involve any assessment of statistical confidence so using the FC approach alone may not be optimal [11,16]. Although it is a straightforward and intuitive way to detect oligonucleotides using the dual fold-change criterion, the approach does not engage any evaluation of the significance of differential hybridisation in the presence of biological and experimental variation, which might differ from probe to probe. We have therefore developed inferential statistics herein through a method called the probewise one-sample statistical test (POST) for the assessment of the differential oligoprobe variation observed in terms of statistical power and measures of confidence. We first define an MA-value ρij for the examination of a signal variant in the single trait experiment, for the value is calculated by the following formula: (15) to exactly correspond with the experimental attribute of . The MA-value is named after the MA plot, a very useful tool in cDNA and GeneChip® microarray data analysis [17,18,19], and is the average intensity ratio between parental samples and F2 bulks in a base 2 logarithmic scale with a mnemonic for subtraction and a mnemonic for addition. The POST then uses the MA-value and a single sample t-test to statistically assess differentially hybridised oligonucleotides between the parent group and the offspring group and to test in a probe-set i whether or not there is significant difference between an interrogated probe k and the other probes in that probe-set, in terms of their log ratios. As a test statistic, the average of the MA-values of each of the probe-pairs except the probe k is denoted by ρik and determined by: (16) where ni = #(Bi) - 1 is the sample size in the examined probe-set i. Suppose that the sampling distribution of ρik is normal so that the random variable (17) has a Student’s t-distribution with ni - 1 degrees of freedom. Where Sik is the standard deviation of the sample of the log ratios in the i-th probe-set excluding the MA-value of the oligoprobe k. The last step performed by the POST is to asymptotically compute the p-value converting the value of Tik into a probability that expresses how likely the oligonucleotides in question are to be differentially hybridised. To visualize the results of this probewise testing of single oligonucleotides, a filter with a Volcano Plot output was also developed. The volcano plot is an effective and easy-to-interpret scatter plot for the selection of DEGs [11]. In the POST, the plot shows the negative common logarithm (base 10) of the p-value versus the average intensity ratio in the form of the binary logarithm (base 2), i.e., average fold-change ratio. Probe-pairs with large log ratios and low p-values are easily detectable in the view and a list of potential SFP markers can be generated. Another approach for statistical inference using a different measure based on intensity difference has also been implemented in the POST to identify and evaluate significantly variable oligonucleotides within an experimental group. Basically, the approach is a methodology analogous to that of testing between two groups, but it is more focused on variation within a single group. Since a potential SFP marker could be due to oligonucleotide target regions within the test genome with deletions or duplications or nucleotide differences with respect to the design probe-pairs, we propose using intensity difference rather than the traditional intensity ratio to determine significant differences in intensity between the signal of array elements within either the parent group G1 or the offspring group G2. We name the intensity difference the D-value, in contrast to the MA-value, and define it in compliance with the trait of interest as below: (18) Similar to statistical tests between groups, the sample mean of the D-value would be the statistic to test whether the intensity difference of the oligoprobe under interrogation is significantly different from that of the other signals in the same probe-set of G1 or G2; meanwhile, an ad hoc test procedure within G1 or G2 also assumes that the population distribution is at least approximately normal and proceeds with the probe-wise strategy. However, there are practical issues that need to be addressed. The majority of intensity signals are likely to be affected by poor hybridisation of the target genome to the heterologous oligonucleotide microarray, leading to the presence of a few or even one possible SFP within a probe-set. The exact number per probe-set will be dependent upon the evolutionary distance between the target species and design array, the rate of evolution of the individual gene represented by the probe-set and the array design itself. Thus, the sample mean is in general a good estimator for the central value of the data distribution of δij when statistical testing is performed according to the probe-wise strategy. But for those probe-sets which have two or more possible SFPs, the mean is no longer an appropriate measure of location under the probe-wise procedure since it will be susceptible to an extreme value. Accordingly, the γ-trimmed mean (0 < γ < 0.5) is employed instead of the mean as the statistic in this version of POST. More mathematically, let , …, k - 1, k + 1, …, #(Bi} and let δi(1) ≤ δi(2) ≤… δi(ni) be the observations of Δik written in ascending order. We define the sample γ-trimmed mean δik to account for probe-specific fluctuations in a probe-set i and its value is calculated by (19) where h = [γni] is the value of γni rounded down to the nearest integer. Then, let sik2 be the sample γ-Winsorized variance in the data of Δik and consider the finite-sample Student-t statistic analogue, the γ-trimmed mean can be studentized by sik as the form of (20) Tukey and McLaughlin [20] suggested a reasonably accurate approximation of the distribution of tik using a Student’s t-distribution with ni - 2h - 1 degrees of freedom. Also, Patel et al. [21] further introduced a scaled Student-t variate a(ni,h)tik and proposed approximating the distribution of a(ni,h)tik with a Student’s t distribution having v(ni,h) degrees of freedom, where a(ni,h) = 1 + 16h0.5e2h-ni for small-samples (ni < 18) t-type statistics and v(ni,h) has a slight variation depending on γ in their investigation. Given γ = 0.05, 0.10, 0.15, 0.20 or 0.25 we apply the Tukey-McLaughlin suggestion and Patel’s refined approximation to each of tik for the calculation of the p-value, and the asymptotic p-value accompanied with the intensity difference can therefore be prepared for the volcano plot filter and output. To better reveal detection of large-magnitude changes in the output, the POST used the square-root-transformation of the D-value into the fold-change difference FCDij defined as follows: (21) which produces a symmetric distribution of intensity differences under the assumption that most oligonucleotides are not differentially hybridised, so that the modified volcano plot using fold-change differences is still able to plot changes in both directions, showing equidistance from the centre. Due to the experimental design, the POST tests the inferential statistics on individual oligonucleotides within the parent group and within the offspring group respectively, colouring the plotted points in accordance to the group that they belong to. The colour scheme can be employed as a third dimension of information, for ease of filtering and the setting of parameters. By constructing the coloured volcano plots of G1 and G2, one can quickly identify the most-meaningful changes in hybridisation signal strength focused on the feature of interest. 3. Results and Discussion 3.1. Software Implementation Pigeons is a standalone GUI program for the Windows platform under the .NET framework to analyze Affymetrix GeneChip® data generated from cross-species experiments and the current version number is 1.2.1, released in late-June 2012. The software is able to read most recent or current Affymetrix .CEL file types, including version 3, version 4 and Command Console version 1 (the latest one at the time of program development). It is focused around visualization and interactive studies of data (Figure 2). This computer program is a freeware license so it is free of charge to download and to fully execute for research uses. The .NET Framework version 3.5 or greater is required to install the program. 2 MB of free hard disk space is the minimum to execute the program while 200 MB would be better if data/image file export is required. The golden rule of thumb is that the more RAM the better the capacity, and the faster the microprocessor the quicker the response. At least 1 GB RAM and an Intel® Pentium® M-class processors or better are recommended, although slower CPU speeds with 512 MB system memory will still work in most circumstances. This computer software has successfully been tested on Windows 2000, Windows XP, Windows Vista and Windows 7. Figure 2 Software Snapshots. Pigeons is a tab-page based standalone graphical user interface (GUI) program. There are three tab-pages for the three main applications in the main form. Each application can be used either separately or jointly. Other tools in a menu strip are also tab-page associated, that is, their availability depends on the application currently being performed. (A) Central Applications. The three main applications are: (i) Pigeon Filter; (ii) Pigeon Mining/Image and (iii) Pigeon Query. These are executed after the completion of two core components; (iv) File Reading; and (v) Data Preprocessing; (B) Statistical Analyses. Several essential tools can also be called from the menu strip. They are: (i) Dual fold-change (DFC); (ii) Probewise one-sample statistical test (POST); (iii) Twin Volcano Plot; (iv) Volcano Plot; (v) Box Plot dialog-box; and (vi) Box Plot output. Pigeons is a tab-page-reliant program with the availability of the functions in the main form depending on the tab-page currently presented. There are three tab-pages inside the Windows form. Pigeon Filter is an application to implement the ATM method for the removal of poorly hybridising probe-pairs and to make a probe-masking CDF file (Figure 2(A-i)). Pigeon Mining & Image is developed to perform the DFC analysis approach and the POST statistical filters are used to find potential SFP markers (Figure 2(A-ii,2B-i)). There are two POST-based graphical summary tools within Pigeon Mining (Figure 2(B-ii)). While the Volcano Plot (VP) is used to test differential variation between groups of parents and F2 hybrids using the binary average fold-change ratio (Figure 2(B-iv)), the Twin Volcano Plot (TVP) has been designed based on statistical tests within the groups (Figure 2(B-iii)). Results acquired by either the DFC or the POST can be exported as lists and as graphical representations for probe-sets to assist in the interpretation of oligo-level data at the DNA or RNA level. Pigeon Query is an interface for quick probe-set retrieval from datasets (Figure 2(A-iii)). Besides the three main applications, a couple of essential upstream tools are also involved in this software package—they are data preprocessing (Figure 2(A-v)) and a box-and-whisker plot (Figure 2(B-v, 2B-vi)). The Exponential-Normal Convolution Model was utilised for background correction in this program to adjust for systematic effects that arise from variation in the Affymetrix platform [18]. Pigeons employs quantile normalization to address the comparability of intensity distributions between arrays [19]. Then, one can use the box-and-whisker plot, a significant quality control tool, to examine the data before and after data preprocessing. This exploratory data analysis conducts a check for evaluating any extraordinary chip distributions and to verify if a normalization procedure has been effective. A user manual has been provided and built within an installer program so that users can access it from the start menu of MS Windows after the Pigeons has successfully been installed on a local machine. The software with its manual (the current release number version 1.2.1) can be freely downloaded at http://affymetrix.arabidopsis.info/xspecies/pigeons. 3.2. Case Studies of ATM Here, we surveyed a number of previous studies which focused on transcriptome analysis of heterologous species through the across species microarray approach, and compared the cut-off values chosen to make species-specific CDF files in those studies with the ATM’s suggestion based on gDNA hybridisation intensity thresholds (Table 1). Brassica oleracea L. (case 1) and Thlaspi caerulescens (case 2) were hybridised onto the Arabidopsis thaliana ATH1-121501 GeneChip® Arrays [3,4] whereas the two animals (case 4 and 5) were hybridised onto the Human U133 Plus 2.0 Genome Arrays [22,23]. In the third case, the Affymetrix Rice Genome Array was used to investigate transcriptomic profiling related to drought stress in Musa [7]. In the original Xspecies approach, i.e., the first case, a probe mask created at a cut-off value of 400 was determined systemically and empirically by generating 13 custom CDF files with a series of gDNA hybridisation intensity thresholds and each CDF was assessed in turn. The probe mask file excluded 68% of the probe-pairs but retained 96% of the available probe-sets, and this was used to study transcriptional response under phosphorus stress. This empirical method of determining the cut-off value was also applied to the second and the fourth cases, which selected the preferred hybridisation intensity thresholds of 300 and of 100, respectively. The same probe selection strategy but subtly different considerations were taken in account in the third and fifth cases. The authors of these two studies determined the hybridisation intensity threshold used to create a probe mask file that was able to detect the maximum possible number of Differentially Expressed Genes (DEGs) even though Hammond et al. showed that there was a significant loss of available probe-sets for transcriptomic profiling at the higher end of the cut-off value [3]. As a result, the selected cut-offs used in Banana and Sheep were at the value of 550 and of 450 respectively. microarrays-03-00001-t001_Table 1Table 1 Summaries of case studies. Species Selected Cut-off Automated Threshold Mapping (ATM) Reference % Suggested Cut-off Target Interval Tolerance Interval Brassica oleracea L. 400 2.17 391.34 a [351,426] a [272,454] a Hammond et al. 2005 [3] Thlaspi caerulescens 300 10.54 331.63 a [297,363] a [234,387] a Hammond et al. 2006 [4] Musa (Banana) 550 10.47 492.40 b [399,586] b [305,698] b Davey et al. 2009 [7] Equine (Horse) 100 5.93 94.07 a [82,106] a [65,119] a Graham et al. 2010 [22] Ovine (Sheep) 450 6.93 481.20 b [381,582] b [284,694] b Graham et al. 2011 [23] The cut-off values to mask the intensity signals were examined from 0 to 1,000 with an increment of 1 in all cases. These data sets were then tested under the ATM framework with cluster validation methods to generate the ATM three-tuple result for comparison to the previous publications. % denotes relative difference in cut-off and was calculated from the absolute value of difference between the selected and the suggested cut-off, divided by the selected cut-off value. a ATM was accompanied by a cluster validation procedure using Fukuyama-Sugeno’s index; b The partition entropy was applied as a cluster validity index into the ATM algorithm. Since FCM is an unsupervised process, we introduce two cluster validity indices to accompany the ATM framework to indicate the reliability of clustering results and to cover two different aspects of choosing gDNA hybridisation intensity thresholds. The two cluster validation measures are Fukuyama-Sugeno’s index [24] and partition entropy [15]. In our studies, the first index was exploited in case 1, 2 and 4 whilst the second one was utilised where there was a desire to gain a larger number of differentially regulated transcripts as was the case in 3 and 5; the 3-tuple results of ATM are summarised in Table 1. We found that the hybridisation intensity thresholds selected to understand the transcriptome results in the five cases were all located in the target interval, and they were generally in the vicinity of the cut-off values suggested by ATM. The relative difference in the hybridisation intensity cut-offs were from 2.2% up to 10.5%. Out of the five species, the numerical suggestion of 391.34 by ATM was very close to the biologist’s choice of 400 in Brassica oleracea L.—the original research paper presenting the heterologous gDNA hybridisation probe selection approach. The restriction imposed by the researchers of having less than 3% removed probe-sets, led to the selection of an optimal cut-off of 300. This imposed constraint explains the fact that the value of the researcher’s selection was reasonably different to the suggested value given by ATM (331.63) being very near the value of 297, the low end of the target area. ATM was initially developed to find the optimal cut-off of a vector valued retention function (Figure 1) and in practice, the probe mask filter developed using this numerical optimum was able to allow the discover of changes of gene expression in heterologous species. The practical consequence can substantially be shown by the means of the above studies, particularly the first, second and fourth cases. The difference in thresholds between the experienced researcher selection and the ATM’s suggestion in Banana and Sheep was by 6.93% and 10.47%. Not surprisingly, both were higher than those in the other three species, due to the selected stringent criterion for detecting the maximum number of differentially expressed transcripts. By having studied the five non-model plants/animals using model species oligonucleotide arrays, we believe that ATM is valid for the determination of gDNA hybridisation intensity thresholds. The proposed approach can provide fast and objective intensity thresholds, in comparison with the empirical method. When ATM is in operation, we strongly recommend making use of the Fukuyama-Sugeno’s index for transcriptomic and genomics analysis. This index is best for research activities where there is no direct interest in the evaluation of the expressed genes in an experiment, for example, as with finding SFP markers. If the number of DEGs is, however, the major consideration, the partition entropy approach will be a good cluster validity index for this biological purpose. 3.3. Examples of an SFP Screen Besides the generation of an optimal probe mask, a complete solution containing biological and algorithmic approaches to SFP interrogation has been proposed in this article. While DFC is a biology-oriented method and conventionally uses two fold-change with a gDNA hybridisation intensity threshold, POST is a statistically-based and newly-developed procedure with graphical summary filters from two aspects of the test approach. To evaluate these approaches, we examined bambara groundnut genotypes from an F2 offspring derived from a cross between two contrasting parental genotypes. The offspring were bulked according to the trait “number of branches per plant”. Bambara groundnut (Vigna subterranea (L.) Verdc.) is an underutilised indigenous African crop species and an important food legume grown widely in sub-Saharan Africa and has been shown to be highly inbreeding. At present, limited sequence resources exist, which means that the Xspecies is a valid approach. The gDNA-based probe-selection using heterologous oligonucleotide microarrays allows us to interrogate thousands of SFPs in parallel and, through the current design, should allow us to efficiently discover markers in a genomic region associated with a specific phenotype. As an illustration of this point, we selected the agronomic trait “number of branches per plant” in a cross between a wild accession with a spreading habit and a cultivated accession with a bunched habit [13,14]. Cross-hybridisation of bambara groundnut DNA from the two parental landrace genotypes VSSP11 (few stem per plant) and DipC (many stem per plant) were conducted using the Affymetrix Arabidopsis ATH1 GeneChip®. Meanwhile, two bulks from F2 individuals (10 individuals each, representing the high and low stem number extremes from 96 individual F2 plants) were hybridised separately onto the Arabidopsis ATH1 GeneChip® array. The experiment was therefore composed of four gDNA hybridisation chips and their relationship could be represented as , as defined in the methodology section. The probe-level raw data were then background-adjusted and quantile-normalized using the RMA method [18,19] so that these preprocessed intensity signals could be carried over into high level analyses. Figure 3 Filtering on Volcano Plots. The customised Volcano-plot tools depicting estimated fold-change (x-axis) and statistical significance (−log10P-value, y-axis) were created by means of the POST inferential statistics for filtering on screening of the single oligonucleotides related to the trait of interest. Each point represents an oligonucleotide probe, and the black crosses corresponded to large fold-changes with a p-value less than the significance level or the user-defined value under a number of filtering criteria. (A) Volcano Plot (VP). This is an example of applying the POST approach to test between groups of parents and F2 hybrid bulks using the binary average fold-change ratio, the MA-value; (B) Twin Volcano Plot (TVP). This is an illustration of another version of POST—testing oligonucleotide probes within a parental group and within an offspring group, respectively, followed by plotting the two graphical summaries together in different colours. Light-gray spots were the output of the parental group and gray ones represented the group of F2 hybrid bulks. The fold-change difference was defined by transforming the intensity difference D-value into its square root, and was used as a measure to identify the significant intensity differences in the plot. Figure 3 illustrates two graphical filters, VP and TVP, generated by the POST’s two different visual outputs based on an interrogation of the statistically significant differential hybridisation between the two bulks of bambara groundnut in relation to the trait “stem number”. To correct for multiple testing, we implemented an approach based on controlling False Discovery Rate (FDR), as proposed by Benjamini and Hochberg [25]. The BH adjusted p-values were transformed into inverse significances in both VP and TVP, and the suspected SFPs can be filtered and highlighted by the graphical outputs under a number of conditions. Since the samples of the F2 offspring act as a cross-checking mechanism in our experimental design, the fold-change of the offspring (FCF2) is used as one of the filtering parameters. Additionally, the optimal hybridisation threshold cut-off of the gDNA hybridisation intensity produced by ATM and the cut-off of the parental fold-change used in DFC can be optionally selected to increase the sensitivity of the graphical filters. The 7,903 differentially hybridised signals were summarised (BH adjusted p < 0.05, MA ≥ 0.75, MA ≤ -0.75, FCF2 ≥ 1.5) when the POST procedure was performed between the group of parents and of F2 samples (Figure 3(A)). The lower levels of hybridisation of features will be more likely to show a significant difference between parental genotypes by chance than high level differences in hybridisation, although the latter could represent repetitive elements within the bambara groundnut genome. Due to the scale of the binary fold-change ratio, this phenomenon is quite common in microarray data analysis. The same preprocessed data set was tested using the other version of POST to examine intensities within groups, followed by filtering potential SFPs using the coloured TVP (Figure 3(B)). Interestingly, there were only 59 probe-pairs (BH adjusted p < 0.05, FCD ≥ 8, FCD ≤ -8, FCF2 ≥ 1.5) detected as statistically differentially hybridised using the probewise strategy. The sharply reduced number from thousands to dozens shows that the D-value is highly selective against low intensity signals and that the design of TVP, disjointed testing on two groups with a process of filtering in relation to each other, was much more sensitive than the approach of VP based on the average fold-change ratio. To have a deeper understanding of the practical effects of using different approaches for SFP detection, various conditions of VP, TVP and DFC were systemically examined and are briefly described in Table 2. Two-fold change is normally the cut-off accepted in microarray analysis. However, the value of 1.5 was adopted rather than 2 for the cut-off of F2 in our illustration since the stringent conditions used led to very little in dual fold-change analysis and the hybridisation molecule in this case is genomic DNA, rather than dealing with expression values for RNA. As such, we might expect there to be a similar “dosage” of each gene in the individual genotypes, in the absence of wide-spread duplications. There were four instances inspected using VP and TVP, respectively whereas two cases were considered in DFC. Initial filtering parameters were fixed in the four instances of VP (BH adjusted p < 0.05, MA ≥ 0.75, MA ≤ -0.75) and TVP (10% trimmed mean, BH adjusted p < 0.05, FCD ≥ 8, FCD ≤ -8) and in the two instances of DFC (FCP ≥ 2, FCF2 ≥ 1.5). ATM with Fukuyama-Sugeno’s index producing the three-tuple suggestion (93.04, [81,106], [63,120]) of gDNA hybridisation intensity cut-offs for the cases of VP3, 4 and DFC2. Only the perfect match features of the ATH1 GeneChip® was considered in these investigations. When filtering on VP and TVP using initial conditions of x and y axis without extra parameters, we found that VP1 identified more than ten thousand potential SFPs. This was eight times the number using TVP1. This large difference was similar to our findings in Figure 3. We also noticed that the number of differentially hybridised features significantly declined from VP1 to VP2 and very dramatically dropped from VP1 to VP3. These results reveal that the gDNA hybridisation intensity threshold is an essential parameter in the VP filter and low signal hybridised probe-pairs were largely generated in the experiment. This is consistent with the phylogenetic distance between Vigna subterranea L and Arabidopsis thaliana. When all conditions were applied in VP4 and TVP4, there were approximately equivalent numbers of potential SFPs identified in the two cases, 10 and 8, respectively. An analogous situation between VP1 and VP3 could be found in the investigation of DFC as well. While 3,360 differentially hybridised features were detected in DFC1, very surprisingly, there were just 5 probable SFPs discovered in DFC2—the lowest number out of ten examined conditions. This implies that dual fold-change analysis would be the most stringent approach among the three methods. From the outcomes of VP4, TVP4 and DFC2, where few SFPs were identified we can conclude that the Affymetrix ATH1 GeneChip might not be the best array for heterologous genomic DNA hybridisation with a view to interrogation of the bambara groundnut genome, due to the distant evolutionary relationship between Arabidopsis thaliana and bambara groundnut. microarrays-03-00001-t002_Table 2Table 2 Screening for differentially hybridised oligonucleotides by filtering on two types of volcano plots and dual fold-change analysis under a number of criteria. Method Filtering Criteria Number of potentiallydifferential hybridization d VP p-value a MA-value FCF2 TH b,c Probe-pairs Probe-Sets VP1 <0.05 ≥|0.75| - - 13,694 10,492 VP2 <0.05 ≥|0.75| ≥1.5 - 7903 6722 VP3 <0.05 ≥|0.75| - >93.04 125 124 VP4 <0.05 ≥|0.75| ≥1.5 >93.04 10 10 TVP e p-value a FCD-value FCF2 FCP Probe-pairs Probe-Sets TVP1 <0.05 ≥|8.0| - - 1,637 1,563 TVP2 <0.05 ≥|8.0| ≥1.5 - 59 59 TVP3 <0.05 ≥|8.0| - >2 50 50 TVP4 <0.05 ≥|8.0| ≥1.5 >2 8 8 DFC FCP FCF2 TH b,c Probe-pairs Probe-Sets DFC1 ≥2 ≥1.5 - 3,360 3,132 DFC2 ≥2 ≥1.5 >93.04 5 5 The total number of interrogated probe-pairs and probe-sets is 250,103 and 22,746 respectively. Abbreviations. VP: volcano plot; TVP: twin volcano plot; DFC: dual fold-change analysis; FCP: the cut-off of parent fold-change; FCF2: the cut-off of F2 fold-change; TH: the genomic DNA hybridisation intensity threshold; MA-value: binary average fold-change ratio; FCD-value: fold-change difference as the square-root-transformation of the D-value. a Benjamini-Hochberg adjusted p-values were calculated for multiple testing correction; b The mask of multiple chips was applied. A technique where each signal is extracted from the minimal intensity of four gDNA chips in the single trait experiment to create a pseudo array that will be analysed under the ATM framework; c Fukuyama-Sugeno’s index was used to generate ATM-suggested gDNA hybridisation intensity threshold; d SFPs were examined on the Perfect Match probe datasets in all cases; e 10% trimmed mean, γ = 0.1, of intensity difference was used. Among the ten instances, VP4, TVP2, TVP4 and DFC2 were selected to acquire more dependable SFPs through Euler diagram analysis. Since TVP2 (bcef) and VP4 (abde) were proper supersets of TVP4 (ef) and DFC2 (de) respectively, we can produce a simplified version of the 4 unit diagram (Figure 4). As seen in Table 2, DFC takes advantage of the hard cut-off values of FCP and genomic DNA hybridisation intensity and this approach has a limitation—it may cause possible oligonucleotides to be omitted where they detect repetitive elements within the genome of an investigated species. The set constructed by subtracting DFC2 from VP4 would be able to overcome this potential limitation of DFC. So would the difference between TVP2 and TVP4. The intersection of four units, e, is a focus from which the most probable candidates can be found. In the example, 3 suspected probe-pairs were found in this intersection (Figure 4(A)) although one of them was not considered as a potential SFP since its square root intensity difference was not much greater than the FCD cut-off (data not shown). An area, b, where the overlap between VP4 and TVP2 but excludes DFC2 is another focus. The elements of this area have potential as their parental fold-changes approach the cut-off value and the signal intensities are not at the low end of the range. To take 258467_at_680_81 as an example, its parental fold-change was 1.96 (564/288), with strong hybridisation and a ratio very near to the cut-off of 2. There were 2 and 5 oligoprobes discovered in the sets of d and f (Figure 4(A)), respectively, and both d and f were associated with FCP & FCF2. Of the two possibilities for SFPs, the latter seemed more likely. Although the identified oligoprobes exceeded the ATM’s suggested threshold and the cut-off based on the two fold-change parameter, they did not have a particularly large intensity difference (data not shown) so should probably not be selected as candidates. On the other hand the partition f has potentially large FCD-values with signal intensities slightly smaller than the gDNA hybridisation intensity threshold based on the ATM suggestion. Out of the 5 filtered entities, there was only one having very poor hybridisation (42 vs. 93.04), and this was discarded. The partition built by deducting the intersection of the four units from TVP4 is able to complement another potential constraint of DFC—the hard cut-off value of gDNA hybridisation intensity. When it comes to the area where TVP2 excludes VP4 & TVP4, there were 47 candidates, the largest number in the Euler diagram, detected as statistically significant variable probe-pairs (Figure 4(A)). However, we did not consider any of these as potential SFPs. The reason is that nearly all elements of this set have a much smaller parental fold-change than the given cut-off. Similarly, most discovered probes in the portion where VP4 excludes TVP2 & DFC2 have either small intensity differences or small parental fold-change. In this analysis, there was one probe, 265228_s_at_195_89, belonging to this type of set and we regarded it as a candidate because of its strong hybridisation and reasonable parental ratio of FC (1822/962). The Euler diagram was then updated to show the situation of retained candidates in the units (Figure 4(B)). Eventually, this informed selection enables us to produce a final list of potential SFPs for further validation in vitro. Through this small-size demonstration, an optimal strategy based on the Euler diagram for the selection of differentially hybridised oligonucleotides using POST and DFC has been summarised (Figure 4(C)). Using this strategy, researchers could determine a final candidate list. Firstly, we suggest neglecting the subsets c and d and picking the elements of the intersection of four-set Euler diagram e. Next, the two buffers, b and f, need to be thoroughly examined as to whether there are any elements whose parental fold-change (for b) and signal intensities (for f) approximate to the predefined cut-off values, respectively, to find statistically significant variable probe-pairs. Finally, partition a should be checked to see if those signals which have strong hybridisation as well as a parental fold-change approaching the cut-off. In addition, there is some opportunity to identify a probe-set having differentially hybridised probe-pairs with more than or equal to a two-fold difference in this partition. Ideally, a probe-set containing multiple SFPs ought to be detected in the intersection of TVP and VP if the trimmed mean percentage γ can be carefully chosen. In our example, γ = 0.1 was used, implying the detection of two SFPs in the same set, and we did not discover any probe-sets with this observable property, arguing against differences between the two parental genotypes (and their offspring) involving the complete absence of probe-sets or their duplication in one genotype only. Making the most of VP, TVP and DFC, the recognition of differentially hybridised oligonucleotides associated with the phenotypic region in a non-model species could be increased. Figure 4 Euler Diagram Analysis. This was an example to show how potential SFPs can be selected by the POST and the DFC using Pigeons. The four-set diagram was established according to VP4 (abde), DFC2 (de), TVP2 (bcef) and TVP4 (ef) illustrated in Table 2, where lowercase letters stand for the portions of the four filtering methods. (A) SFP Candidates. Numbers in the partitions indicate the number of detected probe-pairs that can be recognised as potential SFPs; (B) Final Candidates. After careful selection and consideration portion by portion, potentially differentially hybridised oligonucleotides could be determined. They were e:264674_at_473_177, 257321_at_566_65; b:258467_at_680_81; f:244964_at_665_15, 255530_at_691_371, 257050_at_8_423 and 266293_at_656_319; a:265228_s_at_195_89; (C) Optimal strategy for potential SFP selection. Where √: candidates; ×: elimination, ≈FCP: the parental fold-change value is just below cut-off; «FCP: the parental fold-change value is significantly below cut-off, small D: little intensity difference; ≈FCD: the fold-change difference value is slightly above cut-off; «TH: poor hybridisation; ≈TH: the signal intensity is a little lower than the value of gDNA hybridisation intensity threshold; 2↑SFPs: there are more than or equal to two potential SFPs found in the same probe-set. 4. Conclusions Oligonucleotide microarrays have been verified as a powerful high-throughput technology to study plant genomics and transcriptomics. While most arrays are designed for model and major species investigation, there is limited availability of designed microarray platforms for the study of minor crop species that might currently be important food sources in some countries and have potential for future food production more widely. With the advent of the high density oligonucleotide arrays, Xspecies can be used to investigate the transcriptomes of underutilised plants. We have developed several computational algorithms and statistical methods to accompany this oligonucleotide probe-based cross-species platform for the analysis of oligoprobe selection/parsing and for finding potential SFP in minor crop species. These methods have been packaged in a computer program, named Pigeons, focused around visualization and interactive studies of the datasets at the probe level. A number of case studies and an illustration of the analysis of an underutilised crop dataset using Pigeons have also been performed to show the effectiveness and the usefulness of the proposed methods. Acknowledgments The authors would like to acknowledge Zoe Philips for hybridisation of the Affymetrix Arrays and we appreciate the comments received from the referees. Conflicts of Interest The authors declare no conflict of interest. ==== Refs References 1. Wang J. Computational biology of genome expression and regulation—A review of microarray bioinformatics J. Environ. Pathol. Toxicol. Oncol. 2008 27 157 179 10.1615/JEnvironPatholToxicolOncol.v27.i3.10 18652564 2. Kumar R.M. The widely used diagnostics “DNA microarrays”—A review Am. J. Infect. Dis. 2009 5 214 225 10.3844/ajidsp.2009.214.225 3. Hammond J.P. Broadley M.R. Craigon D.J. Higgins J. Emmerson Z.F. Townsend H.J. White P.J. May S.T. Using genomic DNA-based probe-selection to improve the sensitivity of high-density oligonucleotide arrays when applied to heterologous species Plant Methods 2005 1 10 10.1186/1746-4811-1-10 16280083 4. Hammond J.P. Bowen H.C. White P.J. Mills V. Pyke K.A. Baker A.J. Whiting S.N. May S.T. Broadley M.R. A comparison of the Thlaspi caerulescens and Thlaspi arvense shoot transcriptomes New Phytol. 2006 170 239 260 10.1111/j.1469-8137.2006.01662.x 16608451 5. Graham N.S. Broadley M.R. Hammond J.P. White P.J. May S.T. Optimising the analysis of transcript data using high density oligonucleotide arrays and genomic DNA-based probe selection BMC Genomics 2007 8 344 10.1186/1471-2164-8-344 17908303 6. Broadley M.R. White P.J. Hammond J.P. Graham N.S. Bowen H.C. Emmerson Z.F. Fray R.G. Iannetta P.P.M. McNicol J.W. May S.T. Evidence of neutral transcriptome evolution in plants New Phytol. 2008 180 587 593 10.1111/j.1469-8137.2008.02640.x 18801004 7. Davey M.W. Graham N.S. Vanholme B. Swennen R. May S.T. Keulemans J. Heterologous oligonucleotide microarrays for transcriptomics in a non-model species; A proof-of-concept study of drought stress in Musa BMC Genomics 2009 10 436 10.1186/1471-2164-10-436 19758430 8. Kreyszig E. Advanced Engineering Mathematics 10th ed. John Wiley & Sons Hoboken, NJ, USA 2011 790 842 9. Xu R. Wunsch D. II. Survey of clustering algorithms IEEE Trans. Neural Netw. 2005 16 645 678 10.1109/TNN.2005.845141 15940994 10. Schena M. Shalon D. Heller R. Chai A. Brown P.O. Davis R.W. Parallel human genome analysis: Microarray-based expression monitoring of 1,000 genes Proc. Natl Acad. Sci. USA 1996 93 10614 10619 10.1073/pnas.93.20.10614 8855227 11. Cui X. Churchill G.A. Statistical tests for differential expression in cDNA microarray experiments Genome Biol. 2003 4 210 10.1186/gb-2003-4-4-210 12702200 12. Kooperberg C. Aragaki A. Strand A.D. Olson J.M. Significance testing for small microarray experiments Stat. Med. 2005 24 2281 2298 10.1002/sim.2109 15889452 13. Mayes S. Stadler S. Basu S. Murchie E. Massawe F. Kilian A. Roberts J.A. Mohler V. Wenzel G. Beena R. BAMLINK—A cross disciplinary programme to enhance the role of bambara groundnut (Vigna subterranea L. Verdc.) for food security in Africa and India Acta Hortic. 2009 806 137 150 14. Basu S. Mayes S. Davey M. Roberts J.A. Azam-Ali S.N. Mithren R. Pasquet R.S. Inheritance of “domestication” traits in bambara groundnut (Vigna subterranea L. Verdc.) Euphytica 2007 157 59 68 10.1007/s10681-007-9396-4 15. Bezdek J. Pattern Recognition with Fuzzy Objective Function Algorithms 1st ed. Plenum Press New York, NY, USA 1981 95 154 16. Jeffery I.B. Higgins D.G. Culhane A.C. Comparison and evaluation of methods for generating differentially expressed gene lists from microarray data BMC Bioinform. 2006 7 359 10.1186/1471-2105-7-359 17. Dudoit S. Yang Y.H. Callow M.J. Speed T.P. Statistical methods for identifying genes with differential expression in replicated cDNA microarray experiments Stat. Sin. 2002 12 111 139 18. Irizarry R.A. Hobbs B. Collin F. Beazer-Barclay Y.D. Antonellis K.J. Scherf U. Speed T.P. Exploration, normalization, and summaries of high density oligonucleotide array probe level data Biostatistics 2003 4 249 264 10.1093/biostatistics/4.2.249 12925520 19. Bolstad B.M. Irizarry R.A. Astrand M. Speed T.P. A comparison of normalization methods for high density oligonucleotide array data based on variance and bias Bioinformatics 2003 19 185 193 10.1093/bioinformatics/19.2.185 12538238 20. Tukey J.W. McLaughlin D.H. Less vulnerable confidence and significance procedures for location based on a single sample: Trimming/Winsorization 1 Sankhya A 1963 25 331 352 21. Patel K.R. Mudholkar G.S. Fernando J.L.I. Student’s t approximations for three simple robust estimators J. Am. Stat. Assoc. 1988 83 1203 1210 22. Graham N.S. Clutterbuck A.L. James N. Lea R.G. Mobasheri A. Broadley M.R. May S.T. Equine transcriptome quantification using human GeneChip arrays can be improved using genomic DNA hybridisation and probe selection Vet. J. 2010 186 323 327 10.1016/j.tvjl.2009.08.030 19786357 23. Graham N.S. May S.T. Daniel Z.C.T.R. Emmerson Z.F. Brameld J.M. Parr T. Use of the Affymetrix Human GeneChip array and genomic DNA hybridisation probe selection to study ovine transcriptomes Animal 2011 5 861 866 10.1017/S1751731110002533 22440025 24. Fukuyama Y. Sugeno M. A New Method of Choosing the Number of Clusters for the Fuzzy C-Mean Method Available online:http://citeseer.uark.edu:8080/citeseerx/showciting;jsessionid=1AF0955F44EC87078947AADEDE29D50C?cid=664813 (accessed on 10 December 2013) 25. Benjamini Y. Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing J. R. Stat. Soc. B 1995 57 289 300

PMC005xxxxxx/PMC5003453.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays3010024microarrays-03-00024ReviewCopy Number Variation in Chickens: A Review and Future Prospects Wang Xiaofei *Byers Shannon Department of Biological Sciences, Tennessee State University, 3500 John A. Merritt Blvd., Nashville, TN 37209, USA; E-Mail: shannon.byers@rocketmail.com* Author to whom correspondence should be addressed; E-Mail: xwang@tnstate.edu; Tel.: +1-615-963-2541.05 2 2014 3 2014 3 1 24 38 15 12 2013 22 1 2014 23 1 2014 © 2014 by the authors; licensee MDPI, Basel, Switzerland.2014This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).DNA sequence variations include nucleotide substitution, deletion, insertion, translocation and inversion. Deletion or insertion of a large DNA segment in the genome, referred to as copy number variation (CNV), has caught the attention of many researchers recently. It is believed that CNVs contribute significantly to genome variability, and thus contribute to phenotypic variability. In chickens, genome-wide surveys with array comparative genome hybridization (aCGH), SNP chip detection or whole genome sequencing have revealed a large number of CNVs. A large portion of chicken CNVs involves protein coding or regulatory sequences. A few CNVs have been demonstrated to be the determinant factors for single gene traits, such as late-feathering, pea-comb and dermal hyperpigmentation. The phenotypic effects of the majority of chicken CNVs are to be delineated. copy number variationphenotypic variabilitychicken ==== Body 1. Introduction Genomes vary among different individuals of the same species, even among different cells within the same individual of multicellular organisms. Variations include differences at a single nucleotide position up to entire sets of chromosomes. Small scale variations, involving a single or a short stretch of nucleotide positions, are often discovered by sequencing. Various methods are available for detection of these small scale differences and are routinely studied for a long time. Large scale variations are visible under microscope, thus are discovered and studied routinely with microscopy during karyotyping. Variations of DNA sequences of intermediate scale, from 1,000 bp (kb) to a few million base pairs (mb), now known as copy number variation (CNV) [1,2], were ignored by researchers for some time due to the unavailability of suitable research tools. New DNA technology, especially microarray and sequencing, has permitted the convenient detection of CNVs in various genomes in the last decade, leading to the booming of CNV studies on various species, especially humans. Ongoing research into CNV will likely change the landscape of SNP-centric genome-wide association studies (GWAS) since CNV regions (CNVRs) show more inclusions and complex genetic variants than SNP sites. Comparing among genomes of different cells within the same individual, or different individuals of the same or closely related species, DNA sequence differences can be considered as the result of nucleotide substitution, insertion, deletion, inversion, translocation, or combinations of these events. CNVs basically involve the insertion or deletion of DNA sequences, or the combination of both. An event of insertion leads to the gain of DNA segment, while an event of deletion leads to the loss of a segment. Although many CNVs involve the activity of transposable elements [3], the gain or loss of transposable elements is not considered in this category of genomic variability [4]. Because there is no convenient method to detect inversion and translocation of intermediate sizes, the scale of such structural changes is currently unknown. In the last several years, there has been an increasing interest in the study of CNVs in the chicken. The chicken is an important farm animal species. It has served as an important protein source by providing table eggs and meat for human nutrition since its domestication in Southeast Asia over 8,000 years ago [5]. In some cultures, the chicken has also been used for ritualistic activities. Due to readily available and easy handling, the chicken has been used in fundamental biological studies, including embryonic development, genetics, immunology, oncology and virology studies. The chicken is an excellent example of the unique genome arrangement in avian species, where there are a few pairs of large chromosomes called macrochromosomes and a large number of tiny chromosomes called microchromosomes [6]. Studies of chicken CNV are driven by interests from not only the poultry science point of view, but also the basic biological perspective. Since the time when genomic tiling arrays became available for chickens [7] several years ago, a good number of studies on chicken CNV have been published. Here, we review various aspects of chicken CNVs, especially the genome-wide distribution, known phenotypic effects and areas for future research. 2. CNV Distribution in the Chicken Genome The majority of chicken CNVs were detected with array comparative genome hybridization (aCGH), with which two DNA samples were labeled respectively with different fluorochromes, often cy3 and cy5, then equal amounts of labeled DNA were co-hybridized to a whole genome tiling array [8,9,10]. The hybridization signals were recorded from the tiling array using laser scanners, and the signal intensities for each probe from each DNA sample were compared [11]. The hybridization signals are supposed to have equal intensity from each of the two respectively labeled DNA samples. Sophisticated computational algorithms were then used to determine whether a region of the genome has different signal intensities. Unequal signal intensities suggest gain or loss of DNA copies in the region of genome in question. The aCGH method has been applied in studies of human and many other species [12,13,14,15]. SNP arrays have also been used to detect CNVs. Although SNP arrays are primarily designed for SNP genotyping, detection of CNV with SNP arrays is possible because abnormal hybridization occurs when a SNP probe is located in a CNV locus. Various computer programs were developed to detect CNVs from SNP genotyping data [16,17]; thus, no additional wet laboratory work is needed. As such, the detection of CNVs from SNP data added value to genome-wide association study (GWAS) with SNP arrays. This approach has been especially widely used in human population studies [18,19,20]. A few studies using SNP arrays have also been analyzed for CNV in farm animals [21,22], including chickens [23]. Reports for genome–wide detection of CNVs in chickens are accumulating (Table 1). The first study of chicken CNV with aCGH was reported by Griffin and coauthors [7] in an attempt to establish inter-species genomic rearrangement. They identified 12 CNVs between the domesticated chicken and red jungle fowl (the wild ancestor of domesticated chickens), and several other CNVs between red jungle fowl and turkey. Since this study examined only one broiler and one Leghorn chicken, the prevalence of the CNVs could not be inferred. Comparisons between the chicken and duck, and between the chicken and zebra finch revealed more inter-species CNVs along the chicken chromosome [7,24,25]. Such regions could potentially harbor CNV within the chicken because some inter-species CNVRs were also found as intra-specific CNVRs in the chicken. microarrays-03-00024-t001_Table 1Table 1 Current reports on genome-wide analysis of copy number variation (CNV) in chickens. Author [ref] Method No. of Birds Breed or Line * No. of CNVRs Griffin et al. 2008 [7] aCGH 3 Red jungle fowl (1), Leghorn (1), broiler (1) 12 Skinner et al. 2009 [24] aCGH 1 Red jungle fowl 32 Volker et al. 2010 [25] aCGH 1 Red jungle fowl 32 Wang et al. 2010 [26] aCGH 10 Cornish rock (4), Rhode island red (4), Leghorn (2) 96 Jia et al. 2012 [23] SNP chip 746 White leghorn, dwarf 315 Wang et al. 2012 [27] aCGH 18 Broiler (6), Leghorn (6), Chinese local breed (6) 327 Luo et al. 2013 [28] aCGH 6 62 (2), 73 (2) and cross hybrid (2) 32 Fan et al. 2013 [29] Sequencing 2 Silkie (1), local breed (1) 8,839 Crooijmans et al. 2013 [30] aCGH 64 Leghorn, broiler (15 lines in total) 1,556 * Values in parentheses are numbers of birds from the breed or line. We were the first to report intra-specific CNVRs in chickens [26]. Our study examined ten individual chickens and identified 96 CNVs corresponding to approximately 1.3% of the chicken genome, 27 of which were observed in more than one individual [26]. Since there are many well distinguished varieties of chickens, it would be of great interest if the genetic architecture that defines varieties could be elucidated. Several studies reported so far have paid attention to this issue [26,27,30]. However, each of these studies examined a small number of animals from each variety. Thus, it is uncertain regarding variety-variety specificity of particular CNVs. Until now, Crooijmans et al. [30] have reported the largest number of CNVs in chickens. They identified a total of 3,154 CNVs, which were grouped into 1,556 CNVRs based on overlapping. Those CNVs were identified from 64 birds of 15 commercial and experimental lines. This study could confirm 50% of high confidence CNVRs and 23.9% of single observation CNVRs reported by Wang et al. [26], and 21% of the 238 CNVRs reported by Wang et al. [27]. Reports by others [23,28] showed additional CNVs among chickens of different variety/lines. Using whole genome sequencing approach, Fan and his colleagues reported 8,839 putative CNVs with size >2 kb, but those with size >5 kb were listed with chromosomal coordinates [29]. Taken together, currently reported chicken CNVs amount to 3,961, which aggregate to 1,876 CNVRs (see Supplement file). These CNVRs were found on GGA1-28, E64, W and Z (Table 2). The CNVRs reported on random chromosomal segments [28] in the galGal3 assembly were not counted in this statistic, due to the uncertainty about the order of probes within the random segment of the assembly. The chicken genome assembly galGal3 does not contain information on GGA29-31 and 33-38. Likely, probes on these chromosomes are represented in the unknown random chromosomal segments. The current known chicken CNVRs encompassed 8.3% of the chicken genome, or 9.6% of the ordered assembly (Table 2). There are huge differences among chromosomes in terms of the fraction of DNA sequences involved CNVRs. On GGA1-15, CNVRs involve 5–14% of all DNA sequences. On the majority of other microchromosomes, figures are similar. Exceptions are GGA16, 25, E64, and W. GGA16 harbors the complex gene family histocompatibility proteins. Despite the fact that GGA16 is physically similar to GGA15 and GGA17, its assembled sequence is less than 4%. In contrast to aCGH studies so far, where a small number of samples from each line/breed of chickens were analyzed, CNV identification from SNP genotyping data by Jia et al. [23] analyzed 746 chickens, in which 417 birds were found to harbor CNVs. Some birds were found to have no CNV in the SNP data, which is in sharp contrast to aCGH analysis. The discrepancy is understandable because (1) the 60 K SNP arrays had much lower probe density [31], while aCGH arrays contained 244,000 to 385,000 probes for the chicken genome; (2) detection of CNVs from aCGH data is based on competitive hybridization of two samples on the same array, while detection from SNP data would have to compare across different arrays. Another sharp contrast between aCGH and SNP data is the frequencies of CNVRs. Among the 315 distinct CNVs detected with SNP arrays, only five CNVs (CNVRs) had a frequency greater than 5% and none greater than 10% in the chicken population. While in aCGH studies, frequencies of 130 CNVs were greater than 10% (calculated from [30]). Such difference could result from the use of references and also probe densities. In the aCGH study by Crooijmans et al. [30], a red jungle fowl was used as the reference to which all other chickens were compared, while SNP chip study by Jia et al. [23], CNV calls ought be made in reference to the two experimental chicken lines. Regardless of analysis with aCGH or SNP genotyping arrays, boundaries of each CNV are uncertain due to the noise of DNA hybridization and the possibility of multiple breakpoints. Comparisons between different array platforms are hindered by differences in probe locations. Many CNVs within the same CNVR, detected by different array platforms, may actually be the same allele. Another issue with hybridization detection of CNVs is the uncertainty of zygosity state. Neither aCGH nor SNP genotyping may tell whether a test sample is homozygous or heterozygous. Whole genome sequencing, on the other hand, has the potential to accurately map the breakpoints [32,33], as has been shown in the study by Fan et al. [29]. It is also possible that the whole genome sequencing approach can distinguish between homozygote and heterozygote. Currently, this approach faces challenges of high levels of false positive and high cost. microarrays-03-00024-t002_Table 2Table 2 Distribution of CNV regions (CNVRs) on chicken chromosome. Chromosome No of CNVRs Total CNVR size (bp) Assembled chromosme size (bp) in galGal3 % in CNVRs 1 364 11,375,702 200,994,015 5.66 2 311 8,125,223 154,873,767 5.25 3 180 5,540,459 113,657,789 4.87 4 159 4,609,754 94,230,402 4.89 5 123 6,862,048 62,238,931 11.03 6 93 2,262,075 37,400,442 6.05 7 66 4,203,267 38,384,769 10.95 8 51 2,875,956 30,671,729 9.38 9 54 1,443,118 25,554,352 5.65 10 43 1,412,549 22,556,432 6.26 11 57 1,955,749 21,928,095 8.92 12 34 2,906,682 20,536,687 14.15 13 43 1,362,013 18,911,934 7.20 14 41 1,136,391 15,819,469 7.18 15 35 1,551,252 12,968,165 11.96 16 1 432,778 432,983 99.95 17 22 719,234 11,182,526 6.43 18 16 1,743,280 10,925,261 15.96 19 16 524,800 9,939,723 5.28 20 31 1,596,835 13,986,235 11.42 21 16 771,271 6,959,642 11.08 22 5 267,594 3,936,574 6.80 23 18 1,223,166 6,042,217 20.24 24 12 635,342 6,400,109 9.93 25 1 2,026,539 2,031,799 99.74 26 10 998,206 5,102,438 19.56 27 18 1,438,608 4,841,970 29.71 28 11 491,924 4,512,026 10.90 E64 1 44,645 49,846 89.57 W 1 257,546 259,642 99.19 Z 43 28,708,659 74,602,320 38.48 Total 1,876 99,502,665 1,031,932,289 9.64 3. CNV and Phenotypic Variation A CNV may affect phenotypic characteristics through various mechanisms. If a CNV is involved in protein coding, it may directly alter the protein function. If a CNV involves the regulatory region of a functional protein gene, it may alter when, where and how much of the gene is transcribed. The effect of CNV can even extend to half a megabase away [34]. It is also possible that a CNV imposes very little effect on the phenotype. While genome-wide survey seeks to provide a comprehensive map of CNVs, functional analysis provides insight into the effect of various CNVs on phenotype. In humans, the phenotypic impact of many CNVs has been demonstrated (see review by Henrichsen et al. [35]). Known phenotypes associated with CNV in chickens include pea-comb [36], late-feathering on chromosome Z [37], dark brown plumage color [38] and dermal hyperpigmentation [39,40]. The functional consequences of the overwhelming majority of the chicken CNVs are yet to be revealed. As have been observed in many species, individual chickens carrying most of the CNVs appear “normal.” Because CNVs often involve large genomic regions (several kb to several mb), a large proportion of reported CNVs involve protein coding or functional RNA. Inter-species aCGH studies have shown that there are more CNVs that involve coding genes than CNVs involving solely noncoding sequences [7,24,25], regardless if the species in comparison are closely related (between turkey and chicken) or more distantly related (between chicken and duck or between chicken and zebra finch). Similarly, inter-species CNVs have similar partitions: more involve coding sequences than solely noncoding sequence. A popular method for analysis of gene content in CNVRs is to determine enrichment of specific gene ontology (GO) terms. The list of genes in CNVRs is compared against a background gene list in a database using computational tools [41,42]. In the chicken, a striking feature is the enrichment of cytoskeletal protein genes in CNVRs [27,30], especially the keratin super family. Crooijmans et al. [30] suggested that such enrichment may be related to the over-representation of keratin genes in aves when compared to mammals. Gene enrichment findings of avian species are in sharp contrast to those in mammals in which enriched genes include those that respond to stimulus, antigen processing and defense [12,43]. Likely, differences are due to species-specific biology. It is also necessary to improve the background dataset so that more accurate analysis can be done. It is common to many species that the majority of CNVs have low frequencies. Chickens are no exception. Most chicken CNVs were observed only once among the birds studied. For example, among the 96 CNVRs described by Wang et al. [26], 70 were detected in only one bird. Similarly, among the 3,154 CNVs reported by Crooijmans et al. [30], 2,210 CNVs (70%) were observed in only one bird. Some CNVs found in only one chicken in one study could be corroborated by other studies. The observed low frequencies are partly attributable to uncertainty about CNV boundaries, false negative and false positive CNV calls in aCGH. On the other hand, because some CNVs may have significant disadvantages over the individual’s phenotype, they could be under selective pressure for elimination. Thus, their frequency could not reach higher level. Pea-Comb Phenotype: The pea-comb phenotype exhibits reduced size of comb and wattles in the chicken [36]. It is one of the two epistatic genes interacting with each other in classic genetics textbooks. When a chicken carries the dominant pea comb (P) allele at one locus and dominant rose comb allele (R) at another locus simultaneously, it develops walnut comb. The pea-comb is advantageous in cold climate because it reduces heat loss and makes chickens less susceptible to frost lesions [36]. Through linkage analyses using dense genetic markers and segregating families, the pea comb gene has been found within the interval containing SOX5 on GGA1 [36,44]. The pea-comb phenotype results from a CNV: a massive amplification of a duplicated segment near evolutionary conserved non-coding sequences in intron 1 of the SOX5 transcription factor, signifying that the duplicated expansion interferes with SOX5 expression and the regulation of gene expression during differentiation of cells crucial for the development of comb and wattles [36]. SOX5 encodes a member of the SRY-related HMG family of transcription factors and known to enrich for ECNS. It plays a role in cell fate and differentiation, skeletal development, chondrocyte development and extracellular matrix production. The interval harboring the code for pea-comb trait has been defined as 67,831,796–68,456,921 bp on chromosome 1 [36]. Interestingly, the rose comb phenotype also results from a type of structural variation. It has been demonstrated to be caused by an inversion of about 7 mb on GGA7, resulting in ectopic expression of homeodomain protein MNR2, another transcription factor [45]. Late Feathering: The late feathering phenotype has been widely used in commercial poultry operations for sexing of chicks at hatch. This trait is determined by a partially dominant K allele of a sex-linked locus on GGAZ. Early studies showed that the phenotype may involve an endogenous viral gene ev21. However, detailed mapping studies indicate that the K allele results from a partial duplication of the prolactin receptor gene (PRLR) and the sperm flagellar protein 2 gene (SPEF2) from the k+ allele [37,46]. This mutation causes reduced fertility and retarded development of fly feathers. Late-feathering birds not only have increased PRLR mRNA expression, but also altered mRNA levels of other genes during early development; many of them are keratin-related genes [47]. Nonetheless, transcripts of the partial duplicated dPRLR were found in a wide array of tissues. Its encoded protein, a truncated version of PRLR lacking a 149-aa C-terminal tail, can be potently activated by prolatin [48], suggesting the duplicated PRLR may be involved in a wide range of physiological activities. Dark Brown Plumage: A 2011 study by Gunnarsson and coauthors found that dark brown (DB) plumage color mutation in chickens reduces the expression of black eumelanin and enhances the expression of red pheomelanin in certain parts of the plumage [38]. They demonstrated that the causal mutation factor is an 8.3 kb deletion upstream of the SOX10 transcription start site. The SOX10 transcription factor plays a role in melanocyte development essential for melanocyte migration and survival. Deletion in this locus is thought to reduce SOX10 expression, down-regulating expression of key enzymes in pigment synthesis like tyrosinase. Lower tyrosinase activity causes a more pheomelanistic or reddish plumage color, the characteristic feature of the DB phenotype. Dermal Hyperpigmentation: Silkie chickens originate in China. These birds have unique phenotypes, including elongated feathers on the head, fluffy plumage, dark blue skin, viscera, bones and ears, feathered legs and feet, and five toes on each foot. In some cultures, people believe that the dark colored bone and skin render the chicken meat to have special therapeutical capability. The dark color in all connective tissues results from unusual melanogenesis, a condition called dermal hyperpigmentation or fibromelanosis [49]. A dominant FM allele was shown to be responsible for extensive pigmentation of the dermal layer of skin and the internal connective tissue [39]. The causal FM is an inverted duplication and junction of two genomic regions separated by more than 400 kb in wild chickens. The duplicated regions contained endothelin 3 (EDN3) gene that promotes melanoblast proliferation [40]. EDN3 expression is thought to be increased in the developing Silkie embryos while melanoblasts are migrating. Elevated levels of expression are found in adult skin tissue. Comparison of four chicken breeds from Asia and Europe also displaying dermal hyperpigmentation revealed that the same structural variant regulated this phenotype across all chicken breeds. This genomic rearrangement causes a specific monogenic trait in chickens, illustrating how novel mutations with phenotypic effects have been reused during breed formation in domestic animals. 4. CNV and Complex Trait Most important traits are complex, often measured in quantitative terms. This is especially true for agricultural traits such as growth rate, disease resistance, feed conversion, egg production, etc. In humans, there are wide interests in evaluation of the relationship between copy number variants and complex traits. Studies have demonstrated that CNVs contribute to drug response [50,51]. There are also results suggesting that an association between candidate CNVs and complex traits may be disappointing [52,53]. Studies on the relationship between CNV and complex traits in farm animals are lagging behind. In cattle, studies have suggested that CNVs may be associated with parasite resistance [54,55] and residual feed intake [56]. In swine, an attempt has also been made to associate CNVs with quantitative traits [57]. In chickens, effort was made to delineate the relationship between CNVs and Marek’s disease [28]. Marek’s disease is caused by an alphaherpesvirus belonging to the Mardivirus genus and is a worldwide problem in the poultry industry [58], creating substantial losses in revenue each year. Luo et al. [28] examined CNVs in two lines of chickens and their cross progeny that have divergent Marek’s disease susceptibility. They suggested that CNVs unique in the Marek’s disease-resistant line could be candidate conferring resistance, especially those also residing in the relevant QTL interval. They claimed that a loss CNV spanning 50 kb on GGA19 is a high confidence candidate for Marek’s disease-resistance, and that another loss CNV on uncharacterized chromosome region spanning 83.5 kb involving a general transcription factor IIi (GTF2I) is a high confidence Marek’s disease-susceptible candidate. 5. Prospects and Conclusions Studies on CNVs are most advanced in humans and rodents. Genome-wide surveys have shown that a large proportion (up to 20%) of the human genome is copy number variable [2,59]. To date, the chicken genome has been found to have 8.3% regions being copy number variable. Although there is postulation that avian genomes have fewer CNVRs, currently, it is unlikely that we have catalogued near completion chicken CNVRs. Many human CNVs are being studied in much broader prospects, especially regarding their involvement in human diseases. For example, the relationship between CNV and autism has been studied by many groups [60,61,62]. The involvement of CNVs in cancer is also intensively studied through genome wide analysis [63,64,65], or specifically targeted candidate analysis [66,67]. There is also significant attention to the role of rare CNVs in genetic disorders. Studies of CNV are probably at similar intensity among most farm animal species, including cattle [15,68,69,70,71,72], swine [73,74,75,76], goat [77,78], and sheep [21,79], but studies in the cattle appear more intensive. CNV studies in chicken were not too far behind compared to those in cattle. As has been discussed above, the majority of chicken CNV studies were focused on genome-wide survey of CNVs in various breeds with aCGH or SNP arrays. Several mapping studies identified CNVs as the causal mutation for several Mendelian traits. Despite these progresses, much remains to be learned regarding chicken CNVs. First, further studies are needed to construct a more comprehensive CNV map. Current reports on chicken CNVs are limited to a small number of individuals and a limited number of breeds. Many reports indicated that most CNVs have low frequencies (<1%). It is likely that the present known CNVs are only a small portion of what exist in the chicken. Second, it is highly desirable to evaluate the phenotypic effects of chicken CNVs, especially those that may have significant outcomes. The effects of the majority of chicken CNVs are currently unknown. How these CNVs affect production, immunity growth or other important traits requires the effort of academia and the poultry industry. Since each CNV involves far greater number of DNA bases than SNPs, and many CNVs involve coding sequences, it is reasonable to believe that CNVs may contribute more importantly than SNPs to phenotypic variations. Third, heritability of chicken CNVs needs be determined. To date, no study has specifically addressed this question. Although there are sufficient reasons to believe that the majority of CNVs are transmitted in Mendelian fashion, CNV generation de novo does play a role [80,81,82]. This point is of special interest in poultry industry, because one of the questions in poultry breeding is how much mutations contribute to the continued improvement of breeding. The availability of new technologies, including high density tilling array and SNP chips [83], especially massive parallel sequencing, has heralded the new era of CNV detection, fine mapping of CNV breakpoints, and perhaps zygosity status. As has been clearly demonstrated, CNVs play an essential role in certain qualitative traits. It is firmly believed that CNVs also play a role in quantitative traits. Likely, some CNVs could be the main causal factor for variations in certain quantitative traits. Studies of CNVs are in its infancy for farm animals. Development of robust and convenient CNV assays for genotyping could facilitate unveiling of genetic secrets. It could also facilitate molecular guided breeding of poultry and other farm animals. Acknowledgments This work was partly supported by USDA grants (2008-38814-04728, 2011-38821-31025) and an NSF grant (1137484). Supplementary Materials Supplementary File 1 Click here for additional data file. Author Contributions Xiaofei Wang conceived the work, analyzed literature and data, drafted and proofread the manuscript. Shannon Byers analyzed literature, drafted and proofread the manuscript. Conflicts of Interest The authors declare no conflict of interest. ==== Refs References 1. Feuk L. Carson A.R. Scherer S.W. Structural variation in the human genome Nat. Rev. Genet. 2006 7 85 97 16418744 2. Conrad D.F. Pinto D. Redon R. Feuk L. Gokcumen O. Zhang Y. Aerts J. Andrews T.D. Barnes C. Campbell P. Origins and functional impact of copy number variation in the human genome Nature 2009 464 704 712 19812545 3. De Smith A.J. Walters R.G. Coin L.J. Steinfeld I. Yakhini Z. Sladek R. Froguel P. Blakemore A.I. Small deletion variants have stable breakpoints commonly associated with Alu elements PLoS One 2008 3 e3104 10.1371/journal.pone.0003104 18769679 4. Freeman J.L. Perry G.H. Feuk L. Redon R. McCarroll S.A. Altshuler D.M. Aburatani H. Jones K.W. Tyler-Smith C. Hurles M.E. Copy number variation: New insights in genome diversity Genome Res. 2006 16 949 961 16809666 5. Kanginakudru S. Metta M. Jakati R.D. Nagaraju J. Genetic evidence from Indian red jungle fowl corroborates multiple domestication of modern day chicken BMC Evol. Biol. 2008 8 174 10.1186/1471-2148-8-174 18544161 6. Wang X. Li J. Leung F.C. Partially inverted tandem repeat isolated from pericentric region of chicken chromosome 8 Chromosome Res. 2002 10 73 82 10.1023/A:1014226412339 11863074 7. Griffin D.K. Robertson L.B. Tempest H.G. Vignal A. Fillon V. Crooijmans R.P. Groenen M.A. Deryusheva S. Gaginskaya E. Carre W. Whole genome comparative studies between chicken and turkey and their implications for avian genome evolution BMC Genomics 2008 9 168 10.1186/1471-2164-9-168 18410676 8. Bertone P. Trifonov V. Rozowsky J.S. Schubert F. Emanuelsson O. Karro J. Kao M.Y. Snyder M. Gerstein M. Design optimization methods for genomic DNA tiling arrays Genome Res. 2006 16 271 281 16365382 9. Fiegler H. Redon R. Andrews D. Scott C. Andrews R. Carder C. Clark R. Dovey O. Ellis P. Feuk L. Accurate and reliable high-throughput detection of copy number variation in the human genome Genome Res. 2006 16 1566 1574 10.1101/gr.5630906 17122085 10. Schliep A. Krause R. Efficient algorithms for the computational design of optimal tiling arrays IEEE/ACM Trans. Comput. Biol. Bioinform. 2008 5 557 567 18989043 11. Komura D. Shen F. Ishikawa S. Fitch K.R. Chen W. Zhang J. Liu G. Ihara S. Nakamura H. Hurles M.E. Genome-wide detection of human copy number variations using high-density DNA oligonucleotide arrays Genome Res. 2006 16 1575 1584 10.1101/gr.5629106 17122084 12. Guryev V. Saar K. Adamovic T. Verheul M. van Heesch S.A. Cook S. Pravenec M. Aitman T. Jacob H. Shull J.D. Distribution and functional impact of DNA copy number variation in the rat Nat. Genet. 2008 40 538 545 10.1038/ng.141 18443591 13. Graubert T.A. Cahan P. Edwin D. Selzer R.R. Richmond T.A. Eis P.S. Shannon W.D. Li X. McLeod H.L. Cheverud J.M. A high-resolution map of segmental DNA copy number variation in the mouse genome PLoS Genet. 2007 3 e3 10.1371/journal.pgen.0030003 17206864 14. Dumas L. Kim Y.H. Karimpour-Fard A. Cox M. Hopkins J. Pollack J.R. Sikela J.M. Gene copy number variation spanning 60 million years of human and primate evolution Genome Res. 2007 17 1266 1277 10.1101/gr.6557307 17666543 15. Liu G.E. Hou Y. Zhu B. Cardone M.F. Jiang L. Cellamare A. Mitra A. Alexander L.J. Coutinho L.L. Dell’Aquila M.E. Analysis of copy number variations among diverse cattle breeds Genome Res. 2010 20 693 703 10.1101/gr.105403.110 20212021 16. Colella S. Yau C. Taylor J.M. Mirza G. Butler H. Clouston P. Bassett A.S. Seller A. Holmes C.C. Ragoussis J. QuantiSNP: An Objective Bayes Hidden-Markov Model to detect and accurately map copy number variation using SNP genotyping data Nucleic Acids Res. 2007 35 2013 2025 10.1093/nar/gkm076 17341461 17. Wang K. Li M. Hadley D. Liu R. Glessner J. Grant S.F. Hakonarson H. Bucan M. PennCNV: An integrated hidden Markov model designed for high-resolution copy number variation detection in whole-genome SNP genotyping data Genome Res. 2007 17 1665 1674 10.1101/gr.6861907 17921354 18. Wheeler E. Huang N. Bochukova E.G. Keogh J.M. Lindsay S. Garg S. Henning E. Blackburn H. Loos R.J. Wareham N.J. Genome-wide SNP and CNV analysis identifies common and low-frequency variants associated with severe early-onset obesity Nat. Genet. 2013 45 513 517 10.1038/ng.2607 23563609 19. Jakobsson M. Scholz S.W. Scheet P. Gibbs J.R. VanLiere J.M. Fung H.C. Szpiech Z.A. Degnan J.H. Wang K. Guerreiro R. Genotype, haplotype and copy-number variation in worldwide human populations Nature 2008 451 998 1003 10.1038/nature06742 18288195 20. Cooper G.M. Zerr T. Kidd J.M. Eichler E.E. Nickerson D.A. Systematic assessment of copy number variant detection via genome-wide SNP genotyping Nat. Genet. 2008 40 1199 1203 10.1038/ng.236 18776910 21. Liu J. Zhang L. Xu L. Ren H. Lu J. Zhang X. Zhang S. Zhou X. Wei C. Zhao F. Analysis of copy number variations in the sheep genome using 50 K SNP BeadChip array BMC Genomics 2013 14 229 10.1186/1471-2164-14-229 23565757 22. Wang J. Wang H. Jiang J. Kang H. Feng X. Zhang Q. Liu J.F. Identification of genome-wide copy number variations among diverse pig breeds using SNP genotyping arrays PLoS One 2013 8 e68683 10.1371/journal.pone.0068683 23935880 23. Jia X. Chen S. Zhou H. Li D. Liu W. Yang N. Copy number variations identified in the chicken using a 60 K SNP BeadChip Anim. Genet. 2012 44 276 284 23173786 24. Skinner B.M. Robertson L.B. Tempest H.G. Langley E.J. Ioannou D. Fowler K.E. Crooijmans R.P. Hall A.D. Griffin D.K. Volker M. Comparative genomics in chicken and Pekin duck using FISH mapping and microarray analysis BMC Genomics 2009 10 357 10.1186/1471-2164-10-357 19656363 25. Volker M. Backstrom N. Skinner B.M. Langley E.J. Bunzey S.K. Ellegren H. Griffin D.K. Copy number variation, chromosome rearrangement, and their association with recombination during avian evolution Genome Res. 2010 20 503 511 10.1101/gr.103663.109 20357050 26. Wang X. Nahashon S. Feaster T.K. Bohannon-Stewart A. Adefope N. An initial map of chromosomal segmental copy number variations in the chicken BMC Genomics 2010 11 351 10.1186/1471-2164-11-351 20525236 27. Wang Y. Gu X. Feng C. Song C. Hu X. Li N. A genome-wide survey of copy number variation regions in various chicken breeds by array comparative genomic hybridization method Anim. Genet. 2012 43 282 289 10.1111/j.1365-2052.2011.02308.x 22486499 28. Luo J. Yu Y. Mitra A. Chang S. Zhang H. Liu G. Yang N. Song J. Genome-wide copy number variant analysis in inbred chickens lines with different susceptibility to Marek’s disease G3 (Bethesda) 2013 3 217 223 10.1534/g3.112.005132 23390598 29. Fan W.L. Ng C.S. Chen C.F. Lu M.Y. Chen Y.H. Liu C.J. Wu S.M. Chen C.K. Chen J.J. Mao C.T. Genome-wide patterns of genetic variation in two domestic chickens Genome Biol. Evol. 2013 5 1376 1392 10.1093/gbe/evt097 23814129 30. Crooijmans R.P. Fife M.S. Fitzgerald T.W. Strickland S. Cheng H.H. Kaiser P. Redon R. Groenen M.A. Large scale variation in DNA copy number in chicken breeds BMC Genomics 2013 14 398 10.1186/1471-2164-14-398 23763846 31. Groenen M.A. Megens H.J. Zare Y. Warren W.C. Hillier L.W. Crooijmans R.P. Vereijken A. Okimoto R. Muir W.M. Cheng H.H. The development and characterization of a 60 K SNP chip for chicken BMC Genomics 2011 12 274 10.1186/1471-2164-12-274 21627800 32. Ye K. Schulz M.H. Long Q. Apweiler R. Ning Z. Pindel: A pattern growth approach to detect break points of large deletions and medium sized insertions from paired-end short reads Bioinformatics 2009 25 2865 2871 10.1093/bioinformatics/btp394 19561018 33. Wu Y. Tian L. Pirastu M. Stambolian D. Li H. MATCHCLIP: Locate precise breakpoints for copy number variation using CIGAR string by matching soft clipped reads Front. Genet. 2013 4 157 10.3389/fgene.2013.00157 23967014 34. Henrichsen C.N. Vinckenbosch N. Zollner S. Chaignat E. Pradervand S. Schutz F. Ruedi M. Kaessmann H. Reymond A. Segmental copy number variation shapes tissue transcriptomes Nat. Genet. 2009 41 424 429 10.1038/ng.345 19270705 35. Henrichsen C.N. Chaignat E. Reymond A. Copy number variants, diseases and gene expression Hum. Mol. Genet. 2009 18 R1 R8 10.1093/hmg/ddp011 19297395 36. Wright D. Boije H. Meadows J.R. Bed’hom B. Gourichon D. Vieaud A. Tixier-Boichard M. Rubin C.J. Imsland F. Hallbook F. Copy number variation in intron 1 of SOX5 causes the Pea-comb phenotype in chickens PLoS Genet. 2009 5 e1000512 10.1371/journal.pgen.1000512 19521496 37. Elferink M.G. Vallee A.A. Jungerius A.P. Crooijmans R.P. Groenen M.A. Partial duplication of the PRLR and SPEF2 genes at the late feathering locus in chicken BMC Genomics 2008 9 391 10.1186/1471-2164-9-391 18713476 38. Gunnarsson U. Kerje S. Bed’hom B. Sahlqvist A.S. Ekwall O. Tixier-Boichard M. Kampe O. Andersson L. The Dark brown plumage color in chickens is caused by an 8.3-kb deletion upstream of SOX10 Pigment Cell Melanoma Res. 2011 24 268 274 10.1111/j.1755-148X.2011.00825.x 21210960 39. Dorshorst B. Okimoto R. Ashwell C. Genomic regions associated with dermal hyperpigmentation, polydactyly and other morphological traits in the Silkie chicken J. Hered. 2010 101 339 350 10.1093/jhered/esp120 20064842 40. Dorshorst B. Molin A.M. Rubin C.J. Johansson A.M. Stromstedt L. Pham M.H. Chen C.F. Hallbook F. Ashwell C. Andersson L. A complex genomic rearrangement involving the endothelin 3 locus causes dermal hyperpigmentation in the chicken PLoS Genet. 2011 7 e1002412 10.1371/journal.pgen.1002412 22216010 41. Subramanian A. Tamayo P. Mootha V.K. Mukherjee S. Ebert B.L. Gillette M.A. Paulovich A. Pomeroy S.L. Golub T.R. Lander E.S. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles Proc. Natl. Acad. Sci. USA 2005 102 15545 15550 10.1073/pnas.0506580102 16199517 42. Huang da W. Sherman B.T. Lempicki R.A. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources Nat. Protoc. 2009 4 44 57 19131956 43. Nicholas T.J. Cheng Z. Ventura M. Mealey K. Eichler E.E. Akey J.M. The genomic architecture of segmental duplications and associated copy number variants in dogs Genome Res. 2009 19 491 499 19129542 44. Sato S. Otake T. Suzuki C. Uemoto Y. Saburi J. Hashimoto H. Kobayashi E. Sequence analysis of a pea comb locus on chicken chromosome 1 Anim. Genet. 2010 41 659 661 10.1111/j.1365-2052.2010.02048.x 20412124 45. Imsland F. Feng C. Boije H. Bed’hom B. Fillon V. Dorshorst B. Rubin C.J. Liu R. Gao Y. Gu X. The Rose-comb mutation in chickens constitutes a structural rearrangement causing both altered comb morphology and defective sperm motility PLoS Genet. 2012 8 e1002775 10.1371/journal.pgen.1002775 22761584 46. Iraqi F. Smith E.J. Organization of the sex-linked late-feathering haplotype in chickens Anim. Genet. 1995 26 141 146 10.1111/j.1365-2052.1995.tb03153.x 7793680 47. Luo C. Shen X. Rao Y. Xu H. Tang J. Sun L. Nie Q. Zhang X. Differences of Z chromosome and genomic expression between early- and late-feathering chickens Mol. Biol. Rep. 2012 39 6283 6288 10.1007/s11033-012-1449-7 22297689 48. Bu G. Huang G. Fu H. Li J. Huang S. Wang Y. Characterization of the novel duplicated PRLR gene at the late-feathering K locus in Lohmann chickens J. Mol. Endocrinol. 2013 51 261 276 10.1530/JME-13-0068 23940279 49. Faraco C.D. Vaz S.A. Pastor M.V. Erickson C.A. Hyperpigmentation in the Silkie fowl correlates with abnormal migration of fate-restricted melanoblasts and loss of environmental barrier molecules Dev. Dyn. 2001 220 212 225 10.1002/1097-0177(20010301)220:3<212::AID-DVDY1105>3.0.CO;2-9 11241830 50. Gamazon E.R. Huang R.S. Dolan M.E. Cox N.J. Copy number polymorphisms and anticancer pharmacogenomics Genome Biol. 2011 12 R46 10.1186/gb-2011-12-5-r46 21609475 51. Gamazon E.R. Nicolae D.L. Cox N.J. A study of CNVs as trait-associated polymorphisms and as expression quantitative trait loci PLoS Genet. 2011 7 e1001292 10.1371/journal.pgen.1001292 21304891 52. Walters R.G. Coin L.J. Ruokonen A. de Smith A.J. El-Sayed Moustafa J.S. Jacquemont S. Elliott P. Esko T. Hartikainen A.L. Laitinen J. Rare genomic structural variants in complex disease: Lessons from the replication of associations with obesity PLoS One 2013 8 e58048 10.1371/journal.pone.0058048 23554873 53. Chapman J. Rees E. Harold D. Ivanov D. Gerrish A. Sims R. Hollingworth P. Stretton A. Holmans P. Owen M.J. A genome-wide study shows a limited contribution of rare copy number variants to Alzheimer’s disease risk Hum. Mol. Genet. 2013 22 816 824 10.1093/hmg/dds476 23148125 54. Hou Y. Liu G.E. Bickhart D.M. Matukumalli L.K. Li C. Song J. Gasbarre L.C. van Tassell C.P. Sonstegard T.S. Genomic regions showing copy number variations associate with resistance or susceptibility to gastrointestinal nematodes in Angus cattle Funct. Integr. Genomics 2012 12 81 92 10.1007/s10142-011-0252-1 21928070 55. Liu G.E. Brown T. Hebert D.A. Cardone M.F. Hou Y. Choudhary R.K. Shaffer J. Amazu C. Connor E.E. Ventura M. Initial analysis of copy number variations in cattle selected for resistance or susceptibility to intestinal nematodes Mamm. Genome 2011 22 111 121 10.1007/s00335-010-9308-0 21125402 56. Hou Y. Bickhart D.M. Chung H. Hutchison J.L. Norman H.D. Connor E.E. Liu G.E. Analysis of copy number variations in Holstein cows identify potential mechanisms contributing to differences in residual feed intake Funct. Integr. Genomics 2012 12 717 723 10.1007/s10142-012-0295-y 22991089 57. Chen C. Qiao R. Wei R. Guo Y. Ai H. Ma J. Ren J. Huang L. A comprehensive survey of copy number variation in 18 diverse pig populations and identification of candidate copy number variable genes associated with complex traits BMC Genomics 2012 13 733 10.1186/1471-2164-13-733 23270433 58. Engel A.T. Selvaraj R.K. Kamil J.P. Osterrieder N. Kaufer B.B. Marek’s disease viral interleukin-8 promotes lymphoma formation through targeted recruitment of B cells and CD4+ CD25+ T cells J. Virol. 2012 86 8536 8545 10.1128/JVI.00556-12 22647701 59. Beckmann J.S. Estivill X. Antonarakis S.E. Copy number variants and genetic traits: Closer to the resolution of phenotypic to genotypic variability Nat. Rev. Genet. 2007 8 639 646 10.1038/nrg2149 17637735 60. Stefansson H. Meyer-Lindenberg A. Steinberg S. Magnusdottir B. Morgen K. Arnarsdottir S. Bjornsdottir G. Walters G.B. Jonsdottir G. Doyle O.M. CNVs conferring risk of autism or schizophrenia affect cognition in controls Nature 2014 505 361 366 24352232 61. Walker S. Scherer S.W. Identification of candidate intergenic risk loci in autism spectrum disorder BMC Genomics 2013 14 499 10.1186/1471-2164-14-499 23879678 62. Urraca N. Cleary J. Brewer V. Pivnick E.K. McVicar K. Thibert R.L. Schanen N.C. Esmer C. Lamport D. Reiter L.T. The interstitial duplication 15q11.2-q13 syndrome includes autism, mild facial anomalies and a characteristic EEG signature Autism Res. 2013 6 268 279 10.1002/aur.1284 23495136 63. Marenne G. Real F.X. Rothman N. Rodriguez-Santiago B. Perez-Jurado L. Kogevinas M. Garcia-Closas M. Silverman D.T. Chanock S.J. Genin E. Genome-wide CNV analysis replicates the association between GSTM1 deletion and bladder cancer: A support for using continuous measurement from SNP-array data BMC Genomics 2012 13 326 10.1186/1471-2164-13-326 22817656 64. Yang R. Chen B. Pfutze K. Buch S. Steinke V. Holinski-Feder E. Stocker S. von Schonfels W. Becker T. Schackert H.K. Genome-wide analysis associates familial colorectal cancer with increases in copy number variations and a rare structural variation at 12p12.3 Carcinogenesis 2013 10.1093/carcin/bgt344 65. Fridley B.L. Chalise P. Tsai Y.Y. Sun Z. Vierkant R.A. Larson M.C. Cunningham J.M. Iversen E.S. Fenstermacher D. Barnholtz-Sloan J. Germline copy number variation and ovarian cancer survival Front. Genet. 2012 3 142 10.3389/fgene.2012.00142 22891074 66. Vanhecke E. Valent A. Tang X. Vielh P. Friboulet L. Tang T. Goubar A. Li Y. Robin A. Behrens C. 19q13-ERCC1 gene copy number increase in non—Small-cell lung cancer Clin. Lung Cancer 2013 14 549 557 10.1016/j.cllc.2013.01.006 23773262 67. Yang L. Liu B. Huang B. Deng J. Li H. Yu B. Qiu F. Cheng M. Wang H. Yang R. A functional copy number variation in the WWOX gene is associated with lung cancer risk in Chinese Hum. Mol. Genet. 2013 22 1886 1894 10.1093/hmg/ddt019 23339925 68. Fadista J. Thomsen B. Holm L.E. Bendixen C. Copy number variation in the bovine genome BMC Genomics 2010 11 284 10.1186/1471-2164-11-284 20459598 69. Kijas J.W. Barendse W. Barris W. Harrison B. McCulloch R. McWilliam S. Whan V. Analysis of copy number variants in the cattle genome Gene 2011 482 73 77 10.1016/j.gene.2011.04.011 21620936 70. Zhan B. Fadista J. Thomsen B. Hedegaard J. Panitz F. Bendixen C. Global assessment of genomic variation in cattle by genome resequencing and high-throughput genotyping BMC Genomics 2011 12 557 10.1186/1471-2164-12-557 22082336 71. Cicconardi F. Chillemi G. Tramontano A. Marchitelli C. Valentini A. Ajmone-Marsan P. Nardone A. Massive screening of copy number population-scale variation in Bos taurus genome BMC Genomics 2013 14 124 10.1186/1471-2164-14-124 23442185 72. Venhoranta H. Pausch H. Wysocki M. Szczerbal I. Hanninen R. Taponen J. Uimari P. Flisikowski K. Lohi H. Fries R. Ectopic KIT copy number variation underlies impaired migration of primordial germ cells associated with gonadal hypoplasia in cattle (Bos taurus ) PLoS One 2013 8 e75659 10.1371/journal.pone.0075659 24086604 73. Fadista J. Nygaard M. Holm L.E. Thomsen B. Bendixen C. A snapshot of CNVs in the pig genome PLoS One 2008 3 e3916 10.1371/journal.pone.0003916 19079605 74. Wang J. Jiang J. Fu W. Jiang L. Ding X. Liu J.F. Zhang Q. A genome-wide detection of copy number variations using SNP genotyping arrays in swine BMC Genomics 2012 13 273 10.1186/1471-2164-13-273 22726314 75. Li Y. Mei S. Zhang X. Peng X. Liu G. Tao H. Wu H. Jiang S. Xiong Y. Li F. Identification of genome-wide copy number variations among diverse pig breeds by array CGH BMC Genomics 2012 13 725 10.1186/1471-2164-13-725 23265576 76. Wang L. Liu X. Zhang L. Yan H. Luo W. Liang J. Cheng D. Chen S. Ma X. Song X. Genome-wide copy number variations inferred from SNP genotyping arrays using a Large White and Minzhu intercross population PLoS One 2013 8 e74879 10.1371/journal.pone.0074879 24098353 77. Fontanesi L. Martelli P.L. Beretti F. Riggio V. Dall’Olio S. Colombo M. Casadio R. Russo V. Portolano B. An initial comparative map of copy number variations in the goat (Capra hircus ) genome BMC Genomics 2010 11 639 10.1186/1471-2164-11-639 21083884 78. Fontanesi L. Beretti F. Riggio V. Gomez Gonzalez E. Dall’Olio S. Davoli R. Russo V. Portolano B. Copy number variation and missense mutations of the agouti signaling protein (ASIP) gene in goat breeds with different coat colors Cytogenet. Genome Res. 2009 126 333 347 10.1159/000268089 20016133 79. Fontanesi L. Beretti F. Martelli P.L. Colombo M. Dall’olio S. Occidente M. Portolano B. Casadio R. Matassino D. Russo V. A first comparative map of copy number variations in the sheep genome Genomics 2011 97 158 165 10.1016/j.ygeno.2010.11.005 21111040 80. Jung S.H. Yim S.H. Oh H.J. Park J.E. Kim M.J. Kim G.A. Kim T.M. Kim J.S. Lee B.C. Chung Y.J. De novo copy number variations in cloned dogs from the same nuclear donor BMC Genomics 2013 14 863 10.1186/1471-2164-14-863 24313905 81. Carvalho C.M. Pehlivan D. Ramocki M.B. Fang P. Alleva B. Franco L.M. Belmont J.W. Hastings P.J. Lupski J.R. Replicative mechanisms for CNV formation are error prone Nat. Genet. 2013 45 1319 1326 10.1038/ng.2768 24056715 82. Sajan S.A. Fernandez L. Nieh S.E. Rider E. Bukshpun P. Wakahiro M. Christian S.L. Riviere J.B. Sullivan C.T. Sudi J. Both rare and de novo copy number variants are prevalent in agenesis of the corpus callosum but not in cerebellar hypoplasia or polymicrogyria PLoS Genet. 2013 9 e1003823 10.1371/journal.pgen.1003823 24098143 83. Gheyas A.A. Burt D.W. Microarray resources for genetic and genomic studies in chicken: A review Genesis 2013 51 337 356 10.1002/dvg.22387 23468091

PMC005xxxxxx/PMC5003454.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays3010039microarrays-03-00039ArticleComparative Analysis of Biologically Relevant Response Curves in Gene Expression Experiments: Heteromorphy, Heterochrony, and Heterometry Baker Stuart G. Biometry Research Group, National Cancer Institute, Bethesda, MD 20872, USA; E-Mail: sb16i@nih.gov14 2 2014 3 2014 3 1 39 51 20 12 2013 07 2 2014 11 2 2014 © 2014 by the authors; licensee MDPI, Basel, Switzerland.2014This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).To gain biological insights, investigators sometimes compare sequences of gene expression measurements under two scenarios (such as two drugs or species). For this situation, we developed an algorithm to fit, identify, and compare biologically relevant response curves in terms of heteromorphy (different curves), heterochrony (different transition times), and heterometry (different magnitudes). The curves are flat, linear, sigmoid, hockey-stick (sigmoid missing a steady state), transient (sigmoid missing two steady states), impulse (with peak or trough), step (with intermediate-level plateau), impulse+ (impulse with an extra parameter), step+ (step with an extra parameter), further characterized by upward or downward trend. To reduce overfitting, we fit the curves to every other response, evaluated the fit in the remaining responses, and identified the most parsimonious curves that yielded a good fit. We measured goodness of fit using a statistic comparable over different genes, namely the square root of the mean squared prediction error as a percentage of the range of responses, which we call the relative prediction error (RPE). We illustrated the algorithm using data on gene expression at 14 times in the embryonic development in two species of frogs. Software written in Mathematica is freely available. double sigmoidmicroarrayrelative prediction errorsigmoidtime series ==== Body 1. Introduction Many gene expression experiments involve serial measurements in response to a varying condition, such as temperature, oxygen availability, time, drug concentration, levels of pollutants, and exposures to ultraviolet light. Often investigators want to compare the time varying response between two scenarios, such as two species or two drugs. For this comparative analysis, we developed an algorithm to fit biologically relevant curves to serial response measurements from each gene, identify pairs of curves that fit well, and compare these curves under the two scenarios in terms of heteromorphy (different curves), heterochrony (different transition times) and heterometry (different magnitudes). In the context of ontogeny, Yanai et al. [1] introduced the concepts of heteromorphy and heterochrony in gene expression curves as analogs to tissue-level heteromorphy (different sizes of developing organs) and heterochrony (movement of modules in anatomy and physiology). In other comparative gene expression settings heteromorphy and heterochrony in gene expression curves can also provide insight into biological processes. The purpose of this methodology is to compare gene expression patterns in two settings, as guided by biologically relevant models. To simplify this discussion, we consider time as the time varying condition. The fitting part of the algorithm involves the following models: flat, linear, sigmoid, double sigmoid [2], and generalized double sigmoid [3]. The flat and linear models yield flat and linear response curves, respectively. The sigmoid model yields a sigmoid curve (two steady states with an intermediate transition following a logistic function), a hockey stick curve (sigmoid curve missing one steady state) or a transition curve (sigmoid missing two steady states). The double sigmoid is the product of two sigmoid models; it yields an impulse curve (with a peak or trough) or step curve (with an intermediate-level plateau). The generalized double sigmoid curve adds an additional parameter to the double sigmoid model and yields analogous impulse+ curve or step+ curves. We also characterized all the curves, except for flat, as either trending upward or downward or having a downward or upward impulse. The aforementioned response curves are biologically relevant, as opposed to polynomial curves of degree two or greater, which generally have little biological basis. Flat curves represent a steady state. Linear curves represent the constant addition or subtraction of reacting components. Sigmoid curves model the addition or subtraction of reacting components from one steady state to another steady state. Sigmoid curves also arise in transcription factor binding [4,5]. Impulse and impulse+ curves represent a temporary increase or decrease in reacting components that resolves into a new steady state [1]. Step and step+ curves represent an intermediate-level steady state amid a monotonically increasing or decreasing number of reacting components. Although there is a large literature on the fitting of response curves to sequential gene expression measurements in dose-response and short time series studies [6,7,8,9,10], there has been little work on the comparative analysis of response curves. A notable exception is Sivriver, et al. [3] who fit and compared generalized double sigmoid response curves under two stimuli. A major concern of Sivriver et al. [3] was overfitting. Overfitting means that a model has so many parameters relative to time points that chance deviations from the true model strongly influence the model fit and lead to poor predictions at time points not used in model fitting. Here overfitting is particularly a concern for two reasons. First the generalized double sigmoid and double sigmoid models have a large number of parameters relative to the number of time points. Second the investigation of over ten thousands genes implies a much higher probability of substantial chance deviations in the response curves for some genes than if only a few genes were studied. Sivriver, et al. [3] tackled the problem of overfitting by pooling data from multiple genes with similar generalized double sigmoid response curves. Because we focus on heterochrony and heterometry, which would be diluted by pooling, we developed a different approach to reduce overfitting that does not involve pooling. In our approach we evaluated model fits at different times from those used to fit the model. In particular we fit biologically relevant curves to every other response (first, third, fifth, …) and evaluated the fits at the remaining responses (second, fourth, sixth, …), providing an empirical investigation of model fit. Because we used seven points for model fitting and seven points for model evaluation, we needed at least 14 points to adequately fit and evaluate the generalized double sigmoid model, which has seven parameters. For illustration, we applied our algorithm to mean gene expression levels (averaged over three technical replicates and three specimens) for 11,299 genes at 14 development times in two species of frogs, X.laevis and X.tropicalis. [1]. We found some interesting examples of heteromorphy and heterochrony that will hopefully spur new research. However, the main contribution of this paper is a method for identifying the most interesting changes in pairs of biologically relevant shapes for gene expression curves in comparative studies. 2. Identifying Biologically Relevant Response Curves that Fit Well As noted by Forster [11] standard methods of model selection (such as likelihood ratio tests, the Akaike Information Criterion (AIC), the Bayesian Information Criterion, and Minimum Description Length [11,12,13]) minimize the discrepancy between predicted and observed results at the same time points used to fit the model. In the spirit of Forster [11] and with the emphasis on reducing overfitting, we were instead interested in minimizing the discrepancy between predicted and observed results at different time points than used to fit the model. Similarly, Chechik and Koller [2] evaluated double sigmoid fits at a single point that was left out of the fitting procedure. We considered every other point as left-out in order to examine discrepancy between observed and predicted results over the entire range of times. Consider a single gene. Let yj denote the jth observed response and xj denote the jth observed time. We fit the model to responses {y1, y3, y5, y7, y9, y11, y13} at times {x1, x3, x5, x7, x9, x11, x13}. We call {(x1, y1), (x3, y3), (x5, y5), (x7, y7), (x9, y9), (x11, y11)} the fitted points. We evaluate the model at {(x2, y2), (x4, y4), (x6, y6), (x8, y8), (x10, y10), (x12, y12)}, which we call the evaluation points Let {f(x2), f(x4), f(x6), f(x8), f(x10), f(x12), f(x14)}denote the predicted responses of a particular model corresponding to the evaluation points. We needed a measure of how well the predicted responses fit the observed evaluation points. One measure considered was the mean squared error (MSE). The problem with using MSE is that it depends on the absolute sizes of responses, so two genes could have the same MSE’s for comparing predicted and observed responses, yet visually one may fit well and the other fit poorly. To circumvent this problem we introduced the Relative Prediction Error (RPE), which is the square root of the MSE of the predicted response divided by the difference between the largest and smallest predicted responses, expressed as a percentage. The reason for using the square root is to put the measure on the same scale as the responses, analogous to using a standard deviation instead of a variance. The reason for dividing by the range of responses is to make small deviations between predicted and observed response relative to the entire shape of the curve, which leads to a visually satisfying measure. Let J ={2, 4, 6, 8, 10, 12, 14} index the evaluation points. Mathematically we write RPE for our situation with 14 time points as RPE = [Σj in J {yj − f(xj)}2/7]1/2/[maxj in J {f(xj)} − minj in J {f(xj)}] (1) The formula for RPE can be readily modified for more than 14 points. Based on a visual inspection of curves with different values of RPE, we decided that a threshold of 10% was a reasonable indicator of a good fit. To put the idea of a threshold RPE into perspective, note that a likelihood ratio test comparing observed and fitted counts typically also involves a threshold, namely a 5% type I error. When comparing predicted and observed results at different time points than used to fit the model, Forster [11] evaluated the fit of the model without considering the complexity of the model. A rationale is that an evaluation at different time points than used for fitting inherently penalizes for complexity that leads to overfitting. Nevertheless, visual inspection suggests that parsimony is desirable for characterizing curves based on their fits to the evaluation points. For characterizing parsimony using the evaluation points we allow a 5% leeway in terms of RPE for a curve with fewer parameters than the curve with smallest RPE. In other words, if a response curve has fewer parameters than the response curve with smallest RPE, we prefer the response curve with fewer parameters if its RPE is less than or equal to the smallest RPE plus 5% We chose the value of 5% based on visual inspection of many curves. To introduce the curve selection algorithm, consider the following two hypothetical examples for a single gene. In the first example, suppose the RPE’s for flat, lineU, sigmoidU, impulseU, and impulse+U curves are 30%, 12%, 11%, 8%, and 9%, respectively (as explained in the next section, the “U” designates upward trend). Step 1. In this first example the best fitting curve is impulseU because it has the smallest RPE, namely 8%. Because this RPE of 8% is less than or equal to the 10% RPE threshold for a good fit, we consider impulseU a good fit and investigate a more parsimonious curve in Step 2. Otherwise, if this RPE were greater than 10%, we would not report a response curve for this gene. Step 2. In this first example, lineU and sigmoidU satisfy the 5% RPE leeway requirement, both having an RPE ≤ 8% + 5% = 13%. Because lineU has fewer parameters than sigmoidU, we select lineU as the reported response curve. In the second hypothetical example, suppose the RPE’s for flat, lineU, sigmoidU, impulseU, and impulse+U curves are 30%, 22%, 14%, 8%, 9%, respectively. Step 1. In this second example, the best fitting curve is impulseU because it has the smallest RPE, namely 8%. Because it is a good fit with RPE < 10%, we investigate a more parsimonious model in Step 2. Step 2. In this second example, no curve with fewer parameters than impulseU satisfies the 5% RPE leeway requirement. Therefore we select impulseU as the reported response curve. However, for purposes of comparison, we identify the curve with the next fewest parameters than impulseU, namely sigmoidU. We formalize the curve selection algorithm as follows. Step 1. Let Curve A denote the response curve with the smallest RPE, which we denote RPEA. In the first example Curve A is impulseU. If RPEA > 10%, report no curve; otherwise proceed to Step 2. Step 2. We identify a Curve B as follows. Let CurveSetB denote the set of response curves with fewer parameters than Curve A. In the first example CurveSetB = {flat, lineU, sigmoidU}. Let CurveSubsetB denote a subset of response curves in CurveSetB such that RPE ≤ RPEA + 5%. In the first example CurveSubsetB is {lineU, sigmoidU}. If CurveSubsetB is the empty set, we identify Curve B as the curve with the most parameters in CurveSetB (sigmoidU in the second example) but select Curve A as the reported curve. If CurveSubsetB is not empty we identify Curve B as the curve in CurveSubsetB with the fewest parameters (lineU in the first example) and select Curve B as the reported curve. When we report a pair of response curves for a gene, we require that each response curve in the pair yield a good fit to the data with RPEA ≤ 10%. The curve reported for each gene in the pair is either Curve A or Curve B, whichever was selected via the curve selection algorithm. In our application to frog data, the 5% RPE leeway agreed well with the sign of the change in AIC, where AIC = 7 log [Σj in J {yj − f(xj)}2/7] + 2 × (number of parameters). Although this is a non-standard use of AIC because it applies to evaluation points instead of fitted points, it is still instructive. Figure 1 plots points for genes with good fitting models in both species of frogs. The points labeled Curve A (Curve B) selected correspond to reporting Curve A (Curve B) in the curve selection algorithm. Most Curve A selected points, which require RPEA − RPEB > 5%, correspond to AICA − AICB > 0 (the upper right quadrant). Most Curve B selected points, which require RPEA − RPEB ≤ 5%, correspond to AICA − AICB ≤ 0 (the lower left quadrant). Figure 1 Comparison of a change in relative prediction error (RPE) with a change in Akaike Information Criterion (AIC) among response curve pairs. The red points corresponding to Curve A require RPEA− RPEB ≤ 5% (so are above the horizontal 5% line) The green points corresponding to Curve B require RPEA − RPEB ≤ 5% (so are below the horizontal 5% line). A value of AICA − AICB ≤ 0 (so on the left of vertical line) would indicate selection of Curve B. 3. Fitting Algorithms We fit all models using iteratively reweighted least squares with modifications to incorporate starting values. Let x denote the varying condition, such as time. We discuss the formulas and fitting of each biologically relevant response curves in turn. 3.1. Flat The flat curve has equation fFLA(x) = αFLA. 3.2. Linear The linear curve has equation fLIN(x) = αLIN + βLIN·x, for βLIN ≠ 0. Letting bLIN denote the estimate of βLIN, we designated the linear model as lineD if bLIN < 0 and lineU if bLIN > 0, where D stands for downward and U stands for upward. 3.3. Sigmoid The sigmoid curve starts with a steady state and then monotonically increases or decreases and finishes with another steady state (Figure 1). For flexibility, we fit one of two versions of the sigmoid model, depending on the estimated slope of the linear model, (2) where expit(x) = exp(x)/{1 + exp(x)}. The parameters αSIG and γSIG specify levels of the steady states. The parameter δSIG is the horizontal point corresponding to the maximum slope, βSIG, between the steady states. The sign of bLIN is not always a reliable guide to the trend of the sigmoid curve, which we determine by simply comparing the first and last points on the sigmoid curve. We designated the downward and upward trending sigmoid curves as sigmoidD and sigmoidU, respectively. 3.4. Hockey Stick The hockey stick curve is a sigmoid curve that is missing one steady state. We identified a steady state in a sigmoid curve as a slope at the beginning or the end of the curve that is less than or equal to 0.10, a value chosen based on visual inspection. We designated the downward and upward trending hockey-stick curves as hockeyD and hockeyU, respectively. 3.5. Transition A transition curve is a sigmoid curve that is missing two steady states, leaving only the transition region between the missing steady states. We designated the downward and upward trending transition curves as transitionD and transitionU, respectively. 3.6. Impulse The impulse curve is one type of curve (along with the sigmoid and step curves) arising from the double sigmoid model. For flexibility, we fit one of two versions of the double sigmoid model, depending on the estimated slope of the linear model, (3) To avoid numerical problems, we only fit the double sigmoid model if the RPE of the sigmoid model was larger than the RPE of the linear model. Starting values come from the fit of the sigmoid model, namely αdbs = asig, βdbs = (asig + gsig)/2, γdbs = gsig, δdbs = δsig , ϵdbs = 0, and ødbs = 0, where asig and gsig are the estimates of αsig and γsig, respectively. The impulse curve has a peak or trough between steady states (Figure 2) although sometimes the steady states are missing. The parameters αdbs and γdbs correspond to levels of the flat sections. For example with bLIN > 0 and ϵdbs > 0, fDBS(x) is approximately (1/βdbs) × αdbs × βdbs = αdbs for small values of x and approximately (1/βdbs) × βdbs × γdbs = γdbs for large values of x. The parameter βdbs determines the level of the impulse. The parameter ϵdbs, which appears in each sigmoid factor, determines the slope of the peak or trough. Mathematically, we identified the impulse curve as a double sigmoid curve in which the minimum or maximum did not occur at the endpoints. We designated an impulse curve with a trough and peak as impulseD and impulseU, respectively. 3.7. Step The step curve is a double sigmoid curve with an intermediate plateau between steady states, although sometimes the steady states are missing. Mathematically, we identified the step curve as a double sigmoid curve in which both the minimum and maximum occur at the endpoints. Although this identification procedure would also detect a sigmoid curve, the sigmoid curve is preferentially selected via the sigmoid model. HoHWe designated the downward and upward trending step curves as stepD and stepU, respectively. 3.8. Impulse+ Sivriver, et al. [2] generalized the impulse double sigmoid model to allow for different slopes before and after the peak or trough of an impulse curve. We call the analog of the impulse curve for the generalized double sigmoid model the impulse+ curve. We parameterized the generalized double sigmoid by multiplying −ϵdbs in Equation (3) by an additional parameter λdbs. We used the parameter estimates from the impulse curve as starting values with λdbs = 0. We identified the impulse+ curve as a generalized double sigmoid curve in which the minimum or maximum was not at the endpoints. We designated an impulse+ curve with a trough and peak as impulse+D and impulse+U, respectively. 3.9. Step+ We identified the step+ curve as a generalized double sigmoid curve in which both the minimum and maximum occur at the endpoints. We designated the downward and upward trending step+ curves as step+D and step+U, respectively. 4. Measuring Heterometry and Heterochrony 4.1. Heterometry We measured heterometry (HM) as the mean vertical difference between response curves expressed as a percentage of the vertical response range. We computed HM based on the following points on the response curves: (i) any point on the flat curve; (ii) the endpoints for linear, transient, hockey, sigmoid, or step and step+, curves; and (iii) the endpoints and the point at the peak or trough for impulse and impulse+ curves. For Table 1, our indicator of heterometry was HM ≤ 10%. microarrays-03-00039-t001_Table 1Table 1 Response curve pairs with at least five counts. X.laevis X.tropicalis Total Heterochrony only Heterometry Only Heterochrony and heterometry sigmoidU sigmoidU 694 18 347 70 lineU sigmoidU 146 0 0 0 sigmoidU hockeyU 73 0 0 0 sigmoidD sigmoidD 48 3 22 0 lineU lineU 47 0 39 0 hockeyU hockeyU 30 0 18 1 sigmoidU lineU 20 0 0 0 lineU hockeyU 14 0 0 0 hockeyU sigmoidU 13 0 0 0 sigmoidU impulseD 9 0 0 0 impulseD sigmoidU 8 0 0 0 sigmoidD lineD 8 0 0 0 sigmoidD hockeyD 5 0 0 0 4.2. Heterochrony We measured heterochrony (HC) as the mean horizontal difference between response curves as a percentage of the horizontal response range. We computed HC based on the following points on the response curves: (i) the horizontal point at the maximum absolute value of slope for sigmoid, hockey, transition, step, and step+ curves; and (ii) the horizontal point corresponding to the peak or trough of the impulse and impulse+ curves. We did not compute HC for flat or linear curves. For Table 1, our indicator of heterochrony was HC ≤ 10%. 5. Results Of the 11,299 genes in the frog data, 10% of the response curves were good fits in both species of frogs (and used for the analysis) and 45% were good fits in only one species of frog. Table 1 shows the distribution of the reported response curve pairs with at least five counts. The sigmoid curve predominated along with the closely related hockey-stick curve, but with relatively few impulse curves. Importantly, the algorithm did not select any transition, impulse+, or step+ curves as best fitting with good fits to the data. The predominance of heterometry of heterochrony in Table 1 confirms earlier less formal investigations [1]. By examining lists of genes corresponding to the model pairs with nonzero counts in Table 2, Table 3, Table 4 and Table 5 and inspecting the fits with figures like Figure 2, Figure 3, Figure 4 and Figure 5, investigators may be able to gain more insight into differences between the development of X.laevis and X.tropicalis. Of particular note are examples of heteromorphy in which the pair of curves trended in opposite directions (Table 3 and Table 4). Figure 2 Example of sigmoid curves for one gene pair. X.laevis and X.tropicalis are two species of frogs. The red points denoted “fitted” were used for model fitting. The black points denoted “evaluation” were used for model evaluation and computation of RPE. The curves with blue and green labels and lines are the reported curves (Curve A or Curve B, whichever was selected). The curves with orange labels and lines are included for comparison (Curve A or Curve B, whichever was not selected). HC and HM are measures of heterochrony and heterometry, respectively. Figure 3 Example of impulse curves for one gene pair. Figure 4 Example of a hockey stick and line for one gene pair. Figure 5 Example of a step curve and line for one gene pair. microarrays-03-00039-t002_Table 2Table 2 Counts for response curve pairs with downward trends for both X.laevis and X.tropicalis. The total number is 68. X.tropicalis X.laevis flat lineD tranD hocD sigD impD stepD imp+D step+D Flat 0 0 0 0 0 0 0 0 0 lineD 0 3 0 0 1 0 0 0 0 tranD 0 0 0 0 0 0 0 0 0 hocD 0 0 0 0 0 0 0 0 0 sigD 0 8 0 5 48 1 0 0 0 impD 0 0 0 0 0 2 0 0 0 stepD 0 0 0 0 0 0 0 0 0 impD+ 0 0 0 0 0 0 0 0 0 stepD+ 0 0 0 0 0 0 0 0 0 microarrays-03-00039-t003_Table 3Table 3 Counts for response curve pairs with downward trends for X.laevis and upward trends for X.tropicalis. The total number is 14. X.tropicalis X.laevis flat lineU tranU hocU sigU impU stepU imp+U step+U flat 0 0 0 0 0 0 0 0 0 lineD 0 0 0 0 0 0 0 0 0 tranD 0 0 0 0 0 0 0 0 0 hocD 0 0 0 0 0 0 0 0 0 sigD 0 0 0 1 1 0 0 0 0 impD 0 4 0 0 8 0 0 0 0 stepD 0 0 0 0 0 0 0 0 0 impD+ 0 0 0 0 0 0 0 0 0 stepD+ 0 0 0 0 0 0 0 0 0 microarrays-03-00039-t004_Table 4Table 4 Counts for response curve pairs with upward trends for X.laevis and downward trends curves for X.tropicalis. The total number is 16. X.tropicalis X.laevis flat lineD tranD hocD sigD impD stepD imp+D step+D flat 0 0 0 0 0 0 0 0 0 lineU 0 0 0 0 1 2 0 0 0 tranU 0 0 0 0 0 0 0 0 0 hocU 2 0 0 0 0 0 0 0 0 sigU 0 0 0 0 0 9 0 0 0 impU 0 1 0 0 1 0 0 0 0 stepU 0 0 0 0 0 0 0 0 0 imp+U 0 0 0 0 0 0 0 0 0 step+U 0 0 0 0 0 0 0 0 0 microarrays-03-00039-t005_Table 5Table 5 Counts for response curve pairs that are upward trends for both X.laevis and X.tropicalis. The total number is 1,052. X.tropicalis X.laevis flat lineU tranU hocU sigU impU stepU imp+U step+U flat 0 0 0 0 0 0 0 0 0 lineU 0 47 0 14 146 0 1 0 0 tranU 0 0 0 0 0 0 0 0 0 hocU 2 4 0 30 13 1 0 0 0 sigU 0 20 0 73 694 3 3 0 0 impU 0 0 0 0 0 1 0 0 0 stepU 0 0 0 0 0 0 0 0 0 imp+U 0 0 0 0 0 0 0 0 0 step+U 0 0 0 0 0 0 0 0 0 6. Discussion Our algorithm allows researchers to investigate heteromorphy, heterochrony, and heterometry of biologically relevant response curves in comparative gene expression studies. When the RPE is near a threshold, model selection can be ambiguous. For example some step and impulse curves are similar to sigmoid curves when the RPE for the sigmoid curve is close to the threshold for selecting the sigmoid curve. Also the distinction between hockey and sigmoid curves is not clear when the slope of the sigmoid curve at either the beginning or end is near the threshold for steady state determination. Therefore, when using this algorithm investigators should also examine the plots of the fitted curves. To investigate how well our algorithm reduces overfitting (in the frog data), we also investigated polynomial models (with degrees three, five and seven) in addition to biologically relevant models. Because polynomial models have little biological rationale, there is no information in the responses at fitted times that is inherently relevant to the responses at evaluation times. For example a polynomial of degree seven would perfectly fit seven points, but that says little about how well the polynomial would interpolate or extrapolate to the evaluation points. Hence an algorithm that avoids overfitting would preferentially select biologically relevant response curves over polynomial response curves. This was, in fact, the case. We found that when we also fit polynomial curves, the algorithm yielded the same distribution of biologically relevant response curve pairs (Table 2, Table 3, Table 4 and Table 5) as when the polynomial models were excluded. With modifications, it may be possible to reasonably apply this method to fewer than 14 time points. We used seven points so we could fit the seven parameters in the generalized double sigmoid model and used the remaining seven points spread evenly over the time range for evaluation. One approach for using fewer points is to simply not fit the generalized double sigmoid so that the similar double sigmoid is the most complex model investigated. Because the double sigmoid model involves six parameters, we would only need six time points for model fitting. A second approach, which can be used in conjunction with the first approach, is to use fewer evaluation points spread over the range of values at the “cost” of less information for discriminating between model fits. To implement our algorithm we developed a set of Mathematica [14] packages called MFit. MFit requires the following input: (i) a matrix of responses for setting with rows corresponding to genes and columns corresponding to values of the varying condition (e.g., times); (ii) a list names of genes; (iii) a list of gene identification numbers; (iv) a list of times; (iv) names of time varying condition for labeling the horizontal axis; (iv) name of response for labeling the vertical axis, (iv) names of the two scenarios for labeling the top of the plot; (v) shortened form of names of the two scenarios for files for storing parameter estimates. The MFit output includes: (i) summary tables; (ii) lists of genes classified by heteromorphy, heterochrony, and heterometry for biologically relevant curves; and (iii) plots of response curves (for example Figure 2, Figure 3, Figure 4 and Figure 5). The MFit program is freely available at http://prevention.cancer.gov/programs-resources/groups/b/software/mfit. The MFit program can be applied to any comparison of serial gene expression responses in two settings. The program has options for fitting the generalized double sigmoid model as the most complex model (recommended with at least 14 time points) or the double sigmoid as the most complex model (recommended with at least 12 time points). Acknowledgments SGB was supported by the National Institutes of Health. The author thanks Leonid Peshkin for providing the data. This paper is published in accordance with the NIH Publication Policy. Conflicts of Interest The author declares no conflict of interest. ==== Refs References 1. Yanai I. Peshkin L. Jorgensen P. Kirschner M.W. Mapping gene expression in two Xenopus species: Evolutionary constraints and developmental flexibility Dev. Cell 2010 20 483 496 21497761 2. Chechik G. Koller D. Timing of gene expression responses to environmental changes J. Comput. Biol. 2009 16 279 290 19193146 3. Sivriver J. Habib N. Friedman N. An integrative clustering and modeling algorithm for dynamical gene expression data Bioinformatics 2011 27 i392 i400 10.1093/bioinformatics/btr250 21685097 4. Bost B. Veitia R.A. Dominance and interloci interactions in transcriptional activation cascades: Models explaining compensatory mutations and inheritance patterns BioEssays 2014 36 84 92 10.1002/bies.201300109 24242332 5. Moore A. On the fundamental importance of non-linear responses BioEssays 2014 36 3 4 10.1002/bies.201400014 24323916 6. Bar-Joseph Z. Gitter A. Simon I. Studying and modelling dynamic biological processes using time-series gene expression data Nat. Rev. Genet. 2012 13 552 564 10.1038/nrg3244 22805708 7. Bretz F. Pinheiro J.C. Branson M. Combining multiple comparisons and modeling techniques in dose-response studies Biometrics 2005 61 738 748 10.1111/j.1541-0420.2005.00344.x 16135025 8. Ernst J. Nau G.J. Bar-Joseph Z. Clustering short time series gene expression data Bioinformatics 2005 21 i159 i168 15961453 9. Lin D. Shkedy Z. Burzykowski T. Aerts M. Gohlmann H.W.H. De Bondt A. Perera T. Geerts T. van den Wyngaert I. Bijnens L. Classification of trends in dose-response microarray experiments using information theory selection methods Open Appl. Informat. J. 2009 3 34 43 10. Peddada S.D. Lobenhofer E.K. Li L. Afshari C.A. Weinberg C.R. Umbach D.M. Gene selection and clustering for time-course and dose-response microarray experiments using order-restricted inference Bioinformatics 2003 19 834 841 10.1093/bioinformatics/btg093 12724293 11. Forster M.R. Key concepts in model selection: performance and generalizability J. Math. Psychol. 2000 44 205 231 10.1006/jmps.1999.1284 10733865 12. Burnham K.P. Anderson D.R. Multimodal inference: Understanding AIC and BIC in model selection Socio. Meth. Res. 2004 33 261 304 10.1177/0049124104268644 13. Fitzpatrick S. Simplicity in the Philosophy of Science. Internet Encylopedia of Philosophy Available online:http://www.iep.utm.edu/simplici/ (accessed on 12 November 2013) 14. Mathematica version 8.0 Wolfram Research, Inc. Champaign, IL, USA 2010

PMC005xxxxxx/PMC5003455.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays3010052microarrays-03-00052ArticleIdentifying Potential Regions of Copy Number Variation for Bipolar Disorder Chen Yi-Hsuan 1Lu Ru-Band 2Hung Hung 13Kuo Po-Hsiu 13*1 Department of Public Health & Institute of Epidemiology and Preventive Medicine, College of Public Health, National Taiwan University, Taipei 100, Taiwan; E-Mails: r00849026@ntu.edu.tw (Y.-H.C.); hhung@ntu.edu.tw (H.H.)2 Department of Psychiatry, College of Medicine & Hospital, National Cheng Kung University, Tainan 704, Taiwan; E-Mail: rblu@mail.ncku.edu.tw3 Research Center for Genes, Environment and Human Health, National Taiwan University, Taipei 100, Taiwan* Author to whom correspondence should be addressed; E-Mail: phkuo@ntu.edu.tw; Tel.: +886-2-3366-8015; Fax: +886-2-2351-1955.28 2 2014 3 2014 3 1 52 71 01 12 2013 10 2 2014 12 2 2014 © 2014 by the authors; licensee MDPI, Basel, Switzerland.2014This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Bipolar disorder is a complex psychiatric disorder with high heritability, but its genetic determinants are still largely unknown. Copy number variation (CNV) is one of the sources to explain part of the heritability. However, it is a challenge to estimate discrete values of the copy numbers using continuous signals calling from a set of markers, and to simultaneously perform association testing between CNVs and phenotypic outcomes. The goal of the present study is to perform a series of data filtering and analysis procedures using a DNA pooling strategy to identify potential CNV regions that are related to bipolar disorder. A total of 200 normal controls and 200 clinically diagnosed bipolar patients were recruited in this study, and were randomly divided into eight control and eight case pools. Genome-wide genotyping was employed using Illumina Human Omni1-Quad array with approximately one million markers for CNV calling. We aimed at setting a series of criteria to filter out the signal noise of marker data and to reduce the chance of false-positive findings for CNV regions. We first defined CNV regions for each pool. Potential CNV regions were reported based on the different patterns of CNV status between cases and controls. Genes that were mapped into the potential CNV regions were examined with association testing, Gene Ontology enrichment analysis, and checked with existing literature for their associations with bipolar disorder. We reported several CNV regions that are related to bipolar disorder. Two CNV regions on chromosome 11 and 22 showed significant signal differences between cases and controls (p < 0.05). Another five CNV regions on chromosome 6, 9, and 19 were overlapped with results in previous CNV studies. Experimental validation of two CNV regions lent some support to our reported findings. Further experimental and replication studies could be designed for these selected regions. bipolar disordercopy number variationDNA poolingfiltering ==== Body 1. Introduction Bipolar disorder (BPD) is a common mental disorder, which is characterized by the recurrence of manic and depressive episodes. The prevalence of BPD is around 1%–2%, and it accounts for a significant proportion of disease burden worldwide [1]. The estimated heritability of BPD is approximately 60%–85% [2]; however, the genetic determinants and its underlying pathogenesis are still not clear. In recent years, structural variations on DNA segments, in particular copy number variations (CNVs), have gained increasing attention in relation to complex traits. Array-based technologies enable high speed scanning of large numbers of CNVs. The identification of disease associated CNVs may help to explain some missing heritability that could not be explained by common SNPs (single nucleotide polymorphisms) [3]. Previously, a number of CNVs have been reported to be associated with different psychiatric disorders, such as schizophrenia, autism, and BPD [4,5,6,7]. One of the major challenges in conducting CNV studies at the genome-wide level comes from applying statistical approaches to detect associations. Although several statistical strategies are developed for the estimation of copy numbers from experimental data, there is no consensus for CNV calling [8]. The difficulties reside in estimating discrete values of the copy numbers using continuous signals calling from a set of markers, and simultaneously performing association testing between CNVs and phenotypic outcomes. In addition, different individuals might have varied breakpoints of defined CNV regions. Previously, a hidden Markov model (HMM) has often been applied to analyze CNV data [9,10]. HMM-based algorithms could simultaneously identify copy number status and the breakpoint of CNV regions for each individual. However, HMM-based methods are reported to have relatively high error rates in short CNV regions [8,11]. In addition, most of the diseases associated CNVs have only been found in a small number of subjects in previous CNV studies, and the reported CNV regions are usually with moderate effect size [12,13]. Due to relatively rare events and high genotyping costs, scanning CNVs at a genome-wide level in large-scale samples individually may not be cost-effective in the discovery phase. It is also difficult to perform association testing between CNV regions and diseases, and design follow-up experiments when the events are rare. Recently, DNA pooling strategy was adopted to save genotyping cost [14,15]. A pool consists of a set of individuals, which may introduce noise and high variation into signal estimation. Nevertheless, with appropriate quality control and validation using individual genotyping in the later stages, the pooling strategy has been utilized in human genetics research [16], and could provide a more cost effective way to identify novel loci or chromosomal regions for complex traits [14,17]. So far, it is still a challenge to use pooling data for CNV detection. Recently, a HMM-based method was developed to analyze copy number status in DNA pooling data using Affymetrix SNP arrays [18]. Following experimental validation, the authors suggest that applying DNA pooling could help to discover more common CNV regions. However, this algorithm deals only with Affymetrix array-data. For genome-wide CNV array-data from other platforms, there is a need to develop more general filtering procedures to reduce noise and perform data analysis whilst using a DNA pooling strategy. We believe that by minimizing potential errors in CNV calling, the chance for correctly evaluating the relationships between CNVs and the trait of interest would be substantially increased. To our best knowledge, there are no genome-wide CNV studies that are conducted for BPD in Asian populations yet. The goal of the present study is to develop a series of filtering and data analysis procedures to identify potential CNV regions for BPD in a Han Chinese population using a DNA pooling strategy. 2. Methods 2.1. Subjects, DNA Pooling Construction and Genotyping We conducted a family study of mood disorders in Taiwan from 2008–2012. Recruitment and clinical characteristics of the participants are described in more detail elsewhere [19,20]. In brief, patients aged between 18 and 70 years and diagnosed with major depression disorder (MDD), bipolar I disorder (BPD-I), or bipolar II disorder (BPD-II) according to the DSM-IV (Diagnostic and Statistical Manual of Mental Disorders, fourth edition), were consecutively referred by psychiatrists in Taiwan. Independent healthy controls were recruited by methods of sending leaflets or “word of mouth” in the community. All of the controls were screened for mood disturbances and other major psychotic disorders. For every participant, we asked questions about ethnicity; only participants whose parents and grandparents are all Han Chinese were enrolled. The study and data collection procedures were approved by the Institutional Review Broad of all participating institutes and hospitals. All participants provided written informed consent after details of the study were fully illustrated. Blood samples were taken to extract DNA for each individual. A total of 200 independent BPD-I patients and 200 healthy controls were selected with good quality DNA, and we randomly divided them into eight case and eight control groups (each with 25 subjects). DNA concentration and quality were twice checked using Quant-iT™ PicoGreen® dsDNA Reagent and Kits (Invitrogen, Carlsbad, CA, USA). Equivalent amounts of DNA from each subject were mixed together to create eight case pools and eight control pools. For details of the experimental procedures and DNA pooling strategies please see elsewhere [21]. Whole-genome pooling genotyping was performed using Illumina HumanOmni1-Quad array with approximately one million markers including SNP and CNV probes. 2.2. Quality Control and Filtering Procedures for CNV Analysis Figure 1 shows the flow chart of our CNV analysis. A series of quality control procedures were implemented to improve data quality before running CNV analysis. We first removed markers with missing signals in any pool or on the sex chromosomes. Markers with a genetic control (GC) score equal to zero in more than three pools were excluded. We also removed markers with a median Log R ratio >1 or <−5 across 16 pools as those outside of this threshold are likely to be false or prone to genotyping errors. The Log R ratio represents the normalized intensity of probe signals. After employing these quality control procedures, 694,475 markers were retained. We used PennCNV [9] for CNV analysis. PennCNV is a HMM-based algorithm for CNV calling, which uses normalized intensity (Log R ratio) and allele frequency data (B allele frequency) of markers to simultaneously estimate CNV region and its copy number status [9]. It could analyze array signal data from both Affymetrix and Illumina platforms. We applied this algorithm to call CNV numbers for each pool. If the copy number is equal to two, the Log R ratio is approximately zero. If the copy number is greater (a gain CNV status) or less (a loss CNV status) than two, the corresponding Log R ratio is higher or lower than zero, respectively. Figure 1 The flow chart of the criteria for copy number variation (CNV) analysis. The estimated CNV regions for each pool were then identified using PennCNV. We set a series of criteria to obtain informative CNV regions. First, regions with less than 20 markers were filtered out to avoid false-positive results in short regions with PennCNV analysis. A large number of gain CNV regions were predicted from 16 pools. To further reduce the likelihood of obtaining false positive findings in the gain CNV regions, we applied other criteria for the gain CNV regions by Log R ratio to increase data quality; (1) If the mean Log R ratio within the identified CNV region is less than 0.02, we filtered out this region as the intensity around zero indicating a high potential to be a normal CNV status (copy number = 2); (2) If the standard deviation of Log R ratio within the identified CNV region is greater than 0.2, we filtered out this region. The second procedure is also suggested by PennCNV for the quality control of individual samples [9]. Because PennCNV tends to split large CNV regions into multiple smaller regions, a merge procedure for adjacent CNV calls was applied in the next step [9]. We performed a gap cleaning procedure to merge neighboring CNVs where the ratio of the gap length and sum of the two neighboring CNV lengths is less than 0.2. At this step, we had obtained around three thousand estimated CNV regions with length ranging between 1.49 kb and 1021.08 kb in the 16 pools. To make comparisons of CNV results possible across pools, we took the union of each defined CNV region for all the pools. In total, there were 2243 unique CNV regions in case pools and 2426 unique CNV regions in control pools. In addition, different CNV calling algorithm could result in different calling results; therefore, we also used QuantiSNP [10] to analyze potential CNV regions identified by PennCNV. For small sample size, several strategies were implemented to conduct association testing. In the present study, we first constructed the Han Chinese CNV map. We used the published CHB (Han Chinese in Beijing) CNV regions in HapMap and 2 CNV databases from Lin et al. [22] and Lou et al. [23]. We selected more informative CNVs according to the different CNV patterns between case and control pools. The informative CNVs were defined based on the following criteria. (1) the CNV regions were only found in case pools but not in the Han Chinese CNV map; these CNVs were defined as important regions in cases; (2) the CNV regions were only found in control pools and also reported in the Han Chinese CNV map; these CNVs were defined as important regions in controls; (3) the CNV regions were shown in both case and control pools, but the frequency difference in the two groups is large (>3); these CNVs were defined as enriched in cases or controls; (4) the CNV regions were found in both case and control pools, however the CNV status (gain/loss) was different in the two groups. These selected CNV regions were potential targets for BPD and were included in the following analyses. 2.3. Association Testing for CNV Regions with BPD We first conducted CNV burden analysis between case and control pools, which is employed in previous studies [24]. We conducted burden analyses stratified by CNV types (gain or loss) and size (length ≥100 kb or ≥500 kb). Secondly, using a data integration framework, we had previously built a candidate gene database for BPD, and obtained 164 prioritized susceptible loci, namely BPDgenes [25]. We mapped genes for the potential CNV regions and compared them to the BPDgenes. For the mapped genes in the CNV regions that overlapped with BPDgenes, we tested the signal differences between case and control pools. We calculated the median signal of the markers within the defined CNV regions. Both t-test and Wilcoxon test were used to evaluate signal associations between cases and controls. Thirdly, a functional enrichment analysis for the mapped genes in selected CNV regions was performed using WEB-based GEne SeT AnaLysis Toolkit (WebGestalt) [26]. We adopted multiple testing corrections using the Benjamini-Hochberg method for each analyzed pathway with the significance level p < 0.05. Finally, we searched the literature to identify previously reported CNV regions in BPD. For associated CNV regions that were found in more than one patient in previous studies, we compared them with our potential CNV regions. These results are summarized in Appendix Table A1. 2.4. CNV Validation by RT-qPCR Due to the genome-wide scale of the CNV identification, we were only able to conduct experimental validation for a few CNV regions for the proof of principle of our filtering procedures and data analysis results. Two CNV regions were selected for validation using real-time quantitative polymerase chain reaction (RT-qPCR). The first region was the 6q27 CNV (Results Section) that replicated results from two previous studies, the second was a CNV region on chromosome 3p14.2 (Results Section) showing signal differences between CNV carriers and non-carriers, including a BPD candidate gene PTPRG. We performed RT-qPCR using Taqman Copy number assays (Chr.3 Hs04761773_cn and Chr. 6 Hs03602538_cn) and Taqman Copy number reference assays (Applied Biosystems, Foster City, CA, USA). Individuals in the carrier case-pool and non-carrier control-pool (25 cases and 25 controls) were tested in each assay, and the RT-qPCR was carried out in triplicate. Sequence Detection Software (SDS) was used for exporting the threshold cycle (Ct) data and further analyzing differences in Ct values (ΔCt) between the test locus and the control locus. Copy number variation was analyzed with the CopyCaller software [27]. We used Student’s t-test to compare raw copy number signals calculated from ∆Ct values to determine the statistical significance of predicted copy-number differences in cases and controls. The significant threshold was defined by p < 0.05. 3. Results Results of the CNV burden analysis for the potential CNV regions are displayed in Table 1. There was a relatively higher proportion of loss CNV regions (≥100 kb) in BPD patients than in controls, though the difference did not reach statistical significance (Wilcoxon p-value = 0.105). There was no significant difference in other types of the CNVs between BPD cases and controls. Table 2 shows the results of gene mapping in the selected CNV regions. There were 882 CNV regions that were only found in case pools and not mapped to the Han Chinese CNV map (Important regions in cases). In contrast, 94 CNV regions were only found in control pools and were mapped to the Han Chinese CNV map (Important regions in controls). In addition, two CNV regions were enriched in cases for which the frequency of case pools having this CNV is three more than that in control pools, and 26 CNV regions were enriched in controls. Only one CNV region had a different CNV status, with loss status in cases and gain status in controls. In total, 1247 genes were mapped to these selected CNV regions, and 30 of them were overlapped with the prioritized genes in the BPDgenes [25]. We compared the CNV signal differences between cases and controls and focus on the regions that had affected genes mapped to the BPDgenes (i.e., 30 CNV regions). We presented the CNV regions that exhibited signal differences between cases and controls with p-value <0.2 using t-test or Wilcoxon test, with these CNVs regarded as the top priority regions for BPD in our samples (see Table 3). Two of the CNV regions showed a significant difference (at p < 0.05) using either a t-test or Wilcoxon test. This included CNV regions on chromosomes 11 and 22. Functional enrichment analysis was performed for the 1247 CNV genes to explore their biological information. These mapped genes were significantly enriched (adjusted p-value less than 0.05) in 21 GO terms (see Table 4). In addition, the enrichment analysis was conducted using the database of disease associated genes in WebGestalt. Two disease categories (bipolar disorder and mood disorder) reached statistical significance (p < 0.05) (Appendix Table A2). microarrays-03-00052-t001_Table 1Table 1 CNV burden analysis in BPD case pools and control pools. CNV size CNV Sample No. of Mean CNVs Wilcoxon type Group unique CNVs per pool p-value ≥100 kb Both Controls 1441 346.25 0.645 Patients 1446 307.25 ≥100 kb Gain Controls 1438 345.5 0.645 Patients 1434 304.375 ≥100 kb Loss Controls 3 0.75 0.105 Patients 12 2.875 ≥500 kb Both Controls 43 16 0.798 Patients 44 15.5 ≥500 kb Gain Controls 43 16 0.798 Patients 44 15.5 ≥500 kb Loss Controls 0 0 NA Patients 0 0 Abbreviation: CNV, copy number variation; BPD, Bipolar Disorder. Lastly, we compared our CNV results to findings in the previous studies of BPD [7,24,28,29,30,31,32,33]. We found that five of the 1005 selected CNV regions (6q16.3, 6q27, 9q34.3, and two regions on 19p12) were also identified in previous studies (see Table 5). All of the overlapped regions were only found in cases. Four were gain CNV regions, and one region on 19p12 was loss CNV status. Priebe et al. found a CNV region at 6q27 that was overrepresented in bipolar patients with age at onset ≤21 [28]. This region was also reported to be enriched in affected members of a three-generation Amish pedigree of European descent [29]. Two CNV regions on chromosome 19 were both located at 19p12. Grozeva et al. found that these CNV regions were associated with BPD in Wellcome Trust Case Control Consortium samples [30]. In addition, Bergen et al. reported a duplication CNV region (9q34.3) in 14 patients with BPD, 17 patients with schizophrenia, and 11 normal controls [31]. Association testing of this region was significant when combining the two patient groups. Moreover, McQuillin et al. reported CNV regions at 6q16.3 in 2 BPD cases in British samples [32]. The rest of the 1000 CNV regions were not reported in more than one individual in previous studies. microarrays-03-00052-t002_Table 2Table 2 Potential CNV regions related to bipolar disorder and the information of mapping genes. Potential CNV regions No. of CNV (Gain/Loss) Mean CNV length (kb) No. of mapped Genes in CNV regions a Genes overlapped with the list in BPDgenes b Important regions in cases 882 (859/23) 120.52 982 ANK3, ARNTL, ASTN2, CHST11, CSMD2, DACH, DLG2, DPP10, DSCAM, GRIK1, HTR6, KALRN, MCTP1, MYO3B, NALCN, NOS1, OPCML, OR6S1, PARK2, PDLIM5, PLCB1, PTPRG, SLC39A3, SYN3, TGFB2, UGT1A10, VAV3 Important regions in controls 94 (94/0) 91.36 164 DMGDH Regions enriched in cases 2 (2/0) 447.74 0 None Regions enriched in controls 26 (25/1) 244.49 86 CSMD2, OPCML Different CNV status in cases and controls (1 Gain CNV in 1 control /1 Loss CNV in 1 case) 253.96 15 None a Genes that are partial or fully included in the CNV regions; b 164 prioritized loci for BPD in the BPDgenes [25]. microarrays-03-00052-t003_Table 3Table 3 Signal differences between case and control pools for identified CNV regions. Chr Position a CNV type b Length (kb) Affected Genes c p-value (t test) p-value (Wilcoxon test) 1 34,268,681–34,936,979 Gain in 6 controls and 2 cases 668.30 CSMD2, C1orf94 0.192 0.169 3 61,681,785–61,928,141 Gain in 1 case 246.36 PTPRG 0.171 0.234 4 95,487,295–95,868,284 Gain in 3 case 380.99 PDLIM5, ENH, ENH1, LIM 0.646 0.161 9 118,292,450–118,450,577 Gain in 1 case 158.13 ASTN2, KIAA0634, bA67K19.1 0.157 0.169 11 13,224,130–13,256,233 Gain in 1 case 32.10 ARNTL, BMAL1, BMAL1c, JAP3, MGC47515, MOP3, PASD3, TIC, bHLHe5 0.007 * 0.010 * 12 103,611,282–103,669,104 Gain in 1 case 57.82 CHST11, C4ST, C4ST-1, C4ST1, DKFZp667A035, FLJ41682, HSA269537 0.113 0.065 22 31,480,536–31,564,931 Gain in 1 case 84.40 SYN3, TIMP3 0.029 * 0.021 * Abbreviation: Chr, Chromosome; a Position were assembled by NCBI build 36 (UCSC hg 18); b Gain or loss CNV type in case or control pools; c The bold labels the affected Genes in the CNV regions that are mapped to BPDgenes [25]; * p-value is smaller than 0.05. In experimental validation of CNV at chromosome 3p14.2, two BPD individuals and one control subject out of 25 cases and 25 controls were found to have gain CNV at PRPRG region. The signal intensity of copy number value in the gain CNV that compares with a CNV status of 2 was 2.52 ± 0.002 (mean ± SD) versus 1.98 ± 0.033 (mean ± SD). The difference reached statistical significance (p < 0.05) using t test. Thus, confirmatory RT-qPCR experiments lent further support for this CNV validation. For the chromosome 6q27 region, this region was overlapped with CNV findings reported in previous BPD studies but did not show signal difference between our cases and controls. Experimental results showed that no individual (out of 25 cases and 25 controls) was validated to have gain CNV. The signal intensity of copy number value in cases and controls was 1.99 ± 0.03 and 1.92 ± 0.05, respectively. microarrays-03-00052-t004_Table 4Table 4 Gene Set Enrichment Analysis of genes mapped to potential CNV regions. Enriched GO category Database ID p-value a Adjusted p-value b O c N d Biological process biological adhesion GO:0022610 1.16 × 10−8 1.5 × 10−5 93 905 cell adhesion GO:0007155 2.13 × 10−8 1.5 × 10−5 92 903 cell-cell adhesion GO:0016337 4.03 × 10−5 0.0190 41 374 Cellular component neuron projection GO:0043005 1.03 × 10−7 2.2 × 10−5 69 628 synapse GO:0045202 1.13 × 10−5 0.0008 51 478 cell projection GO:0042995 9.19 × 10−6 0.0008 102 1173 axon GO:0030424 1.85 × 10−5 0.0010 34 276 dendrite GO:0030425 2.61 × 10−5 0.0011 39 341 cell projection part GO:0044463 4.52 × 10−5 0.0016 59 610 synaptic membrane GO:0097060 0.0001 0.0031 26 208 synapse part GO:0044456 0.0002 0.0053 38 361 neuron spine GO:0044309 0.0004 0.0086 20 153 dendritic spine GO:0043197 0.0004 0.0086 20 153 cell periphery GO:0071944 0.0010 0.0195 267 3989 keratin filament GO:0045095 0.0013 0.0227 10 57 plasma membrane GO:0005886 0.0017 0.0227 260 3905 postsynaptic density GO:0014069 0.0016 0.0227 15 111 cytoskeleton GO:0005856 0.0017 0.0227 119 1613 dendritic spine head GO:0044327 0.0016 0.0227 15 111 postsynaptic membrane GO:0045211 0.0028 0.0333 20 178 presynaptic membrane GO:0042734 0.0028 0.0333 9 53 a p-value was derived from Fisher’s exact test; b p-value was adjusted by Benjamini-Hochberg method; c O: number of genes in the gene set and also in the GO category; d N: number of reference genes in the pathway category. microarrays-03-00052-t005_Table 5Table 5 Comparison of our CNV results of bipolar disorder with findings in previous studies. Present Study Previous Studies Location Position a CNV type Length (kb) Affected Genes No. of cases/controls Position a CNV type Length (kb) References 6q16.3 101966969:102040222 Gain 73.250 GRIK2 1/0 101953625:102624651 Unknown 671.027 [32] 6q27 168320777:168376820 Gain 56.044 KIF25, FERM, 1/0 168090000:168330000 Gain 240.000 [28,29] MILT4, DACT2 9q34.3 138149942:138217164 Gain 67.233 None 2/0 136600001:140273252 Gain 3673.252 [31] 19p12 20091264:2029165 Gain 200.402 ZNF682, ZNF90, 1/0 20001614:20177979 Gain/Loss 176.365 [30] ZNF486 19p12 24193894:24282139 Loss 88.246 ZNF254 2/0 24013968-24295825 Gain 281.857 [30] a Position were assembled by NCBI build 36 (UCSC hg 18). 4. Discussion The current study applied a series of filtering and data analysis procedures to identify CNV regions that are related to bipolar disorder using a DNA pooling strategy. Several filtering methods have previously been developed for array data to increase detection power, in particular for microarray gene expression data. A screening threshold is usually set based on the variance of expression signals, where probes with low variance are excluded as non-informative markers [34,35]. The filtering procedures become even more important while adopting a DNA pooling strategy. In a genome-wide association study using pooled DNA, SNP quality control filters are set based on the indicators calculated from pooled intensity [36]. Similar concepts are adopted in our filtering scheme for pooling CNV data. The criteria we set for a CNV analysis with small sample size could assist for CNV identification by reducing the potential impact of experimental noise to explore the relationships between CNVs and the trait of interest. A number of potential CNV regions are reported for BPD in our Han Chinese samples that may warrant further investigation. Higher CNV burden is often observed in patients with psychiatric disorders when compared with healthy controls [37]. However, the reported specific CNV regions, even in large-scale Caucasian samples, are rarely replicated [24,28,30,32]. In the present study, we found that the burden of loss status CNVs in BPD cases is higher than in controls, though the comparisons did not reach statistical significance. Similar findings are reported in Zhang et al., which conducted a genome-wide CNV study of BPD in European Americans. They found that the number of singleton deletion CNVs in BPD cases is significantly higher than those in controls (p = 0.007) [38]. Other studies reported fewer CNVs with loss status in BPD cases than controls. For instance, in a young adult British sample, McQuillin et al. found that BPD subjects have significantly fewer deletion CNVs, with the size ranging from 200–500 kb compared to controls (p = 0.039), while fewer singleton duplication CNVs with the size over 100 kb are also found in BPD cases (p = 0.03) [32]. In addition, large (≥500 kb) inherited duplication CNVs are also found to be enriched in familial BPD cases (p = 0.03) [24]. The distinct findings of excessive deletion or duplication CNVs among BPD patients in the previous studies may result from the differences in sample populations, clinical characteristics of BPD cases, the CNV detection platforms, and CNV analysis criteria. Some studies advocate to subgroup BPD patients to obtain genetically more homogeneous groups. Age at onset is an often considered feature. Two CNV studies divided BPD cases into early or late onset subgroups by the age of onset (AO) of BPD diagnosis [24,28]. Comparing with healthy controls, one study reported that the rate of de novo CNVs is significantly higher in the patients group with AO ≤ 18 [24], whilst the other reported a higher frequency of microduplication CNVs in patients with AO ≤ 21 [28]. Two regions with duplication CNVs—the 6q27 and 10q11 CNV regions—are especially noted for early onset BPD [28]. To stratify BPD patients into subgroups based on relevant clinical characteristics could be considered in future CNV studies to reduce heterogeneity among patients. Through a series of data analysis procedures and a comprehensive literature search, we identified several CNV regions in relation to BPD. Some of the regions are reported in previous CNV studies, and some are novel regions. Novel CNV regions may be ethnic group specific and provide additional clues for exploring the pathogenesis of BPD in Han Chinese population. At the first stage of data screening, we found approximately 1000 novel CNV regions. Using a gene prioritization framework, we had previously built a gene database for BPD. The top list in the BPDgenes has a higher combined score, and thus higher confidence to be associated with bipolar illness. There are 30 genes in our identified CNV regions that are mapped to the BPDgenes (see Table 2), and the CNV regions that encompass these genes are considered high priority for further association testing. We reported signal differences between cases and controls for CNV regions on several chromosomes (see Table 3). These CNV regions have a higher potential to be related with BPD, and genes mapped to these regions are candidate genes for BPD. For instance, ARNTL is a circadian gene and has been found to be associated with BPD in Caucasian samples [39,40]. Gene PTPRG has previously been found to be associated with schizoaffective disorder [41]. We conducted RT-qPCR experimental validation for the PTPRG gene region, and the gain CNV status was validated, for which the signal intensity was higher in BPD cases than in controls. Research on the functional properties of these affected genes in potential CNV regions and how they link to the etiology of BPD may help point to a direction for the development of a new drug target. In addition to novel regions, there are five CNV regions (located at 6q16.3, 6q27, 9q34.3, 19p12) overlapped with the results from previous studies (see Table 5). One CNV region, 6q16.3, is reported to be associated with BPD in both ours and the study conducted by McQuillin et al. [32]. Gene GRIK2 is mapped to this CNV region. This gene is essential for brain development [42]. Previously, polymorphisms in GRIK2 gene have been reported to exhibit associations with obsessive-compulsive disorder [43] and autism spectrum disorders [44]. A loss CNV was found in our cases in 19p12 while a gain CNV in the same region was reported in Grozeva et al. [30]. Olsen et al. conducted a meta-analysis for three CNV regions—6q27 and 19p12 (two CNVs)—that are overrepresented in patients with affective disorder in three case-control studies [45]. However, the association testing is not significant for the three CNVs. As very few individuals possess either of these CNVs, a reliable test is not easy to perform for testing CNV associations with disease outcomes. Further replicated studies with larger sample size are needed to verify the relationship between candidate CNVs and BPD. The heterogeneity of genetic architecture across populations often leads to diverse genetic findings on the phenotypic outcomes of interest [46,47]. The diversity of CNVs in different ethnic groups has also been noted previously [48,49]. Among CNV studies in BPD, to our best knowledge, we are the first to conduct a genome-wide level of CNV analysis in an Asian population. For identified potential CNV regions for BPD, we also compared our results with findings from previous studies in different samples. The CNV regions that reported consistently in ours and previous studies may represent common risk regions across populations for BPD. If validated by experiments, novel CNV regions that were only reported in our study may indicate population specific genetic components for BPD, such as the CNV region on chromosome 3p14.2. By conducting functional analysis using GO terms for our mapped CNV genes, we found that the top three enriched pathways were involved in biological adhesion, cell adhesion, and neuron projection (Table 4). Several other genetic association studies, but not CNV studies, have also performed pathway analysis for BPD related candidate genes. The top significant GO pathways reported in Chang et al., are amine binding, synapse transmission, and transmission of nerve impulse [50]. Another study applied pathway analysis while incorporating information of allele-specific gene methylation [51]. They reported enriched pathways for extracellular matrix in brain, gated ion channel, and neurotransmitter receptor related pathways. Their findings support the involvement of biological functions of cell adhesion and neuronal transmission underlying bipolar illness. Further studies to investigate the interaction and networks among identified molecules for BPD could be conducted to understand the pathophysiology of BPD. There are several limitations in the present study. First, DNA pooling strategy is restricted to the original study design (i.e., for our study, bipolar disorder vs. control) and not flexible for conducting secondary data analysis. If there is a belief of true genetic heterogeneity in disease subtypes or the genetic factors are influenced by other covariates, it is not possible to adjust results for these concerns. Second, employing a series of data filtering steps may cause false-negative findings for certain CNV regions. In addition, because there is no consensus of a standard method for CNV calling, we used a second calling algorithm for our identified CNV regions. In the 12 reported CNV regions (listed Table 3 and Table 5), only the loss CNV region at chr19: 24193894:24282139 were consistently called by both calling algorithms. It is consistent with prior study showing that both algorithms have high reproducibility rates in loss CNV regions, but low rates in gain CNV regions [52]. In future study, applying multiple CNV calling algorithms and conducting experimental validation are desired. Third, due to having a small sample size we could only identify relatively common CNVs. Very rare or de novo CNVs are likely to be ignored. Nevertheless, we conducted power calculation [53] using median Log R ratio within a CNV region. We took one CNV at chr11: 13,224,130–13,256,233 as an example (see Table 3). The power for detection of signal difference between cases and controls can reach 0.87 in the current study. Follow-up individual studies with larger sample sizes should be designed to validate and test associations for identified CNV regions. Fourth, CNV regions that do not have any mapped genes (i.e., in gene desert regions) are not reported as the potential roles of these CNVs are not clear. Lastly, the experimental quality to detect CNV signals is a concern for pooling based design as there are no easy indicators to estimate the accuracy of CNV signal intensities. Other than two selected CNVs for experimental validation, we do not conduct a genome-wide level of validation and independent replication studies for our identified CNV regions. Further large-scale individual and replication studies are needed to investigate the roles of these CNVs and eventually provide clues for the underlying mechanisms of bipolar illness. 5. Conclusions There are many difficulties faced in performing CNV studies as most of the disease associated CNVs are complex, rare and usually with marginal effect size. The heterogeneity of bipolar disorder brings another challenge in explaining the CNV results. Proper data filtering and analysis strategies are recommended in exploring the relationships between CNVs and the trait of interest. We conducted the first pilot study of CNV association with BPD in Han Chinese population and identified several potential CNV regions for BPD. It is important to design further validation experiments and perform basic research for these CNVs to reveal their biological roles and explain their involvement in bipolar illness. Acknowledgments This work was supported by National Science Council (NSC 97-2314-B-002-184-MY2 and NSC 99-2314-B-002-140-MY3) and National Health Research Institute (EX102-9918NC) to P-H Kuo. We thank H.M. Liao, T.P. Lu, and C.K Hsiao who provided informative discussion for the management of CNV data and results interpretation. We also thank S.J. Lin who helped to conduct experimental validation and A. Woolston who assisted for English editing. We especially thank all participants who agreed to join this study. Authors Contributions Yi-Hsuan Chen and Po-Hsiu Kuo conceived and designed the experiments. Ru-Band Lu assisted for data collection. Yi-Hsuan Chen carried out the data analysis. Hung Hung and Po-Hsiu Kuo assisted with interpretation of the data. Yi-Hsuan Chen and Po-Hsiu Kuo drafted the manuscript. All authors read and approved the final manuscript. Conflicts of Interest The authors declare no conflict of interest. Appendices microarrays-03-00052-t006_Appendix Table A1Appendix Table A1 Reported associated CNV regions in previous BPD studies. No. Chromosome Start Position End Position CNV type Length (kb) Gene References 1 * 6q27 168,090,000 168,330,000 duplication 240 KIF25, FERM, MILT4, DACT2 [28,29] 2 * 6q16.3 101,953,625 102,624,651 unknown 671.027 GRIK2 [32] 3 * 9q34.3 136,600,001 14,0273,252 duplication 3,673.252 None [31] 4 * 19p12 20,001,614 20,177,979 both 176.366 ZNF682, ZNF90, ZNF486 [30] 5 * 19p12 24,013,968 24,295,825 duplication 281.858 ZNF254 [30] 6 1 28,399,376 28,842,172 unknown 442.797 DNAJC8, ATPIF1, SESN2, MED18, SNHG3-RCC1, RCC1, TRSPAP1, RAB42, TAF12PHACTR4 [32] 7 1 47,415,160 47,600,013 duplication 184.854 PDZK1IP1; TAL1; STIL; CMPK1 [24] 8 1 144,439,082 144,791,590 unknown 352.509 PDZK1, GPR89A, GPR89C, NBPF11, LOC728912, FAM108A3 [32] 9 1q21.1 142,400,001 148,000,000 both 5,600 None [31] 10 1q25.1 173,769,777 173,978,862 duplication 209.086 TNR [30] 11 1 232,723,219 232,828,069 unknown 104.851 IRF2BP2 [31] 12 2 196,772,221 197,165,580 unknown 393.36 HECW2 [32] 13 3 8,896,559 8,980,146 unknown 83.588 RAD18 [32] 14 3p14 65,649,762 65,848,146 deletion 198.385 MAGI1 [33] 15 3p26 2,124,587 2,955,648 duplication 831.062 CNTN4 [24] 16 3q 120,920,000 121,100,000 deletion 180.001 GSK3beta [7] 17 4q34.3 180,892,619 180,921,485 unknown 28.867 None [31] 18 5 180,098,728 180,099,664 unknown 0.937 OR2Y1 [32] 19 6 56,430,743 56,816,422 unknown 385.68 DST [32] 20 6 57,290,380 57,621,335 unknown 330.956 PRIM2 [32] 21 6 157,140,777 157,572,094 unknown 431.318 ARID1B [32] 22 7 34,935,017 35,044,178 unknown 109.162 DPY19L1 [32] 23 7 75,975,221 76,052,734 unknown 77.514 UPK3B [32] 24 7 88,226,688 89,777,622 unknown 1,550.94 ZNF804B, MGC26647, STEAP1, STEAP2, FLJ21062 [32] 25 7 132,588,362 133,401,053 unknown 812.692 EXOC4 [32] 26 8 13,236,908 13,304,907 unknown 68 DLC1 [31] 27 9 111,037 169,075 unknown 58.039 CBWD1 [32] 28 9 71,289,871 71,308,782 duplication 18.912 None [29] 29 9 134,871,014 134,890,520 unknown 19.507 GTF3C5, GFI1B [31] 30 9q31.1 104,826,097 104,885,068 both 58.972 None [30] 31 10 8,108,359 8,192,845 duplication 84.487 GATA3 [24] 32 10q11 47,010,000 47,170,000 duplication 160 ANTXRL [28] 33 10 50,334,496 50,490,772 unknown 156.277 ERCC6, PGBD3, CHAT, SLC18A3 [32] 34 10 51,497,689 52,053,743 unknown 556.055 FAM21A, FAM21B, ASAH2, SGMS1 [32] 35 12 7,884,583 8,017,012 duplication 132.43 SCL2A3M, SLC2A14 [29] 36 12p11.21 31,202,250 31,301,551 duplication 99.302 OVOS2 [30] 37 12 107,243,140 107,266,950 unknown 23.811 CMKLR1 [31] 38 13 49,932,650 49,982,221 deletion 49.572 AJ412031; AJ412041 [24] 39 13 90,848,887 92,317,488 unknown 1,468.60 GPC5 [32] 40 14 24,044,551 24,047,311 unknown 2.761 CMA1 [32] 41 15q.2 21,905,523 22,023,095 deletion 117.573 None [29] 42 15q13.2 28,000,001 29,000,000 both 1,000 CHRFAM7A, MRMR15 [31] 43 16p13.11 14,700,001 16,700,000 duplication 2,000 None [31] 44 16 15,435,825 15,889,948 unknown 454.124 C16orf45, KIAA0430, NDE1, MYH11, C16orf63 [32] 45 16 15,950,934 16,296,168 unknown 345.235 ABCC1, ABCC6, NOMO3, [32] 46 16 16,333,234 16351940 unknown 18.707 LOC339047 [32] 47 16 68,705,029 69,071,678 unknown 366.65 PDPR, MGC34761, EXOSC6, AARS, DDX19B, DDX19A, ST3GAL2, FUK [32] 48 17 36,465,156 36,477,177 deletion 12.022 KRTAP2-4; KRTAP2-4 [24] 49 17q25.1 68,400,001 72,200,000 duplication 3800 None [31] 50 18p11.21-11.1 14,694,694 15,092,421 duplication 397.728 ANKRD30B [30] 51 18 27,210,737 27,312,663 unknown 101.927 DSG4, DSG3 [32] 52 19 49,581,647 49,644,505 unknown 62.859 ZNF285A, ZNF229 [32] 53 19 58,644,961 58,689,358 unknown 44.398 ZNF761, ZNF813, ZNF765, ZNF331 [32] 54 21q11.2 13,200,001 15,300,000 both 2,100 ANKRD21, LOC441956, LIPI, RBM11 [31] 55 21 36,429,132 36,440,730 unknown 11.599 CBR3 [32] * No. 1–5 CNV regions were overlapped with our findings. microarrays-03-00052-t007_Appendix Table A2Appendix Table A2 Enriched gene set of the mapped genes in the potential CNV regions. Functional Category Genes on CNV p-value c Adjusted p-value d O a N b Bipolar Disorder (DB_ID:PA447199) PCDH17, ANK3, GABRR3, GABRG2, NOS1, CNTNAP2, ADCYAP1, PTPRG, NRXN1, PCLO, TACR1, TCF4, JMJD8, ADCY3, CSMD1, DPP10, CNTN5, RELN, NALCN, HTR5A, AGAP1, DFNB31, HTR4, GPC6, ATP8A2, GABRA1, CNTN6, ASTN2, FAT1, ADCY8, ARNTL, RPL14, PPP3CC, NRG1, MAGI1, PDLIM5, MMP16, HTR2A, CHRM2 2.40 × 10 −5 0.0252 39 286 Mood Disorders (DB_ID:PA447209) ANK3, GABRR3, GABRG2, NOS1, SST, CNTNAP2, ADCYAP1, PTPRG, PCLO, GRM5, TACR1, TCF4, HTR6, DPP10, CNR2, CNTN5, GPM6A, NALCN, RELN, HTR5A, GRIK1, DFNB31, AGAP1, CNTN6, GABRA1, ASTN2, NXPH1, FAT1, ARNTL, NRG1, PDLIM5, HTR2A, CHRM2 5.53 × 10 −5 0.0290 33 235 a O: number of genes in the gene set and also in the category; b N: number of reference genes in the category; c p-value was derived from Fisher’s exact test; d adjusted-p-value was corrected by Benjamini-Hochberg correction. ==== Refs References 1. Fagiolini A. Forgione R. Maccari M. Cuomo A. Morana B. Dell’Osso M.C. Pellegrini F. Rossi A. Prevalence, chronicity, burden and borders of bipolar disorder J. Affect. Disord. 2013 148 161 169 10.1016/j.jad.2013.02.001 23477848 2. Burmeister M. McInnis M.G. Zollner S. Psychiatric genetics: Progress amid controversy Nat. Rev. Genet. 2008 9 527 540 10.1038/nrg2381 18560438 3. Zhang F. Gu W. Hurles M.E. Lupski J.R. Copy number variation in human health, disease, and evolution Annu. Rev. Genomics Hum. Genet. 2009 10 451 481 10.1146/annurev.genom.9.081307.164217 19715442 4. Tam G.W. Redon R. Carter N.P. Grant S.G. The role of DNA copy number variation in schizophrenia Biol. Psychiatry 2009 66 1005 1012 10.1016/j.biopsych.2009.07.027 19748074 5. Glessner J.T. Wang K. Cai G. Korvatska O. Kim C.E. Wood S. Zhang H. Estes A. Brune C.W. Bradfield J.P. Autism genome-wide copy number variation reveals ubiquitin and neuronal genes Nature 2009 459 569 573 10.1038/nature07953 19404257 6. Sullivan P.F. Daly M.J. O’Donovan M. Genetic architectures of psychiatric disorders: The emerging picture and its implications Nat. Rev. Genet. 2012 13 537 551 10.1038/nrg3240 22777127 7. Lachman H.M. Pedrosa E. Petruolo O.A. Cockerham M. Papolos A. Novak T. Papolos D.F. Stopkova P. Increase in gsk3beta gene copy number variation in bipolar disorder Am. J. Med. Genet. B Neuropsychiatr Genet. 2007 144B 259 265 10.1002/ajmg.b.30498 17357145 8. Zöllner S. Teslovich T.M. Using gwas data to identify copy number variants contributing to common complex diseases Stat. Sci. 2009 24 530 546 10.1214/09-STS304 9. Wang K. Li M. Hadley D. Liu R. Glessner J. Grant S.F. Hakonarson H. Bucan M. Penncnv: An integrated hidden markov model designed for high-resolution copy number variation detection in whole-genome snp genotyping data Genome Res. 2007 17 1665 1674 17921354 10. Colella S. Yau C. Taylor J.M. Mirza G. Butler H. Clouston P. Bassett A.S. Seller A. Holmes C.C. Ragoussis J. Quantisnp: An objective bayes hidden-markov model to detect and accurately map copy number variation using SNP genotyping data Nucleic Acids Res. 2007 35 2013 2025 10.1093/nar/gkm076 17341461 11. Wineinger N.E. Tiwari H.K. The impact of errors in copy number variation detection algorithms on association results PLoS One 2012 7 e32396 10.1371/journal.pone.0032396 22523537 12. Conrad D.F. Pinto D. Redon R. Feuk L. Gokcumen O. Zhang Y. Aerts J. Andrews T.D. Barnes C. Campbell P. Origins and functional impact of copy number variation in the human genome Nature 2010 464 704 712 10.1038/nature08516 19812545 13. McCarroll S.A. Altshuler D.M. Copy-number variation and association studies of human disease Nat. Genet. 2007 39 S37 S42 10.1038/ng2080 17597780 14. Sham P. Bader J.S. Craig I. O’Donovan M. Owen M. DNA pooling: A tool for large-scale association studies Nat. Rev. Genet. 2002 3 862 871 10.1038/nrg930 12415316 15. Butcher L. Meaburn E. Liu L. Fernandes C. Hill L. Al-Chalabi A. Plomin R. Schalkwyk L. Craig I. Genotyping pooled DNA on microarrays: A systematic genome screen of thousands of snps in large samples to detect qtls for complex traits Behav. Genet. 2004 34 549 555 10.1023/B:BEGE.0000038493.26202.d3 15319578 16. Zuo Y. Zou G. Wang J. Zhao H. Liang H. Optimal two-stage design for case-control association analysis incorporating genotyping errors Ann. Hum. Genet. 2008 72 375 387 10.1111/j.1469-1809.2007.00419.x 18215207 17. Wang H. Thomas D.C. Pe’er I. Stram D.O. Optimal two-stage genotyping designs for genome-wide association scans Genet. Epidemiol. 2006 30 356 368 10.1002/gepi.20150 16607626 18. Lin C.H. Huang M.C. Li L.H. Wu J.Y. Chen Y.T. Fann C.S. Genome-wide copy number analysis using copy number inferring tool (CNIT) and DNA pooling Hum. Mutat. 2008 29 1055 1062 10.1002/humu.20760 18470944 19. Tsai H.C. Lu M.K. Yang Y.K. Huang M.C. Yeh T.L. Chen W.J. Lu R.B. Kuo P.H. Empirically derived subgroups of bipolar i patients with different comorbidity patterns of anxiety and substance use disorders in han chinese population J. Affect. Disord. 2012 136 81 89 10.1016/j.jad.2011.08.015 21906818 20. Wu P.J. Chang S.M. Lu M.K. Chen W.J. Yang Y.K. Yeh T.L. Liao S.C. Lu R.B. Kuo P.H. The profile and familiality of personality traits in mood disorder families J. Affect. Disord. 2012 138 367 374 10.1016/j.jad.2012.01.015 22331025 21. Kuo P.H. Chuang L.C. Liu J.R. Liu C.M. Huang M.C. Lin S.K. Sun H.F.S. Hsieh M.H. Hung H. Lu R.B. Identification of novel loci for bipolar I disorder in a multi-stage genome-wide association study Progr. Neuro. Psychopharmacol. Biologic. Psychiatr. 2014 in press 22. Lin C.H. Lin Y.C. Wu J.Y. Pan W.H. Chen Y.T. Fann C.S. A genome-wide survey of copy number variations in han chinese residing in taiwan Genomics 2009 94 241 246 10.1016/j.ygeno.2009.06.004 19559783 23. Lou H. Li S. Yang Y. Kang L. Zhang X. Jin W. Wu B. Jin L. Xu S. A map of copy number variations in chinese populations PLoS One 2011 6 e27341 10.1371/journal.pone.0027341 22087296 24. Malhotra D. McCarthy S. Michaelson J.J. Vacic V. Burdick K.E. Yoon S. Cichon S. Corvin A. Gary S. Gershon E.S. High frequencies of de novo CNVs in bipolar disorder and schizophrenia Neuron 2011 72 951 963 10.1016/j.neuron.2011.11.007 22196331 25. Kao C.F. Kuo P.H. Risk and information evaluation of prioritized genes for complex traits: Application to bipolar disorder Am. J. Med. Genet. B 2014 submitted 26. Wang J. Duncan D. Shi Z. Zhang B. Web-based gene set analysis toolkit (webgestalt): Update 2013 Nucleic Acids Res. 2013 41 W77 W83 10.1093/nar/gkt439 23703215 27. CopyCaller® Software v2.0 Available online: http://www.appliedbiosystems.com/absite/us/en/home/support/software/real-time-pcr/copycaller.html (accessed on 15 January 2014) 28. Priebe L. Degenhardt F.A. Herms S. Haenisch B. Mattheisen M. Nieratschker V. Weingarten M. Witt S. Breuer R. Paul T. Genome-wide survey implicates the influence of copy number variants (CNVs) in the development of early-onset bipolar disorder Mol. Psychiatry 2012 17 421 432 21358712 29. Yang S. Wang K. Gregory B. Berrettini W. Wang L.S. Hakonarson H. Bucan M. Genomic landscape of a three-generation pedigree segregating affective disorder PLoS One 2009 4 e4474 19214233 30. Grozeva D. Kirov G. Ivanov D. Jones I.R. Jones L. Green E.K. St Clair D.M. Young A.H. Ferrier N. Farmer A.E. Rare copy number variants: A point of rarity in genetic risk for bipolar disorder and schizophrenia Arch. Gen. Psychiatry 2010 67 318 327 10.1001/archgenpsychiatry.2010.25 20368508 31. Bergen S.E. O’Dushlaine C.T. Ripke S. Lee P.H. Ruderfer D.M. Akterin S. Moran J.L. Chambert K.D. Handsaker R.E. Backlund L. Genome-wide association study in a swedish population yields support for greater CNV and mhc involvement in schizophrenia compared with bipolar disorder Mol. Psychiatry 2012 17 880 886 10.1038/mp.2012.73 22688191 32. McQuillin A. Bass N. Anjorin A. Lawrence J. Kandaswamy R. Lydall G. Moran J. Sklar P. Purcell S. Gurling H. Analysis of genetic deletions and duplications in the university college london bipolar disorder case control sample Eur. J. Hum. Genet. 2011 19 588 592 10.1038/ejhg.2010.221 21206513 33. Karlsson R. Graae L. Lekman M. Wang D. Favis R. Axelsson T. Galter D. Belin A.C. Paddock S. Magi1 copy number variation in bipolar affective disorder and schizophrenia Biol. Psychiatry 2012 71 922 930 10.1016/j.biopsych.2012.01.020 22381734 34. Hackstadt A.J. Hess A.M. Filtering for increased power for microarray data analysis BMC Bioinformatics 2009 10 11 10.1186/1471-2105-10-11 19133141 35. Donner Y. Feng T. Benoist C. Koller D. Imputing gene expression from selectively reduced probe sets Nat. Methods 2012 9 1120 1125 10.1038/nmeth.2207 23064520 36. Chiang C.W. Gajdos Z.K. Korn J.M. Kuruvilla F.G. Butler J.L. Hackett R. Guiducci C. Nguyen T.T. Wilks R. Forrester T. Rapid assessment of genetic ancestry in populations of unknown origin by genome-wide genotyping of pooled samples PLoS Genet. 2010 6 e1000866 10.1371/journal.pgen.1000866 20221249 37. Malhotra D. Sebat J. CNVs: Harbingers of a rare variant revolution in psychiatric genetics Cell 2012 148 1223 1241 10.1016/j.cell.2012.02.039 22424231 38. Zhang D. Cheng L. Qian Y. Alliey-Rodriguez N. Kelsoe J.R. Greenwood T. Nievergelt C. Barrett T.B. McKinney R. Schork N. Singleton deletions throughout the genome increase risk of bipolar disorder Mol. Psychiatry 2009 14 376 380 10.1038/mp.2008.144 19114987 39. Mansour H.A. Talkowski M.E. Wood J. Chowdari K.V. McClain L. Prasad K. Montrose D. Fagiolini A. Friedman E.S. Allen M.H. Association study of 21 circadian genes with bipolar i disorder, schizoaffective disorder, and schizophrenia Bipolar Disord. 2009 11 701 710 10.1111/j.1399-5618.2009.00756.x 19839995 40. Shi J. Wittke-Thompson J.K. Badner J.A. Hattori E. Potash J.B. Willour V.L. McMahon F.J. Gershon E.S. Liu C. Clock genes may influence bipolar disorder susceptibility and dysfunctional circadian rhythm Am. J. Med. Genet. B Neuropsychiatr. Genet. 2008 147B 1047 1055 10.1002/ajmg.b.30714 18228528 41. Hamshere M.L. Green E.K. Jones I.R. Jones L. Moskvina V. Kirov G. Grozeva D. Nikolov I. Vukcevic D. Caesar S. Genetic utility of broadly defined bipolar schizoaffective disorder as a diagnostic concept Br. J. Psychiatry 2009 195 23 29 10.1192/bjp.bp.108.061424 19567891 42. Glessner J.T. Reilly M.P. Kim C.E. Takahashi N. Albano A. Hou C. Bradfield J.P. Zhang H. Sleiman P.M.A. Flory J.H. Strong synaptic transmission impact by copy number variations in schizophrenia Proc. Natl. Acad. Sci. USA 2010 107 10584 10589 10.1073/pnas.1000274107 20489179 43. Sampaio A.S. Fagerness J. Crane J. Leboyer M. Delorme R. Pauls D.L. Stewart S.E. Association between polymorphisms in grik2 gene and obsessive-compulsive disorder: A family-based study CNS Neurosci. Ther. 2011 17 141 147 10.1111/j.1755-5949.2009.00130.x 20370803 44. Kim S.A. Kim J.H. Park M. Cho I.H. Yoo H.J. Family-based association study between grik2 polymorphisms and autism spectrum disorders in the korean trios Neurosci. Res. 2007 58 332 335 10.1016/j.neures.2007.03.002 17428563 45. Olsen L. Hansen T. Djurovic S. Haastrup E. Albrecthsen A. Hoeffding L.K. Secher A. Gustafsson O. Jakobsen K.D. Nielsen F.C. Copy number variations in affective disorders and meta-analysis Psychiatr. Genet. 2011 21 319 322 10.1097/YPG.0b013e3283463deb 21451435 46. Ioannidis J.P. Ntzani E.E. Trikalinos T.A. “Racial” differences in genetic effects for complex diseases Nat. Genet. 2004 36 1312 1318 10.1038/ng1474 15543147 47. Greene C.S. Penrod N.M. Williams S.M. Moore J.H. Failure to replicate a genetic association may provide important clues about genetic architecture PLoS One 2009 4 e5639 10.1371/journal.pone.0005639 19503614 48. Freeman J.L. Perry G.H. Feuk L. Redon R. McCarroll S.A. Altshuler D.M. Aburatani H. Jones K.W. Tyler-Smith C. Hurles M.E. Copy number variation: New insights in genome diversity Genome Res. 2006 16 949 961 10.1101/gr.3677206 16809666 49. Armengol L. Villatoro S. Gonzalez J.R. Pantano L. Garcia-Aragones M. Rabionet R. Caceres M. Estivill X. Identification of copy number variants defining genomic differences among major human groups PLoS One 2009 4 e7230 10.1371/journal.pone.0007230 19789632 50. Chang S.H. Gao L. Li Z. Zhang W.N. Du Y. Wang J. Bdgene: A genetic database for bipolar disorder and its overlap with schizophrenia and major depressive disorder Biol. Psychiatry 2013 74 727 733 10.1016/j.biopsych.2013.04.016 23764453 51. Chuang L.-C. Kao C.-F. Shih W.-L. Kuo P.-H. Pathway analysis using information from allele-specific gene methylation in genome-wide association studies for bipolar disorder PLoS One 2013 8 e53092 23326387 52. Lin P. Hartz S.M. Wang J.C. Krueger R.F. Foroud T.M. Edenberg H.J. Nurnberger J.I. Jr. Brooks A.I. Tischfield J.A. Almasy L. Copy number variation accuracy in genome-wide association studies Hum. Hered. 2011 71 141 147 10.1159/000324683 21778733 53. Rosner B.A. Fundamentals of Biostatistics Cengage Learning Singapore 2011

PMC005xxxxxx/PMC5003456.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays3010072microarrays-03-00072ArticleCan Archival Tissue Reveal Answers to Modern Research Questions?: Computer-Aided Histological Assessment of Neuroblastoma Tumours Collected over 60 Years Chetcuti Albert 1Mackie Nicole 2Tafavogh Siamak 3Graf Nicole 2Henwood Tony 2Charlton Amanda 2Catchpoole Daniel 1*1 Tumour Bank, The Children’s Cancer Research Unit, Kid’s Research Institute, The Children’s Hospital at Westmead, Westmead, NSW 2145, Australia; E-Mail: a.chetcuti@hotmail.com2 Histopathology Department, The Children’s Hospital at Westmead, Westmead, NSW 2145, Australia; E-Mails: nmackie7@gmail.com (N.M.); nicole.graf@health.nsw.gov.au (N.G.); tony.henwood@health.nsw.gov.au (T.H.); amanda.charlton@health.nsw.gov.au (A.C.)3 Faculty of Engineering and Information Technology, The University of Technology Sydney, Sydney, NSW 2007, Australia; E-Mail: siamak.tafavogh@student.uts.edu.au* Author to whom correspondence should be addressed; E-Mail: daniel.catchpoole@health.nsw.gov.au; Tel.: +61-2-9845-1205; Fax: +61-2-9845-3078.28 2 2014 3 2014 3 1 72 88 20 1 2014 13 2 2014 24 2 2014 © 2014 by the authors; licensee MDPI, Basel, Switzerland.2014This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Despite neuroblastoma being the most common extracranial solid cancer in childhood, it is still a rare disease. Consequently, the unavailability of tissue for research limits the statistical power of studies. Pathology archives are possible sources of rare tissue, which, if proven to remain consistent over time, could prove useful to research of rare disease types. We applied immunohistochemistry to investigate whether long term storage caused any changes to antigens used diagnostically for neuroblastoma. We constructed and quantitatively assessed a tissue microarray containing neuroblastoma archival material dating between 1950 and 2007. A total of 119 neuroblastoma tissue cores were included spanning 6 decades. Fourteen antibodies were screened across the tissue microarray (TMA). These included seven positive neuroblastoma diagnosis markers (NB84, Chromogranin A, NSE, Ki-67, INI1, Neurofilament Protein, Synaptophysin), two anticipated to be negative (S100A, CD99), and five research antibodies (IL-7, IL-7R, JAK1, JAK3, STAT5). The staining of these antibodies was evaluated using Aperio ImageScope software along with novel pattern recognition and quantification algorithms. This analysis demonstrated that marker signal intensity did not decrease over time and that storage for 60 years had little effect on antigenicity. The construction and assessment of this neuroblastoma TMA has demonstrated the feasibility of using archival samples for research. tissue microarrayarchival tissueneuroblastomaimmunohistochemistryimage analysis ==== Body 1. Introduction Neuroblastoma is the most common extracranial solid tumour in childhood and the most common cancer found in children under the age of five [1,2]. It is considered that neuroblastoma arises when the differentiation of these migratory cells of the neural crest are blocked [3,4,5,6]. The differences in outcome for patients with neuroblastoma are striking. Infants younger than 1 year have a good prognosis, even in the presence of metastatic disease, whereas older patients with advanced stage neuroblastoma fare poorly, despite aggressive multimodality therapy [7,8]. Fewer than half of these patients are cured, even with the use of high-dose therapy followed by autologous bone marrow or stem cell rescue [8]. The molecular basis underlying the variability in tumour growth, clinical behaviour, and responsiveness to therapy remains largely unknown [9]. However, the review by Riley and co-authors [10] identify a range of informative molecular and biological markers that are now routinely used to direct the clinical management and treatment of neuroblastoma patients. When compared to adult cancers, neuroblastoma is a very rare disease with current figures in Australia indicating that approximately 35–40 new cases are diagnosed each year [11]. With these relatively low numbers, obtaining sufficient samples to conduct research is challenging in countries like Australia with a relatively low population. Translational research requires suitable collections of tissue samples to be made available for these investigations, yet whilst samples collected from multiple institutes to accumulate sufficient numbers may be feasible, quality assurance requirements many not be guaranteed across cohorts collected from institutions that have diverse biobanking practices. Archived formalin-fixed paraffin-embedded (FFPE) tissues stored in hospital histopathology departments represent valuable collections of biospecimens that allow modern research questions to be investigated in rare tumours. It is queried, however, whether biospecimens prepared and stored over many decades allow a range of tests for protein and DNA elements to be performed to a similar standard as those prepared using current protocols [12,13,14,15]. As the consistency of laboratory results obtained from samples stored over lengthy periods of time is one of the major concerns of biobanks, we compared the marker staining results from FFPE biopsies of neuroblastoma tumours stored between 1950 and 2007 within a single paediatric hospital. To enable the direct comparison of such a diverse collection of samples, a tissue microarray (TMA) containing all materials tested was constructed [16]. Array sections were immunohistochemically (IHC) evaluated for current positive and negative diagnostic markers as well as with novel research-specific antibodies. It was reasoned that protein antigenicity throughout the tissue section could be lost over time. In this context, our principal measure for staining quality was signal intensity, which, by using tissue microarrays, can be systematically and objectively compared across multiple samples treated under exactly the same experimental conditions. Computer-aided digital analysis has been applied to quantitatively determine signal intensity for the antibodies used in each tissue sample on the array [17,18,19,20]. We have demonstrated that despite long term storage of these neuroblastoma specimens, the tissues performed to an equivalent level, and thus open the potential for a large amount of archival material to be used for research purposes into rare disease. 2. Experimental Section 2.1. Neuroblastoma Cases and TMA Construction An extensive search of The Children’s Hospital at Westmead Histopathology department records was conducted, and established a total of 422 neuroblastoma cases treated between 1950 and 2007 for which archival paraffin embedded material was stored. Individual neuroblastoma cases with sufficient viable tumour tissue were assessed by clinical pathologists for inclusion in the tissue microarray. Tissue processing conditions, including fixative and fixation time, were sought from the hospital records, but these were incomplete. Forty nine chosen samples were de-identified to maintain patient privacy and meet local legal and ethical requirements. Neuroblastoma cases were separated in decades (1950–1959, 1960–1969, 1970–1979, 1980–1981, 1990–1999, and 2000–2007) and cases were selected for inclusion in the TMA if sufficient material to include 2–4 cores/case was possible. A total of 20 cores per decade were included in the TMA layout; 120 samples in total (Figure 1). Neuroblastoma tissue cores from the same decade were distributed evenly across the TMA in a 4 × (6 × 5 array) block pattern (Figure 1B). Figure 1 (A) Haematoxylin and eosin stained section of the TMA, containing 119 neuroblastoma tissue cores from 49 individuals and 38 control tissues. (B) Map of the TMA layout. A variety of normal and control tissues were included. Normal control and control tumour tissues were also included in the TMA. These included the following tissue types: liver, testis, pancreas, skin, kidney, placenta, lung, lymph node, muscle, tonsil, thymus, brain, adrenal, breast, spleen, prostate, rhabdomyosarcoma, and Ewing’s sarcoma. Each control tissue was included in duplicate. The TMA was constructed using a Manual Tissue Arrayer MTA-1 (Beecher Instruments, WI, USA) by following the standard instructions [21]. All cores included in the TMA were 1.0 mm in diameter. A donor paraffin block was prepared using paraplast wax (Leica Microsystems, Sydney, Australia) and cored prior to the addition of each recipient tissue core. Following the completion of the TMA, it was annealed using the following procedure: The TMA block was annealed via five cycles of heating at 60 °C for 10 min and cooling at room temperature for 20 min. Following each heating and cooling cycle, the TMA block surface was lightly pressed by hand with a clean glass slide. Serial sections from the TMA block were cut at a thickness of 4 µm using a standard microtome, mounted onto positively charged Superfrost glass microscope slides (Menzel-Glaser, Germany) and dried in at 65 °C. Every tenth section was stained with haematoxylin and eosin using a Vision Biosystems Autostainer (Leica Microsystems) to assess the quality of the cores throughout the TMA block. 2.2. Immunohistochemistry All immunohistochemical staining was performed using the Bond Max Vision BioSystems (Leica Microsystems) automated immunohistochemistry stainer. The following primary antibodies and dilutions were used: Neuroblastoma Marker (clone NB84a, Leica Microsystems) at 1/1000 dilution; Chromogranin A (clone LK2H10, Leica Microsystems) at 1/200 dilution; Neuron Specific Enolase (NSE) (clone 22C9, Leica Microsystems) at 1/100 dilution; Ki-67 (clone MIB-1, Dako, Melbourne, Australia) at 1/200 dilution; INI1 (clone BAF47, BD Biosciences, Sydney, Australia) at 1/500 dilution; Neurofilament Protein (clone 2F11, Dako) at 1/1600 dilution; S100 (clone Z0311, Dako) at 1/1000 dilution; CD99 (clone 12E7, Leica Microsystems) at 1/1200 dilution; Synaptophysin (clone Z66, Invitrogen, Melbourne, Australia) at 1/150 dilution; Interleukin-7 (clone H-151, Santa Cruz Biotechnology, Santa Cruz, CA, USA) at 1/50 dilution; Interleukin-7 receptor (clone H-215, Santa Cruz Biotechnology) at 1/50 dilution, Janus kinase 1 (clone H-106, Santa Cruz Biotechnology) at 1/50 dilution, Janus kinase 3 (clone B-12, Santa Cruz Biotechnology) at 1/50 dilution; and Signal transducer and activator of transcription 5 (clone N-20, Santa Cruz Biotechnology) at 1/50 dilution. The following antigen retrieval methods were used: For Chromagranin A, Ki-67, INI1, Neurofilament Protein, and CD99 antibodies, heat induced epitope retrieval was performed using the Bond Epitope Retrieval Solution 1 (AR9961, Leica Microsystems). For the Synaptophysin, IL-7, IL-7R, JAK1, JAK3, and STAT5 antibodies heat induced epitope retrieval was performed using the Bond Epitope Retrieval Solution 2 (AR9640, Leica Microsystems). For the NB84 antibody, a pronase enzyme retrieval method was used. No antigen retrieval was used for the NSE or S100 primary antibodies. Primary antibody binding was detected using the Bond Polymer Refine Detection Kit (Leica Microsystems). Slides were counterstained with hematoxylin, cover slipped and permanently mounted using Histomount (National Diagnostics, Adelaide, Australia). 2.3. Evaluation of TMA Immunohistochemistry 2.3.1. Digital Image Acquisition Stained TMA slides were scanned on an Aperio ScanScope CS virtual microscope (Aperio, CA, USA). Each stained tissue microarray slide was scanned at 400× absolute magnification to yield high resolution (0.25 μm2/pixel) images. 2.3.2. Computer-Aided Image Analysis: Cell Counting To determine whether each tissue disc contained equivalent numbers of tumour cells, we applied two imaging algorithms developed by and previously described us [19,20] for images of haematoxylin and eosin stained tissue array slides. Briefly, the first algorithm enhances the fluctuating intensity quality of the images by reducing the wide range of colour variation within the images. It partitions the image into several mosaics and normalises the colour of all the pixels within each mosaic, thereby reducing the sensitivity of the downstream analysis to noisy images and improving their performance when identifying and extracting the regions of interests. The second algorithm exploits the differences between the colours, morphological features, and geometrical properties to identify and extract histological elements, such as cellular regions, areas of stroma and neuropil, red blood cells, and background spaces. Cellular regions and areas of stroma/neuropil are presented as percentages of the total area covered by tissue spot. 2.3.3. Computer-Aided Image Analysis: Pixel Counting for Signal Intensity Optimised positive pixel counting algorithms (v9.1, Aperio, Vista, CA, USA) were used for digital analysis, and measured using ImageScope v10.2.2.2352 software (Aperio). An index was calculated to determine the overall signal intensity for all individual tissue spots for each IHC antibody stain based on previously reported approaches [22,23]. The IHC index is based on the addition of weighted proportions as follows: IHC Index = (n1/T) × 1 + (n2/T) × 2 + (n3/T) × 3 (1) where T is the total number of pixels per spot, n is the number of stained pixels per spot, and the subscripts 1, 2, 3 represent weak, moderate, and strong staining, respectively (Figure 2A). An IHC index of 0.0 indicates no signal and no protein expression, whilst an index of 3.0 indicates that all pixels are strongly positive and expression is high throughout the sample. This approach was modified to determine the distribution of antibody staining signal intensity across each neuroblastoma tissue core. The positive pixel counting algorithm was adjusted so that the ‘weak’ and ‘moderate’ intensity bins would capture only one tenth of a specific signal range. The algorithm was run iteratively with these bin regions shifted across the intensity spectrum until the staining intensity was partitioned into 10 equal subdivisions from negative staining (>230) to maximal positive staining (<30) (Figure 2B). The signal intensity for each subdivision was averaged across each decadal storage period to determine changes to the distribution of antigenicity of selected proteins over time. Figure 2 (A) Examples showing mark up areas for positive pixel counting with indicative IHC index. Colours represent the bin (negative–strong) with which a pixel is placed, depending on the intensity of light transmitted through the slide at that point. (B) To determine the distribution of staining intensity across the image, the algorithm was used four times, shifting the binning constraints of the positive pixel counting each time to segregate the pixel distribution into 10 subdivisions. A high value (>230) represents no antibody staining, whilst a low value (<30) represents maximal antibody staining. 2.4. Statistical Analysis For immunohistochemistry results, Pearson’s correlation test was used to compare TMA core staining across the collection time period. A p-value of <0.05 was considered statistically significant. 3. Results and Discussion 3.1. Neuroblastoma TMA Core Morphology In this study, it was important to demonstrate that all tissue cores on the array represented the equivalent proportion of cellular and stromal structures, which is commonly typical for neuroblastoma tumours. Hematoxylin and eosin stained TMA sections were visually assessed for general tissue morphology and histological quality (NG, AC, TH). All cores from each time period demonstrated intact cellular architecture and clarity of staining such that all histological structures of the neuroblastoma tumours could be determined (Figure 3). Despite the absence of full records of tissue processing conditions, no gross fixation or processing artefacts were noted. Figure 3 Selection of six neuroblastoma tumours, collected between 1950–2007. Staining with haematoxylin and eosin, and IHC using a range of antibodies demonstrated maintenance of general staining attributes over storage time. A sample from 1950 showed unique histology with half the sample representing differentiated ganglioneuroma morphology (right) and half undifferentiated neuroblastoma (left). Quantitative analysis of each core using our specialist pattern recognition algorithms demonstrate that the average percentage of each core represented by either cellular or stroma/neuropil is 38.6 ± 12.7% and 54.4 ± 13.3% respectively. Samples across each decade demonstrated equivalent distribution of cellular proportions, although it is clear that samples selected from the 2000’s were generally more cellular whilst those from the 1990s had more stromal elements represented than other decades (Figure 4). We consider this bias to be a random anomaly within the selection of samples included in the TMA and does not reflect a systematic change in sample processing or surgical practice. Figure 4 Proportion of cellular (left) and stromal/neuropil (right) structures within each sample of neuroblastoma tissue collected in each decade (1950–2000). 3.2. Immunohistochemistry Having demonstrated the histological equivalency of the tumour samples selected for inclusion on the tissue microarray, the specificity of antibody staining was considered. We compared the staining observed in the control tissue cores as well as the neuroblastoma tissue for each section. Cell specific staining was observed for all antibodies used (Figure 3). As shown in Figure 5, strong staining for the positive diagnosis marker NB84 was seen in old (1953) and modern (2003) neuroblastoma tissue. Conversely, no staining was observed for the negative marker CD99 in old or modern sections (Figure 5). Visually, samples collected in the 1950s demonstrated the least intense positive staining and higher background staining, whilst those from between 1960–1990 performed comparably well with moderate levels of positive staining and weak negative background staining. Generally, samples collected from 2000 onwards demonstrated the most distinct staining pattern with strong positive signals as well as no staining for negative markers. Despite these observed distinctions, all samples performed adequately using modern immunohistochemical methodologies, yielding measurable results. 3.3. Computer-Aided Image Analysis To identify trends in antigenicity for each protein versus tissue age, the total signal intensity (IHC index) of each TMA tissue core was plotted against collection date (Figure 6), with regression analysis used to explore any changes is antigenicity over time (Table 1). Furthermore, the average distribution of each antibody’s signal across tissue samples collected from each decade is shown in Figure 6. In general, the majority of proteins examined did not demonstrate changes in antigenicity over the 60 year collection period that could be related to a storage related decline in tissue quality. The results of the image analysis for each individual antibody stain are described below. Figure 5 Immunohistochemical staining with antibodies against NB84, NSE, S100A, CD99, IL-7, and JAK3 in neuroblastoma tissue collected in 1953 or 2003. Figure 6 Pearson’s correlation for each neuroblastoma tissue core versus collection date. Left column: Plot of the average positive pixel signal intensity bins per algorithm subdivision for each decade. Right Column: Scatter plots of the IHC index for each tissue core compared to collection date. 3.3.1. Neuroblastoma Marker NB84 Diagnostically, this antibody indicates the presence of neuroblastoma cells. The anti-neuroblastoma marker antibody NB84 stains an uncharacterised 57 kD molecular protein found in most normal human tissues, including epithelial and endothelial cells. This antibody staining was predominately cytoplasmic and medium to strong in all neuroblastoma cores [24]. In control tissue, strong NB84 staining was found in prostate epithelium, liver, kidney, placenta, lung, breast, spleen, and adrenal tissues. Negative NB84 staining was found in muscle, pancreas, skin, rhabdomyosarcoma, thymus, Ewing’s sarcoma, brain, lymph node, and tonsil (data not shown). A statistically significant trend (p < 0.001) in the IHC index over time suggested that total antigenicity for NB84 decreased with extended storage (Figure 6(right); Table 1). It must be noted, however, that the newer samples represented on the array were previously shown to be more cellular (Figure 4) and this result may be biased accordingly. Hence, to compensate for this potential bias, we worked with the assumption that all the staining signal for NB84 would be found in the cellular regions and corrected the IHC index for the percentage of the image represented by cells. This resulted in the fact that the trend for a decreasing signal with storage time disappeared (r = 0.186; Figure 7). The distribution of signal over each core was notably more pronounced towards the darker intensity for the most recent samples (2000’s) compared to all others (Figure 6(left)). This confirmed the results in Figure 5 that the most recent NB84 staining had more contrast, with pixels showing strong to maximal intensity indicative of specific tissue elements being heavily stained. Samples greater than 10 years old, however, demonstrated a more uniform distribution of staining intensity, albeit fainter and more diffuse. microarrays-03-00072-t001_Table 1Table 1 Pearson’s correlation for antibody antigenicity compared to collection date. Marker Linear Regression (R) p value Diagnostic NB84 0.432 <0.001 Chromogranin A −0.138 0.144 NSE 0.209 0.026 Ki-67 0.177 0.066 INI7 0.072 0.452 Neurofilament Protein −0.050 0.607 S100 −0.258 0.006 CD99 −0.200 0.039 Synaptophysin −0.109 0.265 Research IL-7 −0.033 0.724 IL-7R 0.037 0.698 JAK1 0.006 0.951 JAK3 0.065 0.501 STAT5 −0.018 0.855 Figure 7 IHC index for NB84 and NSE corrected for cellular content. 3.3.2. Neuron Specific Enolase The anti-Neuron Specific Enolase (NSE) antibody stains the 47 kD component of the gamma-gamma enolase isoenzyme, which is involved in glycolytic metabolism found in all neurons. This antibody staining was strong in all neuroblastoma cores and was predominately cytoplasmic in nature [25]. In control tissue cores, strong staining was found in brain, adrenal, kidney, rhabdomyosarcoma, spleen, and skin, but was negative in liver, thymus, tonsil, prostate, pancreas, Ewing’s sarcoma, breast, testis, placenta, lung, kidney, and muscle. Like the NB84 results, the IHC index used to quantitate NSE staining demonstrated a significant decreasing trend over time (r = 0.209, p = 0.026) (Figure 6(right); Table 1). Similarly, however, when corrected for cellular content of each sample, this decreasing trend was lost (r = 0.05) (Figure 7). The distribution of NSE intensity across each tissue spot was shifted towards greater intensity and contrast in the most recent samples (Figure 6(left)). 3.3.3. S100A The anti-S100 antibody stains the S100A protein and is a positive marker for Schwann cells of the peripheral nervous system and highlights Schwannian stroma in differentiated ganglioneuroblastoma (Figure 3; 1950’s) as well as sustentacular cells in rare olfactory neuroblastoma [26,27]. This antibody staining is low in undifferentiated ‘stroma-poor’ neuroblastoma tumours that were selected for inclusion in the tissue microarray. The IHC index confirmed low S100 expression in all samples, especially those collected between 1960 and 2005. Although the majority of neuroblastoma cores showed low level S100 staining, samples collected during the 1950s shown some weak to moderate positive staining (Figure 6(right)). 3.3.4. CD99 The anti-CD99 antibody stains a 32 kDa transmembrane glycoprotein, encoded by the MIC2 gene and is used as a diagnostic marker for Ewing’s sarcoma. This antibody, however, stains negative in neuroblastoma cores. In control tissue cores, strong membrane and medium cytoplasmic staining was found in Ewing’s sarcoma, prostate, and spleen tissues, as anticipated [28]. The IHC index demonstrated all neuroblastoma cores as negative for CD99 staining across all decade (Figure 6(right)). 3.3.5. Research Antibodies: Interleukin-7, Interleukin-7 Receptor, Stat5, JAK1, and JAK3 The interleukin-7 protein is a 20 kDa growth factor cytokine that is known to induce neuronal cell differentiation [29] and is purported to have a functional role in neuroblastoma [30]. Members of the IL7 signalling pathway, IL7-receptor, JAK1, JAK3, and Stat5 were also examined across this array. All antibodies showed measurable cellular staining with the IHC index demonstrating consistency in intensity across all neuroblastoma tissue collected over six decades (Table 1). This is exemplified by IL-7 (r = −0.033, p = 0.724) and JAK3 (r = 0.065, p = 0.501) (Figure 6(right); Table 1). The intensity distribution across each tissue spot for all antibodies was shifted towards greater intensity and contrast in the most recent samples, although this difference was only modest (Figure 6(left)). 3.3.6. Chromogranin A, Ki-67, INI1, Neurofilament Protein and Synaptophysin. The following additional diagnostic antibodies were investigated: Anti-chromogranin A antibody stains a 68 kD acidic protein, which is widely expressed in neural tissues and neuroendocrine tumours [31,32]. The anti-Ki-67 antibody stains Ki-67, which is a nuclear protein that is preferentially expressed during all active phases of the cell cycle [31]. The anti-Integrase Interactor 1 (INI1) antibody stains the BAF47 component of the SWI/SNF5 complex. SWI/SNF complexes facilitate gene activation and transcription factor binding by altering repressive chromatin structures in an ATP-dependent manner [33]. The anti-Neurofilament Protein antibody stains a 47 kDa BAF (BRG1-associated factors) protein. This antibody labels neurons (axons) of the central and peripheral nervous system and is useful for the identification of tumours with neuronal differentiation [31]. The anti-Synaptophysin antibody stains a 38 kDa glycosylated polypeptide and is a positive neuroblastoma marker [31]. All antibodies produced measurable and observable staining across all neuroblastoma cores. However, no statistically consistent change in staining signal intensity was observed (Figure 4; Table 1). 4. Conclusions The question of ‘tissue quality’ is an undefined and vexed one. Study designs often require all tissue specimens to be collected in the same fashion to exclude any storage or tissue handling activity that can confound later results. Such requirements are difficult to comply with when investigating rare diseases, and low donor numbers limit the statistical power required for modern research. We have demonstrated, using tissue microarray technology, that despite long term storage within a single hospital’s archive, 60 year old paraffin-embedded neuroblastoma specimens retain their antigenicity for currently used diagnostic markers. Our results concur with the findings of Litlekalsoy et al. [13] and Camp et al. [14], who performed similar studies on individual cohorts of 144 urinary bladder carcinomas or 38 breast carcinomas dating back from the present to 1932. Both concluded that the successful detection of specific proteins will enhance large retrospective investigations into rare diseases. Our study strengthens such findings by using TMA technology and subjecting every slide to exactly the same immunohistochemistry procedure at the same time. Visual assessment of tissue staining was avoided to limit subjectivity in reporting. Therefore, we used computer-aided analysis as our primary tool for measuring staining signal to objectively demonstrate consistency in staining intensity across all ages of sample obtained from a single institute. Variability in the signal intensity across the arrays was seen; most notably, staining distribution within samples collected most recently, indicating a greater level of signal contrast. However, this change in signal distribution and image quality did not persist over time, with samples older than 10 years generally yielding more consistent results. Time in storage for these FFPE samples does not appear to have a major effect on sample quality, but rather histological techniques, such as sample handling protocols, tissue fixation, and antigen retrieval conditions [34,35], appear to be stronger influences on how tissue stains. Indeed the generally tight clustering of all signals from our most recent samples (2000’s) indicates that present day tissue handling, immunohistochemistry techniques, and antibody quality have tissue improved to such an extent that more standardized results are obtained. These findings add support to the option for researchers to access archival material when conducting large retrospective studies into rare diseases and promise that meaningful answers to modern day questions can be obtained with a high degree of accuracy. Acknowledgments We thank the Histopathology Department at The Children’s Hospital at Westmead for their very kind supply of paraffin blocks and for technical assistance. The Children’s Hospital at Westmead Tumour Bank (http://tumourbank.chw.edu.au/) is supported by The Kids Cancer Project and The Cancer Institute of New South Wales. Author Contributions Al.C. designed the study, developed and performed the algorithm generating the IHC index, undertook the data analysis and preparation for the manuscript. N.M. constructed the tissue microarray, performed all histology applications including immunohistochemistry staining and scanned the slides. S.T. designed and applied the cell counting algorithm. T.H., N.G. and Am.C. undertook the pathology review of the neuroblastoma tumours, coring location selection and provided critical comments during the study design, experimental conduct and drafting of the manuscript. D.C. conceived the project, provided oversight of the study direction, critically reviewed the experimental outcomes and drafted the final manuscript. Conflict of Interest The authors declare no conflict of interest. ==== Refs References 1. Miller R.W. Young J.L. Nokavic B. Childhood cancer Cancer 1994 75 395 405 10.1002/1097-0142(19950101)75:1+<395::AID-CNCR2820751321>3.0.CO;2-W 8001010 2. Weinstein J. Katzenstein H. Cohn S. Advances in the diagnosis and treatment of neuroblastoma Oncologist 2003 8 278 292 10.1634/theoncologist.8-3-278 12773750 3. Brodeur G.M. Pritchard J. Berthold F. Carlsen N.L. Castel V. Castelberry R.P. De Bernardi B. Evans A.E. Favrot M. Hedborg F. Revisions of the international criteria for neuroblastoma diagnosis, staging, and response to treatment J. Clin. Oncol. 1993 11 1466 1477 8336186 4. Evans A.E. D’Angio G.J. Propert K. Anderson J. Hann H.W. Prognostic factors in neuroblastoma Cancer 1987 59 1853 1859 10.1002/1097-0142(19870601)59:11<1853::AID-CNCR2820591102>3.0.CO;2-F 3567848 5. Shimada H. Ambros I.M. Dehner L.P. Hata J. Joshi V.V. Roald B. Stram D.O. Gerbing R.B. Lukens J.N. Matthay K.K. The international neuroblastoma pathology classification Cancer 1999 86 364 372 10.1002/(SICI)1097-0142(19990715)86:2<364::AID-CNCR21>3.0.CO;2-7 10421273 6. Maris J.M. Recent advances in neuroblastoma New Engl. J. Med. 2010 362 2202 2211 10.1056/NEJMra0804577 20558371 7. Brodeur G.M. Neuroblastoma: Biological insights into a clinical enigma Nat. Rev. Canc. 2003 3 203 216 10.1038/nrc1014 8. Matthay K.K. Villablanca J.G. Seeger R.C. Stram D.O. Harris R.E. Ramsay N.K. Swift P. Shimada H. Black C.T. Brodeur G.M. Gerbing R.B. Reynolds C.P. Treatment of high risk neuroblastoma with intensive chemotherapy, radiotherapy, autologous bone marrow transplantation, and 13-cis-retinoic acid New Engl. J. Med. 1999 341 1165 1173 10.1056/NEJM199910143411601 10519894 9. Schmidt M.L. Lukens J.N. Seeger R.C. Brodeur G.M. Shimada H. Gerbing R.B. Stram D.O. Perez C. Haase G.M. Matthey K.K. Biologic factors determine prognosis in infants with stage IV neuroblastoma: A prospective Children’s Cancer Group study J. Clin. Oncol. 2000 18 1260 1268 10715296 10. Riley R.D. Heney D. Jones D.R. Sutton A.J. Lambert P.C. Abrams K.R. Young B. Wailoo A.J. Burchill S.A. A systematic review of molecular and biological tumor markers in neuroblastoma Clin. Canc. Res. 2004 10 4 12 10.1158/1078-0432.CCR-1051-2 11. Youlden D. Baade P. Ward L. Valery P. Hassall T. Green A. Aitken J.F. Childhood Cancer Incidence in Australia, 1983–2006 Viertel Centre for Research in Cancer Control, Cancer Council Queensland and the Australian Paediatric Cancer Registry Brisbane, Australia 2009 12. Camp R.L. Charette L.A. Rimm D.L. Validation of tissue microarray technology in breast carcinoma Lab. Investig. 2000 80 1943 1949 10.1038/labinvest.3780204 11140706 13. Litlekalsoy J. Vatne V. Jens G. Hostmark J.G. Laerum O.D. Immunohistochemical markers in urinary bladder carcinomas from paraffin-embedded archival tissue after storage for 5–70 years BJU Int. 2007 99 1013 1019 10.1111/j.1464-410X.2006.06699.x 17437436 14. Cronin M. Pho M. Dutta D. Stephans J.C. Shak S. Kiefer M.C. Esteban J.M. Baker J.B. Measurement of gene expression in archival paraffin-embedded tissues Am. J. Pathol. 2004 164 35 42 10.1016/S0002-9440(10)63093-3 14695316 15. Sugimura H. Mori H. Nagura K. Kiyose S. Hong T. Isozaki M. Igarashi H. Shinmura K. Hasegawa A. Kitayama Y. Tanioka F. Fluorescence in situ hybridization analysis with a tissue microarray: ‘FISH and chips’ analysis of pathology archives Pathol. Int. 2010 60 543 550 20618731 16. Kononen J. Bubendorf L. Kallioniemi A. Barlund M. Schraml P. Leighton S. Torhorst J. Mihatsch M.J. Sauter G. Kallioniemi O.P. Tissue microarrays for high-throughput molecular profiling of tumor specimens Nat. Med. 1998 4 844 847 10.1038/nm0798-844 9662379 17. Catchpoole D.R. Mackie N. Chetcuti A. McIver S. Henwood T. Kan A. Graf N. Arbuckle S. Tape transfer sectioning of tissue microarrays may lead to false positive immunohistochemistry staining Biotech. Histochem. 2011 86 421 428 10.3109/10520295.2010.527859 21091080 18. Tafavogh S. Navarro K.F. Catchpoole D.R. Kennedy P.J. Segmenting cellular regions of neuroblastoma tumor and splitting overlapping cells using shortest path between convex regions of cell contours Artif. Intell. Med. 2013 7885 171 175 19. Tafavogh S. Catchpoole D.R. Kennedy P.J. Determining cellularity status of tumors based on histopathology using hybrid image segmentation Proceedings of the 2012 International Joint Conference on Neural Networks (IJCNN) Brisbane, Australia 10–15 June 2012 1 8 20. Tafavogh S. Navarro K.F. Catchpoole D.R. Kennedy P.J. Non-parametric and integrated framework for segmenting and counting neuroblastic cells within neuroblastoma tumor images Med. Biol. Eng. Comput. 2013 51 645 655 10.1007/s11517-013-1034-9 23359256 21. Ryan D. Mulrane L. Rexhepaj E. Gallagher W.M. Tissue microarrays and digital image analysis Drug Safety Evaluation: Methods and Protocols, Methods in Molecular Biology Gautier J.-C. Springer Science+Business Media LLC Philadelphia, PA, USA 2011 Volume 691 97 112 22. Chetcuti A. Aktas S. Mackie N. Ulger C. Toruner G. Alkan M. Catchpoole D. Expression profiling reveals MSX1 and EphB2 expression correlates with the invasion capacity of Wilms tumors Pediatr. Blood Canc. 2011 57 950 957 10.1002/pbc.23003 23. Krajewska M. Smith L.H. Rong J. Huang X. Hyer M.L. Zeps N. Iacopetta B. Linke S.P. Olson A.H. Reed J.C. Krajewski S. Image analysis algorithms for immunohistochemical assessment of cell death events and fibrosis in tissue sections J. Histochem. Cytochem. 2009 57 649 663 10.1369/jhc.2009.952812 19289554 24. Miettinen M. Chatten J. Paetau A. Stevenson A. Monoclonal antibody NB84 in the differential diagnosis of neuroblastoma and other small round cell tumors Am. J. Surg. Pathol. 1998 22 327 332 10.1097/00000478-199803000-00007 9500774 25. Tsokos M. Linnoila R.I. Chandra R.S. Triche T.J. Neuron-specific enolase in the diagnosis of neuroblastoma and other small, round-cell tumors in children Human Pathol. 1984 15 575 584 10.1016/S0046-8177(84)80012-X 6373565 26. Misugi K. Aoki I. Kikyo S. Sasaki Y. Tsunoda A. Nakajima T. Immunohistochemical study of neuroblastoma and related tumors with anti-S-100 protein antibody Pediatr. Pathol. 1985 3 217 226 10.3109/15513818509078783 3912746 27. Sugita H. Kusano K. Tokunaga O. Mineta T. Abe M. Harada H. Shigemori M. Olfactory neuroepithelioma: An immunohistochemical and ultrastructural study Neuropathology 2006 26 400 408 10.1111/j.1440-1789.2006.00703.x 17080716 28. Parham D.M. Neuroectodermal and neuroendocrine tumors principally seen in children Am. J. Clin. Pathol. 2001 115 S113 S128 11993686 29. Rozental R. Morales M. Mehler M.F. Urban M. Kremer M. Dermietzel R. Kessler J.A. Spray D.C. Changes in the properties of gap junctions during neuronal differentiation of hippocampal progenitor cells J. Neurosci. 1998 18 1753 1762 9465000 30. Prasad L. Gayagay A. Charlton A. Henwood A. Graf N. Arbuckle S. Catchpoole D. Expression of interleukin-7 and its signalling intermediates in human neuroblastoma tumours Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research Chicago, IL, USA 31 March–4 April USA 2012 31. Hirose T. Scheithauer B.W. Lopes M.B.S. Gerber S.H.A. Altermatt H.I. Harner S.G. VandenBerg S.R. Olfactory neuroblastoma: An immunohistochemical, ultrastructural, and flow cytometric study Cancer 2006 76 4 19 8630875 32. Wilson B.S. Lloyd R.V. Detection of chromogranin in neuroendocrine cells with a monoclonal antibody Am. J. Pathol. 1984 115 458 468 6375394 33. Takeuchi T. Nicole S. Misaki A. Furihata M. Iwata J. Sonobe H. Ohtsuki Y. Expression of SMARCF1, a truncated form of SWI1, in neuroblastoma Am. J. Pathol. 2001 158 663 672 10.1016/S0002-9440(10)64008-4 11159203 34. Chiriboga L. Osman I. Mikhail M. Lau C. Tissue microarrays, tread carefully Lab. Investig. 2004 84 1677 10.1038/labinvest.3700172 15545964 35. Gomes L. Mackie N. Catchpoole D.R. Henwood T. Test and teach—In a fix about immunohistochemistry on 60 year old tissue blocks? J. Histotechnol. 2008 31 183 184 10.1179/his.2008.31.4.183

PMC005xxxxxx/PMC5003457.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays3010089microarrays-03-00089EditorialAcknowledgement to Reviewers of Microarrays in 2013 Microarrays Editorial Office MDPI AG, Klybeckstrasse 64, CH-4057 Basel, Switzerland28 2 2014 3 2014 3 1 89 90 © 2014 by the authors; licensee MDPI, Basel, Switzerland.2014This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/). ==== Body The editors of Microarrays would like to express their sincere gratitude to the following reviewers for assessing manuscripts in 2013: Alzate, Oscar Grigoriev, Andrey Sand, Michael Baxter, Robert C. Grzebyk, Daniel Scharpf, Robert B Belin, Andrea Carmine Guerau-De-Arellano, Mireia Seroussi, E. Benn, Peter Hess, Michael W. Settles, Matt Bicciato, Silvio Johnson, Nick Shanahan, Hugh P. Bort, Juan A. Hernández Karn, Thomas Shaw-Smith, Charles Bovolenta, Matteo Keele, John W. Simplício, Ana Luísa Brown, Chad C. Khalyfa, Abdelnaby Sloan Bena, Frederique Bruzzone, Roberto Kotzot, Dieter Söderberg, Ola Burroughs, A. Maxwell Labarge, Mark Sokilde, Rolf Carman, John G. Larson, Nicholas B. Spruck, Charles Chang, Jeffrey T. Leclerc, E. Srebniak, Malgorzata Chechik, Gal Lin, Tong Tang, Hao Chen, Chung-Yung Liu, Zhenqiu Tognini, Paola Chu, Jen-hwa Longy, Michel Underhill, Gregory H. Coleman, William B. Lopez, Arturo Van Den Veyver, Ignatia B. Descombes, Patrick Mascaux, Celine Vonhof, Maarten Dittmer, Dirk Mattheisen, Manuel Wang, Ying Dubnau, Josh Merdes, Gunter Wapner, Ronald J. Eckel-Passow, Jeanette E. Meyer, Swanhild U. Weston, Michael Fardo, David Montazeri, Zahra Winegarden, Neil Fredlund, Erik Müller, Volkmar Yamada, Masumi Galluzzi, L. Nakas, Christos Yamamoto, Norio George, Saji Novelli, Antonio Yang, Ning Giangreco, Adam Nykter, Matti Zavadil, Jiri Girirajan, Santhosh Orr, Megan Zhang, Chunxiang Glück, Stefan Pérez Jurado, Luis A. Zhang, Yinan Gouas, Laetitia Pergament, Eugene Zink, Daniele Griffin, Darren K. Queiroz, Luzia Helena

PMC005xxxxxx/PMC5003459.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2030171microarrays-02-00171ReviewComparative Analysis of CNV Calling Algorithms: Literature Survey and a Case Study Using Bovine High-Density SNP Data Xu Lingyang 12Hou Yali 3Bickhart Derek M. 4Song Jiuzhou 2Liu George E. 1*1 Bovine Functional Genomics Laboratory, BARC, BA, USDA-ARS, Beltsville, MD 20705, USA; E-Mail: xulingyang2008@gmail.com2 Department of Animal and Avian Sciences, University of Maryland, College Park, MD 20742, USA; E-Mail: songj88@umd.edu3 Laboratory of Disease Genomics and Individualized Medicine, Beijing Institute of Genomics, Chinese Academy of Sciences, Beijing 100029, China; E-Mail: houyali1210@gmail.com4 Animal Improvement Programs Laboratory, BARC, BA, USDA-ARS, Beltsville, MD 20705, USA; E-Mail: Derek.Bickhart@ars.usda.gov* Author to whom correspondence should be addressed; E-Mail: George.Liu@ars.usda.gov; Tel.: +1-301-504-9843; Fax: +1-301-504-8414.25 6 2013 9 2013 2 3 171 185 02 5 2013 04 6 2013 05 6 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Copy number variations (CNVs) are gains and losses of genomic sequence between two individuals of a species when compared to a reference genome. The data from single nucleotide polymorphism (SNP) microarrays are now routinely used for genotyping, but they also can be utilized for copy number detection. Substantial progress has been made in array design and CNV calling algorithms and at least 10 comparison studies in humans have been published to assess them. In this review, we first survey the literature on existing microarray platforms and CNV calling algorithms. We then examine a number of CNV calling tools to evaluate their impacts using bovine high-density SNP data. Large incongruities in the results from different CNV calling tools highlight the need for standardizing array data collection, quality assessment and experimental validation. Only after careful experimental design and rigorous data filtering can the impacts of CNVs on both normal phenotypic variability and disease susceptibility be fully revealed. copy number variation (CNV)algorithmsegmental duplicationsingle nucleotide polymorphism (SNP)cattle genome ==== Body 1. Introduction Genomic structural variation, including copy number variation (CNV), has been extensively studied in humans [1,2,3,4,5] and rodents [6,7,8,9]. Initial CNV reports have also been released for domesticated animals, including dog [10,11,12], cattle [13,14], chicken [15,16], pig [17,18], sheep [19,20], and goat [21] amongst others. Recent bovine CNV studies have generated several cattle CNV maps based on the data from Illumina Bovine SNP50K microarrays [22,23,24,25]. CNVs can be identified using various approaches, including comparative genomic hybridization (CGH) arrays, SNP arrays, and DNA sequencing. In spite of the increasing adoption of next-generation sequencing, microarrays will continue to be the primary platform for CNV detection in the near future. Compared to other approaches, the advantages of SNP arrays include their relative low cost and high throughput. Substantial genotyping data have been produced from genome-wide association studies, which can be directly exploited for CNV analysis. Dozens of human and mouse CNV studies have demonstrated that some CNVs are associated with phenotypic traits and diseases [26,27,28,29]. Efforts to explore the association between cattle CNV and economical traits have been published [30,31,32], even though the actual functional mechanisms are not yet well defined. 2. CNV Detection Using SNP Arrays SNP arrays were initially designed to genotype thousands of SNPs across the genome concurrently. Their applications have now expanded to include CNV detection using additional information such as the probe hybridization signal on each individual chip. The most well-known SNP microarrays are available from commercial vendors such as Illumina and Affymetrix [33,34]. Both companies sell competing arrays and continue to offer ever increasing coverage for detecting SNPs and CNVs simultaneously. However, one important consideration is the inherent bias of the SNP chip coverage against areas of the genome known to frequently harbor CNVs. For example, common copy number polymorphisms (CNPs) may cause a SNP to be rejected when the SNP fails standard inheritance checks and Hardy-Weinberg tests [35]. Segmental duplications (SDs), defined as >1 kb stretches of duplicated DNA with high sequence identity in a species, were shown to be one of the catalysts and hotspots for CNV formation [36,37,38]. Although the current microarray platforms offer some detection power in SD regions, calls within these regions are often affected by low probe density and cross-hybridization of repetitive sequence. In addition, only a relative copy number (CN) increase or decrease is reported with respect to the reference samples in SD regions. This poses a particular problem in the detection of CNVs in SD regions as the test individual’s copy number may differ from that of the reference by a smaller proportion than is detectable using array-based calling criteria. Although analyses of a subset of CNVs provided evidence of linkage disequilibrium with flanking SNPs [39], a significant portion of CNVs fell in genomic regions not well covered by SNP arrays, such as SD regions, and thus were not genotyped [40,41,42]. Since SNP chips are primarily designed for their use in SNP genotyping, some background noise that does not affect SNP calling may cause problems for CNV calling algorithms. For example, SNP data is typically normalized against a reference population in order to reduce between-array variations and probe-specific hybridization effects. The assumption that the large majority of reference samples have the same two copies does not hold for common CNV regions. At these regions, the normalization should be further optimized to derive correct parameters. Several new array designs have incorporated CNV detection, for example, monomorphic probes in common CNV regions are included on more recent Illumina and Affymetrix SNP array platforms. 3. Algorithms for CNV Detection Undoubtedly, microarray development has spurred the advances in computational analysis methodology in quantitative fields of biology. A wide range of CNV discovery tools has been developed based on data derived from SNP arrays, such as cnvPartition [43], Birdsuite [44], PennCNV [45], and amongst others. In this section, we briefly introduce these CNV detection tools. cnvPartition: Illumina data can be initially viewed, processed and exported using the proprietary GenomeStudio program (Illumina, CA, USA). In addition to quality checking and genotype calling, the program calculates several important input values for CNV discovery. The log R ratio (LRR), i.e., log2(Robserved/Rexpected), is calculated from the observed normalized intensity of a sample and expected normalized intensity, which is calculated from linear interpolation of canonical genotype clusters. The B allele frequency (BAF, normalized measure of relative signal intensity ratio of the B and A alleles) is calculated from the difference between the actual value and the expected position of the cluster group. LRR and BAF are used by many CNV detection algorithms. cnvPartition is offered as a plug-in for the GenomeStudio program, where it uses LRR and BAF to assess copy number using 14 different Gaussian distribution models between zero and four copies. cnvPartition also uses a likelihood-based method to compute the confidence score for each CNV call. Given the integration of cnvPartition into Illumina proprietary software (GenomeStudio), cnvPartition is currently unable to process and analyze Affymetrix chip data. Birdsuite: Affymetrix SNP array data from older chips must first be analyzed in the Genotyping Console program provided by Affymetrix for initial quality checks and controls. Data from the newer Affymetrix chip can be processed by additional programs contained in the Birdsuite package [44]. The Canary module of Birdsuite genotypes the known common CNVs using an Expectation-Maximization (EM) algorithm while the Birdseye module detects novel CNVs by using a Hidden Markov Model (HMM) with a Viterbi algorithm calculating emission states. For Affymetrix SNP arrays, there are other freely available CNV detection programs, such as GADA [46], Cokgen [47], iPattern [26] in addition to Birdsuite. For details about these programs, please see these published reviews [35,48,49]. The developers of Birdsuite have mentioned future plans for Illumina platform support [50] but current options only include a beta version for Illumina 610 array platforms. PennCNV and QuantiSNP: PennCNV and QuantiSNP are two freely available programs developed based on HMMs [45,51]. Both programs can process Illumina and Affymetrix SNP data. PennCNV incorporates multiple sources of information, including LRR and BAF at each SNP marker, the distance between neighboring SNPs and the allele frequency of SNPs. PennCNV also integrates a computational approach by fitting regression models with GC content to overcome “genomic waves” [52,53]. Additionally, PennCNV is capable of considering pedigree information (a parents-offspring trio) to improve call rates and accuracy of breakpoint prediction as well as to infer chromosome-specific SNP genotypes in CNVs. Finally, PennCNV also reports data quality control measurements for each CNV dataset. QuantiSNP, by contrast, uses an Objective Bayes approach [51] to infer copy number states based on the LogR ratio and the B allele frequency for each SNP marker. Whereas the PennCNV algorithm uses a transition matrix to model realistic copy number transitions between SNP probes [45], QuantiSNP calculates Bayesian probabilities for each SNP marker pair and then uses a HMM to join markers to form CNVs. Another significant difference between the two programs is that PennCNV is an open-source project whereas QuantiSNP was written for MatLab, which may limit availability to users that may not have a MatLab license. Finally, QuantiSNP is no longer under active development as listed on its webpage [54]. Approaches originally developed for array CGH: Several tools for CNV detection, which were originally developed for array CGH CNV calling, have been modified for SNP array analysis. However, these methods normally do not consider BAF information, which is the preferred data source to use for CNV calling in SNP data. For example, the Circular Binary Segmentation (CBS) method was designed to convert noisy intensity values into neighboring segments of distinct assigned copy numbers using dynamic programming [55]. DNAcopy is a widely used R implementation of the CBS method. Other commercial CNV detection tools: Other commercially available programs include Partek Genomics Suite, Nexus Copy Number software and Golden Helix SNP & Variation Suite (SVS). The strength of these commercial tools include their graphical user interfaces, streamlined pipelines for analysis and work flow, optimized computational speed as well as technical support. These factors are very important to labs with limited bioinformatics support; however, commercial companies often do not utilize some of the latest methods developed in the academic environment. For this study, we have chosen to look in detail at the Golden Helix SVS [56]. The SVS Copy Number Analysis Module (CNAM) employs a segmentation algorithm using only the signal intensity data to detect CNVs on either a per-sample (univariate) or multi-sample (multivariate) basis. According to its online manual, the univariate method, which considers only one sample at a time, is designed for detecting rare and/or large CNVs. The multivariate method, which considers all samples simultaneously, is designed for detecting small, common CNVs. Comparing univariate and multivariate methods: Although the exact algorithm of each method is proprietary, Breheny et al. explored the strengths and weaknesses of two similar approaches using both simulations and real data [57]. In their study, the univariate method (the CNV-level testing, i.e., across markers within one sample) involves estimating, at the level of the individual genome, the underlying copy number at each location. Once this is completed, tests are performed to determine the association between copy number state and phenotype. The multivariate method (the pooled marker-level testing across samples) carries out association testing first between the phenotypes and raw intensities at the level of the individual marker, and then aggregates neighboring test results to identify CNVs associated with the phenotype. Accounting for multiple comparisons across SNP markers is more straightforward, as a multiple-comparison correction (e.g., Bonferroni, permutation) can directly control the family-wise error rate (FWER) of the overall procedure [58]. False discovery rates can be calculated to account for multiple comparisons with the CNV-level testing method [59]; however, this is more complicated and somewhat conservative. Partially overlapping CNVs across cell lines introduce dependence across the tests, thereby reducing the effective number of independent tests. Breheny et al. confirmed that that the univariate method/CNV-level testing has greater power to detect associations involving large, rare CNVs, while the multivariate method/pooled marker-level testing has greater power to detect associations involving small, common CNVs. It is important to understand these tradeoffs. Several recent papers have proposed to develop methods capable of simultaneously pooling information across both markers and samples for CNV detection and association studies [60,61,62,63,64]. CNV quality score: Many programs like cnvPartition, Birdsuite, PennCNV and QuantiSNP reported CNV quality scores, which are quantitative values indicating CNV confidences. Although their exact meanings and interpretations depend on each algorithm and they are often not reported in microarray studies. These CNV quality scores are important for constructing CNV regions, which can then be used in association studies. 4. Comparing the CNV Detection Algorithms Using Human Data As shown in Table 1, at least 10 comparisons of the strengths and weaknesses of these array platforms and CNV calling tools have been published using human CNV data. Although published results are quickly outdated as new platforms and tools are introduced, a general theme is consistent across these comparisons. The first of these is the lack of a standard approach to collecting the data and the lack of standardized reference samples; this makes it difficult to compare CNV results across different studies [65]. The second is that CNV results also differ substantially depending on CNV detection methods [35,49]. For example, as the most comprehensive study on this topic, Pinto et al. have systematically compared CNV detection on 11 microarray platforms to evaluate data quality and CNV calling, reproducibility, concordance across array platforms and laboratories, breakpoint accuracy and analysis tool variability [49]. It is surprising that reproducibility in replicate experiments is <70% for most platforms and different analytic tools applied to the same raw data typically yield CNV calls with <50% concordance. The authors attributed these poor reproducibility observations to these facts: (1) large CNVs often overlap with SDs in complex genomic regions (as we described before) and (2) large CNVs also lead to call fragmentation (a single CNV is detected as multiple smaller variants). This led the authors to conclude that, “the striking differences between CNV calls from different platforms and analytic tools highlight the importance of careful assessment of experimental design in discovery and association studies and of strict data curation and filtering in diagnostics” [49]. microarrays-02-00171-t001_Table 1Table 1 Survey of recent comparison studies of copy number variation (CNV) detection. Authors Year Algorithm Data Platform Vendor Conclusion Comment Lai [66] 2005 CGHseq, Quantreg, CLAC, GLAD, CBS, HMM, Wavelet, Lowess, ChARM, GA and ACE Simulation and empirical samples for Glioblastoma array CGH Custom cDNA array Several general characteristics of future program development were suggested. Earlier programs for array CGH. Baross [67] 2007 CNAG, dChip, CNAT, GLAD Simulation and empirical mental retardation 100K Affymetrix SNP array SNP array Affymetrix Multiple programs were needed to find all real aberrations. False positive deletions was substantial, but could be greatly reduced by using the SNP genotype information to confirm loss of heterozygosity. Winchester [35] 2009 Birdsuite, CNAT, GADA, PennCNV, QuantiSNP NA12156, NA15510 SNP array Affymetrix, Illumina Multiple predictions from different software. Use software designed for the platform. Dellinger [68] 2010 CBS, cnvFinder, cnvPartition, GALD, Nexus, PennCNV and QuantiSNP Simulation and empirical samples from Singapore cohort study of the risk factors for Myopia SNP array Illumina QuantiSNP outperformed other methods based on ROC curve residuals over most datasets. Nexus Rank and SNPRank have low specificity and high power. Nexus Rank calls oversized CNVs. PennCNV detects one of the fewest numbers of CNVs. The normalized singleton ratio (NSR) is proposed as a metric for parameter optimization. Tsuang [69] 2010 PennCNV, QuantiSNP, HMMSeg, and cnvPartition 48 Schizophrenia samples SNP array Illumina Both guidelines for the identification of CNVs inferred from high-density arrays and the establishment of a gold standard for validation of CNVs are needed. Given the variety of methods used, there will be many false positives and false negatives. Zhang [70] 2011 Birdsuite, Partek Genomics Suite, HelixTree, and PennCNV-affy ~1,000 Bipolar + 270 HapMap samples SNP array Affymetrix Birdsuite and Partek had higher positive predictive values. Poor overlap between 2 gold standards (Kidd et al. and Conrad et al.). Marenne [71] 2011 cnvPartition, PennCNV, and QuantiSNP 96 pair samples from Spanish Bladder Cancer/EPICURO study SNP array Illumina PennCNV was the most reliable algorithm when assessing the number of copies. Current calling algorithms should be improved for high performance CNV analysis in genome-wide scans. Pinto [49] 2011 Birdsuite, cnvFinder, cnvPartition, dCHIP, ADM-2 (DNA Analytics), Genotyping Console (GTC), iPattern, Nexus Copy Number, Partek Genomics Suite, PennCNV, QuantiSNP 6 samples in triplicate on 11 array platforms array CGH, SNP array, and BAC array Agilent, NimbleGen, Affymetrix, and Illumina Different analytic tools applied to the same raw data typically yield CNV calls with <50% concordance. Moreover, reproducibility in replicate experiments is <70% for most platforms. The CNV resource presented here allows independent data evaluation and provides a means to benchmark new algorithms. CNV calls are disproportionally affected by genome complexity as they tend to overlap SDs and a single CNV is detected as multiple smaller variants. Koike [48] 2011 Birdsuite, Birdseye, PennCNV, CGHseg, DNAcopy HapMap samples SNP array Affymetrix Hidden Markov model-based programs PennCNV and Birdseye (part of Birdsuite), or Birdsuite show better detection performance. Segmental duplications and interspersed repeats (LINEs) are involved in CNVs. Eckel-Passow [72] 2011 Affymetrix Power Tools (APT), Aroma.Affymetrix, PennCNV and CRLMM 1,418 GENOA (Genetic Epidemiology Network of Atherosclerosis)/FBPP (Family Blood Pressure Program) samples SNP array Affymetrix Recommended trying multiple algorithms, evaluating concordance/discordance and subsequently consider the union of regions for downstream association tests. Advocated that software developers need to provide guidance with respect to evaluating and choosing optimal settings in order to obtain optimal results for an individual dataset. 5. Comparing CNV Detection Algorithms Using Bovine High-Density SNP Data We performed an analysis of CNVs based on Illumina BovineHD chips, which contain more than 750,000 SNP markers [73], using PennCNV. As a consequence of the higher SNP count, more CNVs were identified with higher resolution boundaries. In order to provide an additional comparison of CNV detection methods, we have tested three additional tools to call CNVs on the same BovineHD dataset: cnvPartition version 3.6.1, Golden Helix SVS 7.0 and DNAcopy [55]. These four tools were applicable to our dataset (Illumina bead array), available to us (due to existing commercial licensing or free availability) and were not designed specifically for human-based array studies. In order to perform an accurate and fair comparison of calls across the different methods, our PennCNV calls were derived from the same 630 animals of 27 cattle breeds on the cattle reference assembly UMD3.1 without using trio information [73]. We carried out cnvPartion calling using the default parameters as recommended by Illumina. For the Golden Helix SVS7.0, we used the SVS DSF Export Plug-In 4.1 to export LRRs from the GenomeStudio project. We then utilized CNAM to process the DSF file under the univariate option (minimum 3 markers/segment, a significance level of p = 0.005 for 2,000 pairwise permutations). We also performed DNAcopy analysis based on LRR. Finally, CNV segments were then filtered with a minimum of 3 probes for all 4 tools and a minimum of absolute segment mean values of 0.3 for SVS and DNAcopy. microarrays-02-00171-t002_Table 2Table 2 CNVs and CNVRs identified using PennCNV, cnvPartition, SVS, and DNAcopy. Tool Event Count Gain Loss Average Length PennCNV CNV 46,751 (74.2) 17,796 (28.2) 28,955 (46.0) 2,334,244,479 (49,929) CNVR 3,364 a 1,382 b 2,376 c 147,476,461 (43,840) cnvPartition CNV 16,566 (26.3) 5,021 (8.0) 11,545 (18.3) 2,191,528,246 (132,291) CNVR 1,298 a 541 b 916 c 172,378,730 (132,803) SVS CNV 92,463 (146.8) 205 (0.3) 92,258 (146.4) 2,234,601,290 (24,168) CNVR 7,099 a 78 b 7,056 c 151,471,634 (21,337) DNAcopy CNV 41,858 (66.4) 4,469 (7.1) 37,389 (59.3) 1,863,930,368 (44,530) CNVR 5,961 a 1,457 b 5,284 c 194,287,154 (32,593) Numbers in parentheses are values normalized by sample counts, except in the case of the parentheses values in the “Average Length” column, which are average lengths normalized by CNV counts. a These numbers represent non-redundant CNVR counts after merging both gain and loss CNVs identified across all 630 samples. b Gain CNV events were merged separately. c Loss CNV events were merged separately. A summary of CNV and CNVR results derived from all 630 samples is shown in Table 2. Detailed results can be found in the four worksheets of Supplementary Table 1. Compared to PennCNV results, CNVs and CNVRs in cnvPartition results are fewer and ~3 times longer (45 kb vs. 130 kb, respectively). While PennCNV and cnvPartition have loss/gain ratios of ~1.7 and DNAcopy has a ratio of 3.6, SVS has a ratio over 90, suggesting SVS is more sensitive to loss events than to gain events. Additionally, both SVS and DNAcopy CNVRs (average length approximately 20 kb and 30 kb, respectively) are shorter than PennCNV (~40 kb), and significantly shorter than cnvPartition CNVRs (~132 kb). Similar observations were also obtained when each subspecies/group (i.e. taurine, indicine, composite (taurine × indicine) and African breeds) was processed separately, confirming the above results (data not shown). When we compared CNV calls across subspecies/groups for all four CNV calling methods, CNV counts per sample were higher in African and indicine breeds, intermediate in composite breeds, and lower in taurine breeds, agreeing with our previous results using PennCNV [73]. We then compared the CNVRs from the four datasets derived from our calling programs based on the UMD3.1 cattle reference assembly (Figure 1). Approximately 50 Mb of core CNVRs are shared among the four CNVR sets. We calculated concordances using the ratios of between intersections and unions for both counts and lengths (Table 3 and Figure 1). PennCNV shared more regions (108 Mb or 50.82%) by length and 43.80% by count with cnvPartition than with any other tools. Therefore, we have observed that tools based on similar algorithms and input data (both LRR and BAF) seem to share more common regions. By contrast, PennCNV and SVS shared 24.88% or 60 Mb in length and 13.74% by count. This comparison was consistent with a recent publication based on PennCNV and SVS using human autism samples [74]. When we applied different filtering criteria requiring a minimum of five or 12 probes, the overlap of calls from these two methods increased slightly, ranging from 24–52% by the number of nucleotides that overlapped. We also evaluated the overlaps between loss CNV events across four datasets for each individual sample. The number of bases overlapped by CNVs from each dataset ranged from 26–48%, which agreed with our CNVR overlapping results. Figure 1 Comparisons of CNVR results identified by PennCNV, cnvPartition, SVS, and DNAcopy based on genomic location in UMD3.1. The overlap lengths of CNVRs were indicated in Mb. microarrays-02-00171-t003_Table 3Table 3 Overlaps among CNVRs across 4 CNV detection tools. Count Length (base pair) Tool1 Tool2 Intersection a Union a Percentage Intersection b Union b Percentage PennCNV cnvPartition 1,420 3,242 43.80% 107,775,740 212,079,451 50.82% PennCNV DNAcopy 2,355 6,970 33.79% 93,149,061 248,614,554 37.47% PennCNV SVS 1,264 9,199 13.74% 59,557,597 239,390,498 24.88% cnvPartition DNAcopy 1,284 5,975 21.49% 79,825,624 286,840,260 27.83% cnvPartition SVS 981 7,416 13.23% 56,569,347 267,281,017 21.16% DNAcopy SVS 2,332 10,728 21.74% 88,864,805 256,893,983 34.59% a These numbers represent intersections and unions of two CNVR datasets by count. b These numbers represent intersections and unions of two CNVR datasets by length in base pair. 6. Conclusions Like other published comparisons of CNV calling methods, we observed large variations in calls made by different programs. As pointed out previously, hybridization studies will generate both false positive and false negative results, regardless of how the data are analyzed [75]. As shown in Table 1, many authors recommended using multiple CNV calling algorithms instead of just one [35]; however, although the net effect of this strategy decreases the false negative rate, it also increases the false positive rate. With next generation sequencing projects producing better CNV calling standards, such as the 1,000 human genomes project [5] and our recent effort [76], we should be able to better estimate the false positive and false negative rates for each tool. Large incongruities in the results from different CNV calling tools highlight the need for standardizing array data collection, quality assessment and experimental validation. This is extremely true for other species like cattle, for which there is no gold standard of CNV calls to compare data against. Only after careful experimental design and rigorous data filtering can the impacts of CNVs on both normal phenotypic variability and disease susceptibility be fully revealed. Acknowledgments We thank members of the Illumina Bovine HD SNP Consortium for sharing their data. The bovine data will be available to scientists interested in non-commercial research upon signing a Materials Transfer Agreement (MTA). We would also like to thank Reuben Anderson and Alexandre Dimtchev for technical assistance. G.E.L. was supported by NRI/AFRI grants no. 2011-67015-30183 from the USDA NIFA and Project 1265-31000-098-00 from USDA-ARS. Mention of trade names or commercial products in this article is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the US Department of Agriculture. The USDA is an equal opportunity provider and employer. Conflict of Interest The authors declare no conflict of interest. ==== Refs References 1. Sebat J. Lakshmi B. Troge J. Alexander J. Young J. Lundin P. Maner S. Massa H. Walker M. Chi M. Large-scale copy number polymorphism in the human genome Science 2004 305 525 528 10.1126/science.1098918 15273396 2. Redon R. Ishikawa S. Fitch K.R. Feuk L. Perry G.H. Andrews T.D. Fiegler H. Shapero M.H. Carson A.R. Chen W. Global variation in copy number in the human genome Nature 2006 444 444 454 10.1038/nature05329 17122850 3. Conrad D.F. Pinto D. Redon R. Feuk L. Gokcumen O. Zhang Y. Aerts J. Andrews T.D. Barnes C. Campbell P. Origins and functional impact of copy number variation in the human genome Nature 2009 464 704 712 19812545 4. Altshuler D.M. Gibbs R.A. Peltonen L. Dermitzakis E. Schaffner S.F. Yu F.L. Bonnen P.E. de Bakker P.I.W. Deloukas P. Gabriel S.B. Integrating common and rare genetic variation in diverse human populations Nature 2010 467 52 58 10.1038/nature09298 20811451 5. Mills R.E. Walter K. Stewart C. Handsaker R.E. Chen K. Alkan C. Abyzov A. Yoon S.C. Ye K. Cheetham R.K. Mapping copy number variation by population-scale genome sequencing Nature 2011 470 59 65 10.1038/nature09708 21293372 6. Graubert T.A. Cahan P. Edwin D. Selzer R.R. Richmond T.A. Eis P.S. Shannon W.D. Li X. McLeod H.L. Cheverud J.M. A high-resolution map of segmental DNA copy number variation in the mouse genome PLoS. Genet. 2007 3 e3 10.1371/journal.pgen.0030003 17206864 7. Guryev V. Saar K. Adamovic T. Verheul M. van Heesch S.A. Cook S. Pravenec M. Aitman T. Jacob H. Shull J.D. Distribution and functional impact of DNA copy number variation in the rat Nat. Genet. 2008 40 538 545 10.1038/ng.141 18443591 8. She X. Cheng Z. Zollner S. Church D.M. Eichler E.E. Mouse segmental duplication and copy number variation Nat. Genet. 2008 40 909 914 10.1038/ng.172 18500340 9. Yalcin B. Wong K. Agam A. Goodson M. Keane T.M. Gan X.C. Nellaker C. Goodstadt L. Nicod J. Bhomra A. Sequence-based characterization of structural variation in the mouse genome Nature 2011 477 326 329 10.1038/nature10432 21921916 10. Chen W.K. Swartz J.D. Rush L.J. Alvarez C.E. Mapping DNA structural variation in dogs Genome Res. 2009 19 500 509 19015322 11. Nicholas T.J. Cheng Z. Ventura M. Mealey K. Eichler E.E. Akey J.M. The genomic architecture of segmental duplications and associated copy number variants in dogs Genome Res. 2009 19 491 499 19129542 12. Nicholas T.J. Baker C. Eichler E.E. Akey J.M. A high-resolution integrated map of copy number polymorphisms within and between breeds of the modern domesticated dog BMC Genomics 2011 12 414 10.1186/1471-2164-12-414 21846351 13. Liu G.E. van Tassell C.P. Sonstegard T.S. Li R.W. Alexander L.J. Keele J.W. Matukumalli L.K. Smith T.P. Gasbarre L.C. Detection of germline and somatic copy number variations in cattle Dev. Biol. 2008 132 231 237 14. Liu G.E. Hou Y. Zhu B. Cardone M.F. Jiang L. Cellamare A. Mitra A. Alexander L.J. Coutinho L.L. Dell’aquila M.E. Analysis of copy number variations among diverse cattle breeds Genome Res. 2010 20 693 703 10.1101/gr.105403.110 20212021 15. Volker M. Backstrom N. Skinner B.M. Langley E.J. Bunzey S.K. Ellegren H. Griffin D.K. Copy number variation, chromosome rearrangement, and their association with recombination during avian evolution Genome Res. 2010 20 503 511 10.1101/gr.103663.109 20357050 16. Wang X.F. Nahashon S. Feaster T.K. Bohannon-Stewart A. Adefope N. An initial map of chromosomal segmental copy number variations in the chicken BMC Genomics 2010 11 351 10.1186/1471-2164-11-351 20525236 17. Fadista J. Nygaard M. Holm L.E. Thomsen B. Bendixen C. A snapshot of CNVs in the pig genome PLoS ONE 2008 3 e3916 10.1371/journal.pone.0003916 19079605 18. Ramayo-Caldas Y. Castelló A. Pena R.N. Alves E. Mercadé A. Souza C.A. Fernández A.I. Perez-Enciso M. Folch J.M. Copy number variation in the porcine genome inferred from a 60 k SNP BeadChip BMC Genomics 2010 11 593 10.1186/1471-2164-11-593 20969757 19. Fontanesi L. Beretti F. Martelli P.L. Colombo M. Dall’olio S. Occidente M. Portolano B. Casadio R. Matassino D. Russo V. A first comparative map of copy number variations in the sheep genome Genomics 2011 97 158 165 10.1016/j.ygeno.2010.11.005 21111040 20. Liu J. Zhang L. Xu L. Ren H. Lu J. Zhang X. Zhang S. Zhou X. Wei C. Zhao F. Analysis of copy number variations in the sheep genome using 50 k SNP BeadChip array BMC Genomics 2013 14 229 10.1186/1471-2164-14-229 23565757 21. Fontanesi L. Martelli P.L. Beretti F. Riggio V. Dall’olio S. Colombo M. Casadio R. Russo V. Portolano B. An initial comparative map of copy number variations in the goat (Capra hircus ) genome BMC Genomics 2010 11 639 10.1186/1471-2164-11-639 21083884 22. Hou Y. Liu G.E. Bickhart D.M. Cardone M.F. Wang K. Kim E.S. Matukumalli L.K. Ventura M. Song J. Vanradan P.M. Genomic characteristics of cattle copy number variations BMC Genomics 2011 12 127 10.1186/1471-2164-12-127 21345189 23. Bae J.S. Cheong H.S. Kim L.H. NamGung S. Park T.J. Chun J.Y. Kim J.Y. Pasaje C.F. Lee J.S. Shin H.D. Identification of copy number variations and common deletion polymorphisms in cattle BMC Genomics 2010 11 232 10.1186/1471-2164-11-232 20377913 24. Fadista J. Thomsen B. Holm L.E. Bendixen C. Copy number variation in the bovine genome BMC Genomics 2010 11 284 10.1186/1471-2164-11-284 20459598 25. Seroussi E. Glick G. Shirak A. Yakobson E. Weller J.I. Ezra E. Zeron Y. Analysis of copy loss and gain variations in Holstein cattle autosomes using BeadChip SNPs BMC Genomics 2010 11 673 10.1186/1471-2164-11-673 21114805 26. Pinto D. Pagnamenta A.T. Klei L. Anney R. Merico D. Regan R. Conroy J. Magalhaes T.R. Correia C. Abrahams B.S. Functional impact of global rare copy number variation in autism spectrum disorders Nature 2010 466 368 372 10.1038/nature09146 20531469 27. Cook E.H. Jr. Scherer S.W. Copy-number variations associated with neuropsychiatric conditions Nature 2008 455 919 923 10.1038/nature07458 18923514 28. Sebat J. Lakshmi B. Malhotra D. Troge J. Lese-Martin C. Walsh T. Yamrom B. Yoon S. Krasnitz A. Kendall J. Strong association of de novo copy number mutations with autism Science 2007 316 445 449 10.1126/science.1138659 17363630 29. Aitman T.J. Dong R. Vyse T.J. Norsworthy P.J. Johnson M.D. Smith J. Mangion J. Roberton-Lowe C. Marshall A.J. Petretto E. Copy number polymorphism in Fcgr3 predisposes to glomerulonephritis in rats and humans Nature 2006 439 851 855 10.1038/nature04489 16482158 30. Liu G.E. Brown T. Hebert D.A. Cardone M.F. Hou Y.L. Choudhary R.K. Shaffer J. Amazu C. Connor E.E. Ventura M. Initial analysis of copy number variations in cattle selected for resistance or susceptibility to intestinal nematodes Mamm. Genome 2011 22 111 121 10.1007/s00335-010-9308-0 21125402 31. Hou Y. Liu G.E. Bickhart D.M. Matukumalli L.K. Li C. Song J. Gasberre L.C. van Tassell C.P. Sonstegard T.S. Genomic regions showing copy number variations associate with resistance or susceptibility to gastrointestinal nematodes in Angus cattle Funct. Integr. Genomics 2011 12 81 92 21928070 32. Hou Y. Bickhart D.M. Chung H. Hutchison J.L. Norman H.D. Connor E.E. Liu G.E. Analysis of copy number variations in Holstein cows identify potential mechanisms contributing to differences in residual feed intake Funct. Integr. Genomics 2012 12 717 723 10.1007/s10142-012-0295-y 22991089 33. LaFramboise T. Single nucleotide polymorphism arrays: A decade of biological, computational and technological advances Nucleic Acids Res. 2009 37 4181 4193 10.1093/nar/gkp552 19570852 34. Rincon G. Weber K.L. van Eenennaam A.L. Golden B.L. Medrano J.F. Hot topic: Performance of bovine high-density genotyping platforms in Holsteins and Jerseys J. Dairy Sci. 2011 94 6116 6121 10.3168/jds.2011-4764 22118099 35. Winchester L. Yau C. Ragoussis J. Comparing CNV detection methods for SNP arrays Brief. Funct. Genomic Proteomic 2009 8 353 366 10.1093/bfgp/elp017 19737800 36. Sharp A.J. Locke D.P. McGrath S.D. Cheng Z. Bailey J.A. Vallente R.U. Pertz L.M. Clark R.A. Schwartz S. Segraves R. Segmental duplications and copy-number variation in the human genome Am. J. Hum. Genet. 2005 77 78 88 10.1086/431652 15918152 37. Marques-Bonet T. Girirajan S. Eichler E.E. The origins and impact of primate segmental duplications Trends Genet. 2009 25 443 454 10.1016/j.tig.2009.08.002 19796838 38. Alkan C. Kidd J.M. Marques-Bonet T. Aksay G. Antonacci F. Hormozdiari F. Kitzman J.O. Baker C. Malig M. Mutlu O. Personalized copy number and segmental duplication maps using next-generation sequencing Nat. Genet. 2009 41 1061 1067 10.1038/ng.437 19718026 39. McCarroll S.A. Kuruvilla F.G. Korn J.M. Cawley S. Nemesh J. Wysoker A. Shapero M.H. de Bakker P.I. Maller J.B. Kirby A. Integrated detection and population-genetic analysis of SNPs and copy number variation Nat. Genet. 2008 40 1166 1174 10.1038/ng.238 18776908 40. Estivill X. Armengol L. Copy number variants and common disorders: Filling the gaps and exploring complexity in genome-wide association studies PLoS Genet. 2007 3 1787 1799 17953491 41. Locke D.P. Sharp A.J. McCarroll S.A. McGrath S.D. Newman T.L. Cheng Z. Schwartz S. Albertson D.G. Pinkel D. Altshuler D.M. Linkage disequilibrium and heritability of copy-number polymorphisms within duplicated regions of the human genome Am. J. Hum. Genet. 2006 79 275 290 10.1086/505653 16826518 42. Campbell C.D. Sampas N. Tsalenko A. Sudmant P.H. Kidd J.M. Malig M. Vu T.H. Vives L. Tsang P. Bruhn L. Population-genetic properties of differentiated human copy-number polymorphisms Am. J. Human Genet. 2011 88 317 332 10.1016/j.ajhg.2011.02.004 21397061 43. Illumina—Sequencing and Array-Based Solutions for Genetic Research Available online:http://www.illumina.com (accessed on 6 June 2013) 44. Korn J.M. Kuruvilla F.G. McCarroll S.A. Wysoker A. Nemesh J. Cawley S. Hubbell E. Veitch J. Collins P.J. Darvishi K. Integrated genotype calling and association analysis of SNPs, common copy number polymorphisms and rare CNVs Nat. Genet. 2008 40 1253 1260 10.1038/ng.237 18776909 45. Wang K. Li M. Hadley D. Liu R. Glessner J. Grant S.F. Hakonarson H. Bucan M. PennCNV: An integrated hidden Markov model designed for high-resolution copy number variation detection in whole-genome SNP genotyping data Genome Res. 2007 17 1665 1674 10.1101/gr.6861907 17921354 46. Pique-Regi R. Monso-Varona J. Ortega A. Seeger R.C. Triche T.J. Asgharzadeh S. Sparse representation and Bayesian detection of genome copy number alterations from microarray data Bioinformatics 2008 24 309 318 10.1093/bioinformatics/btm601 18203770 47. Yavas G. Koyuturk M. Ozsoyoglu M. Gould M.P. LaFramboise T. An optimization framework for unsupervised identification of rare copy number variation from SNP array data Genome Biol. 2009 10 R119 10.1186/gb-2009-10-10-r119 19849861 48. Koike A. Nishida N. Yamashita D. Tokunaga K. Comparative analysis of copy number variation detection methods and database construction BMC Genet. 2011 12 29 10.1186/1471-2156-12-29 21385384 49. Pinto D. Darvishi K. Shi X.H. Rajan D. Rigler D. Fitzgerald T. Lionel A.C. Thiruvahindrapuram B. MacDonald J.R. Mills R. Comprehensive assessment of array-based platforms and calling algorithms for detection of copy number variants Nat. Biotechnol. 2011 29 512 520 10.1038/nbt.1852 21552272 50. Birdsuite FAQ Broad Institute of MIT and Harvard Available online:http://www.broadinstitute.org/science/programs/medical-and-population-genetics/birdsuite/birdsuite-faq (accessed on 6 June 2013) 51. Colella S. Yau C. Taylor J.M. Mirza G. Butler H. Clouston P. Bassett A.S. Seller A. Holmes C.C. Ragoussis J. QuantiSNP: An objective bayes hidden-Markov model to detect and accurately map copy number variation using SNP genotyping data Nucleic Acids Res. 2007 35 2013 2025 10.1093/nar/gkm076 17341461 52. Marioni J.C. Thorne N.P. Valsesia A. Fitzgerald T. Redon R. Fiegler H. Andrews T.D. Stranger B.E. Lynch A.G. Dermitzakis E.T. Breaking the waves: Improved detection of copy number variation from microarray-based comparative genomic hybridization Genome Biol. 2007 8 R228 10.1186/gb-2007-8-10-r228 17961237 53. Diskin S.J. Li M. Hou C. Yang S. Glessner J. Hakonarson H. Bucan M. Maris J.M. Wang K. Adjustment of genomic waves in signal intensities from whole-genome SNP genotyping platforms Nucleic Acids Res. 2008 36 e126 10.1093/nar/gkn556 18784189 54. QuantiSNP Available online:http://sites.google.com/site/quantisnp/ (accessed on 6 June 2013) 55. Olshen A.B. Venkatraman E.S. Lucito R. Wigler M. Circular binary segmentation for the analysis of array-based DNA copy number data Biostatistics 2004 5 557 572 10.1093/biostatistics/kxh008 15475419 56. Genetic Association Software, Genome-Wide Association (GWAS) Software for SNP, CNV, and NGS Available online:http://www.goldenhelix.com/SNP_Variation/ (accessed on 6 June 2013) 57. Breheny P. Chalise P. Batzler A. Wang L. Fridley B.L. Genetic association studies of copy-number variation: Should assignment of copy number states precede testing? PLoS ONE 2012 7 e34262 10.1371/journal.pone.0034262 22493684 58. Storey J.D. Tibshirani R. Statistical significance for genomewide studies Proc. Natl. Acad. Sci. USA 2003 100 9440 9445 10.1073/pnas.1530509100 12883005 59. Benjamini Y. Yekutieli D. The control of the false discovery rate in multiple testing under dependency Ann. Stat. 2001 29 1165 1188 10.1214/aos/1013699998 60. Li B. Leal S.M. Methods for detecting associations with rare variants for common diseases: Application to analysis of sequence data Am. J. Hum. Genet. 2008 83 311 321 10.1016/j.ajhg.2008.06.024 18691683 61. Yang H.C. Hsieh H.Y. Fann C.S. Kernel-based association test Genetics 2008 179 1057 1068 10.1534/genetics.107.084616 18558654 62. Baladandayuthapani V. Ji Y. Talluri R. Nieto-Barajas L.E. Morris J.S. Bayesian random segmentation models to identify shared copy number aberrations for array CGH data J. Am. Stat. Assoc. 2010 105 1358 1375 10.1198/jasa.2010.ap09250 21512611 63. Nowak G. Hastie T. Pollack J.R. Tibshirani R. A fused lasso latent feature model for analyzing multi-sample aCGH data Biostatistics 2011 12 776 791 10.1093/biostatistics/kxr012 21642389 64. Glessner J.T. Li J. Hakonarson H. ParseCNV integrative copy number variation association software with quality tracking Nucleic Acids Res. 2013 41 e64 10.1093/nar/gks134 23293001 65. Scherer S.W. Lee C. Birney E. Altshuler D.M. Eichler E.E. Carter N.P. Hurles M.E. Feuk L. Challenges and standards in integrating surveys of structural variation Nat. Genet. 2007 39 S7 S15 10.1038/ng2093 17597783 66. Lai W.R. Johnson M.D. Kucherlapati R. Park P.J. Comparative analysis of algorithms for identifying amplifications and deletions in array CGH data Bioinformatics 2005 21 3763 3770 10.1093/bioinformatics/bti611 16081473 67. Baross A. Delaney A.D. Li H.I. Nayar T. Flibotte S. Qian H. Chan S.Y. Asano J. Ally A. Cao M. Assessment of algorithms for high throughput detection of genomic copy number variation in oligonucleotide microarray data BMC Bioinformatics 2007 8 368 10.1186/1471-2105-8-368 17910767 68. Dellinger A.E. Saw S.M. Goh L.K. Seielstad M. Young T.L. Li Y.J. Comparative analyses of seven algorithms for copy number variant identification from single nucleotide polymorphism arrays Nucleic Acids Res. 2010 38 e105 10.1093/nar/gkq040 20142258 69. Tsuang D.W. Millard S.P. Ely B. Chi P. Wang K. Raskind W.H. Kim S. Brkanac Z. Yu C.E. The effect of algorithms on copy number variant detection PLoS ONE 2010 5 e14456 10.1371/journal.pone.0014456 21209939 70. Zhang D. Qian Y. Akula N. Alliey-Rodriguez N. Tang J. Gershon E.S. Liu C. Accuracy of CNV detection from GWAS data PLoS ONE 2011 6 e14511 10.1371/journal.pone.0014511 21249187 71. Marenne G. Rodriguez-Santiago B. Closas M.G. Perez-Jurado L. Rothman N. Rico D. Pita G. Pisano D.G. Kogevinas M. Silverman D.T. Assessment of copy number variation using the Illumina Infinium 1M SNP-array: A comparison of methodological approaches in the Spanish Bladder Cancer/EPICURO study Hum. Mutat. 2011 32 240 248 10.1002/humu.21398 21089066 72. Eckel-Passow J.E. Atkinson E.J. Maharjan S. Kardia S.L. de Andrade M. Software comparison for evaluating genomic copy number variation for Affymetrix 6.0 SNP array platform BMC Bioinformatics 2011 12 220 10.1186/1471-2105-12-220 21627824 73. Hou Y. Bickhart D.M. Hvinden M.L. Li C. Song J. Boichard D.A. Fritz S. Eggen A. Denise S. Wiggans G.R. Fine mapping of copy number variations on two cattle genome assemblies using high density SNP array BMC Genomics 2012 13 376 10.1186/1471-2164-13-376 22866901 74. Matsunami N. Hadley D. Hensel C.H. Christensen G.B. Kim C. Frackelton E. Thomas K. da Silva R.P. Stevens J. Baird L. Identification of rare recurrent copy number variants in high-risk autism families and their prevalence in a large ASD population PLoS ONE 2013 8 e52239 10.1371/journal.pone.0052239 23341896 75. Carter N.P. Methods and strategies for analyzing copy number variation using DNA microarrays Nat. Genet. 2007 39 S16 S21 10.1038/ng2028 17597776 76. Bickhart D.M. Hou Y. Schroeder S.G. Alkan C. Cardone M.F. Matukumalli L.K. Song J. Schnabel R.D. Ventura M. Taylor J.F. Copy number variation of individual cattle genomes using next-generation sequencing Genome Res. 2012 22 778 790 10.1101/gr.133967.111 22300768

PMC005xxxxxx/PMC5003460.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2030265microarrays-02-00265ArticleKernel-Based Aggregation of Marker-Level Genetic Association Tests Involving Copy-Number Variation Li Yinglei 1Breheny Patrick 2*1 Department of Statistics, University of Kentucky, 311 MDS, 725 Rose Street, Lexington, KY 40536, USA; E-Mail: yinglei.li@uky.edu2 Department of Biostatistics, University of Iowa, N336 CPHB, 105 River Street, Iowa City, IA 52242, USA* Author to whom correspondence should be addressed; E-Mail: patrick-breheny@uiowa.edu; Tel.: +1-319-384-1584.04 9 2013 9 2013 2 3 265 283 02 8 2013 29 8 2013 30 8 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Genetic association tests involving copy-number variants (CNVs) are complicated by the fact that CNVs span multiple markers at which measurements are taken. The power of an association test at a single marker is typically low, and it is desirable to pool information across the markers spanned by the CNV. However, CNV boundaries are not known in advance, and the best way to proceed with this pooling is unclear. In this article, we propose a kernel-based method for aggregation of marker-level tests and explore several aspects of its implementation. In addition, we explore some of the theoretical aspects of marker-level test aggregation, proposing a permutation-based approach that preserves the family-wise error rate of the testing procedure, while demonstrating that several simpler alternatives fail to do so. The empirical power of the approach is studied in a number of simulations constructed from real data involving a pharmacogenomic study of gemcitabine and compares favorably with several competing approaches. CNV association studymultiple testingfamily-wise error ratekernel methods ==== Body 1. Introduction The classical genetic model states that humans have two copies of each gene, one on each chromosome. The sequencing of individual human genomes, however, has revealed that there is an unexpected amount of structural variation present in our genetic makeup. Sections of DNA can be deleted or duplicated, leaving individuals with fewer or more copies of portions of their genome (such individuals are said to have a CNV, or copy-number variation, at that position). Understanding the contribution of CNVs to human variation is one of the most compelling current challenges in genetics [1]. As the coverage of single nucleotide polymorphism (SNP) arrays has increased, it is increasingly possible to use this data to infer the CNV status of individuals. Indeed, recent generations of such chips include probes specifically designed to enable measurement of copy number [2,3]. Although other technologies for determining copy number exist, the use of genotyping chips has a clear advantage in that hundreds of genome-wide association studies have been performed [4], and these studies may be mined for CNV data at no added cost. Many of these studies have been carried out with very large sample sizes, thereby enabling CNV studies on a scale that would otherwise be prohibitively expensive [5,6]. These factors have led to a number of studies using data from high-density genotyping arrays to investigate the nature of copy-number variation and its role in human variation and disease [7,8,9,10,11]. Two general strategies have been proposed for conducting genetic association studies of copy-number variation. The majority of analytic techniques attempt to (1) identify or “call” CNVs for each individual, then (2) carry out association tests of whether individuals with a CNV differ from individuals without a CNV with regard to disease or some other phenotype. An alternative approach is to reverse the order of those two steps: (1) carry out association testing at the single marker level, then (2) aggregate information from neighboring markers to determine CNVs associated with disease/phenotype. The key idea of both approaches is that, because the data is noisy, it is virtually impossible to identify CNV associations from a single marker. Because copy number variants extend over multiple markers in a sufficiently high-density array, however, we are able to carry out inferences regarding CNVs by pooling information across neighboring markers. We refer to these two approaches, respectively, as variant-level testing and marker-level testing. In Breheny et al. [12], the authors explored the relative strengths and weaknesses of the two approaches and reached the conclusion that marker-level approaches were better able to identify associations involving small, common CNVs, while variant-level approaches were better able to identify associations involving large, rare CNVs. One serious complication with variant-level testing is that the estimated CNV boundaries from different individuals do not, in general, coincide. This presents a number of difficulties. Whether or not two individuals with partially overlapping CNVs should be in the same risk group for the purposes of association testing is ambiguous and complicates both the association test itself, as well as attempts to correct for multiple comparisons. With a large sample size, the complexity of partially overlapping CNV patterns quickly becomes daunting. Marker-level testing is an attractive alternative, as the aggregation is carried out on the test results, thereby avoiding the complications posed by partially overlapping CNV call boundaries during the analysis process. We illustrate the idea behind marker-level testing and aggregation in Figure 1, which plots a negative log transformation of the p-values of the marker-level tests vs. position along the chromosome. The details of the hypothesis tests for this example are described in Section 4, but the p-values may arise from any test for association between copy number intensity [13] and phenotype. The salient feature of the plot is the cluster of small p-values between 102.5 and 102.7 Mb. The presence of so many low p-values in close proximity to one another suggests an association between the phenotype and copy-number variation in that region. Figure 1 Illustration of marker-level testing. The -log10(p) values from the marker-level tests are plotted as a function of position along the chromosome. To locate these clusters on a genome-wide scale, Breheny et al. [12] used a marker-level approach based on circular binary segmentation [14,15]. Here, we take a closer look at the problem of aggregating p-values from marker-level tests. We present two main findings. First, we develop a computationally efficient kernel-based approach for p-value aggregation. Second, we analyze the multiple comparison properties of this approach and of p-value aggregation in general. In particular, we demonstrate that naïve aggregation approaches assuming exchangeability of test statistics do not preserve the family-wise error rate (FWER). To solve this problem, we present a permutation-based approach and show that it preserves family-wise error rates while maintaining competitive power. 2. Kernel-Based Aggregation Throughout, we will use i to index subjects and j to index markers. Let Xij denote the intensity measurement for subject i at marker j, let yi denote the phenotype for subject i and let pj denote the p-value arising from a test of association between phenotype and intensity at marker j. Finally, let ℓj denote the location of marker j along the chromosome and J denote the total number of markers; in this article, we use physical distance, but genetic distance could be used instead. Consider the aggregation (1) T(ℓ0)=∑jtjKh(ℓj,ℓ0)∑jKh(ℓj,ℓ0) where tj=f(pj) is a function of the test results for marker j and Kh(ℓj,ℓ0) is a kernel that assigns a weight to pj depending on how far away marker j is along the chromosome from the target location, ℓ0. The parameter h defines the bandwidth of the kernel and, thereby, controls the bias-variance tradeoff—a larger bandwidth pools test results over a larger region and, thereby decreases variance, but potentially introduces bias by mixing test results beyond the boundary of a CNV with those inside the boundary. Although, in principle, one could apply Equation (1) at any arbitrary location ℓ0, we restrict attention here to locations at which a marker is present and for which the bandwidth does not extend beyond the borders of the chromosome, thereby obtaining a finite set of aggregates {Tj}. We will consider transformations f(pj) such that low p-values lead to large values of tj, leading to significance testing at the aggregate level based on the statistic T=maxj{Tj}. In this section, we describe the choice of kernel Kh and transformation f(pj), as well as the issue of incorporating the direction of association for signed tests. 2.1. Choice of Kernel We consider two primary choices with regard to the kernel: shape and definition of bandwidth. First, we may consider varying the shape of the kernel. Two common choices are the flat (“boxcar”) kernel and the Epanechnikov kernel: (2) Flat:Kh(ℓj,ℓ0)=1if ℓj-ℓ0≤h0otherwise (3) Epanechnikov:Kh(ℓj,ℓ0)=341-ℓj-ℓ0h2if ℓj-ℓ0≤h0otherwise Intuitively, the Epanechnikov kernel would seem more attractive, as it gives higher weight to markers near the target location and diminished weight to distant markers where bias is a larger concern. Besides varying the shape of the kernel, we consider two definitions of bandwidth, which we refer to as constant width and constant marker (these concepts are sometimes referred to as “metric” and “adaptive” bandwidths, respectively, in the kernel smoothing literature). In the constant width approach, as illustrated in Equations (2) and (3), the width, h, of the kernel is constant. In contrast, the constant marker approach expands and contracts the range of the kernel as needed, so that there are always k markers in the positive support of the kernel. Specifically, the constant marker approach replaces the scalar h in Equations (2) and (3) with the function hk(ℓ0)=ℓ0-ℓ[k], where ℓ[k] is the location of the kth closest marker to x. For the constant width approach, the number of markers given positive weight by the kernel varies depending on ℓ0. The general tradeoff between the two approaches is that as we vary the target location, ℓ0, constant width kernels suffer from fluctuating variance, because the effective sample size is not constant, whereas constant marker kernels suffer from fluctuating bias, because the size of the region over which test results are pooled is not constant. We investigate the benefits and drawbacks of these various kernels in Section 5. As a point of reference, the flat, constant marker kernel is similar to the simple moving average, although not exactly the same. For example, consider the following illustration. Suppose h=3. At ℓ3, the three nearest neighbors are {p1,p2,p3}, while at ℓ4, the three nearest neighbors are {p4,p5,p6}. Thus, combinations such as {p3,p4,p5} are not considered by the kernel approach. This prevents the method from aggregating test results over inappropriately disperse regions of the chromosome, such as across the centromere. 2.2. Transformations Directly pooling p-values is not necessarily optimal. Various transformations of p may be able to better discriminate true associations from noise. Specifically, we consider the following transformations: (4) p:tj=1-pj (5) Z:tj=Φ-1(1-pj) (6) log:tj=-logpj where the text to the left of the equation is the label with which we will refer to these transformations in later figures and tables. The transformations are constructed in such a way that low p-values produce high values of tj for all three transformations. All three transformations have a long history in the field of combining p-values. Forming a combination test statistic based on the sum of logp values (or equivalently, the log of the product of the p-values) was first proposed by Fisher [16]—the so-called Fisher combination test. The transformation (5) was proposed by Stouffer et al. [17], who also studied the properties of sums of these normal-transformed p-values. Finally, Equation (4) was proposed and studied by Edgington [18]. The theoretical properties of these proposals have since been studied by several authors [19,20,21,22]. Throughout this literature, the majority of work has focused on these scales—uniform, Gaussian and logarithmic—each of which has been shown to have advantages and drawbacks. There is no uniformly most powerful method of combining p-values [23]. The present application differs from the classical work described above in that the borders of the CNVs are not known. Thus, we do not know the appropriate set of p-values to combine. Consequently, we must calculate many combinations, {Tj}, which are partially overlapping and, therefore, not independent, thereby requiring further methodological extensions. The implications of these concerns are addressed in Section 3. 2.3. Direction of Association Some association tests (z-tests, t-tests) have a direction associated with them, while others (χ2-tests, F-tests) do not. As we will see in Section 5, it is advantageous to incorporate this direction into the analysis when it is available, as it diminishes noise and improves detection. We introduce here extensions of the transformations presented in Section 2.2 that include the direction of association. Let sj denote the direction of association for test j. For example, in a case control study, if intensities were higher for cases than controls at marker j, then sj=1. At markers where CNV intensities were higher for controls than cases, sj=-1. The signs are arbitrary; their purpose is to reflect the fact that an underlying, latent CNV affects both phenotype and intensity measures; thus, switching directions of association are inconsistent with the biological mechanism being studied and likely to be noise. When sj is available, we adapt the three transformations from Section 2.2 as follows: (7) p:tj=sj(1-pj) (8) Z:tj=Φ-11+sj(1-pj)2 (9) log:tj=-sjlogpj All of these transformations have the same effect: when pj≈0 and sj=1, tj≫0; when pj≈0 and sj=-1, tj≪0; and when pj≈1, tj≈0, regardless of the value of sj. In other words, the test results combine to give an aggregate value, T(ℓ0), that is large in absolute value only if the test results have low p-values and are consistently in the same direction. 3. Significance Testing and FWER Control 3.1. Exchangeability In any analysis that involves aggregating marker-level test results, it is of interest to be able to quantify the significance of regions like those depicted in Figure 1. This is not trivial, however, as the lack of exchangeability between test results complicates matters and causes various naive approaches to fail. In this section, we illustrate the consequences of non-exchangeability by comparing three approaches to establishing the combined significance of a region with a preponderance of low p-values. One approach, suggested in Breheny et al. [12], is to use circular binary segmentation (CBS); implemented in the R package DNAcopy). This method aggregates neighboring p-values by calculating the t-test statistic comparing the intensity of a given region with that of the surrounding region. The significance of this test statistic is quantified by comparing it to the distribution of maximum test statistics obtained by permuting the {pj} values [14,15]. Crucially, however, this approach assumes that the test results, {pj}, are exchangeable as the justification for permuting them. Alternatively, we may use the kernel methods described in Section 2.1 to aggregate the neighboring test results, thereby obtaining Tmax=maxj{Tj}. One approach to approximating the null distribution of Tmax is to use Monte Carlo integration based on the fact that, under the null distribution, pj∼Uniform(0,1). Thus, for any choice of transformation and kernel in Equation (1), we may generate an arbitrary number of independent draws, {Tmax(b)}b=1B, from the null distribution function, F0, of Tmax and use the empirical cumulative distribution function (CDF) of those draws to obtain the estimate, F^0. Thus, we obtain a test for the presence of a CNV-phenotype association based on p=1-F^0(Tmax). The crucial assumption here is that, under the null distribution function, the p-values are independent. An alternative to the Monte Carlo approach for quantifying the significance of Tmax, described fully in Section 3.2, involves obtaining F^0 by permuting the phenotype prior to aggregation of the marker-level tests. Consider a genomic region in which individuals may have a CNV. The goal of the analysis is to detect CNVs associated with a particular phenotype. Thus, the null hypothesis may hold in one of two ways: (“No CNV”) no individuals with CNVs are present in the sample or (“No association”) individuals with CNVs are present in the sample, but the CNV does not change the probability of developing the phenotype. Table 1 demonstrates that while all three methods have the proper type I error rate in the “No CNV” setting, only the permutation approach preserves the correct type I error in the case where a CNV is present, but not associated with the disease. This is due to the fact that when a CNV is present, although it is still true that the marginal distribution of each pj is uniform (0,1), the CNV introduces correlation between nearby markers, thereby violating the assumptions of exchangeability and independence made by the CBS and Monte Carlo approaches. This phenomenon is also illustrated graphically in Figure 2. Unfortunately, the implementation of CBS provided by DNAcopy does not calculate exact p-values, only whether or not they fall below a cutoff. This is sufficient for Table 1 (where a cutoff of 0.05 was used), but insufficient information to construct the corresponding histogram in Figure 2. microarrays-02-00265-t001_Table 1Table 1 Preservation of Type I error for three methods with nominal α=0.05 in two possible settings for which the null hypothesis holds. The simulated genomic region contained 200 markers, 30 of which were spanned by a copy-number variant (CNV). The CNV was present in either 0% or 50% of the samples, depending on the setting. A detailed description of the simulated data is given in Section 5. Circular Binary Segmentation Kernel Monte Carlo Kernel Permutation No CNV 0.05 0.06 0.06 No Association 0.20 0.54 0.06 Figure 2 Ability (or inability) of the Monte Carlo and permutation approaches to maintain family-wise error rate under the two null scenarios. Valid p-values should appear uniformly distributed under the null hypothesis; this is clearly not the case in the lower left histogram. We make the following additional observations: (1) the CBS approach is somewhat more robust to the exchangeability issue than the Monte Carlo approach; i.e., its type I error rate is not as badly violated. (2) The data simulated here for the “no association” setting are somewhat exaggerated: the CNV was present in 50% of the population, and the signal to noise ratio was about twice as high as that typically observed in real data. In more realistic settings, the violation of type I error rate is not nearly as severe. The results in Table 1 and Figure 2 are intended to be an illustrative counterexample to demonstrate that CBS and kernel Monte Carlo are not guaranteed to preserve the type I error in all settings. (3) Circular binary segmentation was developed for the purpose of detecting CNVs, not aggregating marker-level tests, and its failure to preserve the family-wise error rate in this setting is in no way a criticism of CBS in general. 3.2. Permutation Approach We now formally define the kernel permutation method introduced in Section 3.1 and show that it preserves the family-wise error rate for the problem of CNV association testing. For a given set of test results, {pj}, consider quantifying whether or not the data represent compelling evidence for a CNV-phenotype association using the statistic: (10) Tmax=maxj{Tj} If the tests are directional, with results {pj,sj}, we use Tmax=maxj{Tj} To obtain the null distribution of Tmax, we use a permutation approach, generating up to n! unique draws, {Tmax(b)}b=1B, from the permutation distribution of Tmax, where n is the sample size. The procedure is as follows. At any given iteration, draw a random vector of phenotypes, y(b), by permuting the original vector of phenotypes. Next, carry out marker-level tests of association between the original CNV intensities and the permuted vector of phenotypes, obtaining a vector of permutation test results, {pj(b)}. Finally, apply the kernel aggregation procedure described in Section 2.1 to obtain {Tj(b)} and Tmax(b). We may then use the empirical CDF of these draws from the permutation distribution of Tmax to obtain the estimate F^0. Thus, we obtain a global test for the presence of a CNV-phenotype association based on p=1-F^0(Tmax). By preserving the correlation structure of the original CNV intensities, this approach does not rely on any assumptions of exchangeability or independence across neighboring markers and is thereby able to preserve the type I error rate of the testing procedure, unlike the other approaches described in Section 3.1. We now formally present this result, the proof of which appears in the Appendix. Theorem 1.  Let H0 denote the hypothesis that the phenotype, yi, is independent of the vector of CNV intensities, xi. Then, using the permutation approach described above with any of the kernel aggregation approaches in Section 2.1, for any α∈(0,1), (11) P(Type I error)≤α It is worth pointing out that the above theorem is proven for the case in which all permutations of {yi} are considered. In practice, as it is usually impractical to consider all permutations, only a random subset of these permutations are considered. However, by the law of large numbers, the above conclusion still holds approximately and may be made as precise as necessary by increasing the value of B, the number of permutations evaluated. For the numerical results in Section 4 and Section 5, we use B=1,000. The global test above is of limited practical benefit in the sense that it does not indicate the location of the associated CNV. Thus, we also consider the following equivalent procedure: declare significant evidence for the presence of a CNV-phenotype association at any marker for which Tj>F0-1(1-α), where F0 is again the null distribution of Tmax. Below, we state the corollary to Theorem 1 for the kernel permutation method, viewed as a multiple testing procedure for each marker. Corollary 1.  Let H0j denote the hypothesis that the phenotype, yi, is independent of the CNV intensity at marker j, Xij. Then, under the global null hypothesis that yi is jointly independent of {Xij}, for any α∈(0,1), (12) P(At least one Type I error)≤α using the permutation approach described above and Tj>F0-1(1-α) as the test function for H0j. In other words, the testing procedure described above controls the FWER in the weak sense at level α. It is worth noting that the procedure controls the FWER only in the weak sense—in other words, that it limits the probability of a false declaration of a CNV only under the global null hypothesis that there are no CNVs associated with the outcome. Typically in multiple testing scenarios, strong control is desirable. However, in the case of CNV-phenotype association, strong control is impractical, as it would imply that a method not only identifies CNV-phenotype associations, but can perfectly detect the genomic boundary of any associated CNV. This is an unrealistic requirement; in practice, there is no way to prevent the possibility that a detected CNV-phenotype association may spill over beyond the boundary of the CNV. 4. Gemcitabine Study In this section, we describe a pharmacogenomic study of gemcitabine, a commonly used treatment for pancreatic cancer. We begin by describing the design of the study (this description is similar to that provided in [12]), then analyze data from the study using the proposed kernel-based aggregation method. This data will also be used to simulate measurement errors for the simulation studies in Section 5. The gemcitabine study was carried out on the Human Variation Panel, a model system consisting of cell lines derived from Caucasian, African-American and Han Chinese-American subjects (Coriell Institute, Camden, NJ, USA). Gemcitabine cytotoxicity assays were performed at eight drug dosages (1,000, 100, 10, 1, 0.1, 0.01, 0.001 and 0.0001 uM) [24]. Estimation of the phenotype IC50 (the effective dose that kills 50% of the cells) was then completed using a four parameter logistic model [25]. Marker intensity data for the cell lines was collected using the Illumina HumanHap 550K and HumanHap 510S at the Genotyping Shared Resources at the Mayo Clinic in Rochester, MN, which consists of a total of 1,055,048 markers [26,27]. Raw data were normalized according to the procedure outlined in Barnes et al. [5]. 172 cell lines (60 Caucasian, 53 African-American and 59 Han Chinese-American) had both gemcitabine cytotoxicity measurements and genome-wide marker intensity data. To illustrate the application of the kernel-based aggregation approach, we selected one chromosome (chromosome 3) from the genome-wide data. To control for the possibility of population stratification, which can lead to spurious associations, we used the method developed by Price et al. [28], which uses a principal components analysis (PCA) to adjust for stratification. At each marker, a linear regression model was fit with PCA-adjusted IC50 as the outcome and intensity at that marker as the explanatory variable; these models produce the marker-level tests. We analyzed these data using the kernel-based approach described in Section 2 with a bandwidth of 50 markers and the log transformation. The results are shown in Figure 3. Note the presence of a peak at 102.6 Mb; this genomic region was also illustrated in Figure 1. The red line indicates the FWER-controlled, chromosome-wide significance threshold at the α=0.1 level. As the figure indicates, there is insufficient evidence in this study to establish a CNV association involving response to gemcitabine (p = 0.16) after controlling the chromosome-wide FWER. Other choices of bandwidth and transformation produce qualitatively similar, although somewhat less significant, results. Copy number variation in the region of chromosome 3 at 102.6 Mb, which is in close proximity to the gene, ZPLD1, has been found by Glessner et al. [29] to be associated with childhood obesity. An earlier analysis of this data by Breheny et al. [12] indicated suggestive evidence that this region harbors a CNV association with gemcitabine response, but lacked a formal way to control the error rate at the chromosome-wide level. This example illustrates the need for the more rigorous approach we develop here. The lack of significance in this example is perhaps not surprising, in that 172 subjects is a relatively small sample size for a CNV association study. Figure 3 Analysis of the gemcitabine data (chromosome 3) using the proposed kernel aggregation method. The kernel aggregations, Tj, are plotted against chromosomal position. The red line indicates the cutoff for chromosome-wide FWER significance at the α=0.1 level. 5. Simulations 5.1. Design of Spike-in Simulations In this section, we study the ability of the proposed approach to detect CNV-phenotype associations using simulated CNVs and corresponding intensity measurements. The validity of our conclusions depends on how realistic the simulated data is, so we have given careful thought to simulating this data in as realistic a manner as possible. The spike-in design that we describe here is also described in Breheny et al. [12]. The basic design of our simulations is to use real data from the gemcitabine study described in Section 4, “spike” a signal into it, then observe the frequency with which we can recover that signal. We used circular binary segmentation [14,15] to estimate each sample’s underlying mean intensity at every position along the chromosome, then subtracted the actual intensity measurement from the estimated mean to obtain a matrix of residuals representing measurement error. This matrix, denoted R, has 172 rows (one for each subject) and 70,542 columns (one for each marker). We then used these residuals to simulate noise over short genomic regions in which a single simulated CNV is either present or absent. Letting i denote subjects and j denote markers, the following variables are generated: zi, an indicator for the presence or absence of a CNV in individual i; xij, the intensity measurement at marker j for individual i; and yi, the phenotype. For the sake of clarity, we focus here on a random sampling design in which the outcome is continuous; similar results were obtained from a case-control sampling design in which the outcome is binary. In the random sampling design, the CNV indicator, zi, is generated from a Bernoulli distribution where γ=P(zi=1) is the frequency of the CNV in the population; subsequently, yi|zi is generated from a normal distribution whose mean depends on zi. For each simulated data set, 200 markers were independently selected at random from the columns of R. The measurement error for simulated subject i was then drawn from the observed measurement errors at those markers for a randomly chosen row of R. Thus, within a simulated data set, all subjects are studied with respect to the same genetic markers, but the markers vary from data set to data set. Simulating the data in this way preserves all the features of outliers, heavy-tailed distributions, skewness, unequal variability among markers and unequal variability among subjects that are present in real data. The intensity measurements, {xij}, derive from these randomly sampled residuals. To the noise, we add a signal that depends on the presence of the simulated CNV zi. The added signal is equal to zero unless the simulated genome contains a CNV encompassing the jth marker; otherwise, the added signal is equal to the standard deviation of the measurement error times the signal-to-noise ratio. Our simulations employed a signal-to-noise ratio of 0.8, which corresponded roughly to a medium-sized detectable signal based on our inspection of the gemcitabine data. Note that the phenotype and intensity measurement are conditionally independent given the latent copy-number status zi. An illustration of the spike-in process is given in Figure 4. For the Illumina Human1M-Duo BeadChip, which has a median spacing of 1.5 kb between markers, 200 markers corresponds to simulating a 300 kb genomic region. We varied the length of the CNV from 10 to 50 markers, corresponding to a size range of 15 to 75 kb. For the simulations presented in the remainder of the article, we used a sample size of n=1,000 and an effect size (change in mean divided by standard deviation) for the continuous outcome of 0.4. Figure 4 Illustration of spike-in simulation design. Left: The noise, randomly drawn from among the estimated measurement errors for a single cell line. Middle: The spiked-in signal. Right: The resulting simulated data; the gray shaded region denotes the boundary of the spiked-in CNV. 5.2. Transformations We begin by examining the impact on power of the various transformations proposed in Section 2.2 and Section 2.3. In order to isolate the effect of transformation, we focus here on the “optimal bandwidth” results: the bandwidth of the kernel was chosen to match the number of markers in the underlying CNV. This will lead to the maximum power to detect a CNV-phenotype association, although this approach is clearly not feasible in practice, as the size of an underlying CNV is unknown. Figure 5 Effect of transformation choice on power. Population CNV frequency was set to 10%; optimal bandwidths were used. Lines are colored according to the transformations, which were defined in Equations (4)–(9). The relationship between power and transformation choice is illustrated in Figure 5. The figure illustrates a basic trend that held consistently over many CNV frequencies and bandwidth choices: although the various transformations do not dramatically alter power, the normalizing transformation (Z) is most powerful for signed test results, while the log transformation is most powerful for unsigned test results. In the results that follow, unless otherwise specified, we employ the normalizing transformation for signed test results and the log transformation for unsigned tests. The substantial gain in power attained by incorporating the direction of association is also apparent from Figure 5 by comparing the left and right halves of the figure. 5.3. Choice of Kernel In this section, we examine two aspects of kernel choice: bandwidth implementation (constant-width vs. constant-marker) and kernel shape (flat vs. Epanechnikov), defined in Section 2.1. When all markers are equally spaced, the constant-width and constant-marker kernels are equivalent. To examine the impact on power when markers are unequally spaced, we selected at random a 200-marker sequence from chromosome 3 of the combined set of markers Illumina HumanHap 550K and 510S genotyping chips and spiked in CNVs of various sizes. The optimal bandwidth (either in terms of the number of markers or base pairs spanned by the underlying CNV) was chosen for each method. The left side of Figure 6 presents the results of this simulation. The constant-marker approach is substantially more powerful. When the number of markers is not held constant, the aggregation measure Tj is more highly variable for some values of j than others. This causes the null distribution of Tmax to have thicker tails, which, in turn, increases the p-value for the observed Tmax, thus lowering power. This phenomenon manifests itself most dramatically for small bandwidths. Consequently, throughout the rest of this article, we employ constant-marker kernels for all analyses. Figure 6 Effect of kernel choice on power. Left: Constant-width kernel vs. constant-marker kernel. Right: Flat vs. Epanechnikov kernel. In both plots, population CNV frequency was 10%, test results were unsigned and the log transformation was used. The right side of Figure 6 presents the results of changing the kernel shape from the flat kernel described in Equation (2) to the Epanechnikov kernel described in Equation (3). We make several observations: (1) the shape of the kernel has little impact on power; the two lines are nearly superimposed. (2) The kernel approach is relatively robust to choice of bandwidth; even five-fold differences between the bandwidth and optimal bandwidth do not dramatically decrease power. (3) Nevertheless, the optimal bandwidth does indeed occur when the number of markers included in the kernel matches the true number of markers spanned by the CNV. And (4), the Epanechnikov kernel is slightly more robust to choosing a bandwidth that is too large than the flat kernel is. This makes sense, as the Epanechnikov kernel gives less weight to the periphery of the kernel. 5.4. Kernel-Based Aggregation vs. Variant-Level Testing Lastly, we compare the kernel-based aggregation approach with variant-level testing. To implement variant-level testing, each sample was assigned a group (“variant present” or “variant absent”) on the basis of whether a CNV was detected by CBS. A two-sample t-test was then carried out to test for association of the CNV with the phenotype. This variant-level approach was compared with kernel-based aggregation of marker-level testing for a variety of bandwidths. The results are presented in Figure 7. For rare CNVs (5% population frequency), the power of the variant-level approach and the aggregated marker-level approach are comparable. However, for more common CNVs, the marker-level approach offers a substantial increase in power. For the most part, this increase in power persists even when the bandwidth is misspecified. Only when the bandwidth was much too small (selecting a 10-marker bandwidth for a 50-marker CNV) did the variant-level approach surpass marker-level aggregation. Generally speaking, these results are consistent with the findings reported in Breheny et al. [12], who found that variant-level tests have optimal power relative to marker-level tests when CNVs are large and rare; conversely, marker-level tests have optimal power relative to variant-level tests when CNVs are small and common. This is understandable, given the limited accuracy of calling algorithms for small CNVs. Comparing the results in Figure 7 with the results of Breheny et al. [12], who aggregated marker-level tests by applying CBS to the p-values, as described in Section 3.1, we find that the kernel approach is a substantially more powerful method for aggregating marker-level tests than the change-point testing carried out by CBS. Specifically, Breheny et al. found that the change-point approach had very low power at 5% frequency—much lower than the variant-level approach. On the other hand, in the same setting, we find that the kernel approach is comparable to, and even slightly more powerful than, the variant-level approach. Furthermore, as discussed in Section 3.1, a change-point analysis of marker-level tests also relies on exchangeability, which does not always hold. Thus, the methods developed in this article are both more powerful and achieve better control over the FWER than the CBS change-point analysis described in Breheny et al. [12]. Figure 7 Power comparison of variant-level testing (using CBS for CNV calling) with marker-level testing (using kernel-based aggregation). For the marker aggregation, k is the bandwidth (number of markers included in the kernel). A potential drawback of the kernel approach is the need to specify a bandwidth. This makes the robustness of the method to bandwidth misspecification, as illustrated in Figure 7, particularly important, because in practice, it is difficult to correctly specify the bandwidth a priori. Indeed, it is possible that multiple CNVs associated with the outcome are present on the same chromosome and have different lengths. A method that is not robust to bandwidth will be incapable of detecting both CNVs. Generally speaking, a bandwidth of roughly 30 markers seems to provide good power over the range of CNV sizes that we investigate here. 6. Discussion We have explored the use of a kernel-based method for aggregating tests that possess a spatial aspect, whereby underlying latent features cause nearby tests to be correlated, demonstrated some of the analytical challenges and developed an approach that properly accounts for multiple comparisons in a challenging setting. Our motivation for this work is the problem of association testing involving copy-number variants, but our findings may also be applied to other problems in genome-wide association studies, such as testing for haplotype associations. The computational burden of the method is worth further discussion, due to the permutation testing requirement. For simple tests, such as the linear regression tests we used in the gemcitabine study, the burden is quite manageable. On our machine (using an Intel Xeon 3.6 GHz processor), it took under a second to perform the 70,542 marker-level tests on chromosome 3 and under 0.1 s to perform the kernel aggregation. Carrying out 1,000 permutation tests took 1,000 times longer: 15 min to carry out all the permutation tests and 21 s to perform all the kernel aggregation. Extrapolating, a genome-wide analysis would take 3.5 h. These calculations, however, are for simple marker-level tests and a fairly small sample size (n=172). Larger studies will increase the computation burden linearly (i.e., doubling the subjects should double the computing time), but more complicated marker-level tests based on nonlinear, mixed-effects or mixture models would require substantially more time. Fortunately, the procedure is easy to run in parallel on multiple cores or machines, with each processor carrying out a fraction of the permutation tests. It is also worth pointing out that kernel aggregation may be used as an exploratory tool without the need for permutation testing (the black dots of Figure 3 may be calculated rapidly; the red line is what requires the permutation testing). Nevertheless, ongoing research in our group is focusing on ways to speed up the approach described here with a model-based formulation that avoids the need for permutation testing. The simulation studies of Section 5 address a limited-scale version of a larger question: how do marker-level test aggregation and variant-level testing compare for chromosome-wide and genome-wide analysis? This is an important question and deserves further study. In general, multiplicity is a thorny issue for CNV analyses, as the true locations of CNVs are unknown and can overlap in a number of complicated ways. The issue of how many tests to carry out and adjust for is a challenging question for variant-level testing and a considerable practical difficulty in analysis. In contrast, aggregation of marker-level results avoids this issue altogether. We have shown that the proposed approach is both powerful at detecting CNV associations and rigorously controls the FWER at a genome-wide level—two appealing properties. However, future work analyzing additional studies using kernel aggregation and studying its properties in larger, more complex settings is necessary. Acknowledgments We thank Brooke Fridley and Liang Li at the Mayo Clinic for contribution of the gemcitabine pharmacogenomic study data for this research and three anonymous reviewers for providing a number of helpful comments that led to considerable improvement of this article. Conflicts of Interest The authors declare no conflict of interest. Appendix Proof of Theorem 1. Let P denote the set of all possible permutations of {yi}, F0 the CDF of Tmax over P and F0-1 its generalized inverse. Furthermore, let ϕ(X,y)=1 if Tmax(X,y)>F0-1(1-α) and zero otherwise. Now, note that under the null hypothesis that xi and yi are independent: P(X,y)=∏iP(xi,yi)=∏iP(xi)P(yi)=P(X,y*) for all y*∈P. Thus, E0ϕ(X,y*) is a constant for all y* and E0ϕ(X,y)=1n!∑y*∈PE0ϕ(X,y*)=E01n!∑y*∈Pϕ(X,y*)≤α where the term inside the expectation in the second line is less than or equal to α for all X and y by the construction of the test. ==== Refs References 1. McCarroll S. Extending genome-wide association studies to copy-number variation Hum. Mol. Genet. 2008 17 R135 R142 10.1093/hmg/ddn282 18852202 2. McCarroll S. Kuruvilla F. Korn J. Cawley S. Nemesh J. Wysoker A. Shapero M. de Bakker P. Maller J. Kirby A. Integrated detection and population-genetic analysis of SNPs and copy number variation Nat. Genetics 2008 40 1166 1174 10.1038/ng.238 18776908 3. Perkel J. SNP genotyping: Six technologies that keyed a revolution Nat. Meth. 2008 5 447 453 10.1038/nmeth0508-447 4. Hindorff L. MacArthur J. Wise A. Junkins H. Hall P. Klemm A. Manolio T. A Catalog of Published Genome-Wide Association Studies 2012 Available online: http://www.genome.gov/gwastudies/ (accessed on 1 August 2013) 5. Barnes C. Plagnol V. Fitzgerald T. Redon R. Marchini J. Clayton D. Hurles M. A robust statistical method for case-control association testing with copy number variation Nat. Genetics 2008 40 1245 1252 10.1038/ng.206 18776912 6. Wellcome Trust Case Control Consortium Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls Nature 2007 447 661 678 17554300 7. Cooper G. Coe B. Girirajan S. Rosenfeld J. Vu T. Baker C. Williams C. Stalker H. Hamid R. Hannig V. A copy number variation morbidity map of developmental delay Nat. Genetics 2011 43 838 846 10.1038/ng.909 21841781 8. Komura D. Shen F. Ishikawa S. Fitch K. Chen W. Zhang J. Liu G. Ihara S. Nakamura H. Hurles M. Genome-wide detection of human copy number variations using high-density dna oligonucleotide arrays Genome Res. 2006 16 575 1584 10.1101/gr.5629106 17122084 9. Konishi H. Mohseni M. Tamaki A. Garay J. Croessmann S. Karnan S. Ota A. Wong H. Konishi Y. Karakas B. Mutation of a single allele of the cancer susceptibility gene brca1 leads to genomic instability in human breast epithelial cells Proc. Natl. Acad. Sci. USA 2011 108 17773 17778 10.1073/pnas.1110969108 21987798 10. Redon R. Ishikawa S. Fitch K. Feuk L. Perry G. Andrews T. Fiegler H. Shapero M. Carson A. Chen W. Global variation in copy number in the human genome Nature 2006 444 444 454 10.1038/nature05329 17122850 11. Simon-Sanchez J. Scholz S. del Mar Matarin M. Fung H. Hernandez D. Gibbs J. Britton A. Hardy J. Singleton A. Genomewide snp assay reveals mutations underlying parkinson disease Hum. Mutat. 2008 29 315 322 10.1002/humu.20626 17994548 12. Breheny P. Chalise P. Batzler A. Wang L. Fridley B.L. Genetic association studies of copy-number variation: Should assignment of copy number states precede testing? PLoS One 2012 7 e34262 10.1371/journal.pone.0034262 22493684 13. Peiffer D. Le J. Steemers F. Chang W. Jenniges T. Garcia F. Haden K. Li J. Shaw C. Belmont J. High-resolution genomic profiling of chromosomal aberrations using infinium whole-genome genotyping Genome Res. 2006 16 1136 1148 10.1101/gr.5402306 16899659 14. Olshen A. Venkatraman E. Lucito R. Wigler M. Circular binary segmentation for the analysis of array-based DNA copy number data Biostatistics 2004 5 557 572 10.1093/biostatistics/kxh008 15475419 15. Venkatraman E. Olshen A. A faster circular binary segmentation algorithm for the analysis of array CGH data Bioinformatics 2007 23 657 663 10.1093/bioinformatics/btl646 17234643 16. Fisher R. Statistical Methods for Research Workers Oliver and Boyd Edinburgh, Scotland 1925 17. Stouffer S. Suchman E. Devinney L. Star S. Williams R. Jr. American Soldier: Adjustment during Army Life Princeton University Press Princeton, NJ, USA 1949 Volume 1 18. Edgington E. An additive method for combining probability values from independent experiments J. Psychol. 1972 80 351 363 10.1080/00223980.1972.9924813 19. Feller W. An Introduction to Probability Theory and Its Applications Wiley Weinheim, Germany 1968 Volume 1 20. Good I. On the weighted combination of significance tests J. R. Stat. Soc. Ser. B 1955 17 264 265 21. Hedges L. Olkin I. Statistical Methods for Meta-Analysis Academic Press Waltham, MA, USA 1985 22. Littell R.C. Folks J.L. Asymptotic optimality of fisher’s method of combining independent tests J. Am. Stat. Assoc. 1971 66 802 806 10.1080/01621459.1971.10482347 23. Zaykin D. Zhivotovsky L. Westfall P. Weir B. Truncated product method for combining p -values Genetic Epidemiol. 2002 22 170 185 10.1002/gepi.0042 11788962 24. Li L. Fridley B. Kalari K. Jenkins G. Batzler A. Safgren S. Hildebrandt M. Ames M. Schaid D. Wang L. Gemcitabine and cytosine arabinoside cytotoxicity: Association with lymphoblastoid cell expression Cancer Res. 2008 68 7050 7058 10.1158/0008-5472.CAN-08-0405 18757419 25. Davidian M. Giltinan D. Nonlinear Models for Repeated Measurement Data Chapman & Hall/CRC Boca Raton, FL, USA 1995 26. Li L. Fridley B. Kalari K. Jenkins G. Batzler A. Weinshilboum R. Wang L. Gemcitabine and arabinosylcytosin pharmacogenomics: Genome-wide association and drug response biomarkers PLoS One 2009 4 e7765 10.1371/journal.pone.0007765 19898621 27. Niu N. Qin Y. Fridley B. Hou J. Kalari K. Zhu M. Wu T. Jenkins G. Batzler A. Wang L. Radiation pharmacogenomics: A genome-wide association approach to identify radiation response biomarkers using human lymphoblastoid cell lines Genome Res. 2010 20 1482 1492 10.1101/gr.107672.110 20923822 28. Price A. Patterson N. Plenge R. Weinblatt M. Shadick N. Reich D. Principal components analysis corrects for stratification in genome-wide association studies Nature 2006 38 904 909 10.1038/ng1847 16862161 29. Glessner J.T. Bradfield J.P. Wang K. Takahashi N. Zhang H. Sleiman P.M. Mentch F.D. Kim C.E. Hou C. Thomas K.A. A genome-wide study reveals copy number variants exclusive to childhood obesity cases Am. J. Hum. Genet. 2010 87 661 666 10.1016/j.ajhg.2010.09.014 20950786

PMC005xxxxxx/PMC5003461.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2030208microarrays-02-00208ArticleHydrogel Microwell Arrays Allow the Assessment of Protease-Associated Enhancement of Cancer Cell Aggregation and Survival Loessner Daniela 1Kobel Stefan 2Clements Judith A. 1Lutolf Matthias P. 2Hutmacher Dietmar W. 3*1 Faculty of Health, Institute of Health and Biomedical Innovation (IHBI), Queensland University of Technology (QUT), 60 Musk Avenue, Kelvin Grove 4059, Brisbane, Australia; E-Mails: daniela.lossner@qut.edu.au (D.L.); j.clements@qut.edu.au (J.A.C.)2 Laboratory of Stem Cell Bioengineering, Institute of Bioengineering, École Polytechnique Fédérale de Lausanne (EPFL), Building AI 3138, Station 15, CH-1015 Lausanne, Switzerland; E-Mails: stefan.kobel@roche.com (S.K.); matthias.lutolf@epfl.ch (M.P.L.)3 Faculty of Science and Engineering, IHBI, QUT, 60 Musk Avenue, Kelvin Grove 4059, Brisbane, Australia* Author to whom correspondence should be addressed; E-Mail: dietmar.hutmacher@qut.edu.au; Tel.: +61-7-3138-6077; Fax: +61-7-3138-6030.22 8 2013 9 2013 2 3 208 227 02 7 2013 31 7 2013 13 8 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Current routine cell culture techniques are only poorly suited to capture the physiological complexity of tumor microenvironments, wherein tumor cell function is affected by intricate three-dimensional (3D), integrin-dependent cell-cell and cell-extracellular matrix (ECM) interactions. 3D cell cultures allow the investigation of cancer-associated proteases like kallikreins as they degrade ECM proteins and alter integrin signaling, promoting malignant cell behaviors. Here, we employed a hydrogel microwell array platform to probe using a high-throughput mode how ovarian cancer cell aggregates of defined size form and survive in response to the expression of kallikreins and treatment with paclitaxel, by performing microscopic, quantitative image, gene and protein analyses dependent on the varying microwell and aggregate sizes. Paclitaxel treatment increased aggregate formation and survival of kallikrein-expressing cancer cells and levels of integrins and integrin-related factors. Cancer cell aggregate formation was improved with increasing aggregate size, thereby reducing cell death and enhancing integrin expression upon paclitaxel treatment. Therefore, hydrogel microwell arrays are a powerful tool to screen the viability of cancer cell aggregates upon modulation of protease expression, integrin engagement and anti-cancer treatment providing a micro-scaled yet high-throughput technique to assess malignant progression and drug-resistance. microwell arrayscell aggregatesbioengineered microenvironmentsovarian cancerkallikreinsintegrinspaclitaxel ==== Body 1. Introduction Three-dimensional (3D) in vitro culture approaches mimic more closely the physiological cell-cell and cell-extracellular matrix (ECM) interactions seen in vivo [1,2,3,4,5,6,7]. We have demonstrated that biomimetic hydrogels can be used as 3D cell culture platform to investigate the interplay of ovarian cancer cells with the ECM [8]. Within these synthetic microenvironments ovarian cancer cells form multi-cellular spheroids, an integral step leading to metastatic outgrowth and ultimately malignant progression in vivo. That is, after shedding from the primary tumor, these cells aggregate in order to survive within the abdominal cavity and to escape anti-cancer therapies [9,10]. Little is known about the events promoting ovarian cancer progression and how therapy-resistance occurs [11,12]. Cancer-associated proteases play a crucial role during disease progression [13]. Kallikrein-related (KLK) peptidases are known to contribute to metastatic outgrowth by modification of the tumor microenvironment via degradation of (non-)ECM proteins leading to altered cell-cell and cell-ECM interactions, cell proliferation and survival [14,15,16,17,18,19,20]. Elevated expression of KLK4, KLK5, KLK6, and KLK7 are linked to multi-cellular aggregation of ovarian cancer cells and non-responsiveness of patients to paclitaxel [21,22,23,24,25,26,27]. We have reported that combined expression of KLK4, KLK5, KLK6, and KLK7 in OV-MZ-6 ovarian cancer cells regulates integrin expression, cell adhesion, and promotes a malignant phenotype [28,29]. Of interest to this study is that integrins and integrin-related factors regulate tumor-ECM interactions leading to multi-cellular aggregation and drug-resistance [30,31,32]. Different integrins, in particular β1 integrins, are up-regulated in the advanced stages of the disease and mediate aggregation of ovarian cancer cells and therapy-resistance in patients [33,34,35]. Hence, a concomitant KLK4, KLK5, KLK6, and KLK7 expression might facilitate disease progression and lack of therapy response given that KLKs degrade ECM proteins, and therefore, influence the ECM-integrin binding dynamics. Bioengineered microenvironments have proven to be effective in screening the responsiveness of ovarian cancer cells to paclitaxel, thereby revealing increased survival rates after paclitaxel administration in 3D compared to flat cell cultures [8]. However, 3D systems which allow cell growth upon encapsulation of single cells within a hydrogel material lead often to the formation of different sized spheroids [8]. Hence, the purpose of this study was to allow OV-MZ-6 ovarian cancer cell aggregation of a defined size layered on top of polyethylene glycol-based hydrogel microwell arrays and to assess the efficacy of paclitaxel treatment dependent on aggregate size. Furthermore, we sought to determine the contribution of combined KLK4, KLK5, KLK6, and KLK7 expression and integrins to in OV-MZ-6 cell aggregation and survival upon paclitaxel treatment employing hydrogel microwell arrays as high-throughput microarray platforms [36,37] by performing time-lapse and confocal laser scanning microscopy as well as quantitative image, gene and protein analyses dependent on varying microwell and aggregate size. 2. Experimental Section Fabrication of Hydrogel Microwell Arrays. The fabrication of hydrogel microwell arrays was a multistep soft lithography process as reported previously [36]. Briefly, a topographically structured silicon wafer was fabricated, and then polydimethylsiloxane (PDMS; Dow Corning Corporation, Midland, MI, USA) was cast onto this structure, and finally, hydrogel films were patterned in a stamping step using the PDMS template. A 4-inch silicon wafer was designed using the layout editor of CleWin (PhoeniX, Enschede, The Netherlands). A pattern was selected consisting of eight squares; each square matched the dimensions of a standard 96-well plate, comprising 33 × 33 = 1,000 microwells, with a diameter of 100 µm and a depth of 50 µm per microwell. Additionally, new silicon wavers were designed to produce microwells of varying sizes of 50 × 50, 100 × 100, 150 × 150, 200 × 200 µm. Microwell arrays were formed from polyethylene glycol (PEG) hydrogel precursors by cross-linking two multi-arm PEG macromers (NOF Corporation, Tokyo, Japan), end-functionalized with either thiol (SH) or vinylsulfone (VS) groups [36]. The 8arm-PEG-VS was dissolved in 0.3 M triethanolamine (Sigma-Aldrich, Buchs, Switzerland), and the 4arm-PEG-SH was dissolved in bi-distilled water to obtain 100 µm thin hydrogel films (5% (w/v)) coated onto 8-well chamber µ-slides (ibidi GmbH, Munich, Germany) for a microwell size of 50 × 100 µm or onto 48-well tissue culture plates (Thermo Fisher Scientific Inc., Lausanne, Switzerland) for a microwell size of 50–200 × 50–200 µm. Optional, hydrogel microwell arrays were coated with laminin (0.1 mg/mL; BD Biosciences, Allschwil, Switzerland) or type I collagen (0.1 mg/mL; Sigma-Aldrich), both modified with an N-hydroxylsuccinimide (NHS)-PEG-maleimide linker (JenKem Technology, Allen, TX, USA) as described previously [38]. Cell Aggregate Cultures. The human epithelial ovarian carcinoma cell line OV-MZ-6 was established from malignant tumor fluid (ascites) [39], and stable transfectants, with human KLK4, KLK5, KLK6, and KLK7 full-length cDNA (“OV-KLK”) derived from ovarian cancer tissue and an empty vector plasmid (“OV-Vector”), provided by Viktor Magdolen (Technical University of Munich, Munich, Germany), were cultured as reported previously [29]. At a confluency of 60–80%, cells were harvested with EDTA (0.48 mmol/L; Invitrogen, Lucerne, Switzerland). For cell aggregate cultures, cells (5 × 104 cells/mL) were seeded on top of each square, centrifuged at 800 rpm for 5 min and grown over 120 h in 0.25 mL media (Figure 1(A)). Cell density was adapted accordingly to microwells of varying sizes (100 × 50 µm: 5 × 104 cells/mL, 50 × 50 µm: 5 × 104 cells/mL, 100 × 100 µm: 10 × 104 cells/mL, 150 × 150 µm: 15 × 104 cells/mL, 200 × 200 µm: 20 × 104 cells/mL). For exposure to paclitaxel, a microtubule-stabilizing agent that mediates cell cycle arrest and apoptosis [40], cell aggregates were treated with media containing paclitaxel (0, 1, 10, 100 nM; Invitrogen). Integrin inhibition was achieved using media supplemented with a functional blocking β1 integrin antibody (10 µg/mL; Chemicon/Millipore AG, Zug, Switzerland). Time-Lapse Microscopy. Time-lapse microscopy of hydrogel microwell arrays of varying size was performed to live image cell aggregation and survival as reported previously [36]. Samples were imaged 24 h after seeding using an inverted microscope (Zeiss Axio Observer.Z1 and Zeiss Axiovert) equipped with a motorized scanning stage under sterile humidified atmosphere at 37 °C/5% (v/v) CO2 over 96 h, with images taken every 6 h using a 10× air objective (Figure 1(B); Supplementary file). The resulting phase contrast images were then automatically compiled into a stack using Metamorph (Molecular Devices, Sunnyvale, CA, USA). To identify dead cells, propidium iodide (PI; 1:1,000; Invitrogen) was added to the media and fluorescently imaged at the end of each experiment. Cell aggregates were grown within different sized microwells and visualized at up to 20 different positions per condition. Figure 1 Schematic illustration and image analyses of hydrogel microwell arrays. (A) Cancer cell aggregates within microwells and their collection for subsequent expression analyses using a microinjector depicted by bright field microscopy (top panel). Confocal microscopy of four cell aggregates grown over 96 h within microwells (100 × 50 µm) ± paclitaxel treatment (100 nM); nuclei stained blue with DAPI; dead cells stained red with PI (bottom panel). Scale bars, 30 µm. (B) Time-lapse microscopy of cell aggregation ± paclitaxel treatment was performed over 96 h within microwells. Scale bars, 100 µm. (C) Confocal microscopy of cell aggregates grown over 96 h within microwells ± paclitaxel treatment and 3D reconstructions using Imaris; F-actin filaments stained green with Alexa488-conjugated phalloidin; nuclei stained blue with DAPI. Scale bars, 10 µm. (D) Confocal microscopy of the morphological marker N-cadherin in cell aggregates ± paclitaxel treatment; N-cadherin stained red using a respective primary and secondary Alexa555-conjugated IgG; nuclei stained blue with DAPI. Scale bars, 10 µm. Calculation of Cell Aggregate Area and Number. For cell aggregate number and area calculations, the integrated morphometry analysis tool in Metamorph or ImageJ ([41]) was applied to trace the aggregation number and area using either stacked bright field images or fluorescently labelled aggregates. Maximal projections using separate channels of bright field or fluorescent images were arithmetic processed, set to auto-threshold and gray levels binarized. An integrated morphometry analysis was performed to graphically identify the aggregate area. The aggregate number per microwell indicates the ratio of the number of aggregates per microwell to the number of microwells (seeding efficacy of 75–82%) counted per condition. Averages and standard errors were calculated using Excel (Microsoft, Redmond, WA, USA). For each experiment, 20 different positions per condition containing 60–960 aggregates were analyzed. For calculation of cell aggregation after paclitaxel treatment, only intact, non-lyzed cells (without the appearance of apoptotic bodies) were taken into account. Data are expressed as “relative aggregation (%)”, describing the ratio of the number of aggregates to the number of microwells analyzed per condition, and “relative cell death (%)”, referring to the ratio of the number of aggregates containing death cells (as indicated by PI staining) to the number of viable aggregates (no PI staining). Confocal Laser Scanning Microscopy (CLSM). Cell aggregate cultures were processed as described earlier [8]. Briefly, after 4% (w/v) paraformaldehyde (PFA)/PBS containing 0.1% (v/v) triton-X100 for 30 min, F-actin filaments were stained with Alexa488-conjugated phalloidin (0.1 U/mL; Invitrogen) or rhodamine415-conjugated phalloidin (0.3 U/mL; Invitrogen) and nuclei with a far-red DNA stain (DRAQ5; 5 µM; Alexis Biochemicals/Enzo Life Sciences, Lausen, Switzerland) or 4′6-diamidino-2-phenylindole (DAPI; 2.5 µg/mL; Invitrogen) in 1% (w/v) bovine serum albumin (BSA; Sigma-Aldrich)/PBS for 1 h each at room temperature (Figure 1(C)). For cell marker staining, primary (N-cadherin (1:100; R&D Systems, Minneapolis, MN, USA)) and secondary (Alexa555-conjugated sheep IgG (1:500; Invitrogen)) antibodies 1% (w/v) BSA/PBS were incubated for 1 h each at room temperature (Figure 1(D)). Immunofluorescence was visualized and imaged using a confocal microscope (Leica TCS SP2) with a 20/40× immersion oil objective at three to five different positions per sample covering one to four aggregates. Z-stacks were acquired with constant thickness of 2 µm reconstructing a cross-section profile of 100–150 equidistant XY-scans using the Leica Microsystems LAS AF software to generate maximal projections. 3D reconstructions were built using Imaris ([42]). Real-Time Reverse Transcription Quantitative PCR (RT-qPCR). Equal amounts (1 µg) of total RNA from cell aggregate cultures (extracted using an RNeasy micro kit; Qiagen, Magden, Switzerland) were used for cDNA synthesis. RT-qPCR was performed in triplicate with SYBR® Green chemistry (AB Applied Biosystems/Life Technologies, Compark Circuit, VIC, Australia) on an ABI7300 thermal cycler (AB Applied Biosystems). Reaction setup, using an annealing temperature of 60 °C and 40 cycles, and normalization applying the standard curve method (R2 = 0.96–0.99) were conducted as reported previously [8]. Gene specific primers: ITGA5—forward 5′-catttccgagtctgggccaa-3′, reverse 5′-tggaggcttgagctgagctt-3′; ITGB1—forward 5′-aggtggtttcgatgccatcat-3′, reverse 5′-aagtgaaacccggcatctgtg-3′; PTK2/FAK—forward 5'-GCGCTGGCTGGAAAAAGAGGAA-3', reverse 5'-TCGGTGGGTGCTGGCTGGTAGG-3′; 18S—forward 5′-gatccattggagggcaagtct-3′, reverse 5′-ccaagatccaactacgagcttttt-3′. Western Blotting. Lysates from cell aggregate cultures were collected in lysis buffer (according to RNeasy micro kit, Qiagen) as described earlier [8]. Protein concentrations were determined using protein detection reagents (bicinchoninic acid; Sigma-Aldrich) and 40 μg electrophoresed on 10% SDS-PAGE, transferred onto nitrocellulose membranes and treated with Odyssey® blocking buffer (LI-COR Biosciences, Lincoln, NE, USA). Membranes were incubated with primary (α5 integrin (1:1,000; Chemicon); β1 integrin (1:1,000; Chemicon); caspase8 (1:2,000; BD Biosciences); MT1-MMP (1:500; Chemicon); GAPDH (1:10,000; Abcam, Waterloo, NSW, Australia)) and secondary (IRDye 680/800-conjugated rabbit/mouse IgG (1:5,000; LI-COR Biosciences)) antibodies overnight at 4 °C and 1 h at room temperature, respectively. Images were obtained using the Odyssey® system (LI-COR Biosciences) and densitometrically evaluated. Statistics. Statistical analyses were carried out using ANOVA and Student’s t-test with “R”; results with p-values less than 0.05 were considered to be statistically significant (*/#—P < 0.05; **/##—P < 0.01; ***/##—P < 0.001). 3. Results and Discussion 3.1. Hydrogel Microwell Arrays Allow the Aggregation of Ovarian Cancer Cells We sought to apply high-throughput assays—to our knowledge for the first time—to allow defined aggregation of ovarian cancer cells and monitored this cellular process by confocal laser scanning microscopy (Figure 1(A,C,D)) and live cell microscopy over 96 h (Figure 1(B)) to establish their suitability as a drug screening tool using the clinically applied anti-cancer drug paclitaxel. Cancer cells cultured as single cell suspension (1 × 104 cells/mL) did not form aggregates on top of 3D cultures within microwells, and underwent only one cell division within the first 36 h after seeding (data not shown). Microwells coated with laminin or type I collagen did not increase the cell survival rates of single cell suspensions over 96 h (data not shown). As ovarian cancer cells aggregate in the tumor fluid (ascites) accumulated within the abdominal cavity of patients with advanced disease [10], we increased the number from single cancer cells per microwell (100 × 50 µm) to 5 × 104 cells/mL. Time-lapse and confocal laser scanning microscopy revealed compact aggregate formation after 96 h of 3D culture with negligible cell death as indicated by minor propidium iodide (PI) staining. Upon paclitaxel treatment (100 nM), cell aggregation was dramatically reduced and scattered and cell death increased as indicated by a positive PI staining (Figure 1(A,B)). 3D reconstructions and immunostaining of the morphological marker N-cadherin confirmed compact aggregation without treatment and scattered aggregation upon paclitaxel treatment with the appearance of apoptotic nuclei (Figure 1(C,D)). These results suggest that hydrogel microwell arrays allow cancer cell aggregation. The multi-cellular aggregate population in human ovarian tumor fluid (ascites) is thought to be a critical source for intra-abdominal metastases, and thereby, represents a key target for anti-metastatic interventions. Currently, most chemotherapies are ineffective in preventing aggregate dissemination, and the biological mechanisms leading to their formation remain poorly understood [9,10,43]. To improve our understanding of ovarian cancer biology, controlled in vitro models are needed to accurately mimic the in vivo conditions seen in patients [44]. Ill-advisedly, the terms aggregate and spheroid are inconsistently used throughout the literature, and yet, this definition is critical to the rationale of experimental 3D model approaches. The term aggregate is primarily but not always used to describe and eventually to discriminate loose packages of cells from compact spherical cultures. Aggregates with a size smaller than 150 µm may exhibit cell-cell and cell-matrix interactions. Spheroids comprise a defined cell mass of uniform geometry and physiological gradients at diameters ranging from 200–500 µm that can be manipulated and suited for large scale approaches in preclinical drug testing routines [45]. Both aggregate and spheroid cultures are well suited for developing high-throughput screening technologies [38,45], and their gene expression profiles are more truly indicative of clinical expression profiles than those detected in flat cell cultures [38,45,46]. Flat cell cultures fail to reproduce crucial aspects of carcinogenesis, such as 3D growth and architecture, cell-cell associations and cellular heterogeneity of in vivo samples. In this study, we have provided proof that bioengineered arrays represent a high-throughput platform reflecting 3D growth conditions of ovarian cancer cells and validated their responses by applying a clinically used therapeutic concept in vitro. Ovarian cancer cells grew as floatage-independent as multi-cellular aggregates. Immunostaining of structural components indicated cell-cell interactions within aggregates promoting cell survival. This microarray platform has also been used to re-create biophysical and biochemical microenvironmental cues that control stem cell fate [38], further underlining the suitability of this in vitro assay as a powerful 3D culture model. 3.2. KLK-Expressing Cells Increase Aggregation and Viability upon Paclitaxel Treatment As cancer-associated proteases like kallikrein-related (KLK) peptidases have been attributed to chemoresistance—in particular to taxane-based drugs—in ovarian cancer [23,26,27], we further sought to investigate the effect of paclitaxel using gradually increasing doses (0–100 nM) on cell aggregation. Confocal micrographs represented the aggregate morphology with and without paclitaxel treatment (100 nM): large and compact aggregates were formed in non-treated conditions, whereas paclitaxel exposure caused smaller and scattered aggregates and the presence of apoptotic bodies. Paclitaxel treatment was correlated with a positive PI staining, indicating an increased cell death (Figure 2(A)). Both OV-Vector/OV-KLK cells formed significantly fewer aggregates at higher paclitaxel concentrations (10 nM: OV-Vector 47 ± 8%/OV-KLK 54 ± 11%; 100 nM: OV-Vector 44 ± 8%/OV-KLK 60 ± 6%) compared to a lower dose (1 nM; OV-Vector 56 ± 14%/OV-KLK 56 ± 7%) and non-treated controls (0 nM: OV-Vector 57 ± 10%/OV-KLK 61 ± 8%). Strikingly, OV-KLK cells grew significantly more aggregates at higher paclitaxel concentrations (10, 100 nM) than OV-Vector cells (Figure 2(B), top panel). Cell death in both OV-Vector/OV-KLK cell aggregates was significantly increased at higher paclitaxel concentrations (10 nM: OV-Vector 35 ± 1%/OV-KLK 23 ± 7%; 100 nM: OV-Vector 53 ± 3%/OV-KLK 38 ± 4%) compared to a lower dose (1 nM: OV-Vector 10 ± 1%/OV-KLK 9 ± 3%) and non-treated controls (0 nM: OV-Vector 6 ± 1%/OV-KLK 4 ± 2%). Interestingly, OV-KLK cell aggregates showed significantly less cell death at higher paclitaxel concentrations (10, 100 nM) than OV-Vector cells, indicating an increased cell survival (Figure 2(B), bottom panel). Over the monitored time frame of 96 h no release of trapped cells and uniform aggregation of the trapped cells ± paclitaxel treatment were evident as indicated by time-lapse microscopy (Supplementary file). As integrins are associated with cell survival and chemoresistance [30,31,47], we analyzed the expression levels of β1 integrin (ITGB1) and focal adhesion kinase (FAK), an integrin-related factor, after paclitaxel administration. In both OV-Vector/OV-KLK cell aggregates, ITGB1 and FAK levels were increased upon paclitaxel treatment, with a further upregulation in OV-KLK cell aggregates (Figure 2(C)). These results suggest that hydrogel microwell arrays increase cell aggregation and viability of KLK-expressing cells upon paclitaxel treatment. Figure 2 Cell aggregation, survival and gene expression upon KLK expression and paclitaxel treatment. (A) Confocal microscopy of four cell aggregates grown over 96 h within microwells (100 × 50 µm) ± paclitaxel treatment (100 nM); nuclei stained blue with DAPI; dead cells stained red with PI (top panel); overlaid with Alexa488-conjugated phalloidin to stain F-actin filaments, with dead cells appearing yellow (bottom panel). Scale bars, 30 µm. (B) OV-Vector and OV-KLK cell aggregates were grown over 96 h within microwells and treated with increasing concentrations of paclitaxel (0, 1, 10, 100 nM). Both OV-Vector/OV-KLK cells grew less aggregates at higher paclitaxel concentrations (10, 100 nM) compared to a lower paclitaxel dose (1 nM) and non-treated controls (top panel). Cell death was increased at higher paclitaxel concentrations (10, 100 nM) compared to a lower paclitaxel dose (1 nM) and non-treated controls (bottom panel; n = 3; SEM; * P < 0.05; ** P < 0.01; *** P < 0.001). OV-KLK cells formed more aggregates and showed less cell death at higher paclitaxel concentrations (10, 100 nM) than OV-Vector cells (n = 3; SEM; #: P < 0.05; ##: P < 0.01). (C) Administration of paclitaxel (100 nM) was reflected in increased ITGB1 and FAK levels in both OV-Vector/OV-KLK cell aggregates, with an upregulation upon KLK expression. These findings are in line with our previously reported data, showing that KLK4 and KLK7 promote paclitaxel-induced resistance of ovarian cancer cell aggregates that were formed in a tumor fluid (ascites) mimicking microenvironment [25,26]. It was shown that multi-cellular aggregates, harboring a 3D architecture, are more resistant compared to flat cell cultures [48], and compact aggregates are less responsive to different therapeutic regimes, such as chemotherapies, than scattered aggregates [49]. We have also reported that combined expression of KLK4, KLK5, KLK6, and KLK7 in ovarian cancer cells (OV-KLK) mediates resistance to paclitaxel at higher doses (10, 100 nM) compared to control cells (OV-Vector) when grown as flat cell cultures [28]. When the same cells were grown as aggregates in this study, we observed a similar cell survival effect upon KLK expression and paclitaxel treatment. Interestingly, the expression of β1 integrin was decreased upon KLK expression [28], but upon paclitaxel treatment increased in both KLK-expressing and KLK-deficient aggregates, suggesting a critical function of this integrin in paclitaxel-related resistance, only partially induced by these four KLKs. Integrins and integrin-related factors are required for the responsiveness to anti-cancer drugs that bind to microtubules [50]. Although integrins lack kinase activity, by clustering they recruit and activate kinases, such as FAK. FAK is overexpressed in most ovarian cancers, associated with poor clinical outcome and plays a role in regulating invasion and metastasis [51,52]. Paclitaxel treatment stabilizes microtubule dynamics, thereby inhibiting mitosis [40], and FAK is required for integrin-dependent microtubules stabilization and paclitaxel responsiveness [53]. It was shown that FAK regulates the efficacy of taxane-based drugs in both treatment-sensitive and treatment-resistant cells [54]. We detected increased mRNA levels of FAK in aggregates after paclitaxel treatment, further indicating that FAK is an important cell survival factor in ovarian cancer cells. These findings imply the potential of combinatorial therapeutic approaches including the inhibition of KLKs, integrin and integrin-related factors with cytotoxic drugs for the treatment of ovarian cancer patients, especially those with high KLK levels in their tumors. 3.3. Paclitaxel Treatment Alters Integrin Expression of Tailor-Made KLK-Expressing Cell Aggregates Ovarian cancer cell aggregates derived from the tumor fluid (ascites) of patients with late-stage stage disease range in number (from two to more than 20) and size (from 30–200 µm, even up to 750 µm in diameter) and contain up to 100 cells, suggesting a high patient to patient variability [35,55,56,57]. This high variability in aggregate size is also reflected in in vitro aggregate cultures applying the liquid overlay technique [35,48,57,58] or hanging droplet method [46,59] using different ovarian cancer cell lines [34]. In order to control the cellular microenvironment of hydrogel microwell arrays, photolithography was used to fabricate microwells of varying sizes (50 × 50, 100 × 100, 150 × 150, 200 × 200 µm) to generate aggregates of different sizes (Figure 3(A), top panel). Cell aggregation was confirmed by immunostaining of F-actin filaments and nuclei (Figure 3(A), bottom panel). Figure 3 Effect of microwell size on cell aggregation and survival. (A) Schematic illustration of varying microwell sizes (50 × 50, 100 × 100, 150 × 150, 200 × 200 µm) generated using photolithography (top panel). Fluorescent staining of cell aggregates grown over 96 h within microwells of varying size ± paclitaxel (100 nM); F-actin filaments stained green with Alexa488-conjugated phalloidin; nuclei stained blue with DAPI (bottom panel; scale bars corresponding to respective microwell size). (B) Analyses of aggregation and death of both OV-Vector/OV-KLK cells depending on microwell size relative to total aggregate numbers. While in the larger microwells (100–200 × 100–200 µm) complete cell aggregation (98–100%) and no cell death (0–2%) was detected, the smallest microwells (50 × 50 µm) caused incomplete cell aggregration (73–92%) and cell death (8–27%). Upon paclitaxel treatment both OV-Vector/OV-KLK cells aggregated less (42–96%) in medium sized (100–150 × 100–150 µm) and smallest (50 × 50 µm) microwells and revealed higher cell death (4–58%) rates (n = 3; SEM; * P < 0.05; ** P < 0.01; *** P < 0.001). OV-KLK cells showed higher aggregation and less cell death rates in the smallest microwells (50 × 50 µm) without paclitaxel treatment and less evidence of cell death upon paclitaxel in the largest microwells (200 × 200 µm) compared to OV-Vector cells (n = 3; SEM; # P < 0.05; ## P < 0.01). As paclitaxel is subject to multi-cellular-mediated resistance for ovarian cancer cells [48], we treated tailor-made aggregates with paclitaxel (100 nM), and analyzed the aggregation and death of both OV-Vector/OV-KLK cells relative to the total numbers (Figure 3(B)). While in the larger microwells (100–200 × 100–200 µm) complete cell aggregation (98–100%) and no cell death (0–2%) was detected, the smallest microwells (50 × 50 µm) caused incomplete cell aggregration (73–92%) and cell death (8–27%). Upon paclitaxel treatment both OV-Vector/OV-KLK cells aggregated less (42–96%) in the medium sized (100–150 × 100–150 µm) and smallest (50 × 50 µm) microwells and showed higher cell death (4–58%) rates (Figure 3(B)). Then, we analyzed the aggregate number and area in each microwell size performing time-lapse microscopy (Figure 4(A)). Both OV-Vector/OV-KLK cells formed one aggregate (1.27–1.34 × 103 cm2)/well in the smallest (50 × 50 µm) microwells, whereas in the next larger (100 × 100 µm) microwells, two aggregates (4.33–5.34 × 103 cm2)/well were formed. In the medium sized (150 × 150 µm) and largest (200 × 200 µm) microwells, three aggregates (10.68–19.14 × 103 cm2)/well were formed. Upon paclitaxel treatment both OV-Vector/OV-KLK cells formed one aggregate (0.72–0.80 × 103 cm2)/well in the smallest (50 × 50 µm) microwells, while in the next larger (100 × 100 µm) microwells, two aggregates (2.31–3.78 × 103 cm2)/well were detected. In the medium sized (150 × 150 µm) and largest (200 × 200 µm) microwells, three aggregates (6.98–9.68 × 103 cm2)/well were formed. OV-KLK cells formed larger aggregates in the second smallest (100 × 100 µm) microwells after paclitaxel treatment compared to OV-Vector cells (Figure 4(B)). These results suggest that OV-KLK cells had a higher ability to aggregate and survive with and without paclitaxel in all microwell sizes compared to OV-Vector cells. The administration of paclitaxel reduced aggregate area but not numbers compared to non-treated conditions. Figure 4 Cell aggregation as a function of microwell size, KLK expression and paclitaxel treatment. (A) Bright field microscopy depicted both OV-Vector/OV-KLK cell aggregates at the end time point of time-lapse microscopy carried out over 96 h within microwells of varying sizes (50 × 50, 100 × 100, 150 × 150, 200 × 200 µm) ± paclitaxel treatment (100 nM); dead cells stained red with PI. The tailor-made microwell size corresponds to respective aggregate size. (B) Both OV-Vector/OV-KLK cells formed one aggregate (1.27–1.34 × 103 cm2)/microwell in the smallest (50 × 50 µm) microwells, whereas in the next larger (100 × 100 µm) microwells two aggregates (4.33–5.34 × 103 cm2)/microwell were formed. The medium sized (150 × 150 µm) and largest (200 × 200 µm) microwells caused three aggregates (10.68–19.14 × 103 cm2)/well. Administration of paclitaxel reduced aggregate area compared to non-treated conditions (n = 3; SEM; * P < 0.05; ** P < 0.01). Upon paclitaxel treatment both OV-Vector/OV-KLK cells formed one aggregate (0.72–0.80 × 103 cm2)/microwell in the smallest (50 × 50 µm) microwells, while in the next larger (100 × 100 µm) microwells two aggregates (2.31–3.78 × 103 cm2)/microwell were detected. The medium sized (150 × 150 µm) and largest (200 × 200 µm) microwells caused three aggregates (6.98–9.68 ×103 cm2)/microwell. OV-KLK cells formed larger aggregates in the second smallest (100 × 100 µm) microwells after paclitaxel treatment compared to OV-Vector cells (n = 3; SEM; # P < 0.05). Within these bioengineered microwells, the formation of cell aggregates was achieved in sizes ranging from 50–200 µm. Similar aggregate sizes are described in experimental and clinical samples [35,55,56,57] showing a high cell viability in combination with KLK expression, and the results presented in this study are in line with our former reports [25,26]. Paclitaxel treatment revealed that the aggregate area but not aggregate number was reduced, further corroborating the existence of survival-promoting factors, such as integrins, and multi-cellular-mediated drug resistance mechanisms in ovarian cancer cells [48]. A similar bioengineered approach to the one described here has been used to control the size and shape of embryonic bodies employing microwells of varying diameters ranging from 40–150 µm and heights of 20–35 µm and has proven its potential to investigate differentiation of embryonic stem cells [60]. These findings indicate that hydrogel microwell arrays can be used to control cell aggregation, aggregate size and viability, to study factors involved in the responsiveness of different sized aggregates to anti-cancer drugs and the contribution of KLKs. Integrins are integral in mediating cell survival and chemoresistance, in particular α5/β1 integrins [30,31,47]. Hence, we sought to determine α5/β1 integrin mRNA and protein levels in aggregates of varying size upon paclitaxel treatment (100 nM). While no difference in both OV-Vector/OV-KLK cell aggregates without treatment was found, after paclitaxel treatment ITGA5 was increased in aggregates grown in the largest (150–200 × 150–200 µm) microwells, and ITGB1 was enhanced in aggregates, with highest expression levels in OV-KLK cell aggregates grown in the smallest (50–100 × 50–100 µm) microwells (Figure 5(A)). Western blot and densitometrical analyses showed that α5 and β1 integrins were enhanced after paclitaxel treatment in OV-KLK cell aggregates compared to OV-Vector cells, which only had increased α5 integrin in the smallest (50 × 50 µm) and medium sized (150 × 150 µm) microwells (Figure 5(B)). Interestingly, the biggest (200 × 200 µm) microwells resulted in multiple smaller aggregates per microwell (34%), which have the same integrin expression pattern as the aggregates formed in the smallest (50 × 50 µm) microwells. These results suggest that integrin expression is upregulated upon paclitaxel treatment depending on the aggregate size and partially on KLK expression, especially in smaller (50 µm) and larger (150–200 µm) aggregates. Caspases play an important role in apoptosis induced by anti-cancer drugs [61]. In both OV-Vector/OV-KLK cell aggregates, caspase8 expression followed β1 integrin levels in the smallest (50 × 50 µm) and largest (200 × 200 µm) microwells. OV-Vector cell aggregates showed a downregulation of capsase8 in medium sized (100–150 × 100–150 µm) microwells upon paclitaxel treatment (Figure S1). These results imply an involvement of integrins in paclitaxel-induced apoptosis. However, our findings suggest a bi-functional effect of drug treatment: (i) upregulation of integrins to promote cell aggregate survival, and (ii) upregulation of caspase-8 to mediate cell death, further underlining the fine-tuned balance between drug sensitivity and drug resistance. It was shown that the membrane type 1 matrix metalloproteinase (MT1-MMP) regulates ovarian cancer cell aggregation and disaggregation, and its expression level is increased in aggregates relative to flat cell cultures [43]. Ovarian cancer cell aggregates grown within microwells of varying sizes showed MT1-MMP expression in all aggregate sizes independent of KLK expression and paclitaxel treatment (Figure S1). MT1-MMP can be regulated by integrin clustering which was shown to be stimulated by a 3D collagen type I microenvironment [62]. In addition to MT1-MMP activity [43], other factors, such as contractile forces [59], promote cell aggregation. The simultaneous presence of MT1-MMP and integrins in aggregates grown within hydrogel microwell arrays further indicates their interactive relationship within this microarray platform. Figure 5 Altered expression levels as a function of microwell size, KLK expression and paclitaxel treatment. (A) Levels of ITGA5 were a function of microwell size and KLK expression after paclitaxel treatment (100 nM), with highest expression in aggregates grown in the largest (150–200 × 150–200 µm) microwells. No difference in both OV-Vector/OV-KLK cell aggregates without treatment was detected. Levels of ITGB1 were enhanced after paclitaxel administration in both OV-Vector/OV-KLK cell aggregates, with highest expression on OV-KLK cell aggregates in the smallest (50–100 × 50–100 µm) microwells. No difference in both OV-Vector/OV-KLK cell aggregates without treatment was detected. (B) Western blot and densitometrical analyses demonstrated that α5 and β1 integrin expression was enhanced after paclitaxel treatment in OV-KLK cell aggregates compared to OV-Vector cells which only showed an increase of α5 integrin in the smallest (50 × 50 µm) and medium sized (150 × 150 µm) microwells. 3.4. Blocking of Integrin Function Does Not Affect Cell Aggregation It was shown that β1 integrin regulates the formation of ovarian cancer cell aggregates that were generated using the liquid overlay technique [35,55,56]. Hence, we sought to test whether the formation OV-Vector/OV-KLK cell aggregates produced in hydrogel microwell arrays is dependent on β1 integrin by using a functionally blocking antibody (10 µg/mL). Surprisingly, both OV-Vector/OV-KLK cell aggregate number and area was enhanced with increasing microwell size (150–200 × 150–200 µm) upon integrin inhibition, with more (up to three aggregates/well) and larger (12.84–18.18 × 103 cm2) aggregates being formed compared to non-treated conditions. With decreasing microwell size (50–100 × 50–100 µm), only one to two aggregates/well and smaller aggregates (1.30–5.35 × 103 cm2) were formed. In the medium sized microwells (150 × 150 µm), OV-KLK cells formed significantly larger aggregates after integrin inhibition compared to non-treated conditions (Figure 6(A,B)). Figure 6 Cell aggregation in response to functional blocking of integrins. (A) Fluorescent staining of cell aggregates grown within microwells of varying sizes (50 × 50, 100 × 100, 150 × 150, 200 × 200 µm) over 96 h at the end time point of time-lapse microscopy ± inhibition of β1 integrin using a functional blocking antibody (10 µg/mL); F-actin filaments stained red with rhodamine 415-conjugated phalloidin; nuclei stained blue with DAPI. (B) Analyses of both OV-Vector/OV-KLK cell aggregate number and area revealed that with increasing microwell size (150–200 × 150–200 µm) more (up to three aggregates/microwell) and larger (12.84–18.18 × 103 cm2) aggregates were formed upon β1 integrin inhibition compared to decreasing microwell size (50–100 × 50–100 µm; 1.30–5.35 × 103 cm2), with only one to two aggregates/microwell. In the medium sized microwells (150 × 150 µm), OV-KLK cells formed significantly larger aggregates after β1 integrin inhibition compared to non-treated conditions (n = 3; SEM; * P < 0.05). Different to the study by Casey et al. [35], which reported the inhibition of aggregation using the same blocking β1 integrin antibody after 8 h and 24 h in serum-free media, is that we documented the integrin inhibition over 96 h in serum-containing media. Casey et al. [35] showed that after 8 h aggregate formation was inhibited by the blocking β1 integrin antibody, resulting in none or small aggregates. At 24 h, β1 integrin inhibition continued to partially block aggregate formation, resulting in medium to large aggregates. The incomplete inhibition of the β1 integrin at the 24 h time point suggests that if this integrin is inactivated, ovarian cancer cells might possess a compensatory mechanism to facilitate aggregation. However, the antibody might have been internalized over 24 h and 96 h, eventually enabling ovarian cancer cells to aggregate. Moreover, the presence of the serum-containing media allows the continuous proliferation of cells over a longer period of time. It was suggested that β1 integrin mediates the initial formation of cell aggregates and that multiple integrin-ECM interactions, such as αv integrin/vitronectin [57], are involved in this process. Contrary to Casey et al. [35] we hypothesized that aggregation time and technique are important parameters. Casey et al. [35] demonstrated that NIH:OVCAR5 cells formed stable aggregates within 48 h using the liquid overlay method, whereas the OV-MZ-6 cells used in our study formed compact aggregates for up the 120 h within hydrogel microwell arrays. In our previously published work, we demonstrated that OV-MZ-6 cell spheroids proliferated for up to 28 days [8], underlining the robustness of this cell line when combined with a biomimetic hydrogel in a high-throughput system. Although the capacity to form compact aggregates differs between ovarian cancer cell lines [8,35,57], the aggregates formed in bioengineered microenvironment emerge to be similar to those present in the tumor fluid (ascites) of patients. 4. Conclusions When entering the third dimension, investigators need to consider the design of microenvironments for supporting the cell architecture and the capability to conduct such a system in high-throughput. We provide evidence that hydrogel microwell arrays can be engineered to replicate intricate biological functions the tumor microenvironment by allowing aggregation of ovarian cancer cells, and thus, are well suited to decipher the function of cancer-associated proteases and integrins in disease progression and therapy-resistance. Tailor-made hydrogel microwells increase cell aggregation and insensitivity to paclitaxel treatment, in particular in KLK-expressing cancer cells, and thus, representing events seen in patients with metastatic outgrowth. KLK expression in cancer cell aggregates was accompanied with altered integrin levels and integrin-related factors upon paclitaxel treatment. However, blocking of integrin function did not affect cancer cell aggregation, suggesting that the involvement of other cell surface molecules and/or receptors play an important role. In conclusion, the technology platform presented in this study has the potential to provide an alternative screening tool for the efficacy of novel therapeutics specifically targeting multi-cellular aggregates for intra-abdominal intervention of late-stage disease. Acknowledgments The authors are grateful to Sandrine Roy, Manager of the Microscopy Facility of the Diamantina Institute, University of Queensland and co-workers of the Bioimaging and Optics platform at EPFL, Switzerland, for their assistance with time-lapse and fluorescent microscopy techniques. We thank Sylke Hoehnel, Eva Weber and Jeremy Baldwin for their service with the image analyses. This work was supported by the National Health and Medical Research Council (NHMRC) of Australia (#553045; J.A.C., D.W.H.), the Australian Research Council (#DP110103890; D.L., J.A.C., D.W.H.), an Early Career Research Award (D.L.), Institute of Health and Biomedical Innovation, Queensland University of Technology, a Journal of Cell Science Short-Term Travelling Fellowship (D.L.), the Swiss National Science Foundation (#FN205321-112323/1; M.P.L.) and an European Young Investigator Award (S.K., M.P.L.). Supplementary Materials Supplementary File 1 Click here for additional data file. Conflicts of Interest The authors declare no conflict of interest. Appendix Supplementary files. Time-lapse microscopy of aggregation dependent on paclitaxel treatment. Representative time-lapse experiments (avi-files) of aggregates grown under non-treated (2009.07.30_Overlay_s43_KLK) and treated (2009.07.30_Overlay_s53_KLK+Taxol.avi) conditions are shown using a widefield microscope over 96 h, with images taken every 6 h using a 10× air objective. Figure S1 Altered caspase8 expression levels as a function of microwell size, KLK expression and paclitaxel treatment. Caspase8, indicative of apoptosis, was expressed in aggregates grown in the four different microwell sizes (50 × 50, 100 × 100, 150 × 150, 200 × 200 µm) without paclitaxel treatment and upregulated in OV-KLK cell aggregates upon paclitaxel administration. OV-Vector cell aggregates showed a downregulation of capsase8 in medium sized (100–150 × 100–150 µm) microwells upon paclitaxel treatment. Caspase8 levels did not change in both OV-Vector/OV-KLK cell aggregates in the second largest (150 × 150 µm) and largest (200 × 200 µm) microwells respectively upon paclitaxel treatment. Altered MT1-MMP expression levels as a function of microwell size. MT1-MMPwas expressed in all conditions (50 × 50, 100 × 100, 150 × 150, 200 × 200 µm) independent of KLK expression and paclitaxel treatment. ==== Refs References 1. Inman J.L. Bissell M.J. Apical polarity in three-dimensional culture systems: Where to now? J. Biol. 2010 9 2 10.1186/jbiol213 20092610 2. Abbott A. Cell culture: Biology’s new dimension Nature 2003 424 870 872 10.1038/424870a 12931155 3. Debnath J. Brugge J.S. Modelling glandular epithelial cancers in three-dimensional cultures Nat. Rev. Cancer 2005 5 675 688 10.1038/nrc1695 16148884 4. Hebner C. Weaver V.M. Debnath J. Modeling morphogenesis and oncogenesis in three-dimensional breast epithelial cultures Annu. Rev. Pathol. 2008 3 313 339 10.1146/annurev.pathmechdis.3.121806.151526 18039125 5. Pampaloni F. Reynaud E.G. Stelzer E.H. The third dimension bridges the gap between cell culture and live tissue Nat. Rev. Mol. Cell Biol. 2007 8 839 845 10.1038/nrm2236 17684528 6. Yamada K.M. Cukierman E. Modeling tissue morphogenesis and cancer in 3D Cell 2007 130 601 610 10.1016/j.cell.2007.08.006 17719539 7. Griffith L.G. Swartz M.A. Capturing complex 3D tissue physiology in vitro Nat. Rev. Mol. Cell. Biol. 2006 7 211 224 10.1038/nrm1858 16496023 8. Loessner D. Stok K.S. Lutolf M.P. Hutmacher D.W. Clements J.A. Rizzi S.C. Bioengineered 3D platform to explore cell-ECM interactions and drug resistance of epithelial ovarian cancer cells Biomaterials 2010 31 8494 8506 10.1016/j.biomaterials.2010.07.064 20709389 9. Shield K. Ackland M.L. Ahmed N. Rice G.E. Multicellular spheroids in ovarian cancer metastases: Biology and pathology Gynecol. Oncol. 2009 113 143 148 10.1016/j.ygyno.2008.11.032 19135710 10. Lengyel E. Ovarian cancer development and metastasis Am. J. Pathol. 2010 177 1053 1064 10.2353/ajpath.2010.100105 20651229 11. Agarwal R. Kaye S.B. Ovarian cancer: Strategies for overcoming resistance to chemotherapy Nat. Rev. Cancer 2003 3 502 516 10.1038/nrc1123 12835670 12. Bast R.C. Jr. Hennessy B. Mills G.B. The biology of ovarian cancer: New opportunities for translation Nat. Rev. Cancer 2009 9 415 428 10.1038/nrc2644 19461667 13. Mason S.D. Joyce J.A. Proteolytic networks in cancer Trends Cell Biol. 2011 21 228 237 10.1016/j.tcb.2010.12.002 21232958 14. Borgono C.A. Diamandis E.P. The emerging roles of human tissue kallikreins in cancer Nat. Rev. Cancer 2004 4 876 890 10.1038/nrc1474 15516960 15. Ramani V.C. Haun R.S. The extracellular matrix protein fibronectin is a substrate for kallikrein 7 Biochem. Biophys. Res. Commun. 2008 369 1169 1173 10.1016/j.bbrc.2008.03.021 18343220 16. Obiezu C.V. Michael I.P. Levesque M.A. Diamandis E.P. Human kallikrein 4: Enzymatic activity, inhibition, and degradation of extracellular matrix proteins Biol. Chem. 2006 387 749 759 16800736 17. Michael I.P. Sotiropoulou G. Pampalakis G. Magklara A. Ghosh M. Wasney G. Diamandis E.P. Biochemical and enzymatic characterization of human kallikrein 5 (hK5 ), a novel serine protease potentially involved in cancer progression J. Biol. Chem. 2005 280 14628 14635 10.1074/jbc.M408132200 15713679 18. Ghosh M.C. Grass L. Soosaipillai A. Sotiropoulou G. Diamandis E.P. Human kallikrein 6 degrades extracellular matrix proteins and may enhance the metastatic potential of tumor cells Tumor Biol. 2004 25 193 199 10.1159/000081102 19. Witz C.A. Montoya-Rodriguez I.A. Cho S. Centonze V.E. Bonewald L.F. Schenken R.S. Composition of the extracellular matrix of the peritoneum J. Soc. Gynecol. Investig. 2001 8 299 304 10.1016/S1071-5576(01)00122-8 20. Kenny H.A. Kaur S. Coussens L.M. Lengyel E. The initial steps of ovarian cancer cell metastasis are mediated by MMP-2 cleavage of vitronectin and fibronectin J. Clin. Invest. 2008 118 1367 1379 10.1172/JCI33775 18340378 21. Yousef G.M. Diamandis E.P. The human kallikrein gene family: New biomarkers for ovarian cancer Cancer Treat. Res. 2009 149 165 187 10.1007/978-0-387-98094-2_8 19763436 22. Clements J.A. Willemsen N.M. Myers S.A. Dong Y. The tissue kallikrein family of serine proteases: Functional roles in human disease and potential as clinical biomarkers Crit. Rev. Clin. Lab. Sci. 2004 41 265 312 10.1080/10408360490471931 15307634 23. Obiezu C.V. Diamandis E.P. Human tissue kallikrein gene family: Applications in cancer Cancer Lett. 2005 224 1 22 10.1016/j.canlet.2004.09.024 15911097 24. Oikonomopoulou K. Li L. Zheng Y. Simon I. Wolfert R.L. Valik D. Nekulova M. Simickova M. Frgala T. Diamandis E.P. Prediction of ovarian cancer prognosis and response to chemotherapy by a serum-based multiparametric biomarker panel Brit. J. Cancer 2008 99 1103 1113 10.1038/sj.bjc.6604630 18766180 25. Dong Y. Stephens C. Walpole C. Swedberg J.E. Boyle G.M. Parsons P.G. McGuckin M.A. Harris J.M. Clements J.A. Paclitaxel resistance and multicellular spheroid formation are induced by kallikrein-related peptidase 4 in serous ovarian cancer cells in an ascites mimicking microenvironment PLoS One 2013 8 e57056 10.1371/journal.pone.0057056 23451143 26. Dong Y. Tan O.L. Loessner D. Stephens C. Walpole C. Boyle G.M. Parsons P.G. Clements J.A. Kallikrein-related peptidase 7 promotes multicellular aggregation via the α5 β1 integrin pathway and paclitaxel chemoresistance in serous epithelial ovarian carcinoma Cancer Res. 2010 70 2624 2633 20332224 27. Xi Z. Kaern J. Davidson B. Klokk T.I. Risberg B. Trope C. Saatcioglu F. Kallikrein 4 is associated with paclitaxel resistance in ovarian cancer Gynecol. Oncol. 2004 94 80 85 10.1016/j.ygyno.2004.03.044 15262123 28. Loessner D. Quent V.M. Kraemer J. Weber E.C. Hutmacher D.W. Magdolen V. Clements J.A. Combined expression of KLK4, KLK5, KLK6, and KLK7 by ovarian cancer cells leads to decreased adhesion and paclitaxel-induced chemoresistance Gynecol. Oncol. 2012 127 569 578 10.1016/j.ygyno.2012.09.001 22964375 29. Prezas P. Arlt M.J. Viktorov P. Soosaipillai A. Holzscheiter L. Schmitt M. Talieri M. Diamandis E.P. Kruger A. Magdolen V. Overexpression of the human tissue kallikrein genes KLK4, 5, 6, and 7 increases the malignant phenotype of ovarian cancer cells Biol. Chem. 2006 387 807 811 16800744 30. Zutter M.M. Integrin-mediated adhesion: Tipping the balance between chemosensitivity and chemoresistance Adv. Exp. Med. Biol. 2007 608 87 100 10.1007/978-0-387-74039-3_6 17993234 31. Desgrosellier J.S. Cheresh D.A. Integrins in cancer: Biological implications and therapeutic opportunities Nat. Rev. Cancer 2010 10 9 22 10.1038/nrc2748 20029421 32. Cabodi S. del Pilar Camacho-Leal M. Di Stefano P. Defilippi P. Integrin signalling adaptors: Not only figurants in the cancer story Nat. Rev. Cancer 2010 10 858 870 10.1038/nrc2967 21102636 33. Sawada K. Mitra A.K. Radjabi A.R. Bhaskar V. Kistner E.O. Tretiakova M. Jagadeeswaran S. Montag A. Becker A. Kenny H.A. Peter M.E. Ramakrishnan V. Yamada S.D. Lengyel E. Loss of E-cadherin promotes ovarian cancer metastasis via α5 -integrin, which is a therapeutic target Cancer Res. 2008 68 2329 2339 10.1158/0008-5472.CAN-07-5167 18381440 34. Shield K. Riley C. Quinn M.A. Rice G.E. Ackland M.L. Ahmed N. α2β1 integrin affects metastatic potential of ovarian carcinoma spheroids by supporting disaggregation and proteolysis J. Carcinog. 2007 6 11 10.1186/1477-3163-6-11 17567918 35. Casey R.C. Burleson K.M. Skubitz K.M. Pambuccian S.E. Oegema T.R. Jr. Ruff L.E. Skubitz A.P. β1-integrins regulate the formation and adhesion of ovarian carcinoma multicellular spheroids Am. J. Pathol. 2001 159 2071 2080 10.1016/S0002-9440(10)63058-1 11733357 36. Cordey M. Limacher M. Kobel S. Taylor V. Lutolf M.P. Enhancing the reliability and throughput of neurosphere culture on hydrogel microwell arrays Stem Cells 2008 26 2586 2594 10.1634/stemcells.2008-0498 18669905 37. Roccio M. Gobaa S. Lutolf M.P. High-throughput clonal analysis of neural stem cells in microarrayed artificial niches Integr. Biol. 2012 4 391 400 10.1039/c2ib00070a 38. Gobaa S. Hoehnel S. Roccio M. Negro A. Kobel S. Lutolf M.P. Artificial niche microarrays for probing single stem cell fate in high throughput Nat. Methods 2011 8 949 955 10.1038/nmeth.1732 21983923 39. Mobus V. Gerharz C.D. Press U. Moll R. Beck T. Mellin W. Pollow K. Knapstein P.G. Kreienberg R. Morphological, immunohistochemical and biochemical characterization of 6 newly established human ovarian carcinoma cell lines Int. J. Cancer 1992 52 76 84 10.1002/ijc.2910520115 1500230 40. Dumontet C. Jordan M.A. Microtubule-binding agents: A dynamic field of cancer therapeutics Nat. Rev. Drug Discov. 2010 9 790 803 10.1038/nrd3253 20885410 41. ImageJ Available online:http://rsb.info.nih.gov/ij/ (accessed on 2 July 2013) 42. Imaris Available online:http://www.bitplane.com (accessed on 2 July 2013) 43. Moss N.M. Barbolina M.V. Liu Y. Sun L. Munshi H.G. Stack M.S. Ovarian cancer cell detachment and multicellular aggregate formation are regulated by membrane type 1 matrix metalloproteinase: A potential role in I. p. metastatic dissemination Cancer Res. 2009 69 7121 7129 10.1158/0008-5472.CAN-08-4151 19706774 44. Hutmacher D.W. Loessner D. Rizzi S. Kaplan D.L. Mooney D.J. Clements J.A. Can tissue engineering concepts advance tumor biology research? Trends Biotechnol. 2010 28 125 133 10.1016/j.tibtech.2009.12.001 20056286 45. Hirschhaeuser F. Menne H. Dittfeld C. West J. Mueller-Klieser W. Kunz-Schughart L.A. Multicellular tumor spheroids: An underestimated tool is catching up again J. Biotechnol. 2010 148 3 15 20097238 46. Zietarska M. Maugard C.M. Filali-Mouhim A. Alam-Fahmy M. Tonin P.N. Provencher D.M. Mes-Masson A.M. Molecular description of a 3D in vitro model for the study of epithelial ovarian cancer (EOC) Mol. Carcinog. 2007 46 872 885 10.1002/mc.20315 17455221 47. Helleman J. Jansen M.P. Burger C. van der Burg M.E. Berns E.M. Integrated genomics of chemotherapy resistant ovarian cancer: A role for extracellular matrix, TGFbeta and regulating microRNAs Int. J. Biochem. Cell Biol. 2010 42 25 30 19854294 48. Frankel A. Buckman R. Kerbel R.S. Abrogation of taxol-induced G2-M arrest and apoptosis in human ovarian cancer cells grown as multicellular tumor spheroids Cancer Res. 1997 57 2388 2393 9192815 49. Minchinton A.I. Tannock I.F. Drug penetration in solid tumors Nat. Rev. Cancer 2006 6 583 592 10.1038/nrc1893 16862189 50. Hehlgans S. Haase M. Cordes N. Signalling via integrins: Implications for cell survival and anticancer strategies Biochim. Biophys. Acta 2007 1775 163 180 17084981 51. Sood A.K. Coffin J.E. Schneider G.B. Fletcher M.S. DeYoung B.R. Gruman L.M. Gershenson D.M. Schaller M.D. Hendrix M.J. Biological significance of focal adhesion kinase in ovarian cancer: Role in migration and invasion Am. J. Pathol. 2004 165 1087 1095 10.1016/S0002-9440(10)63370-6 15466376 52. Judson P.L. He X. Cance W.G. van Le L. Overexpression of focal adhesion kinase, a protein tyrosine kinase, in ovarian carcinoma Cancer 1999 86 1551 1556 10.1002/(SICI)1097-0142(19991015)86:6<1551::AID-CNCR23>3.0.CO;2-P 10526262 53. Palazzo A.F. Eng C.H. Schlaepfer D.D. Marcantonio E.E. Gundersen G.G. Localized stabilization of microtubules by integrin- and FAK-facilitated Rho signaling Science 2004 303 836 839 10.1126/science.1091325 14764879 54. Halder J. Landen C.N. Jr. Lutgendorf S.K. Li Y. Jennings N.B. Fan D. Nelkin G.M. Schmandt R. Schaller M.D. Sood A.K. Focal adhesion kinase silencing augments docetaxel-mediated apoptosis in ovarian cancer cells Clin. Cancer Res. 2005 11 8829 8836 10.1158/1078-0432.CCR-05-1728 16361572 55. Burleson K.M. Casey R.C. Skubitz K.M. Pambuccian S.E. Oegema T.R. Jr. Skubitz A.P. Ovarian carcinoma ascites spheroids adhere to extracellular matrix components and mesothelial cell monolayers Gynecol. Oncol. 2004 93 170 181 10.1016/j.ygyno.2003.12.034 15047232 56. Burleson K.M. Boente M.P. Pambuccian S.E. Skubitz A.P. Disaggregation and invasion of ovarian carcinoma ascites spheroids J. Transl. Med. 2006 4 6 10.1186/1479-5876-4-6 16433903 57. Kellouche S. Fernandes J. Leroy-Dudal J. Gallet O. Dutoit S. Poulain L. Carreiras F. Initial formation of IGROV1 ovarian cancer multicellular aggregates involves vitronectin Tumor Biol. 2010 31 129 139 10.1007/s13277-010-0017-9 58. Burleson K.M. Hansen L.K. Skubitz A.P. Ovarian carcinoma spheroids disaggregate on type I collagen and invade live human mesothelial cell monolayers Clin. Exp. Metastasis 2004 21 685 697 10.1007/s10585-004-5768-5 16035613 59. Sodek K.L. Ringuette M.J. Brown T.J. Compact spheroid formation by ovarian cancer cells is associated with contractile behavior and an invasive phenotype Int. J. Cancer 2009 124 2060 2070 10.1002/ijc.24188 19132753 60. Karp J.M. Yeh J. Eng G. Fukuda J. Blumling J. Suh K.Y. Cheng J. Mahdavi A. Borenstein J. Langer R. Khademhosseini A. Controlling size, shape and homogeneity of embryoid bodies using poly(ethylene glycol) microwells Lab Chip 2007 7 786 794 10.1039/b705085m 17538722 61. Cory S. Adams J.M. The Bcl2 family: Regulators of the cellular life-or-death switch Nat. Rev. Cancer 2002 2 647 656 10.1038/nrc883 12209154 62. Barbolina M.V. Stack M.S. Membrane type 1-matrix metalloproteinase: Substrate diversity in pericellular proteolysis Semin. Cell Dev. Biol. 2008 19 24 33 10.1016/j.semcdb.2007.06.008 17702616

PMC005xxxxxx/PMC5003463.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2030170microarrays-02-00170CorrectionCorrection: Gan, L.; Denecke, B. Profiling Pre-MicroRNA and Mature MicroRNA Expressions Using a Single Microarray and Avoiding Separate Sample Preparation. Microarrays 2013, 2, 24-33 Gan Lin Denecke Bernd *Interdisciplinary Centre for Clinical Research Aachen, RWTH Aachen University, Pauwelstr. 30, 52074 Aachen, Germany; E-Mail: lgan@ukaachen.de* Author to whom correspondence should be addressed; E-Mail: bernd.denecke@rwth-aachen.de; Tel.: +49-241-80-89918; Fax: +49-241-80-82124.24 6 2013 9 2013 2 3 170 170 30 5 2013 03 6 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/). ==== Body It came to our attention that a paper has recently been published concerning one of the GEO datasets (GSE34413) we cited in our published paper [1]. The original reference (reference 27) cited for this dataset leads to a paper about a similar study from the same research group [2]. In order to provide readers with exact citation information, we would like to update reference 27 in our previous paper to the new published paper concerning GSE34413 [3]. The authors apologize for this inconvenience. ==== Refs References 1. Gan L. Denecke B. Profiling pre-microrna and mature microrna expressions using a single microarray and avoiding separate sample preparation Microarrays 2013 2 24 33 10.3390/microarrays2010024 2. Kleiber M.L. Laufer B.I. Wright E. Diehl E.J. Singh S.M. Long-term alterations to the brain transcriptome in a maternal voluntary consumption model of fetal alcohol spectrum disorders Brain Res. 2012 1458 18 33 10.1016/j.brainres.2012.04.016 22560501 3. Laufer B.I. Mantha K. Kleiber M.L. Diehl E.J. Addison S.M. Singh S.M. Long-lasting alterations to DNA methylation and ncRNAs could underlie the effects of fetal alcohol exposure in mice Dis. Model. Mech. 2013 10.1242/dmm.010975

PMC005xxxxxx/PMC5003464.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2030228microarrays-02-00228ReviewImproving Pathological Assessment of Breast Cancer by Employing Array-Based Transcriptome Analysis Mihály Zsuzsanna 1Győrffy Balázs 2*1 1st Deptartment of Pediatrics, Semmelweis University, Budapest H-1083, Hungary; E-Mail: zsuzsannamihaly@gmail.com2 Research Laboratory of Pediatrics and Nephrology, Hungarian Academy of Sciences, Budapest H-1083, Hungary* Author to whom correspondence should be addressed; E-Mail: gyorffy@gyer1.sote.hu; Tel.: +36-1-459-1500 (ext. 52792).29 8 2013 9 2013 2 3 228 242 29 7 2013 17 8 2013 22 8 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Breast cancer research has paved the way of personalized oncology with the introduction of hormonal therapy and the measurement of estrogen receptor as the first widely accepted clinical biomarker. The expression of another receptor—HER2/ERBB2/neu—was initially a sign of worse prognosis, but targeted therapy has granted improved outcome for these patients so that today HER2 positive patients have better prognosis than HER2 negative patients. Later, the introduction of multigene assays provided the pathologists with an unbiased assessment of the tumors’ molecular fingerprint. The recent FDA approval of complete microarray pipelines has opened new possibilities for the objective classification of breast cancer samples. Here we review the applications of microarrays for determining ER and HER2 status, molecular subtypes as well as predicting prognosis and grade for breast cancer patients. An open question remains the role of single genes within such signatures. Openly available microarray datasets enable the execution of an independent cross-validation of new marker and signature candidates. In summary, we review the current state regarding clinical applications of microarrays in breast cancer molecular pathology. breast cancermicroarraymolecular subtypeprognosispredictionbiomarker ==== Body 1. Introduction Today, pathological evaluation is still a gold standard for breast cancer diagnosis. Besides the routine assessment it includes the investigation of three molecular markers: estrogen receptor (ESR1), progesterone receptor (PGR) and HER2 receptor. The expression of ESR1 is used to predict response to endocrine therapies, which deliver improved outcome for ESR1 positive patients while no response can be expected in ESR1 negative patients [1]. We must note that only half of ESR1 positive patients will actually benefit from endocrine therapy [2]. PGR, a gene regulated by ESR1, can be used as a surrogate marker for ESR1 status determination. HER2 is used to predict response to trastuzumab, and—similarly to ESR1—about half of HER2 positive patients will respond to anti-HER2 therapy [3]. HER2 positivity was once a sign of worse prognosis, but since the introduction of targeted therapy the expectations have changed: today patients with HER2 positive disease can actually expect a better survival than those with a negative cancer [4]. These markers are also termed negative biomarkers, because they can only be employed to predict the lack of response. At the same time, there are numerous studies questioning the reliability of pathological evaluation. The assessment of ESR1 expression is significantly different among various laboratories [5]. When comparing laboratories from 26 countries, the results for ESR1 IHC were erroneous in up to 30–60% of the analyses [6]. In another comparison of 105 laboratories by an external quality assessment, only 36% delivered reliable results [7]. Although the reproducibility was good for all antibodies used for ESR1 status determination, the different ESR1 antibodies and heat-induced epitope retrieval could influence the final determination of ESR1-positivity [8]. The interobserver agreement in HER2 status determination was questioned by comparing the diagnosis of nine pathologists for the same sample [9]. Similar discrepancies were observed for tumor grade: concordant vote by three pathologists was only 43% [10]. When comparing the diagnoses of two independent pathologists in 405 node negative patients, a discrepancy of 20% was observed [11]. These results were the main driving force behind the development of the first multigene assays enabling an unbiased determination of pathological parameters. The high concordance between microarrays and RT-PCR [12] and the correlation between different microarray platforms [13] has been documented previously. Moreover, the application of unambiguous mapping between genes and probes within one array platform enables the consistent and reliable measurement of gene expression [14]. Based on these observations, microarrays promised an optimal alternative strategy to acquire the diagnoses needed in the pathological determination of breast cancer biomarkers. 2. The Discovery of Molecular Subtypes It has been established that the IHC-based analysis of classical markers can define new subcategories [15] which can differ not only in the proper diagnosis but also in prognosis. In 2001, the intrinsic molecular subtypes including luminal A, luminal B, luminal C, basal-like and HER2 positive cohorts of breast cancer were proposed by Sorlie et al. [16]. After validation of the subtypes in three independent gene-sets, the molecular subtypes were settled to luminal A, luminal B, basal-like, HER2 positive and normal-breast-like [17]. Since response to various systemic therapies differs tremendously among these molecular subgroups, the clinical decision making for the appropriate therapy is also affected. The molecular classification comprises two subtypes (luminal A and B) for ESR1 positive tumors. These luminal subtypes express cytokeratins 8/18, ESR1 and genes associated with an active ESR1 pathway. The luminal A subtype expresses low proliferation rates and it is associated with good prognosis contrary to luminal B, which has high proliferation rates and higher histological grade with worse prognosis. When comparing all four subtypes in over 2,000 patients, luminal A tumors had the lowest rate of relapse while luminal B, HER2 positive, and basal-like subtypes were associated with an increased risk of relapse [18]. In an independent analysis of the patient samples originally used by Sorlie et al. higher drug sensitivity among luminal A patients was suggested to provide the basis for better patients survival as compared to luminal B patients [19]. Tumors of the HER2 positive subtype overexpress HER2 and genes associated with the HER2 amplicon and the HER2 pathways but they are ESR1 negative. Basal-like tumors can be characterized by cytokeratin 5, 17, caveolin 1 and 2, nestin, CD44 and EGFR expression and have the worst prognosis among all subtypes [20]. Furthermore, there is an overlap in definition between triple-negative breast cancer and the basal subtype due to the triple-negative profile of all basal samples [21]. Meanwhile a study warned that identification of luminal cancers and normal breast-like cancers by visual inspection of dendrograms obtained from hierarchical cluster analysis shows suboptimal levels of interobserver agreement while the identification of basal-like and HER2 cancers showed almost perfect interobserver agreement rates [22]. In contrast, a study validated the molecular subtypes in various microarray platforms and confirmed high reproducibility of the classification [23]. Although most of the classical histological types of breast cancer can be correlated to the molecular subtypes, the adenoid cystic and medullar carcinomas display basal-like signature despite the contradictory prognosis of the molecular and histological subtype [24]. Consequently, it seems that molecular subtypes might be divided into additional subgroups with further data from transcriptomic analyses. For example, the lack of ESR1 or PGR receptors in luminal A subtype can define new subgroups with unique clinicopathologic characteristics [25]. Further molecular markers are capable to estimate prognosis in a subtype-independent manner using claudin expression [26]. The subtypes have been extended by a “claudin low” group in which all five claudins display low expression [27]. The claudin-low subtype was a frequent phenomenon in metaplastic and basal-like breast cancer and was a strong predictor of disease recurrence [27]. Other data suggest subtype-specific differences in the relevance of proliferation-associated genes in addition to MKI67 [28]. In summary, the microarray-based characterization of the molecular subtypes can provide more detailed and individualized classification of the patients, making individual patient-tailored therapy possible. However, many of the additional biomarkers identified within the established subtypes have still to be converted from the transcriptomic data into guidelines. 3. Determination of Receptor Status Although IHC remains the gold standard for receptor status determination, a more reliable assessment using Affymetrix microarrays was established by Gong et al. [29]. In this, the receptor status determination is performed using Affymetrix HGU133x platforms by using a cutoff of 1,150 (MAS5 normalized value) for HER2 and 500 for ESR1. The ESR1 and HER2 mRNA expression quantification by microarrays correlate significantly with the corresponding clinical receptor status. Moreover, the array-based results are highly reproducible and are less influenced by the tissue acquisition method. In an analysis comparing microarrays and IHC, the arrays delivered a highly concordant, objective and quantitative assessment of tumor receptor status [30]. Although there is a high correlation between biomarker scoring by IHC and by gene expression, the gene expression determinations for ESR1 and ERBB2 status had higher prognostic power [31]. While the price of a microarrays has fallen in the past years, IHC or FISH still remain cheaper in the case of only four (HER2, ESR1, PGR, MKI67) examined markers. However, the cost-effectiveness depends on the eligible number of biomarkers used to classify the patients regarding prognosis and therapy selection and the relative price of an array-based diagnostic test as such is lower when additional markers are included. By employing microarrays and/or IHC, a number of additional markers have been proposed for predicting response to hormonal therapy in breast cancer including MAPT [32], SLC7A5 [33], TP53 [33], DRG1 [33], CXCL10 [34], MT1 [35], and many more. These markers have been recently evaluated in two large pooled transcriptomic databases where PGR, MAPT and SLC7A5 have been identified as the most promising biomarkers for predicting survival after hormonal therapy [36]. In addition to single genes, epigenetic changes are also increasingly linked to cancer pathogenesis [37]. Among them is the reprogramming of the chromatin landscape which was also linked to endocrine resistance in breast cancer [38]. The development of resistance to endocrine therapy is a slow process involving extensive transcriptional reprogramming reminiscent of cell fate commitment [39]. However, in such scenarios the measurement of the entire transcriptome is not necessary as gene expression signatures of selected pathways can be representative for chromatin reprogramming as it was demonstrated for NOTCH-PBX1 activity [38]. In the near future simultaneous determination of ESR1 and HER2 status with additional marker genes will most probably be part of multigenic breast cancer classification. Even today there are multigene assays employing HER2 and ESR1 with a high significance in the applied scoring for endocrine-treated node-negative breast cancer patients [10]. In addition, an internet-based classification software is already available which is capable of deriving a hormone receptor status by an automated processing of commonly used commercially available genome-wide microarrays [40]. 4. Prognostic Signatures For specific clinico-pathological cohorts of breast cancer patients, new multigenic signatures developed using microarrays (MapQuant DX [41], Mammaprint [42]) and qPCR (Oncotype DX [10], Theros Breast Cancer Gene Expression Ratio Assay [43], PAM50 [44]) are already available as support tools for clinical decision-making. Of these, the first one approved by the FDA as an in vitro diagnostic multivariate index assay (IVDMIA), was the 70-gene signature (Mammaprint) [42]. The 70-gene signature provides a risk prediction (low or high group) of distant recurrence after surgery in lymph node negative patients regardless of their ESR1 status or prior treatment [18]. In the case of node-negative patients, a prediction of low risk equals to a 13% risk of distant metastasis within 10 years while high risk equals to a 56% risk. Consequently, low risk patients may avoid chemotherapy. The scoring of the 70-gene signature was validated in the TRANSBIG trial [42] and in an independent study of the Massachusetts General Hospital [45]. Oncotype DX is a qPCR based test measuring the expression of 21 genes (including 5 housekeeping genes), which are used to calculate a recurrence score ranging between 1 and 100 [10]. The recurrence score provides a risk prediction of recurrence in 10 years for node negative, ESR1 positive patients [10]. A high recurrence score (over 30) is predictive for worse prognosis but it indicates a better response to chemotherapy. The test was also validated in independent clinical studies [46,47,48]. The reproducibility of qPCR-based results was also confirmed by employing Affymetrix microarrays [40]. However, genes have various weights in the analysis and known clinical parameters (ESR1, HER2) have the highest contribution to the final score. Without the application of these weights, the expression signature of the 21 genes is not capable of predicting survival as it was demonstrated in 1,079 breast cancer patients [49]. In addition, when assessing concordance between Oncotype DX results and IHC/FISH, an unacceptably high false negative rate (58%) of Oncotpye DX was observed while all patients designated by Oncotype DX as HER2 positive were also positive by IHC/FISH [50]. To date, the remaining tests have gained limited clinical use. MapQuant DX signature evaluates 98 genes in a molecular diagnostic test for estrogen receptor positive, grade II breast cancer patients to measure tumor proliferation, the risk of metastasis and response to chemotherapy [41]. The H:I ratio, also known as Theros Breast Cancer index is a molecular grade index for ESR1-positive breast cancer patients treated with tamoxifen [43]. It can stratify tamoxifen-treated and untreated breast cancer patients into high and low risk of recurrence which cohorts differ in outcome within 5 years. The H:I ratio was also evaluated in additional cohorts of patients [51,52], however discrepant results were documented in a different study [53]. The Invasive Gene Signature (IGS) [54] and the HER2-Drived Prognostic Predictor (HDP) [55] are based on microarray assays, while the Celera Metastasis Score [56] and the BreastOnc DX [57] are PCR based. We must note here the requirement to adequately identify molecular subtypes of breast cancer which is not achieved by routine IHC panel alone. The qPCR based PAM50—Breast Bioclassifier—test can define these subtypes from FFPE [44] and can therefore provide estimation of prognosis as well [58]. In addition, it is capable of predicting a response to neoadjuvant endocrine therapy of ER-positive tumors [59]. We have summarized the above-described tests in Table 1. microarrays-02-00228-t001_Table 1Table 1 Biomarkers using conventional methods and multigene classification tools for breast cancer. Indication IHC/FISH/RT-PCR-based tests Ref. Microarray-based tests Ref. Endrocine therapy ESR1 * (I) (P) and PGR * (I) (P) ESR1 [29] H:I ratio (tamoxifen) [43] RecurrenceOnline [40] Targeted therapy HER2 * (I) (F) (P) HER2 [29] RecurrenceOnline [40] Grade FoxTop (P) [60] MapQuant DX [41] Chemotherapy response PAM50 (P) [44] MapQuant DX [41] Oncotype DX (P) [61] Prognosis Oncotype DX (P) CURIO (I) Celera Metastasis Score (P) BreastOncPx (P) [10] [26] [56] [57] 70 gene * RecurrenceOnline IGS HDP Rotterdam signature [42] [40] [54] [55] [62] * FDA approved diagnostic biomarkers, (I): IHC, (F): FISH, (P): RT-PCR, ESR1: Estrogen Receptor, PGR: Progesterone Receptor, HER2: Human Epidermal Growth Factor Receptor 2. 5. Solitary Genes of the Signatures The value of single genes within gene expression signatures remains an open question. We will discuss some studies in which biomarker candidate discovery performed by transcriptome analyses was followed by in vitro investigation and clinical validation of its prognostic or predictive power in independent patient cohorts. For endocrine therapy response, MYC has been identified as a key molecule in estrogen independent growth utilizing gene expression signature of long term estrogen-deprived cells. Its potential to predict poor outcome for patients following tamoxifen administration was then evaluated in three independent patient cohorts including 164, 181, and 298 patients [63]. Another biomarker candidate for endocrine therapy response prediction is PUMA. In vitro results indicated the role of PUMA in apoptotic dysregulation which could have an impact on progression and therapy response. Its association with breast cancer specific death and worse outcome of tamoxifen treated patients was confirmed in the case of 148 and 201 patients, respectively [64]. Similar studies were also conducted for chemotherapy response. The PSMB7 gene was discovered as a new biomarker candidate for anthracycline resistance by comparing drug resistant and -sensitive cell lines. Its driver role in doxorubicin resistance was assessed by a combination of gene silencing and drug treatment. Its high expression was linked to unfavorable prognosis in 1,592 breast cancer patients [65]. Recently, new biomarker candidates including the HMGA2 gene of the beta-catenin signal transduction pathway were discovered by ChIP analysis of 55 patients. Its expression predicted relapse-free survival and metastasis in 82 triple negative breast cancer patients [66]. HMGA2 was previously linked to multidrug resistance by a microarray analysis of cancer cell lines [67]. The expression of the IGFBP3 receptor had a strong prognostic value for predicting relapse-free survival time in 478 basal type breast cancer patients. In vitro results also show that IGFBP3 has a significant role in cell proliferation control [68]. The CDK8 gene was identified as a mediator of chemotherapy-induced tumor-promoting paracrine activities in CMV-GFP and NFKB-GFP reporter cell lines. The association between CDK8 and survival was further confirmed in 2897 breast cancer patients [69]. The ATIP3 was identified as predicting overall survival in metastatic breast cancer patients using microarray data of 150 patients and was validated in 162 samples. The role of ATIP3 was investigated in vitro via silencing and in vivo via mouse experiments showing that ATIP3 delays metastatic progression and limits the growth of metastases [70]. Many of these genes could be potential targets of personalized treatments of metastatic breast cancer. However, to date these predictor candidates have not yet made their way into therapeutic protocols, mostly because they are limited to selected drugs while current systemic protocols use a multi-target approach employing several systemic therapy agents sequentially or simultaneously. Most probably findings related to single genes will be even more valuable for the pharmacological drug research when utilizing new molecular targets. 6. Validation Studies The publication of numerous microarray datasets in large-scale public repositories including GEO [71] and EGA [72] enables the cross-validation of findings obtained in different studies. By integrating several independent datasets into a single database it is possible to investigate sub-cohorts which could not include sufficient number of patients by evaluating any of the original datasets alone. Such an integrated online database for evaluating survival-associated biomarkers has already been constructed for breast cancer [73]. In the most recent version of the tool, in addition to single genes combination, signatures can also be evaluated in one analysis [74]. An important issue in such meta-analyses is the reproducibility of previously published, conventional, pathology based biomarkers and subtypes. An example for these is the above discussed molecular subtype. In a recent large scale analysis Park et al., established the subtype distribution in 1,006 patients [75]. In GEO, six datasets are available with determined molecular subtypes, including GSE1456 [76], GSE21653 [77], GSE25066 [78], GSE20711 [79], GSE31519 [80] and GSE17907 [81]. The re-computation of subtypes in these datasets is possible with the use of current StGallen guidelines and validated probe sets of Affymetrix HGU133x arrays (Table 2). By comparing the author reported prevalence of subtypes in the six independent datasets, the immunhistochemistry-results of Park et al. [75], and the re-computed subtypes for patient samples included in the Kaplan-Meier plotter [73], the highest concordance can be observed between the array-based and the IHC based determinations (see Figure 1). In these, basal tumors show the highest overlap and luminal B tumors show the highest discordance. At the same time, in the author-reported prevalence basal tumors are massively over-represented. The most probable reason for this is the predominant investigation of triple negative breast cancers as these patients have the worst prognosis among all molecular subtypes. microarrays-02-00228-t002_Table 2Table 2 The determination of molecular subtypes can be performed using Affymetrix microarrays and the appropriate classification by using the expression of three genes. The clinical difference between the two Luminal B cohorts is still not settled. Probe: Affymetrix HGU133A or HGU133plus2 arrays, N.R.: not relevant ESR1: Estrogen Receptor, HER2: Human Epidermal Growth Factor Receptor 2, MKI67: antigen identified by monoclonal antibody Ki-67. Gene Probe Cutoff Basal Luminal A Luminal B HER2 positive ESR1 205225_at 500 Low High High High Low HER2 216836_s_at 4800 Low Low Low High High MKI67 212021_s_at 470 N.R. Low High N.R. N.R. Figure 1 Reproducibility of the molecular subtypes in three different approaches. Author: as published in GEO by the authors of six datasets. Computed: distribution for all arrays in the database of the Kaplan-Meier plotter [82]. The receptor status determination was performed using Affymetrix HGU133x platforms by using a cutoff of 1,150 (MAS5 normalized value) for HER2 and 500 for ESR1. Park: as in reference [75]. A related question is dichotomization—in other words identification of the optimal cutoff points to determine two cohorts using continuous gene expression data. Researchers often employ packages like SPSS (SPSS Inc., Chicago, IL, USA), GraphPad Prism (GraphPad Software Inc., La Jolla, CA, USA) or WinStat (R. Fitch Software, Bad Krozingen, Germany) to correlate biomarkers with the outcomes or survival. Unfortunately, neither of these packages includes algorithms for cutoff optimization. For the generation of Figure 1, we used established cutoffs for ESR1 and HER2 which have been extensively validated [29,40]. For the proliferation marker MKI67, the median expression of the entire database was used. A more advanced determination of the optimal cutoff is possible by employing online-accessible algorithms specifically designed for cutoff determination [83]. The generated results of classification using multigene signatures are on different scales for 3-category classifiers (for example Oncotpye DX, RecurrenceOnline) and 2-category classifiers (for example Mammaprint, Genomic Grade Index). It is difficult to make comparisons between 3- and 2-category classifiers without biasing against the latter as the application of an intermediate outcome can significantly improve the correlation when only the high- and low-risk cohorts are compared. Although these statistical issues influence the performance of individual tests, they do not cast doubt on the overall performance of the developed diagnostic tools. 7. Conclusions In our review, we summarized the current state of microarray-based advances for breast cancer pathological diagnosis. The true strength of multigene assays lies in the easily accessible independent validation. At present, there are tools available which integrate various diagnostic analyses into one test [40]. Multigene tests have already made their way into the diagnostic procedure and this process can be expected to speed up in the near future. Acknowledgments Our work was supported by the OTKA PD 83154 grant, by the Predict project (grant no. 259303 of the EU Health.2010.2.4.1.-8 call) and by the KTIA EU_BONUS_12-1-2013-0003 grant. Conflicts of Interest The authors declare no conflict of interest. ==== Refs References 1. Davies C. Godwin J. Gray R. Clarke M. Cutter D. Darby S. McGale P. Pan H.C. Taylor C. Wang Y.C. Relevance of breast cancer hormone receptors and other factors to the efficacy of adjuvant tamoxifen: Patient-level meta-analysis of randomised trials Lancet 2011 378 771 784 21802721 2. Early Breast Cancer Trialists’ Collaborative Group Tamoxifen for early breast cancer: An overview of the randomised trials Lancet 1998 351 1451 1467 10.1016/S0140-6736(97)11423-4 9605801 3. Brufsky A. Trastuzumab-based therapy for patients with HER2-positive breast cancer: From early scientific development to foundation of care Am. J. Clin. Oncol. 2010 33 186 195 19675448 4. Dawood S. Broglio K. Buzdar A.U. Hortobagyi G.N. Giordano S.H. Prognosis of women with metastatic breast cancer by HER2 status and trastuzumab treatment: An institutional-based review J. Clin. Oncol. 2010 28 92 98 10.1200/JCO.2008.19.9844 19933921 5. Layfield L.J. Goldstein N. Perkinson K.R. Proia A.D. Interlaboratory variation in results from immunohistochemical assessment of estrogen receptor status Breast J. 2003 9 257 259 10.1046/j.1524-4741.2003.09325.x 12752643 6. Rhodes A. Jasani B. Balaton A.J. Barnes D.M. Miller K.D. Frequency of oestrogen and progesterone receptor positivity by immunohistochemical analysis in 7,016 breast carcinomas: Correlation with patient age, assay sensitivity, threshold value, and mammographic screening J. Clin. Pathol. 2000 53 688 696 10.1136/jcp.53.9.688 11041059 7. Rhodes A. Jasani B. Balaton A.J. Barnes D.M. Anderson E. Bobrow L.G. Miller K.D. Study of interlaboratory reliability and reproducibility of estrogen and progesterone receptor assays in Europe. Documentation of poor reliability and identification of insufficient microwave antigen retrieval time as a major contributory element of unreliable assays Am. J. Clin. Pathol. 2001 115 44 58 10.1309/H905-HYC1-6UQQ-981P 11190807 8. Grabau D.A. Bendahl P.O. Ryden L. Stal O. Ferno M. The prevalence of immunohistochemically determined oestrogen receptor positivity in primary breast cancer is dependent on the choice of antibody and method of heat-induced epitope retrieval—Prognostic implications? Acta Oncol. 2013 10.3109/0284186X.2012.762994 9. Atkinson R. Mollerup J. Laenkholm A.V. Verardo M. Hawes D. Commins D. Engvad B. Correa A. Ehlers C.C. Nielsen K.V. Effects of the change in cutoff values for human epidermal growth factor receptor 2 status by immunohistochemistry and fluorescence in situ hybridization: A study comparing conventional brightfield microscopy, image analysis-assisted microscopy, and interobserver variation Arch. Pathol. Lab. Med. 2011 135 1010 1016 10.5858/2010-0462-OAR 21809992 10. Paik S. Shak S. Tang G. Kim C. Baker J. Cronin M. Baehner F.L. Walker M.G. Watson D. Park T. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer N. Engl. J. Med. 2004 351 2817 2826 10.1056/NEJMoa041588 15591335 11. Kennecke H.F. Speers C.H. Ennis C.A. Gelmon K. Olivotto I.A. Hayes M. Impact of routine pathology review on treatment for node-negative breast cancer J. Clin. Oncol. 2012 30 2227 2231 10.1200/JCO.2011.38.9247 22564990 12. Gyorffy B. Molnar B. Lage H. Szallasi Z. Eklund A.C. Evaluation of microarray preprocessing algorithms based on concordance with RT-PCR in clinical samples PLoS One 2009 4 e5645 10.1371/journal.pone.0005645 19461970 13. Consortium M. Shi L. Reid L.H. Jones W.D. Shippy R. Warrington J.A. Baker S.C. Collins P.J. de Longueville F. Kawasaki E.S. The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements Nat. Biotechnol. 2006 24 1151 1161 10.1038/nbt1239 16964229 14. Li Q. Birkbak N.J. Gyorffy B. Szallasi Z. Eklund A.C. Jetset: Selecting the optimal microarray probe set to represent a gene BMC Bioinformatics 2011 12 474 10.1186/1471-2105-12-474 22172014 15. Sotiriou C. Neo S.Y. McShane L.M. Korn E.L. Long P.M. Jazaeri A. Martiat P. Fox S.B. Harris A.L. Liu E.T. Breast cancer classification and prognosis based on gene expression profiles from a population-based study Proc. Natl. Acad. Sci. USA 2003 100 10393 10398 10.1073/pnas.1732912100 12917485 16. Sorlie T. Perou C.M. Tibshirani R. Aas T. Geisler S. Johnsen H. Hastie T. Eisen M.B. van de Rijn M. Jeffrey S.S. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications Proc. Natl. Acad. Sci. USA 2001 98 10869 10874 10.1073/pnas.191367098 11553815 17. Sorlie T. Tibshirani R. Parker J. Hastie T. Marron J.S. Nobel A. Deng S. Johnsen H. Pesich R. Geisler S. Repeated observation of breast tumor subtypes in independent gene expression data sets Proc. Natl. Acad. Sci. USA 2003 100 8418 8423 10.1073/pnas.0932692100 12829800 18. Wang Y. Yin Q. Yu Q. Zhang J. Liu Z. Wang S. Lv S. Niu Y. A retrospective study of breast cancer subtypes: The risk of relapse and the relations with treatments Breast Cancer Res. Treat. 2011 130 489 498 10.1007/s10549-011-1709-6 21837481 19. Gyorffy B. Serra V. Jurchott K. Abdul-Ghani R. Garber M. Stein U. Petersen I. Lage H. Dietel M. Schafer R. Prediction of doxorubicin sensitivity in breast tumors based on gene expression profiles of drug-resistant cell lines correlates with patient survival Oncogene 2005 24 7542 7551 10.1038/sj.onc.1208908 16044152 20. Colombo P.E. Milanezi F. Weigelt B. Reis-Filho J.S. Microarrays in the 2010s: The contribution of microarray-based gene expression profiling to breast cancer classification, prognostication and prediction Breast Cancer Res. 2011 13 212 10.1186/bcr2890 21787441 21. Valentin M.D. da Silva S.D. Privat M. Alaoui-Jamali M. Bignon Y.J. Molecular insights on basal-like breast cancer Breast Cancer Res. Treat. 2012 134 21 30 10.1007/s10549-011-1934-z 22234518 22. Mackay A. Weigelt B. Grigoriadis A. Kreike B. Natrajan R. A’Hern R. Tan D.S. Dowsett M. Ashworth A. Reis-Filho J.S. Microarray-based class discovery for molecular classification of breast cancer: Analysis of interobserver agreement J. Natl. Cancer Inst. 2011 103 662 673 10.1093/jnci/djr071 21421860 23. Hu Z. Fan C. Oh D.S. Marron J.S. He X. Qaqish B.F. Livasy C. Carey L.A. Reynolds E. Dressler L. The molecular portraits of breast tumors are conserved across microarray platforms BMC Genomics 2006 7 96 10.1186/1471-2164-796 16643655 24. Weigelt B. Horlings H.M. Kreike B. Hayes M.M. Hauptmann M. Wessels L.F. de Jong D. van de Vijver M.J. van’t Veer L.J. Peterse J.L. Refinement of breast cancer classification by molecular characterization of histological special types J. Pathol. 2008 216 141 150 10.1002/path.2407 18720457 25. Park S. Park B.W. Kim T.H. Jeon C.W. Kang H.S. Choi J.E. Hwang K.T. Kim I.C. Lack of either estrogen or progesterone receptor expression is associated with poor survival outcome among luminal A breast cancer subtype Ann. Surg. Oncol. 2013 20 1505 1513 10.1245/s10434-012-2772-x 23192228 26. Szasz A.M. Nemeth Z. Gyorffy B. Micsinai M. Krenacs T. Baranyai Z. Harsanyi L. Kiss A. Schaff Z. Tokes A.M. Identification of a claudin-4 and E-cadherin score to predict prognosis in breast cancer Cancer Sci. 2011 102 2248 2254 10.1111/j.1349-7006.2011.02085.x 21883696 27. Lu S. Singh K. Mangray S. Tavares R. Noble L. Resnick M.B. Yakirevich E. Claudin expression in high-grade invasive ductal carcinoma of the breast: Correlation with the molecular subtype Mod. Pathol. 2013 26 485 495 10.1038/modpathol.2012.187 23222490 28. Milde-Langosch K. Karn T. Muller V. Witzel I. Rody A. Schmidt M. Wirtz R.M. Validity of the proliferation markers Ki67, TOP2A, and RacGAP1 in molecular subgroups of breast cancer Breast Cancer Res. Treat. 2013 137 57 67 10.1007/s10549-012-2296-x 23135572 29. Gong Y. Yan K. Lin F. Anderson K. Sotiriou C. Andre F. Holmes F.A. Valero Booser D. Pippen J.E. Jr. Determination of oestrogen-receptor status and ERBB2 status of breast carcinoma: A gene-expression profiling study Lancet Oncol. 2007 8 203 211 10.1016/S1470-2045(07)70042-6 17329190 30. Roepman P. Horlings H.M. Krijgsman O. Kok M. Bueno-de-Mesquita J.M. Bender R. Linn S.C. Glas A.M. van de Vijver M.J. Microarray-based determination of estrogen receptor, progesterone receptor, and HER2 receptor status in breast cancer Clin. Cancer Res. 2009 15 7003 7011 10.1158/1078-0432.CCR-09-0449 19887485 31. Bastien R.R. Rodriguez-Lescure A. Ebbert M.T. Prat A. Munarriz B. Rowe L. Miller P. Ruiz-Borrego M. Anderson D. Lyons B. PAM50 breast cancer subtyping by RT-qPCR and concordance with standard clinical molecular markers BMC Med. Genomics 2012 5 44 10.1186/1755-8794-5-44 23035882 32. Zoubir M.M.M. Liedtke C. Bidard F. Delaloge S. Corley L. Spielmann M. Pusztai L. André F. Symmans W.F. Predictive biomarkers for preoperative endocrine therapy of stage II-III breast cancer by tissue microarrays J. Clin. Oncol. 2008 26 560 33. Bartlett J.M. Thomas J. Ross D.T. Seitz R.S. Ring B.Z. Beck R.A. Pedersen H.C. Munro A. Kunkler I.H. Campbell F.M. Mammostrat as a tool to stratify breast cancer patients at risk of recurrence during endocrine therapy Breast Cancer Res. 2010 12 R47 10.1186/bcr2604 20615243 34. Hilborn E.S.T. Kot A. Fornander T. Skoog L. Nordenskjöld B. Stål O. Jansson A. The importance of CXCL10 and CXCR3-A in breast cancer Cancer Res. 2011 71 10.1158/0008-5472.SABCS11-P1-06-06 35. Surowiak P. Matkowski R. Materna V. Gyorffy B. Wojnar A. Pudelko M. Dziegiel P. Kornafel J. Zabel M. Elevated metallothionein (MT) expression in invasive ductal breast cancers predicts tamoxifen resistance Histol. Histopathol. 2005 20 1037 1044 16136485 36. Mihaly Z. Kormos M. Lanczky A. Dank M. Budczies J. Szasz M.A. Gyorffy B. A meta-analysis of gene expression-based biomarkers predicting outcome after tamoxifen treatment in breast cancer Breast Cancer Res. Treat. 2013 140 219 232 10.1007/s10549-013-2622-y 23836010 37. Akhtar-Zaidi B. Cowper-Sal-lari R. Corradin O. Saiakhova A. Bartels C.F. Balasubramanian D. Myeroff L. Lutterbaugh J. Jarrar A. Kalady M.F. Epigenomic enhancer profiling defines a signature of colon cancer Science 2012 336 736 739 10.1126/science.1217277 22499810 38. Magnani L. Stoeck A. Zhang X. Lanczky A. Mirabella A.C. Wang T.L. Gyorffy B. Lupien M. Genome-wide reprogramming of the chromatin landscape underlies endocrine therapy resistance in breast cancer Proc. Natl. Acad. Sci. USA 2013 110 E1490 E1499 10.1073/pnas.1219992110 23576735 39. Aguilar H. Sole X. Bonifaci N. Serra-Musach J. Islam A. Lopez-Bigas N. Mendez-Pertuz M. Beijersbergen R.L. Lazaro C. Urruticoechea A. Biological reprogramming in acquired resistance to endocrine therapy of breast cancer Oncogene 2010 29 6071 6083 10.1038/onc.2010.333 20711236 40. Gyorffy B. Benke Z. Lanczky A. Balazs B. Szallasi Z. Timar J. Schafer R. RecurrenceOnline: An online analysis tool to determine breast cancer recurrence and hormone receptor status using microarray data Breast Cancer Res. Treat. 2012 132 1025 1034 10.1007/s10549-011-1676-y 21773767 41. Sotiriou C. Wirapati P. Loi S. Harris A. Fox S. Smeds J. Nordgren H. Farmer P. Praz V. Haibe-Kains B. Gene expression profiling in breast cancer: Understanding the molecular basis of histologic grade to improve prognosis J. Natl. Cancer Inst. 2006 98 262 272 10.1093/jnci/djj052 16478745 42. Van de Vijver M.J. He Y.D. van’t Veer L.J. Dai H. Hart A.A. Voskuil D.W. Schreiber G.J. Peterse J.L. Roberts C. Marton M.J. A gene-expression signature as a predictor of survival in breast cancer N. Engl. J. Med. 2002 347 1999 2009 10.1056/NEJMoa021967 12490681 43. Ma X.J. Wang Z. Ryan P.D. Isakoff S.J. Barmettler A. Fuller A. Muir B. Mohapatra G. Salunga R. Tuggle J.T. A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen Cancer Cell 2004 5 607 616 10.1016/j.ccr.2004.05.015 15193263 44. Parker J.S. Mullins M. Cheang M.C. Leung S. Voduc D. Vickery T. Davies S. Fauron C. He X. Hu Z. Supervised risk predictor of breast cancer based on intrinsic subtypes J. Clin. Oncol. 2009 27 1160 1167 10.1200/JCO.2008.18.1370 19204204 45. Wittner B.S. Sgroi D.C. Ryan P.D. Bruinsma T.J. Glas A.M. Male A. Dahiya S. Habin K. Bernards R. Haber D.A. Analysis of the MammaPrint breast cancer assay in a predominantly postmenopausal cohort Clin. Cancer Res. 2008 14 2988 2993 10.1158/1078-0432.CCR-07-4723 18483364 46. Habel L.A. Shak S. Jacobs M.K. Capra A. Alexander C. Pho M. Baker J. Walker M. Watson D. Hackett J. A population-based study of tumor gene expression and risk of breast cancer death among lymph node-negative patients Breast Cancer Res. 2006 8 R25 10.1186/bcr1412 16737553 47. Esteva F.J. Sahin A.A. Cristofanilli M. Coombes K. Lee S.J. Baker J. Cronin M. Walker M. Watson D. Shak S. Prognostic role of a multigene reverse transcriptase-PCR assay in patients with node-negative breast cancer not receiving adjuvant systemic therapy Clin. Cancer Res. 2005 11 3315 3319 10.1158/1078-0432.CCR-04-1707 15867229 48. Dowsett C.W. Forbes J. Mallon L. Salter J. Cuzick J. Wales C. Forbes J. Mallon L. Salter J. Quinn E. Risk of distant recurrence using Oncotype DX in postmenopausal primary breast cancer patients treated with anastrozole or tamoxifen: A TransATAC study Cancer Res. 2009 69 1059 1061 49. Gyorffy B. Schafer R. Meta-analysis of gene expression profiles related to relapse-free survival in 1079 breast cancer patients Breast Cancer Res. Treat. 2009 118 433 441 10.1007/s10549-008-0242-8 19052860 50. Dabbs D.J. Klein M.E. Mohsin S.K. Tubbs R.R. Shuai Y. Bhargava R. High false-negative rate of HER2 quantitative reverse transcription polymerase chain reaction of the Oncotype DX test: An independent quality assurance study J. Clin. Oncol. 2011 29 4279 4285 10.1200/JCO.2011.34.7963 21990395 51. Ma X.J. Hilsenbeck S.G. Wang W. Ding L. Sgroi D.C. Bender R.A. Osborne C.K. Allred D.C. Erlander M.G. The HOXB13:IL17BR expression index is a prognostic factor in early-stage breast cancer J. Clin. Oncol. 2006 24 4611 4619 10.1200/JCO.2006.06.6944 17008703 52. Jerevall P.L. Brommesson S. Strand C. Gruvberger-Saal S. Malmstrom P. Nordenskjold B. Wingren S. Soderkvist P. Ferno M. Stal O. Exploring the two-gene ratio in breast cancer—Independent roles for HOXB13 and IL17BR in prediction of clinical outcome Breast Cancer Res. Treat. 2008 107 225 234 10.1007/s10549-007-9541-8 17453342 53. Reid J.F. Lusa L. de Cecco L. Coradini D. Veneroni S. Daidone M.G. Gariboldi M. Pierotti M.A. Limits of predictive models using microarray data for breast cancer clinical treatment outcome J. Natl. Cancer Inst. 2005 97 927 930 10.1093/jnci/dji153 15956654 54. Liu R. Wang X. Chen G.Y. Dalerba P. Gurney A. Hoey T. Sherlock G. Lewicki J. Shedden K. Clarke M.F. The prognostic role of a gene signature from tumorigenic breast-cancer cells N. Engl. J. Med. 2007 356 217 226 10.1056/NEJMoa063994 17229949 55. Staaf J. Ringner M. Vallon-Christersson J. Jonsson G. Bendahl P.O. Holm K. Arason A. Gunnarsson H. Hegardt C. Agnarsson B.A. Identification of subtypes in human epidermal growth factor receptor 2—Positive breast cancer reveals a gene signature prognostic of outcome J. Clin. Oncol. 2010 28 1813 1820 10.1200/JCO.2009.22.8775 20231686 56. Cobleigh M.A. Tabesh B. Bitterman P. Baker J. Cronin M. Liu M.L. Borchik R. Mosquera J.M. Walker M.G. Shak S. Tumor gene expression and prognosis in breast cancer patients with 10 or more positive lymph nodes Clin. Cancer Res. 2005 11 8623 8631 10.1158/1078-0432.CCR-05-0735 16361546 57. Tutt A. Wang A. Rowland C. Gillett C. Lau K. Chew K. Dai H. Kwok S. Ryder K. Shu H. Risk estimation of distant metastasis in node-negative, estrogen receptor-positive breast cancer patients using an RT-PCR based prognostic expression signature BMC Cancer 2008 8 339 10.1186/1471-2407-8-339 19025599 58. Chia S.K. Bramwell V.H. Tu D. Shepherd L.E. Jiang S. Vickery T. Mardis E. Leung S. Ung K. Pritchard K.I. A 50-gene intrinsic subtype classifier for prognosis and prediction of benefit from adjuvant tamoxifen Clin. Cancer Res. 2012 18 4465 4472 10.1158/1078-0432.CCR-12-0286 22711706 59. Harvell D.M. Spoelstra N.S. Singh M. McManaman J.L. Finlayson C. Phang T. Trapp S. Hunter L. Dye W.W. Borges V.F. Molecular signatures of neoadjuvant endocrine therapy for breast cancer: Characteristics of response or intrinsic resistance Breast Cancer Res. Treat. 2008 112 475 488 10.1007/s10549-008-9897-4 18327671 60. Szasz A.M. Li Q. Eklund A.C. Sztupinszki Z. Rowan A. Tokes A.M. Szekely B. Kiss A. Szendroi M. Gyorffy B. The CIN4 chromosomal instability qPCR classifier defines tumor aneuploidy and stratifies outcome in grade 2 breast cancer PLoS One 2013 8 e56707 10.1371/journal.pone.0056707 23468873 61. Paik S. Tang G. Shak S. Kim C. Baker J. Kim W. Cronin M. Baehner F.L. Watson D. Bryant J. Gene expression and benefit of chemotherapy in women with node-negative, estrogen receptor-positive breast cancer J. Clin. Oncol. 2006 24 3726 3734 10.1200/JCO.2005.04.7985 16720680 62. Wang Y. Klijn J.G. Zhang Y. Sieuwerts A.M. Look M.P. Yang F. Talantov D. Timmermans M. Meijer-van Gelder M.E. Yu J. Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer Lancet 2005 365 671 679 15721472 63. Miller T.W. Balko J.M. Ghazoui Z. Dunbier A. Anderson H. Dowsett M. Gonzalez-Angulo A.M. Mills G.B. Miller W.R. Wu H. A gene expression signature from human breast cancer cells with acquired hormone independence identifies MYC as a mediator of antiestrogen resistance Clin. Cancer Res. 2011 17 2024 2034 10.1158/1078-0432.CCR-10-2567 21346144 64. Roberts C.G. Millar E.K. O’Toole S.A. McNeil C.M. Lehrbach G.M. Pinese M. Tobelmann P. McCloy R.A. Musgrove E.A. Sutherland R.L. Identification of PUMA as an estrogen target gene that mediates the apoptotic response to tamoxifen in human breast cancer cells and predicts patient outcome and tamoxifen responsiveness in breast cancer Oncogene 2011 30 3186 3197 10.1038/onc.2011.36 21383694 65. Munkacsy G. Abdul-Ghani R. Mihaly Z. Tegze B. Tchernitsa O. Surowiak P. Schafer R. Gyorffy B. PSMB7 is associated with anthracycline resistance and is a prognostic biomarker in breast cancer Br. J. Cancer 2010 102 361 368 10.1038/sj.bjc.6605478 20010949 66. Wend P. Runke S. Wend K. Anchondo B. Yesayan M. Jardon M. Hardie N. Loddenkemper C. Ulasov I. Lesniak M.S. WNT10B/β-catenin signalling induces HMGA2 and proliferation in metastatic triple-negative breast cancer EMBO Mol. Med. 2013 5 264 279 10.1002/emmm.201201320 23307470 67. Gyorffy B. Surowiak P. Kiesslich O. Denkert C. Schafer R. Dietel M. Lage H. Gene expression profiling of 30 cancer cell lines predicts resistance towards 11 anticancer drugs at clinically achieved concentrations Int. J. Cancer 2006 118 1699 1712 10.1002/ijc.21570 16217747 68. Baxter R.C. Insulin-like growth factor binding protein-3 (IGFBP-3): Novel ligands mediate unexpected functions J. Cell Commun. Signal. 2013 7 179 189 10.1007/s12079-013-0203-9 23700234 69. Porter D.C. Farmaki E. Altilia S. Schools G.P. West D.K. Chen M. Chang B.D. Puzyrev A.T. Lim C.U. Rokow-Kittell R. Cyclin-dependent kinase 8 mediates chemotherapy-induced tumor-promoting paracrine activities Proc. Natl. Acad. Sci. USA 2012 109 13799 13804 10.1073/pnas.1206906109 22869755 70. Molina A. Velot L. Ghouinem L. Abdelkarim M. Bouchet B.P. Luissint A.C. Bouhlel I. Morel M. Sapharikas E. di Tommaso A. ATIP3, a novel prognostic marker of breast cancer patient survival, limits cancer cell migration and slows metastatic progression by regulating microtubule dynamics Cancer Res. 2013 73 2905 2915 10.1158/0008-5472.CAN-12-3565 23396587 71. Gene Expression Omnibus Available online:http://www.ncbi.nlm.nih.gov/geo/ (accessed on 26 August 2013) 72. The European Genome-phenome Archive Available online:https://www.ebi.ac.uk/ega/ (accessed on 26 August 2013) 73. Gyorffy B. Lanczky A. Eklund A.C. Denkert C. Budczies J. Li Q. Szallasi Z. An online survival analysis tool to rapidly assess the effect of 22,277 genes on breast cancer prognosis using microarray data of 1,809 patients Breast Cancer Res. Treat. 2010 123 725 731 10.1007/s10549-009-0674-9 20020197 74. Gyorffy B. Lanczky A. Szallasi Z. Implementing an online tool for genome-wide validation of survival-associated biomarkers in ovarian-cancer using microarray data from 1,287 patients Endocr. Relat. Cancer 2012 19 197 208 10.1530/ERC-11-0329 22277193 75. Park S. Koo J.S. Kim M.S. Park H.S. Lee J.S. Lee J.S. Kim S.I. Park B.W. Characteristics and outcomes according to molecular subtypes of breast cancer as classified by a panel of four biomarkers using immunohistochemistry Breast 2012 21 50 57 10.1016/j.breast.2011.07.008 21865043 76. Pawitan Y. Bjohle J. Amler L. Borg A.L. Egyhazi S. Hall P. Han X. Holmberg L. Huang F. Klaar S. Gene expression profiling spares early breast cancer patients from adjuvant therapy: Derived and validated in two population-based cohorts Breast Cancer Res. 2005 7 R953 R964 10.1186/bcr1325 16280042 77. Sabatier R. Finetti P. Cervera N. Lambaudie E. Esterni B. Mamessier E. Tallet A. Chabannon C. Extra J.M. Jacquemier J. A gene expression signature identifies two prognostic subgroups of basal breast cancer Breast Cancer Res. Treat. 2011 126 407 420 10.1007/s10549-010-0897-9 20490655 78. Hatzis C. Pusztai L. Valero V. Booser D.J. Esserman L. Lluch A. Vidaurre T. Holmes F. Souchon E. Wang H. A genomic predictor of response and survival following taxane-anthracycline chemotherapy for invasive breast cancer JAMA 2011 305 1873 1881 10.1001/jama.2011.593 21558518 79. Dedeurwaerder S. Desmedt C. Calonne E. Singhal S.K. Haibe-Kains B. Defrance M. Michiels S. Volkmar M. Deplus R. Luciani J. DNA methylation profiling reveals a predominant immune component in breast cancers EMBO Mol. Med. 2011 3 726 741 10.1002/emmm.201100801 21910250 80. Karn T. Pusztai L. Holtrich U. Iwamoto T. Shiang C.Y. Schmidt M. Muller V. Solbach C. Gaetje R. Hanker L. Homogeneous datasets of triple negative breast cancers enable the identification of novel prognostic and predictive signatures PLoS One 2011 6 e28403 10.1371/journal.pone.0028403 22220191 81. Sircoulomb F. Bekhouche I. Finetti P. Adelaide J. Ben Hamida A. Bonansea J. Raynaud S. Innocenti C. Charafe-Jauffret E. Tarpin C. Genome profiling of ERBB2-amplified breast cancers BMC Cancer 2010 10 539 10.1186/1471-2407-10-539 20932292 82. KM Plotter Available online:http://www.kmplot.com (accessed on 26 August 2013) 83. Budczies J. Klauschen F. Sinn B.V. Gyorffy B. Schmitt W.D. Darb-Esfahani S. Denkert C. Cutoff finder: A comprehensive and straightforward Web application enabling rapid biomarker cutoff optimization PLoS One 2012 7 e51862 10.1371/journal.pone.0051862 23251644

PMC005xxxxxx/PMC5003465.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2030243microarrays-02-00243ReviewFrom High-Throughput Microarray-Based Screening to Clinical Application: The Development of a Second Generation Multigene Test for Breast Cancer Prognosis Brase Jan C. 1*Kronenwett Ralf 1Petry Christoph 1Denkert Carsten 2Schmidt Marcus 31 Sividon Diagnostics GmbH, Nattermannallee 1, 50829, Cologne, Germany; E-Mails: Kronenwett@sividon.com (R.K.); petry@sividon.com (C.P.)2 Institute of Pathology, Charité University Medicine Berlin, Charitéplatz 1, 10117 Berlin, Germany; E-Mail: carsten.denkert@charite.de3 Department of Gynecology and Obstetrics, University of Mainz, Langenbeckstr. 1, 55131 Mainz, Germany; E-Mail: Marcus.Schmidt@unimedizin-mainz.de* Author to whom correspondence should be addressed; E-Mail: brase@sividon.com; Tel.: +49-(0)-221-669561-40; Fax: +49-(0)-221-669561-99.29 8 2013 9 2013 2 3 243 264 17 7 2013 12 8 2013 22 8 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Several multigene tests have been developed for breast cancer patients to predict the individual risk of recurrence. Most of the first generation tests rely on proliferation-associated genes and are commonly carried out in central reference laboratories. Here, we describe the development of a second generation multigene assay, the EndoPredict test, a prognostic multigene expression test for estrogen receptor (ER) positive, human epidermal growth factor receptor (HER2) negative (ER+/HER2−) breast cancer patients. The EndoPredict gene signature was initially established in a large high-throughput microarray-based screening study. The key steps for biomarker identification are discussed in detail, in comparison to the establishment of other multigene signatures. After biomarker selection, genes and algorithms were transferred to a diagnostic platform (reverse transcription quantitative PCR (RT-qPCR)) to allow for assaying formalin-fixed, paraffin-embedded (FFPE) samples. A comprehensive analytical validation was performed and a prospective proficiency testing study with seven pathological laboratories finally proved that EndoPredict can be reliably used in the decentralized setting. Three independent large clinical validation studies (n = 2,257) demonstrated that EndoPredict offers independent prognostic information beyond current clinicopathological parameters and clinical guidelines. The review article summarizes several important steps that should be considered for the development process of a second generation multigene test and offers a means for transferring a microarray signature from the research laboratory to clinical practice. breast cancerEndoPredictmultigene testendocrine therapy ==== Body 1. Background—Establishment and Clinical Validation of First Generation Multigene Tests for Breast Cancer Patients Breast cancer is the most common cancer in women worldwide, with a high number of cancer-related fatalities [1]. The decision on how to best treat a breast cancer patient is generally made based on clinical guidelines. They primarily use standard clinicopathological parameters, like age, tumor size, nodal status, grading and hormone receptor status, to define individual prognosis and to categorize patients into clinical stages. Clinical variables have also been integrated in clinical prediction models, such as Adjuvant!Online [2] and the Nottingham Prognostic Index [3,4]. However, guidelines and clinical prediction models rarely provide unambiguous treatment recommendation and do not fully capture the clinical course of the disease. This is because breast cancer is biologically heterogeneous, and molecular differences can lead to a differing outcome, even among tumors with similar clinical characteristics. One of the most pressing clinical questions in the management of estrogen receptor (ER) positive, human epidermal growth factor receptor (HER2) negative patients not satisfactorily addressed by current guidelines is whether to limit systemic treatment to just endocrine therapy or to employ adjuvant chemotherapy. While chemotherapy has been shown to provide an overall improved therapy outcome, this benefit is known to be limited to a subgroup of the patients. Still, in most countries, the established clinical practice is to treat the vast majority of patients with an anthracycline- and taxane-containing adjuvant chemotherapy, resulting in considerable overtreatment. To address this clinical need, the St. Gallen expert board has recommended since 2009 integrating molecular data into prognostic and predictive models and using validated multigene tests to assist in deciding whether to add chemotherapy to endocrine therapy [5]. The application of gene expression profiling with the use of microarrays has allowed for measuring thousands of mRNAs in parallel to identify markers that reflect molecular heterogeneity. In 2000, Perou et al. identified by unsupervised hierarchical cluster analysis that breast cancer can be subdivided into at least four molecularly distinct subgroups using an intrinsic gene signature [6]. Later, the molecularly distinct subtypes were repeatedly found to be associated with prognosis and response to chemotherapy treatment [7,8,9,10]. Over the past decade, many gene expression signatures have been established, but only a few of these have progressed to commercial availability. Investigators from the Netherlands Cancer Institute (NKI) in Amsterdam developed the first prognostic gene signature (Mammaprint, Agendia) for breast cancer patients in 2002 [11]. The signature is based on the measurement of 70 genes and was established in a retrospective series of 78 tumor samples using global gene expression profiles. A statistical “top-down” approach was applied to determine the most relevant genes that were associated with early recurrence in untreated node-negative breast cancer patients [11]. The performance of the 70-gene signature was subsequently validated in a consecutive series of 295 node-negative and node-positive breast cancer patients from the same institution [12]. However, in this validation study, 46% of all patients received adjuvant endocrine or chemotherapy therapy, and the samples had been partially used to establish the Mammaprint assay. Therefore, the validation study raised some concerns about whether the results could have been biased. The first independent validation was conducted using a multicenter cohort (n = 307) from the international Transbig consortium [13]—none of the patients had received systemic adjuvant therapy. The 70-gene signature was prognostic and identified a low-risk subgroup with 12% distant-metastasis events. Based on these data and other studies [14,15,16], Agendia B.V. developed a prognostic test for commercial use in node-negative breast cancer. The test was later approved by the U.S. Food and Drug Administration. In contrast to the Mammaprint assay, the 21-gene recurrence score (RS; Oncotype DX, Genomic Health) was established based on a candidate gene approach in estrogen-receptor positive (ER+) breast cancer patients [17]. The recurrence score is a multiparameter gene expression test that was initially defined in a combined training set of three sample cohorts, including samples from the clinical trial National Surgical Adjuvant Breast and Bowel Project (NSABP)-B20. The finding cohort encompassed a total of 447 node-negative breast cancer patients. In contrast to the Mammaprint assay, the selection of candidate genes was “hypothesis-driven”, and markers were selected due to their known relevance in breast cancer. Sixteen prognostic genes-of-interest were identified, and five reference genes were selected to normalize the gene expression levels. The continuous risk score can be calculated from the relative RNA abundance of the candidate genes. The sources of RNA are formalin-fixed paraffin-embedded (FFPE) tumor blocks. RNA quantification is accomplished by two-step reverse transcription quantitative PCR (RT-qPCR). The 21 gene panel encompasses genes associated with proliferation, invasion, ER and HER2 expression. The proliferation- and HER2-related genes are weighted highest in the mathematical algorithm and, therefore, dominate the test results. The RS can estimate the likelihood of distant metastasis, grouping patients into three risk categories (low, intermediate and high-risk). The RS was validated in the NSABP-B14 trial using 668 node-negative breast cancer patients treated with tamoxifen only [17]. 51% of the evaluated patients from the NSABP-B14 trial were classified as RS-low-risk. This subgroup had a low distant-metastasis rate of 6.8%. Later, the RS was validated in several other clinical trials (NSABP-B20 [18], Southwest Oncology Group (SWOG)-8814 [19], Arimidex, Tamoxifen, Alone or in Combination (ATAC) [20]). The NSABP-B20 results indicated that RS-high-risk patients have a benefit from adjuvant cyclophosphamide, methotrexate and fluorouracil chemotherapy [18]. However, the performance of the RS might be overestimated in this study, since some of the NSABP-B20 samples were included in the training phase of the RS [17,18]. Similar results were reported from the SWOG-8814 study, a randomized trial encompassing node-positive breast cancer patients treated with tamoxifen with or without anthracycline-based chemotherapy treatment [19]. Still, none of the two validation studies were carried out using a non-inferiority design. Accordingly, it remains elusive if the relative benefit in the high-, intermediate- and low-risk groups is really different. Furthermore, both validation studies encompassed HER2-positive patients. In SWOG-8814, it has been shown that the RS is not predictive for chemotherapy benefit in the relevant subgroup of ER+/HER2− patients. For the NSABP-B20 study, no data have been published for this key patient group. Recently, a biomarker substudy of the ATAC trial suggested that centrally-assessed classical clinical parameters, such as ER, progesterone receptor (PgR), HER2 and Ki67, offer the same prognostic information as the recurrence score [21]. Both tests—Oncotype DX and Mammaprint—help to determine which patients with early stage breast cancer are at lower risk of recurrence. Both multigene assays are carried out in central reference laboratories in Europe and the USA and have now been used in clinics for several years. Decision impact studies and health economic analyses demonstrated that these first generation signatures can be used to reduce healthcare costs and avoid chemotherapy [22,23,24,25,26,27]. Currently, both tests are prospectively evaluated in the Mindact, RxPonder and TailorX trials, respectively [28,29]. 2. Important Aspects for the Establishment and Clinical Validation of Novel Second Generation Multigene Tests The substantial increase of knowledge in breast cancer research in the last decade has resulted in a new understanding of how the disease can be managed and how novel drugs and diagnostic tests need to be developed and used in clinical routine. Evaluations of first generation multigene tests for breast cancer did not clearly answer whether or not prognostic tests are fit-for-purpose and should be routinely applied. For instance, the “Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Working Group” found insufficient evidence to make recommendations for or against the use of first generation multigene tests in 2008/2009 [30]. Several research gaps were identified by the EGAPP working group that originated from the study design, analysis and evaluation of the tests. The research gaps were published to encourage further development and evaluation of novel assays [30]. Some of the important aspects for the development of second generation multigene tests are summarized in the following sections. 2.1. Biomarker and Molecular Subtypes Breast cancer has been recognized to consist of different molecular subtypes. By determining the expression level of ER and HER2, three major subgroups can be defined: ER+/HER2−, ER−/HER2− and HER2+. All three subtypes differ in molecular and clinical characteristics. They are also predictive of patterns of response to systemic treatment or specific targeted agents. For instance, ER+/HER2− breast cancer patients can be treated by antagonizing the activity of estrogen with the selective estrogen-receptor modulator, tamoxifen [31,32]. However, ER+/HER2− breast cancer is a large and heterogeneous subgroup, and frequently, clinical parameters do not allow for deciding whether the patient is sufficiently treated with endocrine therapy only. Combined chemotherapy plus hormonal therapy is, therefore, an additional treatment option. Prognostic tests are urgently needed to allow for tailored treatment strategies in ER+/HER2− breast cancer, since it is well accepted that low absolute risk implies low absolute benefit from the addition of adjuvant chemotherapy [33]. In contrast to that, ER−/HER2− tumors have an increased likelihood of distant recurrence and do not benefit from any targeted intervention developed yet. Chemotherapy is so far the only modality of systemic treatment, and ER−/HER2− tumors seem to benefit the most from cytotoxic regimens [34]. Therefore, almost all patients belonging to this subgroup are currently treated with chemotherapy. Similarly, HER2-overexpressing tumors also show an aggressive behavior, but the clinical outcome can be significantly increased by targeting the extracellular domain of the HER2 receptor using a recombinant monoclonal antibody (trastuzumab) or by HER2 tyrosine kinase inhibitors. ER−/HER2− or HER2+ breast cancer patients were included in the development phase of first generation multigene tests. Therefore, the question was raised whether these assays are of prognostic value once HER2+ and ER−/HER2− breast cancer samples are removed [35]. These first generation multigene tests were never clearly assessed in the different molecular subgroups [36]. Moreover, first generation multigene tests provide little information on ER−/HER2− or HER2+ tumors, since almost all cases in these subgroups are classified as high-risk, due to their general high cell proliferation activity [36]. Biomarkers related to the extracellular environment—especially the adaptive immune system—seem to be more relevant in ER−/HER2− or HER2+ tumors. They appear to be able to identify subgroups of patients with better prognosis or—even more importantly—response to systemic or targeted treatment [37,38,39,40,41]. It is unquestionable that molecular subtypes have already begun to alter the way clinical investigators design clinical trials. Specific subgroups of breast cancer patients are enrolled for specific clinical questions. In line with the experiences from the therapeutic trials and daily clinical management, biomarker studies and second generation multigene tests should also be established in a specific molecular subgroup to account for the remarkable differences among groups. 2.2. ER+/HER2− Breast Cancer is a Chronic Disease—The Importance of Predicting Late Metastases The risk of breast cancer recurrence is well known to span more than ten years. However, molecular subtypes differ in terms of timing of distant recurrence. In contrast to ER−/HER2− and HER2+ breast cancer patients, ER+/HER2− breast cancer patients have an increased risk of developing late recurrences for an indefinite period after diagnosis [42,43]. More than 50% of all relapses in ER+/HER2− breast cancer patients occur later than five years after primary treatment. Several large phase III clinical trials have been initiated to study the effects of extended endocrine therapy. Recently, the aTTom (adjuvant Tamoxifen Treatment offer more) and ATLAS (Adjuvant Tamoxifen: Longer Against Shorter) trial reported a significantly improved outcome after completing 10 years of tamoxifen in comparison to five years of tamoxifen treatment [44]. Additionally, the National Cancer Institute of Canada Clinical Trials Group MA-17 [45,46,47,48,49], NSABP-B33 [50] and Austrian Breast and Colorectal Cancer Study Group (ABCSG)-06a [51] trials demonstrated that suppressing estrogen production by an aromatase inhibitor after the discontinuation of tamoxifen therapy prolongs disease-free survival. Currently, there are close to 20,000 additional breast cancer patients treated in randomized phase III clinical trials investigating endocrine therapy of a longer duration. However, the improved outcome observed in the clinical trials needs to be balanced between competing risks/side effects and individual risk of late recurrence. First generation gene expression tests are largely not suitable to predict late metastases [42,52]. Their prognostic performance seems to be time-dependent and higher in the first five years than between five and 10 years of follow-up [52]. Proliferation markers are the principal driving force of first generation tests, and proliferation lacks prognostic value beyond five years of follow-up. Therefore, novel predictors are required to identify patients at very low risk of developing late metastases to safely avoid the side-effects of extended endocrine therapy. 2.3. Pitfalls in Study Design—The Importance of Unique Clinical Characteristics and Treatment Strategies in the Training and Validation Phase First generation gene signatures were exclusively established in node-negative breast cancer patients. Today, node-positive breast cancer patients with a favorable biology are also strong candidates for omitting chemotherapy treatment. Therefore, several validation studies were initiated to assess whether first generation multigene tests are prognostic in node-positive disease [12,19]. The results of these studies were positive. Still, compared to node-negative breast cancer, the residual risk of recurrence of node-positive patients in the putative low-risk group is considerably higher [19]. This is likely due to the fact that node-positive patients were not included in the training sets of the first generation assays, so the algorithms obtained only exert the technology’s full potential in node-negative patients. Another important aspect for the development of multigene assays is that the treatment strategy should be similar in the training and validation cohorts and should be consistent with the current clinical recommendations. Mammaprint, for instance, was developed and validated in an untreated cohort of breast cancer patients [11,12]. However, according to the current clinical guidelines, all ER+ breast cancer patients should be treated with endocrine therapy. Therefore, an easy transfer of the validation results to the current clinical practice may be illusive. 2.4. Additional Prognostic Information—Clinical and Molecular Parameters Ki-67 is a cellular proliferation marker that has been recently suggested as an immunohistochemistry surrogate to stratify ER+/HER2− breast cancer patients into the intrinsic subgroups, luminal A and luminal B [53]. The St. Gallen consensus panel recommended Ki-67 in 2011 as a marker to decide whether chemotherapy can be safely foregone in patients with ER+/HER2− breast cancer [54]. However, immunohistological determination of Ki-67 expression suffers from intra- and inter-observer variability [55]. The lack of standardization and the unreliable use of a specific cut-off [53] to separate clinical meaningful subgroups has been an obstacle for the marker to make its way from the St. Gallen consensus into major clinical guidelines. New multigene tests should offer independent prognostic information to all common clinicopathological parameters, including histological markers, like Ki-67. The tests should clearly demonstrate additional prognostic information beyond what can be achieved with standard clinical and histological parameters. The ATAC trial recently suggested that the 21-gene recurrence score offers no additional prognostic information when compared to centrally assessed immunohistochemical parameters, including Ki-67 [21]. Although second generation multigene tests should offer additional prognostic information to all clinicopathological parameters, it seems very unlikely that these tests supplant the significant prognostic information of factors that measure the extent of tumor progression and dissemination. Therefore, molecular information should be refined and complemented with the prognostic information available from clinicopathological parameters [30] to establish hybrid scores integrating classical risk factors offering the best prediction accuracy. 2.5. Decentral Testing—The Importance of Analytic Validity of Tests and External Proficiency Testing All first generation multigene tests are provided by a diagnostic service through a central manufacturers’ reference laboratory. Due to their high complexity, a standardized robust performance in local routine laboratories seems to be challenging. However, this service model is an obstacle to wide acceptance in Europe’s decentrally organized healthcare systems—not only because of reimbursement issues. Therefore, second generation multigene tests should also allow decentralized testing in specialized local laboratories in order to provide a comprehensive tissue-based diagnosis by a pathologist. Therefore, the new tests must be compatible with established clinical workflows. However, for reliable high-quality results, performance characteristics and analytical validation data have to be published, and the robustness of decentralized assays have to be shown in external proficiency testing and round-robin trials. 3. The Establishment of EndoPredict—A Second Generation Multigene Test 3.1. Relevant Patient Group in Training and Validation As emphasized before, there is a clinical need for multigene tests to identify those patients with ER+/HER2− breast cancer, who are sufficiently treated with endocrine therapy. Identifying prognostic markers in specific molecular subtypes is pivotal to identifying such patients and has the largest potential to impact clinical decision making. Nevertheless, using the specific group of ER+/HER2− breast cancer patients leads to technical and statistical challenges compared to analyzing pooled patient subgroups. To this end, training and validation series for the EndoPredict—one of the first second generation multigene tests—were carefully selected. A large high-throughput microarray-based screening study was conducted to establish the EndoPredict signature [56]. For training, ER+/HER2− breast cancer samples were selected from four different institutes and two large clinical trials. ER−/HER2− or HER2+ breast cancer patients were excluded from the training and validation series. Additionally, tumor samples were collected from patients with and without axillary lymph-node involvement. All patients were uniquely treated with endocrine therapy only. 3.2. Gene Selection and Algorithm Design Affymetrix HG-U133A microarrays were employed to identify the most relevant prognostic marker genes. The microarray platform is a highly valuable and reliable tool to discover differentially expressed genes. Initial concerns regarding the reproducibility of microarray experiments and mathematical approaches to select candidates had been addressed by the Microarray Quality Control (MAQC) consortium. The MAQC consortium clearly showed that microarrays are useful for identifying differentially expressed markers [57,58,59]. There are different approaches for how to discover a clinically relevant gene signature using high-dimensional gene expression data [60]. The “top-down” approach was used to establish the 70-gene signature (Mammaprint). This is a purely statistical approach by simply looking for genes that are associated with clinical outcome independent of any biological or clinical assumptions. In contrast to that, the “bottom-up” strategy can be employed. It is based on establishing a gene expression signature according to a hypothesis, a specific biological subgroup or a clinical phenotype [60]. Subsequently, the signature is tested against clinical outcome information. Sotiriou and colleagues used the “bottom-up” strategy to establish the Genomic Grade Index (GGI). GGI was established to distinguish the large subgroup of intermediate-grade (grade 2) tumors [61] into prognostic subgroups by using large-scale gene expression profiling data [62]. The GGI signature was prognostic in independent data sets and able to refine the histological grade assessment [63,64]. A “top-down” approach was used at the beginning of the EndoPredict development to screen for prognostic markers. Sequential screening steps were used, and marker lists were continuously reduced to construct a robust final algorithm (Figure 1) [56]. First, gene expression levels assessed by different probe sets were quality controlled, and informative genes were selected using stringent technical filter criteria. Genes with a low expression level or a low dynamic range were omitted from any further analysis. Afterwards, gene candidates were selected that were consistently associated with prognosis using Cox regression in ER+/HER2− breast cancer patients. The first marker set was further enriched by adding candidate genes that are known to be of particular relevance in breast cancer. Additionally, marker genes were analyzed by unsupervised clustering and principal component analysis to elucidate the association of single markers, gene modules and clinical characteristics. Marker genes were also used for bivariate Cox regression analysis using the gene expressions levels of the proliferation marker, TOP2A. The results of the bivariate analysis could help to identify prognostic genes that are not associated with cell cycle processes. Finally, marker genes were selected according to multiple parameters: prognostic performance in univariate and bivariate Cox regression, analytical performance and associated gene modules. In a nutshell, the microarray-based screening study to define EndoPredict was a combination of a top-down approach and a hypothesis-driven candidate selection. Figure 1 Microarray-based screening and platform transfer in the training phase. QC, Quality Control; RT-qPCR, reverse transcription quantitative PCR. Although microarray-based gene expression analysis has evolved dramatically, the technology has not found its way into clinical routine. This is particularly due to the fact that microarrays work best on RNA from fresh-frozen tumor samples. However, the collection and storage of fresh-frozen samples is associated with logistical challenges and may not be applicable outside large and optimally equipped clinical centers. In contrast to that, FFPE tissue sections are generally prepared from every single tumor for its histopathological assessment. There are emerging technologies to carry out gene expression profiling using FFPE tissue and microarrays, but data quality is still an issue, due to the short RNA fragments created by tissue fixation. RT-qPCR is an appreciated alternative to reliably measure candidate genes using FFPE tissue sections. Robust protocols have been established to automatically extract RNA from FFPE tissue sections [65,66,67]. In order to avoid workflow issues associated with fresh tissue, 104 candidate genes from the EndoPredict microarray-based screening studies were transferred from fresh-frozen tissue and microarrays to FFPE and RT-qPCR. This was an essential step to move the promising candidates from the research laboratory stage to clinical validation and application. The final EndoPredict score was established using 63 marker genes that showed a considerably high correlation between microarrays and the RT-qPCR platform. Eight genes of interest were selected for the EndoPredict score. Besides proliferation, the genes chosen cover several cellular processes, such as apoptosis, DNA repair, cell adhesion and cell signaling. Nevertheless, the markers are also co-regulated with genes reflecting two relevant biological modules known to contribute to recurrence risk: proliferation and ER− signaling/differentiation [56,68]. Since breast cancer is a complex disease; even the best gene expression profile cannot mirror the whole clinical course of the disease. Nodal status and tumor size are still important clinical variables that are independently associated with prognosis. Therefore, improvement of prediction accuracy is possible using a multidimensional approach able to integrate tumor biology and disease burden. This had already been a requirement posted by the EGAPP group [30]. The molecular information of EndoPredict was consequently combined with the clinical parameter nodal status and tumor size, resulting in the molecular and clinical risk score, EndoPredict-clin (EPclin). 3.3. Independent Clinical Validation of the EndoPredict Test Complex high-dimensional gene expression data sets are prone to overfitting, since many more explanatory variables per tumor samples are commonly collected than the number of samples used to generate the dataset. As a consequence, mathematical algorithms perfectly suitable in the training set may subsequently fail in another test set [69]. Therefore, multigene tests should be tested and confirmed using independent validation studies only employing samples not used in the definition phase of the mathematical algorithm. To this end, the predefined and locked-down EndoPredict test—including all cut-off values—was validated in three independent clinical trials (n = 2,257, Figure 2). First, EndoPredict was assessed in postmenopausal ER+/HER2− breast cancer patients from the ABCG-6 trial, immediately followed by the same analysis in the ABCSG-8 trial. All patients had been treated with tamoxifen or tamoxifen, followed by anastrozole (ABCSG-6 [n = 378]; ABCSG-8 [n = 1,324]) [56,70]. None of the patients had been treated with adjuvant chemotherapy. EndoPredict was analyzed retrospectively in both phase III trials relying on prospectively pre-specified study objectives and laboratory data, as recommended by Simon et al. [71]. This allows one to generate level I evidence using this “prospective-retrospective” approach, with consistent results in at least two validation studies. Both validation series were blinded to any clinical outcome until the mathematical model and cut-offs had been locked down and the test had been applied to all samples. A statistical analysis plan had been specified before the performance of the assay was evaluated. Figure 2 Training and independent validation series for the EndoPredict test. FEC, fluorouracil, epirubicin and cyclophosphamide; FEC-P, fluorouracil, epirubicin and cyclophosphamide followed by weekly paclitaxel. These trials demonstrated that EndoPredict can precisely identify distant metastases in patient cohorts treated with endocrine therapy only. The EPclin low-risk group had an excellent prognosis, with an estimated risk of recurrence below 5% in both studies [56]. Treatment recommendations are commonly based upon risk of recurrence and the estimated benefits of treatment weighted against adverse events of therapy. Due to the low-risk of recurrence, the absolute benefit of adjuvant chemotherapy does not outweigh its medical risks and adverse events affecting quality of life. Therefore, a risk of 5% allows one to safely forgo chemotherapy [29]. Additionally, the study demonstrated that EndoPredict adds prognostic performance beyond all common clinicopathological parameters, including centrally-analyzed Ki-67 and quantitative ER. A sub-analysis of the ABCSG trials demonstrated that EndoPredict identifies early and late distant metastases [68]. The EP score provided additional prognostic information regarding late recurrence beyond what can be achieved by all common clinical parameters. An explorative analysis of the biological modules enclosed in the EndoPredict score demonstrated that proliferation-associated genes add prognostic information regarding early relapses, but show a less prognostic performance to identify late recurrence events. In contrast to that, ER-signaling genes add additional prognostic information to all clinical parameters for predicting late metastases. Additionally, the EPclin score improved the prognostic performance for predicting late recurrences: EPclin-low-risk patients had an absolute risk of distant metastasis of 1.8% during the period of five years after the end of endocrine therapy. This provides EndoPredict with the potential to identify patients expected to gain little benefit from extended endocrine treatment. The competing health risk of the individual patient needs to be balanced against the observed risk estimation of EndoPredict to decide on extended treatment strategies. The EndoPredict test was also clinically validated in node-positive ER+/HER2− breast cancer patients treated with chemotherapy [72]. This was the third independent “prospective-retrospective” validation in a large biomarker cohort. So far, only a few of the available prognostic tests have been validated in studies enrolling node-positive patients only [19]. Most of these studies showed that the tests also allow one to identify subgroups with a fair prognosis in spite of nodal involvement. However, the putative low-risk patients still have a considerably risk of disease. It may exceed 30% likelihood of distant metastasis within 10 years [19]. In contrast to that, the EndoPredict validation in the Grupo Español de Investigación en Cáncer de Mama (GEICAM)-9906 trial demonstrated that EndoPredict-low-risk patients had a 10-year risk of recurrence below 10%. Multivariate analysis showed that EndoPredict provides additional prognostic information to common clinical variables. The results suggest that EndoPredict provides important information regarding the residual risk of recurrence after a modern, anthracycline and taxane-based regimen of chemotherapy. While the initial validation studies for EndoPredict (ABCSG6 and ABCSG8) only encompassed postmenopausal breast cancer patients, the analysis of the GEICAM-9906 study clearly demonstrated that EndoPredict is prognostic in pre- and post-menopausal breast cancer patients. The excellent prognostic performance in premenopausal patients suggests that EndoPredict can be also used for risk stratification of younger women. Nevertheless, these results should be taken with some care, since all patients in the GEICAM-9906 trial were treated with chemotherapy. EndoPredict was also tested in a neo-adjuvant cohort of ER+/HER2− breast cancer patients treated with anthracycline/taxane-based therapy [73]. Almost all pathological complete response (pCR) events (91%) were classified as EP-high-risk, suggesting that EP-low-risk tumors are resistant towards chemotherapy treatment and do not particularly benefit from cytotoxic therapy. EndoPredict is currently prospectively tested in a neo-adjuvant trial (ABCSG-34) to validate these results. Neo-adjuvant studies are well suited to analyze chemotherapy response in different subgroups, and FFPE tissue, as well as response data is often readily available. Assessing chemotherapy in adjuvant trials is more demanding, as it requires a study with two treatment arms comparing an endocrine and endo-chemotherapy regimen. For ER+/HER2− breast cancer, this type of study has never been properly completed, yet, and published for any RNA-based multigene assay. Available data for ER+/HER2− patients is either insignificant [19] or tainted with results from more aggressive, non-luminal tumors [18]. 3.4. Analytical Validation and Proficiency Testing Before diagnostic tests are ready for wide-scale clinical application, an extensive analytical validation is necessary to ensure analytical validity and high technical reproducibility within and between laboratories. The EndoPredict test was developed according to current clinical laboratory standards. A comprehensive analytical validation study was carried out to demonstrate that EndoPredict allows a robust and precise determination of gene expression levels [74]. The analytical validation was conducted in accordance to the recommendations of the Clinical Laboratory Standards Institute (CLSI). Different breast cancer samples were used in this study to evaluate essential analytical parameters, like RNA input range, limit of detection, precision and inter-laboratory variability. Finally, the analytical parameters were verified in a molecular pathology laboratory, and the results clearly showed that there was no difference in test performance when compared to the manufacturer’s claims. A proficiency testing program with seven molecular pathology laboratories was subsequently initiated to finally prove that EndoPredict showed reproducible performance with good precision and negligible laboratory-to-laboratory variations [75]. The study demonstrated that EndoPredict is the first multigene test for breast cancer patients that can be reliably used in the decentralized setting [75]. EndoPredict seems to be more reproducible than immunohistochemical tests that reported variations in the decentralized setting, due to intra- and inter-laboratory disconcordance [55,76,77]. EndoPredict results were also compared between core biopsies and surgical tissue specimens in a further analytical study [78]. Test results were highly correlated between core biopsies and surgical specimens, indicating that the assay can also be used on core biopsy samples. The study also showed that inflammatory changes induced by biopsy sampling do not affect the test result [78]. This is probably due to the fact that EndoPredict does not contain genes directly associated with inflammation or wound healing. Therefore, tumor areas containing preoperative biopsy-induced changes might be also used to determine the EndoPredict score, obliterating the need for any biopsy channel dissection that may be associated with less robust assays. 3.5. Clinical Utility—Comparison to Clinical Guidelines, Decision Impact and Health Economics The stratification power of three widely accepted international guidelines (German S3 [79], National Comprehensive Cancer Network (NCCN) [80], St. Gallen [54]) were compared with the EndoPredict in 1,702 ER+/HER2− breast cancer patients treated with endocrine therapy alone [81]. All guidelines and EndoPredict identified a low-risk subgroup with excellent prognosis and a metastasis rate of approximately 5% after 10 years of follow-up. However, the three guidelines only assigned 7–19% of the patients to a low-risk group. In contrast, EndoPredict stratified 63% of the analyzed cohort as low-risk. This is an indication of EndoPredict’s higher specificity. Especially patients classified as intermediate/high-risk by clinical guidelines were reclassified by EndoPredict. The results clearly showed that EndoPredict outperformed all conventional parameters and guidelines by identifying a larger set of low-risk patient’s not needing cytotoxic treatment. Overall, EndoPredict seems to identify those women who should or should not receive chemotherapy and could ensure that more women receive the appropriate treatment. The Charité University Hospital recently analyzed the performance of the EndoPredict test and performed a prospective assessment of the impact on treatment decisions in 167 breast cancer cases [82]. The comparison of the treatment decisions before and after knowledge of the EndoPredict test result indicated that 37.7% of all evaluated breast cancer patients received a different adjuvant treatment recommendation as originally made, on the basis of clinical factors alone. 12% of patients were routed to an additional chemotherapy, thus avoiding potential under-treatment, while 25% of patients were directed to endocrine therapy alone, thus avoiding overtreatment. The results were supported by an evaluation conducted at the interdisciplinary breast center of the Technical University of Munich [83]. The decision impact study was carried out to prospectively examine whether EndoPredict affects the oncologist’s and patient’s adjuvant treatment choice. The results of this study also indicated that EndoPredict can indeed change treatment selection beyond standard clinical parameters and adds value to decision making in comparison to guideline-based patient management. Using the EndoPredict test results, in 44% of the 123 consecutive cases of ER+/HER2−, the breast cancer patients’ adjuvant chemotherapy was omitted. The results of both decision impact studies show that chemotherapy treatment can be markedly reduced with EndoPredict. Accordingly, unnecessary side effects and their corresponding costs can be reduced, as well. Indeed, a health economics analysis further proved that the combination of clinical guidelines and EndoPredict significantly reduced the costs associated with managing primary breast cancer and leads to improved “quality adjusted life years” (QALY) [84]. Overall, the use of EndoPredict led to a reduction of treatment costs. 4. EndoPredict—Ready for Prime Time? As mentioned earlier, the “Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Working Group” evaluated first generation multigene tests in 2009. The EGAPP working group concluded that first generation multigene tests should not be regularly applied in clinical routine, since the benefit and risk cannot be reliably assessed [30]. The foregoing sections showed that EndoPredict addressed several research gaps with regard to clinical validity, analytical validity and clinical utility. The EndoPredict test has been validated in three independent prospective-retrospective clinical trials. All studies were carried out in the prospective-retrospective design (category B studies), resulting in a level of evidence of I, according to Simon et al. [71]. This evidence level has been also acknowledged in German guidelines, as well as by international experts [85]. EndoPredict integrates classical risk factors, such as tumor size and nodal status, into a molecular-clinicopathological hybrid score and predicts not only early, but also late, metastasis. EndoPredict is so far the first multigene test that has been successfully validated in proficiency testing, allowing a widespread adoption in routine laboratory work-up of the molecular pathology. Toxicities of chemotherapy treatment can be safely avoided, and costs and quality-of-life issues can be considerably changed. Figure 3 summarizes all important aspects that demonstrated the high clinical and analytical validity, as well as the clinical utility of this multigene test. Figure 3 Summary of the important aspects regarding analytical and clinical validity, as well as clinical utility. CLSI, Clinical Laboratory Standards Institute; CE, Communauté européenne; IVDD, In vitro Diagnostic Directive; EU, European Union; pCR, pathological complete response; QALY, quality adjusted life years. Therefore, a novel evaluation of all first and second generation multigene tests—including EndoPredict—by the EGAPP working group would be desirable in the near future. 5. Perspectives This review demonstrates that a carefully designed and executed series of training, validation and analytical studies is required to transfer a microarray-based gene signature to a clinically useful test. The rapid gain of knowledge in breast cancer diagnosis and therapy has identified additional clinical needs that the development process of new multigene tests has to account for. Here, we focused on the EndoPredict test, but there are also other novel second generation multigene tests that have been recently established. PAM50 and the breast cancer index (BCI), for instance, can be also used to identify patients with low risk of recurrence [86,87,88,89]. Both tests can be also applied for predicting late recurrence events [52]. Although prognostic multigene tests comprise different gene sets, all mentioned gene signatures seem to have prognostic value and single out similar subsets of breast cancer patients. Nevertheless, there is still discordance in risk stratification [90], and multigene tests should be directly compared using the same clinical material to allow an estimation of the performance characteristics. Data from the ATAC trial suggested that PAM50 offers more prognostic information than the 21-gene recurrence score [91]. So far, first and second generation multigene assays help to determine which patients with early stage breast cancer are at lower risk of recurrence, allowing women to safely forgo chemotherapy treatment. In contrast to that, they have no ability to predict the most appropriate treatment scenario in high-risk patients. All multigene tests investigated so far have failed to identify a subgroup with a particular benefit from adding paclitaxel to anthracycline-based chemotherapy treatment [72,92]. Predictive markers for specific cytotoxic agents are needed to select the tailored treatment strategies for high-risk breast cancer patients. Currently, none of the identified predictive markers for selecting individualized chemotherapy strategies in breast cancer has been successfully validated. Additionally, companion diagnostic tests could help to identify subsets of patients likely to respond to novel targeted treatment strategies. The review article has summarized several important steps to be considered to successfully establish, validate and use second generation multigene tests. The process should be generally applicable to transferring other microarray signatures from the research laboratory to clinical practice. Acknowledgements The authors would like to thank all investigators and patients who contributed to the establishment, clinical and technical validation of the EndoPredict test. Conflicts of Interest J.C. Brase, R. Kronenwett and C. Petry are employees of Sividon Diagnostics GmbH. R. Kronenwett, C. Petry and C. Denkert are shareholders of Sividon Diagnostics GmbH. M. Schmidt received speaker’s honoraria from Sividon Diagnostics GmbH. ==== Refs References 1. Siegel R. Naishadham D. Jemal A. Cancer statistics, 2013 CA. Cancer J. Clin. 2013 63 11 30 10.3322/caac.21166 23335087 2. Olivotto I.A. Bajdik C.D. Ravdin P.M. Speers C.H. Coldman A.J. Norris B.D. Davis G.J. Chia S.K. Gelmon K.A. Population-based validation of the prognostic model ADJUVANT! for early breast cancer J. Clin. Oncol. 2005 23 2716 2725 15837986 3. Galea M.H. Blamey R.W. Elston C.E. Ellis I.O. The Nottingham Prognostic Index in primary breast cancer Breast Cancer Res. Treat. 1992 22 207 219 10.1007/BF01840834 1391987 4. D’Eredita G. Giardina C. Martellotta M. Natale T. Ferrarese F. Prognostic factors in breast cancer: The predictive value of the Nottingham Prognostic Index in patients with a long-term follow-up that were treated in a single institution Eur. J. Cancer 2001 37 591 596 10.1016/S0959-8049(00)00435-4 11290434 5. Goldhirsch A. Ingle J.N. Gelber R.D. Coates A.S. Thurlimann B. Senn H.J. Thresholds for therapies: Highlights of the St. Gallen International Expert Consensus on the primary therapy of early breast cancer 2009 Ann. Oncol. 2009 20 1319 1329 10.1093/annonc/mdp322 19535820 6. Perou C.M. Sorlie T. Eisen M.B. van de Rijn M. Jeffrey S.S. Rees C.A. Pollack J.R. Ross D.T. Johnsen H. Akslen L.A. Molecular portraits of human breast tumours Nature 2000 406 747 752 10.1038/35021093 10963602 7. Sorlie T. Perou C.M. Tibshirani R. Aas T. Geisler S. Johnsen H. Hastie T. Eisen M.B. van de Rijn M. Jeffrey S.S. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications Proc. Natl. Acad. Sci. USA 2001 98 10869 10874 10.1073/pnas.191367098 11553815 8. Sorlie T. Tibshirani R. Parker J. Hastie T. Marron J.S. Nobel A. Deng S. Johnsen H. Pesich R. Geisler S. Repeated observation of breast tumor subtypes in independent gene expression data sets Proc. Natl. Acad. Sci. USA 2003 100 8418 8423 10.1073/pnas.0932692100 12829800 9. Rouzier R. Perou C.M. Symmans W.F. Ibrahim N. Cristofanilli M. Anderson K. Hess K.R. Stec J. Ayers M. Wagner P. Breast cancer molecular subtypes respond differently to preoperative chemotherapy Clin. Cancer Res. 2005 11 5678 5685 10.1158/1078-0432.CCR-04-2421 16115903 10. Gluck S. de Snoo F. Peeters J. Stork-Sloots L. Somlo G. Molecular subtyping of early-stage breast cancer identifies a group of patients who do not benefit from neoadjuvant chemotherapy Breast Cancer Res. Treat. 2013 139 759 767 10.1007/s10549-013-2572-4 23756626 11. Van’t Veer L.J. Dai H. van de Vijver M.J. He Y.D. Hart A.A. Mao M. Peterse H.L. van der Kooy K. Marton M.J. Witteveen A.T. Gene expression profiling predicts clinical outcome of breast cancer Nature 2002 415 530 536 10.1038/415530a 11823860 12. Van de Vijver M.J. He Y.D. van’t Veer L.J. Dai H. Hart A.A. Voskuil D.W. Schreiber G.J. Peterse J.L. Roberts C. Marton M.J. A gene-expression signature as a predictor of survival in breast cancer N. Engl. J. Med. 2002 347 1999 2009 10.1056/NEJMoa021967 12490681 13. Buyse M. Loi S. van’t Veer L. Viale G. Delorenzi M. Glas A.M. d’Assignies M.S. Bergh J. Lidereau R. Ellis P. Validation and clinical utility of a 70-gene prognostic signature for women with node-negative breast cancer J. Natl. Cancer Inst. 2006 98 1183 1192 10.1093/jnci/djj329 16954471 14. Mook S. Schmidt M.K. Weigelt B. Kreike B. Eekhout I. van de Vijver M.J. Glas A.M. Floore A. Rutgers E.J. van’t Veer L.J. The 70-gene prognosis signature predicts early metastasis in breast cancer patients between 55 and 70 years of age Ann. Oncol. 2010 21 717 722 10.1093/annonc/mdp388 19825882 15. Wittner B.S. Sgroi D.C. Ryan P.D. Bruinsma T.J. Glas A.M. Male A. Dahiya S. Habin K. Bernards R. Haber D.A. Analysis of the MammaPrint breast cancer assay in a predominantly postmenopausal cohort Clin. Cancer Res. 2008 14 2988 2993 10.1158/1078-0432.CCR-07-4723 18483364 16. Glas A.M. Floore A. Delahaye L.J. Witteveen A.T. Pover R.C. Bakx N. Lahti-Domenici J.S. Bruinsma T.J. Warmoes M.O. Bernards R. Converting a breast cancer microarray signature into a high-throughput diagnostic test BMC Genomics 2006 7 278 10.1186/1471-2164-7-278 17074082 17. Paik S. Shak S. Tang G. Kim C. Baker J. Cronin M. Baehner F.L. Walker M.G. Watson D. Park T. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer N. Engl. J. Med. 2004 351 2817 2826 10.1056/NEJMoa041588 15591335 18. Paik S. Tang G. Shak S. Kim C. Baker J. Kim W. Cronin M. Baehner F.L. Watson D. Bryant J. Gene expression and benefit of chemotherapy in women with node-negative, estrogen receptor-positive breast cancer J. Clin. Oncol. 2006 24 3726 3734 10.1200/JCO.2005.04.7985 16720680 19. Albain K.S. Barlow W.E. Shak S. Hortobagyi G.N. Livingston R.B. Yeh I.T. Ravdin P. Bugarini R. Baehner F.L. Davidson N.E. Prognostic and predictive value of the 21-gene recurrence score assay in postmenopausal women with node-positive, oestrogen-receptor-positive breast cancer on chemotherapy: A retrospective analysis of a randomised trial Lancet Oncol. 2010 11 55 65 10.1016/S1470-2045(09)70314-6 20005174 20. Dowsett M. Cuzick J. Wale C. Forbes J. Mallon E.A. Salter J. Quinn E. Dunbier A. Baum M. Buzdar A. Prediction of risk of distant recurrence using the 21-gene recurrence score in node-negative and node-positive postmenopausal patients with breast cancer treated with anastrozole or tamoxifen: A TransATAC study J. Clin. Oncol. 2010 28 1829 1834 10.1200/JCO.2009.24.4798 20212256 21. Cuzick J. Dowsett M. Pineda S. Wale C. Salter J. Quinn E. Zabaglo L. Mallon E. Green A.R. Ellis I.O. Prognostic value of a combined estrogen receptor, progesterone receptor, Ki-67, and human epidermal growth factor receptor 2 immunohistochemical score and comparison with the Genomic Health recurrence score in early breast cancer J. Clin. Oncol. 2011 29 4273 4278 10.1200/JCO.2010.31.2835 21990413 22. Klang S.H. Hammerman A. Liebermann N. Efrat N. Doberne J. Hornberger J. Economic implications of 21-gene breast cancer risk assay from the perspective of an Israeli-managed health-care organization Value Health 2010 13 381 387 10.1111/j.1524-4733.2010.00724.x 20412544 23. Partin J.F. Mamounas E.P. Impact of the 21-gene recurrence score assay compared with standard clinicopathologic guidelines in adjuvant therapy selection for node-negative, estrogen receptor-positive breast cancer Ann. Surg. Oncol. 2011 18 3399 3406 10.1245/s10434-011-1698-z 21537874 24. Bueno-de-Mesquita J.M. van Harten W.H. Retel V.P. van’t Veer L.J. van Dam F.S. Karsenberg K. Douma K.F. van Tinteren H. Peterse J.L. Wesseling J. Use of 70-gene signature to predict prognosis of patients with node-negative breast cancer: A prospective community-based feasibility study (RASTER) Lancet Oncol. 2007 8 1079 1087 10.1016/S1470-2045(07)70346-7 18042430 25. Geffen D.B. Amir N. Sion-Vardy N. Ariad S. Kachko L. Bayme M. Delgado B. Dyomin V. Argov S. Koretz M. Stage I breast cancer in a regional oncology practice in Israel 2002–2006: Clinicopathologic features, risk estimation and planned therapy of 328 consecutive patients Breast 2009 18 316 321 10.1016/j.breast.2009.08.004 19819143 26. Asad J. Jacobson A.F. Estabrook A. Smith S.R. Boolbol S.K. Feldman S.M. Osborne M.P. Boachie-Adjei K. Twardzik W. Tartter P.I. Does oncotype DX recurrence score affect the management of patients with early-stage breast cancer? Am. J. Surg. 2008 196 527 529 10.1016/j.amjsurg.2008.06.021 18809056 27. Drukker C.A. Bueno-de-Mesquita J.M. Retel V.P. van Harten W.H. van Tinteren H. Wesseling J. Roumen R.M. Knauer M. van’t Veer L.J. Sonke G.S. A prospective evaluation of a breast cancer prognosis signature in the observational RASTER study Int. J. Cancer 2013 133 929 936 10.1002/ijc.28082 23371464 28. Bogaerts J. Cardoso F. Buyse M. Braga S. Loi S. Harrison J.A. Bines J. Mook S. Decker N. Ravdin P. Gene signature evaluation as a prognostic tool: Challenges in the design of the MINDACT trial Nat. Clin. Pract. Oncol. 2006 3 540 551 17019432 29. Cardoso F. van’t Veer L. Rutgers E. Loi S. Mook S. Piccart-Gebhart M.J. Clinical application of the 70-gene profile: the MINDACT trial J. Clin. Oncol. 2008 26 729 735 10.1200/JCO.2007.14.3222 18258980 30. Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Working Group Recommendations from the EGAPP working group: Can tumor gene expression profiling improve outcomes in patients with breast cancer? Genet. Med. 2009 11 66 73 10.1097/GIM.0b013e3181928f56 19125125 31. Fisher B. Dignam J. Bryant J. Wolmark N. Five versus more than five years of tamoxifen for lymph node-negative breast cancer: Updated findings from the National Surgical Adjuvant Breast and Bowel Project B-14 randomized trial J. Natl. Cancer Inst. 2001 93 684 690 10.1093/jnci/93.9.684 11333290 32. Fisher B. Jeong J.H. Bryant J. Anderson S. Dignam J. Fisher E.R. Wolmark N. Treatment of lymph-node-negative, oestrogen-receptor-positive breast cancer: Long-term findings from National Surgical Adjuvant Breast and Bowel Project randomised clinical trials Lancet 2004 364 858 868 10.1016/S0140-6736(04)16981-X 15351193 33. Peto R. Davies C. Godwin J. Gray R. Pan H.C. Clarke M. Cutter D. Darby S. McGale P. Taylor C. Comparisons between different polychemotherapy regimens for early breast cancer: Meta-analyses of long-term outcome among 100,000 women in 123 randomised trials Lancet 2012 379 432 444 10.1016/S0140-6736(11)61625-5 22152853 34. Berry D.A. Cirrincione C. Henderson I.C. Citron M.L. Budman D.R. Goldstein L.J. Martino S. Perez E.A. Muss H.B. Norton L. Estrogen-receptor status and outcomes of modern chemotherapy for patients with node-positive breast cancer JAMA 2006 295 1658 1667 10.1001/jama.295.14.1658 16609087 35. Milburn M. Rosman M. Mylander C. Tafra L. Is oncotype DX recurrence score (RS) of prognostic value once HER2-positive and low-ER expression patients are removed? Breast J. 2013 19 357 364 10.1111/tbj.12126 23701403 36. Desmedt C. Haibe-Kains B. Wirapati P. Buyse M. Larsimont D. Bontempi G. Delorenzi M. Piccart M. Sotiriou C. Biological processes associated with breast cancer clinical outcome depend on the molecular subtypes Clin. Cancer Res. 2008 14 5158 5165 10.1158/1078-0432.CCR-07-4756 18698033 37. Schmidt M. Bohm D. von Torne C. Steiner E. Puhl A. Pilch H. Lehr H.A. Hengstler J.G. Kolbl H. Gehrmann M. The humoral immune system has a key prognostic impact in node-negative breast cancer Cancer Res. 2008 68 5405 5413 10.1158/0008-5472.CAN-07-5206 18593943 38. Schmidt M. Hellwig B. Hammad S. Othman A. Lohr M. Chen Z. Boehm D. Gebhard S. Petry I. Lebrecht A. A comprehensive analysis of human gene expression profiles identifies stromal immunoglobulin к C as a compatible prognostic marker in human solid tumors Clin. Cancer Res. 2012 18 2695 2703 10.1158/1078-0432.CCR-11-2210 22351685 39. Schmidt M. Hengstler J.G. von Torne C. Koelbl H. Gehrmann M.C. Coordinates in the universe of node-negative breast cancer revisited Cancer Res. 2009 69 2695 2698 10.1158/0008-5472.CAN-08-4013 19318558 40. Teschendorff A.E. Gomez S. Arenas A. El-Ashry D. Schmidt M. Gehrmann M. Caldas C. Improved prognostic classification of breast cancer defined by antagonistic activation patterns of immune response pathway modules BMC Cancer 2010 10 604 10.1186/1471-2407-10-604 21050467 41. Bianchini G. Qi Y. Alvarez R.H. Iwamoto T. Coutant C. Ibrahim N.K. Valero V. Cristofanilli M. Green M.C. Radvanyi L. Molecular anatomy of breast cancer stroma and its prognostic value in estrogen receptor-positive and -negative cancers J. Clin. Oncol. 2010 28 4316 4323 10.1200/JCO.2009.27.2419 20805453 42. Esserman L.J. Moore D.H. Tsing P.J. Chu P.W. Yau C. Ozanne E. Chung R.E. Tandon V.J. Park J.W. Baehner F.L. Biologic markers determine both the risk and the timing of recurrence in breast cancer Breast Cancer Res. Treat. 2011 129 607 616 10.1007/s10549-011-1564-5 21597921 43. Jatoi I. Anderson W.F. Jeong J.H. Redmond C.K. Breast cancer adjuvant therapy: Time to consider its time-dependent effects J. Clin. Oncol. 2011 29 2301 2304 10.1200/JCO.2010.32.3550 21555693 44. Davies C. Pan H. Godwin J. Gray R. Arriagada R. Raina V. Abraham M. Alencar V.H. Badran A. Bonfill X. Long-term effects of continuing adjuvant tamoxifen to 10 years versus stopping at 5 years after diagnosis of oestrogen receptor-positive breast cancer: ATLAS, a randomised trial Lancet 2013 381 805 816 10.1016/S0140-6736(12)61963-1 23219286 45. Goss P.E. Letrozole in the extended adjuvant setting: MA.17 Breast Cancer Res. Treat. 2007 105 Suppl 1 45 53 10.1007/s10549-007-9698-1 17912635 46. Goss P.E. Ingle J.N. Martino S. Robert N.J. Muss H.B. Livingston R.B. Davidson N.E. Perez E.A. Chavarri-Guerra Y. Cameron D.A. Impact of premenopausal status at breast cancer diagnosis in women entered on the placebo-controlled NCIC CTG MA17 trial of extended adjuvant letrozole Ann. Oncol. 2013 24 355 361 10.1093/annonc/mds330 23028039 47. Goss P.E. Ingle J.N. Martino S. Robert N.J. Muss H.B. Piccart M.J. Castiglione M. Tu D. Shepherd L.E. Pritchard K.I. Randomized trial of letrozole following tamoxifen as extended adjuvant therapy in receptor-positive breast cancer: Updated findings from NCIC CTG MA.17 J. Natl. Cancer Inst. 2005 97 1262 1271 10.1093/jnci/dji250 16145047 48. Goss P.E. Ingle J.N. Martino S. Robert N.J. Muss H.B. Piccart M.J. Castiglione M. Tu D. Shepherd L.E. Pritchard K.I. Efficacy of letrozole extended adjuvant therapy according to estrogen receptor and progesterone receptor status of the primary tumor: National Cancer Institute of Canada Clinical Trials Group MA.17 J. Clin. Oncol. 2007 25 2006 2011 10.1200/JCO.2006.09.4482 17452676 49. Goss P.E. Ingle J.N. Martino S. Robert N.J. Muss H.B. Piccart M.J. Castiglione M. Tu D. Shepherd L.E. Pritchard K.I. A randomized trial of letrozole in postmenopausal women after five years of tamoxifen therapy for early-stage breast cancer N. Engl. J. Med. 2003 349 1793 1802 10.1056/NEJMoa032312 14551341 50. Mamounas E.P. Jeong J.H. Wickerham D.L. Smith R.E. Ganz P.A. Land S.R. Eisen A. Fehrenbacher L. Farrar W.B. Atkins J.N. Benefit from exemestane as extended adjuvant therapy after 5 years of adjuvant tamoxifen: Intention-to-treat analysis of the National Surgical Adjuvant Breast And Bowel Project B-33 trial J. Clin. Oncol. 2008 26 1965 1971 10.1200/JCO.2007.14.0228 18332472 51. Jakesz R. Greil R. Gnant M. Schmid M. Kwasny W. Kubista E. Mlineritsch B. Tausch C. Stierer M. Hofbauer F. Extended adjuvant therapy with anastrozole among postmenopausal breast cancer patients: Results from the randomized Austrian Breast and Colorectal Cancer Study Group Trial 6a J. Natl. Cancer Inst. 2007 99 1845 1853 10.1093/jnci/djm246 18073378 52. Sgroi D.C. Sestak I. Cuzick J. Zhang Y. Schnabel C.A. Erlander M.G. Goss P.E. Dowsett M. Comparative performance of breast cancer Index (BCI) vs . oncotype Dx and IHC4 in the prediction of late recurrence in hormonal receptor-positive lymph node-negative breast cancer patients: A TransATAC study Cancer Res. 2012 72 10.1158/0008-5472.SABCS12-S1-9 53. Cheang M.C. Chia S.K. Voduc D. Gao D. Leung S. Snider J. Watson M. Davies S. Bernard P.S. Parker J.S. Ki67 index, HER2 status, and prognosis of patients with luminal B breast cancer J. Natl. Cancer Inst. 2009 101 736 750 10.1093/jnci/djp082 19436038 54. Goldhirsch A. Wood W.C. Coates A.S. Gelber R.D. Thurlimann B. Senn H.J. Strategies for subtypes—Dealing with the diversity of breast cancer: Highlights of the St. Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2011 Ann. Oncol. 2011 22 1736 1747 10.1093/annonc/mdr304 21709140 55. Varga Z. Diebold J. Dommann-Scherrer C. Frick H. Kaup D. Noske A. Obermann E. Ohlschlegel C. Padberg B. Rakozy C. How reliable is Ki-67 immunohistochemistry in grade 2 breast carcinomas? A QA study of the Swiss Working Group of Breast- and Gynecopathologists PLoS One 2012 7 e37379 10.1371/journal.pone.0037379 22662150 56. Filipits M. Rudas M. Jakesz R. Dubsky P. Fitzal F. Singer C.F. Dietze O. Greil R. Jelen A. Sevelda P. A new molecular predictor of distant recurrence in ER-positive, HER2-negative breast cancer adds independent information to conventional clinical risk factors Clin. Cancer Res. 2011 17 6012 6020 10.1158/1078-0432.CCR-11-0926 21807638 57. Patterson T.A. Lobenhofer E.K. Fulmer-Smentek S.B. Collins P.J. Chu T.M. Bao W. Fang H. Kawasaki E.S. Hager J. Tikhonova I.R. Performance comparison of one-color and two-color platforms within the MicroArray Quality Control (MAQC) project Nat. Biotechnol. 2006 24 1140 1150 10.1038/nbt1242 16964228 58. Shi L. Reid L.H. Jones W.D. Shippy R. Warrington J.A. Baker S.C. Collins P.J. de Longueville F. Kawasaki E.S. Lee K.Y. The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements Nat. Biotechnol. 2006 24 1151 1161 10.1038/nbt1239 16964229 59. Irizarry R.A. Warren D. Spencer F. Kim I.F. Biswal S. Frank B.C. Gabrielson E. Garcia J.G. Geoghegan J. Germino G. Multiple-laboratory comparison of microarray platforms Nat. Methods 2005 2 345 350 10.1038/nmeth756 15846361 60. Sotiriou C. Pusztai L. Gene-expression signatures in breast cancer N. Engl. J. Med. 2009 360 790 800 10.1056/NEJMra0801289 19228622 61. Furness P.N. Taub N. Assmann K.J. Banfi G. Cosyns J.P. Dorman A.M. Hill C.M. Kapper S.K. Waldherr R. Laurinavicius A. International variation in histologic grading is large, and persistent feedback does not improve reproducibility Am. J. Surg. Pathol. 2003 27 805 810 10.1097/00000478-200306000-00012 12766585 62. Sotiriou C. Wirapati P. Loi S. Harris A. Fox S. Smeds J. Nordgren H. Farmer P. Praz V. Haibe-Kains B. Gene expression profiling in breast cancer: Understanding the molecular basis of histologic grade to improve prognosis J. Natl. Cancer Inst. 2006 98 262 272 10.1093/jnci/djj052 16478745 63. Desmedt C. Giobbie-Hurder A. Neven P. Paridaens R. Christiaens M.R. Smeets A. Lallemand F. Haibe-Kains B. Viale G. Gelber R.D. The Gene expression Grade Index: A potential predictor of relapse for endocrine-treated breast cancer patients in the BIG 1–98 trial BMC Med. Genom. 2009 2 40 10.1186/1755-8794-2-40 64. Loi S. Haibe-Kains B. Desmedt C. Lallemand F. Tutt A.M. Gillet C. Ellis P. Harris A. Bergh J. Foekens J.A. Definition of clinically distinct molecular subtypes in estrogen receptor-positive breast carcinomas through genomic grade J. Clin. Oncol. 2007 25 1239 1246 10.1200/JCO.2006.07.1522 17401012 65. Bohmann K. Hennig G. Rogel U. Poremba C. Mueller B.M. Fritz P. Stoerkel S. Schaefer K.L. RNA extraction from archival formalin-fixed paraffin-embedded tissue: A comparison of manual, semiautomated, and fully automated purification methods Clin. Chem. 2009 55 1719 1727 10.1373/clinchem.2008.122572 19617290 66. Hennig G. Gehrmann M. Stropp U. Brauch H. Fritz P. Eichelbaum M. Schwab M. Schroth W. Automated extraction of DNA and RNA from a single formalin-fixed paraffin-embedded tissue section for analysis of both single-nucleotide polymorphisms and mRNA expression Clin. Chem. 2010 56 1845 1853 10.1373/clinchem.2010.151233 20947696 67. Muller B.M. Kronenwett R. Hennig G. Euting H. Weber K. Bohmann K. Weichert W. Altmann G. Roth C. Winzer K.J. Quantitative determination of estrogen receptor, progesterone receptor, and HER2 mRNA in formalin-fixed paraffin-embedded tissue—A new option for predictive biomarker assessment in breast cancer Diagn. Mol. Pathol. 2011 20 1 10 10.1097/PDM.0b013e3181e3630c 21326033 68. Dubsky P. Brase J.C. Fisch K. Jakesz R. Singer C.F. Greil R. Dietze O. Weber K.E. Petry C. Kronenwett R. The EndoPredict score identifies late distant metastases in ER+/HER2− breast cancer patients Cancer Res. 2012 72 10.1158/0008-5472.SABCS12-S4-3 69. Ein-Dor L. Kela I. Getz G. Givol D. Domany E. Outcome signature genes in breast cancer: Is there a unique set? Bioinformatics 2005 21 171 178 10.1093/bioinformatics/bth469 15308542 70. Dubsky P.C. Jakesz R. Mlineritsch B. Postlberger S. Samonigg H. Kwasny W. Tausch C. Stoger H. Haider K. Fitzal F. Tamoxifen and anastrozole as a sequencing strategy: A randomized controlled trial in postmenopausal patients with endocrine-responsive early breast cancer from the Austrian Breast and Colorectal Cancer Study Group J. Clin. Oncol. 2012 30 722 728 10.1200/JCO.2011.36.8993 22271481 71. Simon R.M. Paik S. Hayes D.F. Use of archived specimens in evaluation of prognostic and predictive biomarkers J. Natl. Cancer Inst. 2009 101 1446 1452 10.1093/jnci/djp335 19815849 72. Martin M. Brase J.C. Ruiz-Borrego M. Krappmann K. Munarriz B. Fisch K. Ruiz A. Weber K.E. Crespo C. Petry C. Prognostic performance of the EndoPredict score in node-positive chemotherapy-treated ER+/HER2− breast cancer patients: results from the GEICAM/9906 trial Cancer Res. 2012 72 10.1158/0008-5472.SABCS12-P2-10-11 73. Brase J.C. Gehrmann M.C. Petry C. Weber K.E. Schmidt M. Kölbl H. Schroth W. Schwab M. Müller V. Jänicke F. The EndoPredict score is a response predictor for neoadjuvant chemotherapy in ER-positive, HER2-negative breast cancer Cancer Res. 2011 71 10.1158/0008-5472.SABCS11-P1-06-26 74. Kronenwett R. Bohmann K. Prinzler J. Sinn B.V. Haufe F. Roth C. Averdick M. Ropers T. Windbergs C. Brase J.C. Decentral gene expression analysis: Analytical validation of the Endopredict genomic multianalyte breast cancer prognosis test BMC Cancer 2012 12 456 10.1186/1471-2407-12-456 23039280 75. Denkert C. Kronenwett R. Schlake W. Bohmann K. Penzel R. Weber K.E. Hofler H. Lehmann U. Schirmacher P. Specht K. Decentral gene expression analysis for ER+/HER2− breast cancer: Results of a proficiency testing program for the EndoPredict assay Virchows Arch. 2012 460 251 259 10.1007/s00428-012-1204-4 22371223 76. Noske A. Loibl S. Darb-Esfahani S. Roller M. Kronenwett R. Muller B.M. Steffen J. von Toerne C. Wirtz R. Baumann I. Comparison of different approaches for assessment of HER2 expression on protein and mRNA level: Prediction of chemotherapy response in the neoadjuvant GeparTrio trial (NCT00544765) Breast Cancer Res. Treat. 2011 126 109 117 10.1007/s10549-010-1316-y 21190079 77. Loibl S. Muller B.M. von Minckwitz G. Schwabe M. Roller M. Darb-Esfahani S. Ataseven B. du Bois A. Fissler-Eckhoff A. Gerber B. Androgen receptor expression in primary breast cancer and its predictive and prognostic value in patients treated with neoadjuvant chemotherapy Breast Cancer Res. Treat. 2011 130 477 487 10.1007/s10549-011-1715-8 21837479 78. Muller B.M. Brase J.C. Haufe F. Weber K.E. Budzies J. Petry C. Prinzler J. Kronenwett R. Dietel M. Denkert C. Comparison of the RNA-based EndoPredict multigene test between core biopsies and corresponding surgical breast cancer sections J. Clin. Pathol. 2012 65 660 662 10.1136/jclinpath-2012-200716 22447922 79. Wöckel A. Kreienberg R. First revision of the German S3 guideline ‘diagnosis, Therapy, and Follow-Up of Breast Cancer’ Breast Care (Basel) 2008 3 82 86 10.1159/000127509 21373209 80. Carlson R.W. Brown E. Burstein H.J. Gradishar W.J. Hudis C.A. Loprinzi C. Mamounas E.P. Perez E.A. Pritchard K. Ravdin P. NCCN task force report: Adjuvant therapy for breast cancer J. Natl. Compr. Canc. Netw. 2006 4 Suppl 1 S1 S26 81. Dubsky P. Filipits M. Jakesz R. Rudas M. Singer C.F. Greil R. Dietze O. Luisser I. Klug E. Sedivy R. EndoPredict improves the prognostic classification derived from common clinical guidelines in ER-positive, HER2-negative early breast cancer Ann. Oncol. 2013 24 640 647 10.1093/annonc/mds334 23035151 82. Muller B.M. Keil E. Lehmann A. Winzer K.J. Richter-Ehrenstein C. Prinzler J. Bangemann N. Reles A. Stadie S. Schoenegg W. The EndoPredict gene-expression assay in clinical practice—Performance and impact on clinical decisions PLoS One 2013 8 e68252 10.1371/journal.pone.0068252 23826382 83. Ettl J. Große Lackmann K. Hapfelmeier A. Klein E. Paepke S. Petry C. Specht K. Höfler H. Kiechle M. Prospective Comparison of uPA/PAI-1 and Endopredict-Clin Score in ER-Positive, HER2-Negative Breast Cancer: Impact on Risk Stratification and Treatment Decisions Proceeding of 2013 ASCO Annual Meeting Chicago, IL, USA 31 May–4 June 2013 84. Blank P. Schwenkglenks M. Dubsky P. Filipits M. Gutzwiller F. Lux M.P. Brase J.C. Kronenwett R. Szucs T.D. Gnant M. Health economic analysis of guideline and gene expression signature-based risk stratification of distant recurrence in early breast cancer patients Ann. Oncol. 2013 24 10.1093/annonc/mdt084.7 85. Weigelt B. Reis-Filho J.S. Swanton C. Genomic analyses to select patients for adjuvant chemotherapy: Trials and tribulations Ann. Oncol. 2012 23 Suppl 10 x211 x218 22987965 86. Nielsen T.O. Parker J.S. Leung S. Voduc D. Ebbert M. Vickery T. Davies S.R. Snider J. Stijleman I.J. Reed J. A comparison of PAM50 intrinsic subtyping with immunohistochemistry and clinical prognostic factors in tamoxifen-treated estrogen receptor-positive breast cancer Clin. Cancer Res. 2010 16 5222 5232 10.1158/1078-0432.CCR-10-1282 20837693 87. Parker J.S. Mullins M. Cheang M.C. Leung S. Voduc D. Vickery T. Davies S. Fauron C. He X. Hu Z. Supervised risk predictor of breast cancer based on intrinsic subtypes J. Clin. Oncol. 2009 27 1160 1167 10.1200/JCO.2008.18.1370 19204204 88. Jankowitz R.C. Cooper K. Erlander M.G. Ma X.J. Kesty N.C. Li H. Chivukula M. Brufsky A. Prognostic utility of the breast cancer index and comparison to Adjuvant! Online in a clinical case series of early breast cancer Breast Cancer Res. 2011 13 R98 10.1186/bcr3038 21999244 89. Jerevall P.L. Ma X.J. Li H. Salunga R. Kesty N.C. Erlander M.G. Sgroi D.C. Holmlund B. Skoog L. Fornander T. Prognostic utility of HOXB13:IL17BR and molecular grade index in early-stage breast cancer patients from the Stockholm trial Br. J. Cancer 2011 104 1762 1769 10.1038/bjc.2011.145 21559019 90. Varga Z. Sinn P. Fritzsche F. von Hochstetter A. Noske A. Schraml P. Tausch C. Trojan A. Moch H. Comparison of EndoPredict and oncotype DX test results in hormone receptor positive invasive breast cancer PLoS One 2013 8 e58483 10.1371/journal.pone.0058483 23505515 91. Dowsett M. Sestak I. Lopez-Knowles E. Sidhu K. Dunbier A.K. Cowens J.W. Ferree S. Storhoff J. Schaper C. Cuzick J. Comparison of PAM50 risk of recurrence score with oncotype DX and IHC4 for predicting risk of distant recurrence after endocrine therapy J. Clin. Oncol. 2013 10.1200/JCO.2012.46.1558 92. Mamounas E.P. Tang G. Paik S. Baehner F.L. Liu Q. Jeong J.H. Kim S.R. Butler S.M. Jamshidian F. Cherbavaz D.B. Association between the 21-gene recurrence score (RS) and benefit from adjuvant paclitaxel (Pac) in node-positive (N+), ER-positive breast cancer patients (pts): Results from NSABP B-28 Cancer Res. 2012 72 10.1158/0008-5472.SABCS12-S1-10

PMC005xxxxxx/PMC5003475.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2020153microarrays-02-00153ArticleMicroarray for Identification of the Chiropteran Host Species of Rabies Virus in Canada Lung Oliver 1*Nadin-Davis Susan 2Fisher Mathew 1Erickson Anthony 1Knowles M. Kimberly 2Furukawa-Stoffer Tara 1Ambagala Aruna 11 Canadian Food Inspection Agency, National Centres for Animal Disease, Lethbridge Laboratory, P.O. Box 640, Lethbridge, AB T1J 3Z4, Canada; E-Mails: Mathew.Fisher@inspection.gc.ca (M.F.); Anthony.Erickson@inspection.gc.ca (A.E.); Tara.Furukawa-Stoffer@inspection.gc.ca (T.F.-S.); Aruna.Ambagala@inspection.gc.ca (A.A.)2 Canadian Food Inspection Agency, Ottawa Laboratory Fallowfield, 3851 Fallowfield Road, P.O. Box 11300, Ottawa, ON K2H 8P9, Canada; E-Mails: Susan.Nadin-Davis@inspection.gc.ca (S.N.-D.); Kim.Knowles@inspection.gc.ca (M.K.K.)* Author to whom correspondence should be addressed; E-Mail: Oliver.Lung@inspection.gc.ca; Tel.: +1-403-382-5589; Fax: +1-403-381-1202.31 5 2013 6 2013 2 2 153 169 13 4 2013 17 5 2013 17 5 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Species identification through genetic barcoding can augment traditional taxonomic methods, which rely on morphological features of the specimen. Such approaches are especially valuable when specimens are in poor condition or comprise very limited material, a situation that often applies to chiropteran (bat) specimens submitted to the Canadian Food Inspection Agency for rabies diagnosis. Coupled with phenotypic plasticity of many species and inconclusive taxonomic keys, species identification using only morphological traits can be challenging. In this study, a microarray assay with associated PCR of the mitochondrial cytochrome c oxidase subunit I (COI) gene was developed for differentiation of 14 bat species submitted to the Canadian Food Inspection Agency from 1985–2012 for rabies diagnosis. The assay was validated with a reference collection of DNA from 153 field samples, all of which had been barcoded previously. The COI gene from 152 samples which included multiple specimens of each target species were successfully amplified by PCR and accurately identified by the microarray. One sample that was severely decomposed failed to amplify with PCR primers developed in this study, but amplified weakly after switching to alternate primers and was accurately typed by the microarray. Thus, the chiropteran microarray was able to accurately differentiate between the 14 species of Canadian bats targeted. This PCR and microarray assay would allow unequivocal identification to species of most, if not all, bat specimens submitted for rabies diagnosis in Canada. batsmicroarrayrabiesChiropteraCOI ==== Body 1. Introduction The emergence of genetic tools to explore species differences has had a tremendous impact on our knowledge of global biodiversity and has in some instances provided greater insight into higher order taxonomic relationships [1]. In particular, due to the mitochondrial genome’s limited intra-specific variation but significant inter-specific divergence, partial nucleotide sequencing (barcoding) of certain mitochondrial loci for the purpose of species discrimination has gained widespread acceptance [2,3]. Two mitochondrial loci, cyt b (cytochrome b) and COI (cytochrome c oxidase subunit I), have gained particular prominence in this role [2,4]. Based on the pioneering work of Hebert and colleagues [5,6], the international barcode of life (IBOL) consortium, which seeks to generate millions of DNA barcodes for thousands of species world-wide, has adopted COI barcodes as the standard for studies on biodiversity throughout the animal kingdom [7]. Barcoding of one mammalian order, the Chiroptera, has increased our knowledge of diversity within this group significantly [8,9]. With approximately 1,240 species worldwide, the chiroptera is the second largest order of mammals, comprising more than 20% of the species in the class Mammalia [10,11,12]. Moreover bats are now recognized as important hosts involved in the emergence and spread of animal and zoonotic viruses belonging to a wide range of families including the Rhabdoviridae (e.g., Rabies virus), Paramyxoviridae (e.g., Nipah and Hendra virus), Coronaviridae (e.g., SARS-CoV), and Filoviridae (e.g., Ebola and Marburg virus) [13,14,15,16]. Indeed many bat species figure prominently as the reservoirs for members of the Lyssavirus genus, family Rhabdoviridae, a group of negative strand RNA-viruses that are the causative agents of the zoonotic disease rabies [15,17]. Rabies virus (RV) is the type species of this genus and the only Lyssavirus known to circulate in the Americas [17]. Throughout the continent, including Canada, multiple RV strains are maintained by specific wildlife hosts, of both carnivora and chiroptera orders, within discrete but overlapping geographic areas and there is a need for tools that discriminate both host species and viral strains [18,19,20]. Currently, hundreds of confirmed rabid bats are reported annually in North America [17] and the persistence of RV in a broad range of native bat species continues to be a public health hazard due to the high fatality rate of this disease and frequent failure of the public to associate bat contact and even superficial bites with potential rabies exposure [21,22]. While timely post-exposure or pre-exposure vaccination can prevent rabies in humans and other mammals, after onset of clinical disease the prognosis is extremely poor with almost 100% mortality [23]. Additionally, the persistence of RV in chiropteran hosts provides a source of potential cross-species infection for domestic animals and can, on rare occasions, result in the establishment of bat RV variants in terrestrial wildlife [24], events which have the potential to undermine rabies control activities in non-flying RV reservoir species. Improved knowledge of the epidemiology and geographical distribution of bat-associated RVs, including the species responsible for disease maintenance, would facilitate better risk management of this disease and limit its impact on the health of humans and domestic animals. In Canada, rabies diagnosis is routinely performed by the Rabies Centre of Expertise of the Canadian Food Inspection Agency (CFIA). Large numbers of bats are submitted annually and species determination is currently based on traditional taxonomic methods in which the specimen is visually examined for anatomical features consistent with each species [25,26,27]. Although 17 species of bats are considered indigenous to Canada [28] three species, the spotted bat (Euderma maculatum), eastern small-footed bat (Myotis leibii), and the western small-footed bat (Myotis ciliolabrum) rarely come in contact with humans due to their scarcity and/or their very limited ranges within Canada. As a result they are virtually never submitted for rabies testing. The pallid bat (Antrozous pallidus) is also uncommon and very infrequently submitted so the vast majority of submitted specimens represent just 13 species. The bat samples received are often in poor condition due to decomposition or injury, or in some cases are incomplete (e.g., head only), thereby making species identification difficult or even impossible. In addition, some species (particularly of the Myotis genus) share highly similar physical features which further challenge species identification of these samples. To address these issues a recent study, which generated barcodes for 260 bat specimens representing 13 Canadian bat species, proposed the use of COI barcoding methods to improve species identification of submitted specimens [29]. Similar strategies involving sequence determination of other mitochondrial loci have been applied to confirm the nature of rabies reservoir species in Latin America [30], and to expand knowledge of the epidemiology of bat lyssaviruses in Europe [31]. However, to facilitate routine use of such an approach in a diagnostic setting it was determined that a highly automated method which could be applied by analysts not expert in complex molecular biology techniques would be highly valuable. Given the capability of DNA microarray technology to undertake massively parallel analyses in widely diverse biological systems [32,33], a project to develop a rapid and robust method for Canadian bat species identification which combines microarray and bat barcoding technologies has been initiated. A number of studies have combined microarray and DNA barcoding methods for identification of mammalian [34,35], insect [36], fish [37], bird [38], and fungal [39] species. Using these same principles this report describes the development of a universal PCR for amplification of the COI gene and the development of a microarray for differentiation of 14 Canadian bat species that represent virtually all specimens submitted to the CFIA for rabies diagnosis for the past 20 years. This study demonstrates the feasibility of this approach as a first step in the development of a novel user-friendly molecular method for bat species identification, a tool which will be invaluable in supporting bat rabies surveillance in Canada. 2. Experimental Section 2.1. Bat Specimens Specimens used in this study were randomly selected from frozen archived bat tissues submitted to CFIA laboratories located in Ottawa, Ontario and Lethbridge, Alberta for rabies diagnosis between the years 1985–2012 (Table 1). Submitted specimens are routinely examined upon arrival at the laboratory by diagnostic staff and assigned a species designation based on specific morphological keys. All of the samples included in this study had also been assigned to species based on COI barcoding as described previously [29]. microarrays-02-00153-t001_Table 1Table 1 Bat species investigated in study. Species Abbreviation Scientific Name Numbers Tested Little brown bat LBB Myotis lucifugus 24 Northern long-eared bat NLB Myotis septentrionalis 13 California bat CLB Myotis californicus 7 Long-legged bat LLB Myotis volans 5 Western long-eared bat LEB Myotis evotis 12 Yuma bat YUB Myotis yumanensis 8 Keen’s bat KEB Myotis keenii 14 Big brown bat BBB Eptesicus fuscus 25 Townsend’s big-eared bat WEB Corynorhinus townsendii 9 Hoary bat HRB Lasiurus cinereus 10 Eastern red bat REB Lasiurus borealis 9 Silver-haired bat SHB Lasionycteris noctivagans 12 Tricolored bat * EPB Perimyotis subflavus 3 Pallid bat PAB Antrozous pallidus 2 Total 153 * Formerly referred to as the Eastern pipistrelle bat (Pipistrellus subflavus). 2.2. Primer and Probe Design A custom database was initially developed that contained 365 American and 685 Canadian bat COI gene sequences obtained from NCBI, Barcode of Life Data Systems (BOLD) [7] and CFIA ([29] and Nadin-Davis, unpublished results). Due to the presence of gaps in the sequence information, several sets of PCR primers were designed to sequence the COI gene of the targeted species. Sequence across the entire or most of the COI gene was obtained for all targeted species. Multiple sequence alignments were performed using Clone Manager Professional Version 9 (Science & Educational Software) and used to identify highly conserved regions which were then used to design PCR primers for amplification of the COI gene of the targeted species. For probe design, COI sequences generated from 13 Canadian bat species for which specimens were initially available were aligned using ClustalXv2 or ClustalXv1.83 [40,41] and then reviewed manually to identify regions that were conserved within a species but which contained multiple substitutions when compared with all other species. For each species at least two discrete regions of the gene were identified and probes targeting these sequences were designed and optimized using the OligoAnalyzer® tool [42]. Once a sequence for the previously unavailable pallid bat COI gene became available, this information was included in subsequent alignments and a pallid bat probe was designed using the program AlleleID (Premier Biosoft International). This probe was subsequently examined for specificity by BLAST analysis against full length COI sequences of the targeted bat species. 2.3. DNA Extraction Approximately 50 mg of brain, lung, spleen, or hair bud were taken from frozen bat specimens for DNA extraction using either of two methods. In most cases the sample was ground in a hexadecyl trimethyl ammonium bromide (Sigma-Aldrich, Oakville, ON, Canada) solution, subjected to phase separation after addition of chloroform and the aqueous phase was then precipitated with alcohol as described previously [43]. The final dried DNA pellet was dissolved in TE buffer (10 mM Tris-HCl, pH 8.0, 0.1 mM EDTA) and stored at −20 °C. Alternatively, some samples were extracted using a DNeasy Blood and Tissue Kit (Qiagen, Toronto, ON, Canada) as per the manufacturer’s recommendations. Nucleic acid concentration was quantified by UV absorbance at 260 nm on a NanoDrop 8000 (Thermofisher Scientific, Toronto, ON, Canada) or Nanovue instrument (GE Healthcare, Mississauga, ON, Canada). 2.4. Slot Blot Analysis Slot blot analysis was used in the preliminary screening of capture probes. COI PCR products, amplified as described previously [29] and purified using Wizard® PCR Preps DNA Purification System (Promega, Madison, WI, USA) according to the manufacturer’s instructions, were normalized to similar concentrations based on their intensity upon gel electrophoresis. These amplicons were then blotted on to Hybond nylon membranes (GE Healthcare) using a Minifold II slot blot system (Schleicher & Schuell, Keene, NH, USA) followed by fixation using UV cross-linking. Oligonucleotide probes were labelled by 3′ tailing with digoxigenin (DIG) using DIG-11-dUTP and terminal transferase (Roche Applied Science, Laval, QC, Canada) as recommended by the manufacturer. The slot blots were incubated in pre-hybridization buffer (5× SSC, 1% casein, 0.1% N-lauryl sarcosine, 0.02% SDS) for 1 h and the DIG-labelled probes were then added and allowed to hybridize overnight at either 37 °C or 42 °C. After hybridization, blots were washed sequentially in 0.1% SDS containing 2× SSC, 1× SSC and 0.5× SSC at the hybridization temperature, and once in Buffer 1 (0.1 M Tris-HCl, 0.15 M NaCl, pH 7.5) at room temperature. All remaining manipulations were conducted at room temperature. The blots were then blocked for 1 h in blocking buffer (Buffer 1 containing 1% casein) prior to application of anti-DIG alkaline phosphatase (AP) conjugate (Roche Applied Science) diluted 1:2,000 in blocking buffer for 30–60 min. Unbound conjugate was removed by washing twice in buffer 1 for 15 min each time, and once in Buffer 3 (0.1 M Tris-HCl, 0.1 M NaCl, 50 mM MgCl2, pH 9.6) for 10 min. Fresh Buffer 3 containing NBT/BCIP substrate (Roche Applied Science) was applied and colour development proceeded in the dark for 1–2 h until it was terminated by washing in water. Blots were subsequently imaged using a Gel Doc system (Bio-Rad Laboratories, Mississauga, ON, Canada). 2.5. PCR Amplification for Microarray Analysis Approximately 200 ng of DNA was used for PCR amplification of the COI gene. All PCR amplifications were performed in 50 μL reactions using Platinum® Taq DNA Polymerase (Life Technologies, Burlington, ON, Canada). PCR consisted of an initial denaturation step at 94 °C for 3 min, followed by 40 cycles of 94 °C for 45 s, 45 °C for 45 s, 72 °C for 1 min, and a final extension step of 72 °C for 10 min. The 50 μL reaction mixture consisted of 2 μL of DNA template, 0.2 μL of enzyme mix in 5 µL 10× reaction buffer, 1.5 μL of 50 mM MgCl2, 1 μL of 10 mM dNTP mixture, and 1 μM of each primer. The COI primers used for PCR amplification were AECOX-1TAG FWD-3.0 and AECOX-R1A (details are provided in Section 3.1). Alternate primers used to amplify a badly decomposed specimen were COX-F4 (5′-TCAACCAATCAYAAAGAYATTGGTAC-3′) and COX-R4 (5′-GTGAAYATATGGTGGGCTCATACGAT-3′). Following PCR, amplicons were visualized either on the QIAxcel instrument (Qiagen) or on a 1% agarose gel. The remainder of the amplified products were fluorescently labelled using the ARES Alexa Fluor 647 Labelling Kit (Life Technologies) as previously described [44]. Briefly, following amplification, PCR products were hydrolyzed to remove RNA and then purified using DNA Clean and Concentrator spin columns (Zymo Research, Irvine, CA, USA). Aminoallyl dUTP was incorporated by random priming and the product was purified using spin columns, and labelled using Alexa Fluor 647 dye. The fluorescently labelled samples were purified using spin columns and stored at 4 °C until use. 2.6. Microarray Capture probe printing, hybridization, washing and reporting were performed as previously described [44]. Briefly, probes in 15 μL Pronto!TM Epoxide Spotting Solution (Corning, Tewksbury, MA, USA) were spotted in triplicate on Epoxide Slides (Corning) at 60% humidity using the VersArray ChipWriter TM Pro printer (Bio-Rad) and SMP3 pins (ArrayIt, Sunnyvale, CA, USA). Following printing, the humidity was increased to 70% overnight and then gradually reduced to ambient humidity the following day. Printed slides were stored in a desiccating chamber until use. Printed slides were blocked by incubation in pre-hybridization buffer (5× SSC, 0.1% SDS and 0.1 mg/mL BSA) for 45 min at 42 °C. Slides were then washed three times in 0.1× SSC and once in Milli-Q water before being dried using a slide centrifuge (ArrayIt), and loaded onto a 24-well hybridization cassette (ArrayIt). Twenty-one microlitres of labelled amplicon was mixed with an equal volume of hybridization buffer (20% formamide, 10× SSC, 0.2% SDS, and 0.2 mg/mL salmon sperm DNA), denatured at 95 °C for 5 min and then cooled to room temperature. The labelled amplicon and hybridization buffer mixture was loaded into individual wells of the hybridization cassette, which was then sealed and incubated overnight at 42 °C. The following day, hybridized slides were washed once in 2× SSC, 0.1% SDS at 42 °C, twice in 1× SSC and twice in 0.1× SSC at room temperature. Finally, slides were dried by centrifugation and scanned using the GenePix 4000B scanner (Molecular Devices, Sunnyvale, CA, USA). Image analysis was performed using GenePix Pro software version 5.0 (Molecular Devices). The reactivity of each sample to the capture probes was represented using the average median fluorescent intensity (MFI) of the replicates. Probe reactivity was considered positive if it had an MFI greater than the mean MFI of the whole probe set plus two times the standard deviation. This cut-off value was calculated individually for each sample. 3. Results and Discussion 3.1. Development of a Universal PCR Primer Pair The 14 indigenous Canadian bat species targeted by this study are listed in Table 1. A previous study of the variation of the COI gene in 13 Canadian bat species required the use of four separate PCR reactions for successful amplification of the COI gene from all the species [29]. Refinement of this PCR to facilitate target amplification by a single universal primer set was a prerequisite for efficient microarray analysis. Accordingly, an alignment of all available 685 Canadian bat COI gene sequences was examined to assist with primer design. This revealed gaps in the required sequence information. Thus multiple sets of PCR primers flanking the COI gene were designed and used to amplify the entire COI gene of single specimens for 13 of the 14 targeted Canadian bat species, and the majority of the COI gene of the remaining species (tricolored bat). The information obtained from sequencing the COI amplicons allowed design of a single primer pair for partial amplification of the COI gene of all 14 species. These COI primers were AECOX-1TAG FWD-3.0 (5′-TAGRTTTACAGYCTAATRCCTACTC-3′) and AECOX-R1A (5′-ACYTCAGGRTGRCCAAARAATCA-3′). 3.2. PCR for Partial Amplification of the Cytochrome c Oxidase Subunit I (COI) Gene of Canadian Bats The PCR developed in this study generated amplicons of 766 bp (Figure 1). The PCR was validated using DNA extracted from 153 Canadian field specimens of the 14 bat species. While many of the DNA samples included in this study were of poor quality based on analysis by gel electrophoresis (data not shown), all but one specimen were successfully amplified, including multiple samples from each target species. Thus these results indicate that for the vast majority of samples that were encountered, the universal PCR primers designed in this study can successfully replace the four separate PCRs [29] previously used to amplify the partial COI gene of the targeted bat species. The one sample that failed to amplify was badly decomposed and a weak product was only generated after switching to alternate primers that produced a slightly different amplicon of approximately the same size (data not shown). As shown in the following sections and Table 2, for the purpose of this application a sufficiently large COI sequence which encompasses significant genetic diversity to allow use of multiple capture probes is necessary for accurate species identification. However given the possibility that other suboptimal field specimens are likely to be encountered, reduction in amplicon size may improve PCR product yields as described in other studies which used smaller microcodes [45,46]. Thus amplification of this region of COI as two smaller multiplexed PCRs each <400 bp could be entertained. Figure 1 Gel image of PCR amplified material from representative samples of 14 Canadian bat species targeted in this study. A primer pair developed in this study was used for the PCR and the targeted cytochrome c oxidase subunit I (COI) regions of all 14 bat species were successfully amplified. microarrays-02-00153-t002_Table 2Table 2 Description of all oligonucleotides evaluated for their use as species-specific probes. Name Specific Target Sequence (5′-3′) Location 1 BB1 2 BBB CTGCCCTGAGTCTGCT 130–145 BB2 2 BBB GTGGACCTGACCATT 465–479 BB-E 2 BBB (eastern) CTTTTATTCGGCGCTTGA 96–113 BB-W 2 BBB (western) TTCTGTTCGGCGCCTGA 97–113 EP1 EPB CGCACACGCCTTTG 215–228 EP2 2 EPB CTTTCTTTTACTCCTAGCAT 362–381 LAS HRB/REB TATTATTGGCATCWTC 370–385 HR1 2 HRB AGTACCACTGATGATTGG 284–301 HR2 2 HRB CCACCTGCCTTGTCC 555–569 RE1 2 REB TCTATAACGTTATCGTGAC 196–214 RE2 2 REB GGTACCCCTTATRATCG 284–300 SH1 SHB GGGCTCTACTTGGAGAT 172–188 SH2 2 SHB CAACTGGTTAGTTCCTCTG 275–293 WE1 2 WEB TGATTAATCCCACTAATGAT 279–298 WE2 2 WEB CTTACATCTGGCTGG 485–497 CL1 2 CLB CTGGGAGACGATCAAATT 180–197 CL2 2 CLB TTTTATAGTTATGCCAATCAT 239–259 LB1 2 LBB CGCTGAGCTAGGTCAA 152–167 LB2 2 LBB TTTCTATTACTGCTGGC 363–379 LE1 LEB TGACATAGCCTTTCC 308–322 LE2 2 LEB CTGTCTTACTCCTTCTC 613–629 LL1 LLB TGTTGGGGGACGATCAGA 178–195 LL2 2 LLB TCTTCTCTCTGCACTTA 478–494 NL1 NLB AACTGGGCCAGCCA 157–170 NL2 2 NLB ATTCGTTTGGTCCGT 587–601 NL3 2 NLB ATTCGTTTGGTCTGT 587–601 KE1 2 KEB TCATAGTTATGCCCATTA 241–258 KE2 KEB TATGCCCATTATAATTGG 248–265 YU1 YUB CCCTTTTAGGGGATG 175–189 YU2 2 YUB TATAGTAATGCCGATTATAATC 242–263 PA 2 PAB TAATGTAATTGTCACAGCA 200–218 1 Location of targeted sequence within the 766 bp amplicon which corresponds to positions 5,307 to 6,072 of a Lasiurus borealis mitochondrial genome (Genbank Accession NC_016873); 2 Probes selected for the minimal microarray panel to differentiate 14 Canadian bat species. 3.3. Evaluation of Probes by Slot Blot Analysis Thirty candidate species-specific oligonucleotide probes were designed based on previously published partial COI gene sequences [29]. This included a broadly reactive LAS probe designed to detect both species of the Lasiurus genus, the eastern red bat (REB) and hoary bat (HRB), included in this study. A total of four probes directed to the big brown bat (BBB) were designed; some of these were designed to differentiate between bats of the western population (restricted to British Columbia) and those of an eastern population that is distributed across the rest of the country (from Alberta to Eastern Canada) as described previously [47]. This collection of probes did not include a probe for the pallid bat (PAB), as information on this species was not available until later in the study when this probe was incorporated into subsequent microarray analysis. The reactivity and specificity of the 30 DIG-labelled probes on slot blots were first examined using COI PCR products generated from representative members of each of the initial 13 species. At a hybridization temperature of 37 °C, all probes exhibited strong binding to their targeted species and many showed either no or very weak cross reactivity with other species, as illustrated for selected probes in Figure 2. Figure 2 Composite diagram of 22 representative slot blots. Similar quantities of COI PCR product generated from 13 Canadian bat species were transferred to nylon membrane and each blot was hybridized at 37 °C with a probe as described. The bat species in which the PCR amplicon was derived and applied to each slot is indicated across the top, and the probes used for each blot are shown on the right hand side. Water was used in place of a PCR product for the negative (-ve) control. The probes included in this representation of the complete data were those chosen for the minimal microarray protocol with the exception of the PA probe that was not evaluated by slot blot analysis. The BB1 probe bound to both big brown bat (BBB) sub-types while BB2 bound strongly only to the eastern BBB type. The BB-E and BB-W probes behaved as expected by detecting only the targeted sub-type. Probes that exhibited some moderate cross-reactivity included: LB1, which bound to the silver-haired bat (SHB) product and CL1 which bound to the northern long-eared bat (NLB) product (Figure 2). The probes for Yuma bat (YUB) and Keen’s bat (KEB) were more problematic; YU2 cross-reacted moderately with both the western-BBB and tricolored bat (EPB) samples (Figure 2) although this appeared to be substantially reduced when the hybridization temperature was raised to 42 °C (data not shown). Both probes for KEB exhibited significant binding to some of the other species; strong binding of the probe KE1 to products of the Lasiurus bats (HRB and REB), as shown in Figure 2, was partly reduced by increasing the hybridization temperature to 42 °C but this did not entirely eliminate cross-reactivity with the HRB sample. However, overall these results were promising and more extensive evaluation of all probes was subsequently conducted in a microarray format. 3.4. Evaluation and Validation of Probes by Microarray All 30 candidate species-specific capture probes from the slot blot study and a new pallid bat probe were printed on microarray slides and screened for their utility in differentiation of the 14 targeted Canadian bat species. The probes were tested using DNA amplified from 153 field specimens. Each of the 14 bat species (Table 1) was represented by multiple specimens. Probes that showed significant cross-reactivity with heterologous species were eliminated. One or more capture probes were selected for each species and all 14 species were accurately differentiated using a set of 23 probes (Figure 3(A)). The sequences of all capture probes and their relative positions within the 766 bp amplicon are presented in Table 2; a subset was identified as a minimal discriminatory panel. For four of the 153 specimens tested, species assignments were inconsistent with the original species designation based on morphological criteria. DNA sequences of amplicons derived from discordant samples were consistent with species designation based on microarray results and with previous barcoding results (data not shown), suggesting errors in the initial phenotypic species designation. Species-specific probes for 13 of the 14 species in the final probe set were highly specific to the intended species and did not show cross reactivity with heterologous species. The Keen’s bat (KEB) probe (KE1) showed some cross reactivity with specimens from two heterologous species, but did not affect accurate species identification. All nine eastern red bat (REB) specimens tested were detected by the RE probes (RE1 and RE2) as expected. However, one REB sample (#116) also showed strong reactivity with the KE1 probe for Keen’s bat (P/N = 99.0) that was not observed with the other eight REB specimens (Figure 3(A)). Sequencing of the amplicon from sample #116 revealed that the KE1 probe binding region had a single mismatch with the penultimate nucleotide of the KE1 probe, while the amplicon from the other eight REB specimens had an additional mismatch with a nucleotide near the center of the KE1 probe (Table 3). Only four out of 42 COI sequences from REB in the custom database had the same sequence as sample #116 in the KE1 probe binding region while 38 out of 42 (90.5%) had two mismatches in the KE1 binding region. The observation that the majority of the REB CO1 sequences present in the database have two mismatches with the KE1 probe suggest that most REB specimens will likely not react with this probe. The existence of a small number of REB sequences with polymorphisms that allows binding to the KE1 probe indicates that samples which react with both RE and KE1 probes should be categorized as an eastern red bat (REB). Figure 3 Summary heat maps of microarray results for field specimens. Specimen number and species are listed above the heat map while probes are listed to the right of the heat map. Positive reactions are indicated in red, and negative reactions in black. Panel A: Heat map of results for all 153 field specimens. A cutoff of two times the standard deviation of the fluorescent intensity for all probes plus the mean fluorescent intensity of all probes was selected for each sample for positivity. Specific probes for each sample are indicated by a yellow box. All specimens were accurately assigned. The outlying signal with the KE1 probe is due to cross-reactivity with an eastern red bat (REB) specimen with a polymorphism which resulted in just a single mismatch to the Keen’s bat probe (KE1). Panel B: Heat map of microarray results for 25 big brown bat (BBB) field specimens within the panel. A cutoff of 0.25 times the standard deviation of the fluorescent intensity for all probes plus the mean fluorescent intensity of all probes was selected for each sample for positivity. All BBB specimens were accurately classified into western (British Columbia) and eastern (rest of Canada) populations. microarrays-02-00153-t003_Table 3Table 3 Polymorphism observed in the KE1 capture probe binding region of eastern red bats (REB). Sample/Probe Sequence (5′-3′) REB #116 TTA TAG TTA TGC CCA TTA 1 REB #115 TTA TAG TCA TGC CCA TTA 1 KE1 Probe TCA TAG TTA TGC CCA TTA 1,2 KEB #51 TCA TAG TTA TGC CCA TTA 1 The sequence represented by REB #116 contains a single mismatch (shown in bold letters) with the KE1 probe and is observed in approximately 9.5% of the REB sequences in the database. The sequence represented by REB #115 contains two mismatches (shown in bold letters) with the KE1 probe and is observed in approximately 90.5% of the REB sequences in the database. 2 The polymorphism in the middle of the KE1 probe binding region (shown underlined) is a determinant of reactivity with the KE1 probe. Collectively, capture probes for a particular species detected all specimens of that species. However, amplicons from a few of the species did not react with all probes for that species. This is likely due to polymorphisms that exist within the probe binding regions. For example, recent barcoding data divided all Canadian big brown bats (BBB) (Eptesicus fuscus) into two distinct populations (western and eastern) [47]. A capture probe designed from available sequences of BBB from British Columbia (BB-W) and a probe designed to sequences of BBB from the rest of Canada (BB-E) were able to differentiate the two populations using Mean + 0.25SD as a cutoff for positivity (Figure 3(B)). It has been suggested that certain members of the Myotis genus, notably M. keenii, M. evotis and M. lucifugus, are not clearly differentiated by mitochondrial genome analyses and are accordingly often referred to as the M. lucifugus complex [48]. However, in a previous study of Canadian bat barcoding, specimens of these three species were clearly assigned to discrete phylogenetic clades with the exception of just one specimen that could not unequivocably be designated as M. keenii or M. evotis [29]. The present study has employed a subset of the collection examined previously so that the phylogeny of all of these samples was already known. Unfortunately the unclassified specimen was not available for inclusion in this study but interrogation of its sequence revealed that it contained no mismatches with probe LE2, one mismatch with KE2 and two mismatches with both LB probes; accordingly it would most likely have been identified as M. evotis. Apart from this single sample, at least from a Canadian perspective, identification of these three species by barcoding and microarray analysis appears to be straightforward and a relatively large number of specimens belonging to these species have been tested in the current study. However, further COI characterization of specimens of these species throughout their ranges would be helpful in further defining the degree of inter-specific and intra-specific sequence diversity exhibited at this locus. The PCR and microarray assay described can be used to genetically identify Canadian bat specimens to the species level using DNA extracted from small amounts of tissue from bat carcasses. This method could complement traditional taxonomic methods of species assignment especially when specimens are deteriorated, incomplete, or are of species that share highly related morphological features. Although only 14 of the 17 indigenous bat species were included in this study it is anticipated that as genetic information on the three remaining species is acquired, extra probes will be added to the microarray to allow their detection. Moreover, it is quite possible that climate change will result in the future northwards movement of bat species from the USA into Canada such that expansion of the array to enable identification of additional bat species will be required. One of the advantages of the microarray is the large number of capture probes that can be incorporated to provide additional redundancy and to increase assay resolution for differentiation of a larger number of species. This feature could potentially broaden the application of this technology to areas such as wildlife survey and conservation. Iatrogenic incursions of bats from further afield is also possible though probably highly infrequent; such specimens would likely give negative or inconclusive results with the microarray described and will likely require traditional barcoding or high density resequencing arrays for sequence characterization. A number of prior studies have combined barcoding and microarray methods, mostly by employing traditional glass slide microarrays, to facilitate species identification. The scope of these prior studies varied widely. In one study Affymetrix technology was used to develop methods for the identification of a wide range of mammalian species [35]; another study targeted a range of European fish species [37]. Some studies have focused on more limited numbers of species restricted to specific regions and/or animal orders [34,38] while others have explored the utility of microarrays to identify species likely to transmit infectious disease agents [36,38]. While the present study has used a similar approach to design and validate a tool for a very specific purpose, further development of this technique could enlarge its applicability. One possibility is to expand the array for use in other jurisdictions by adding probes for the differentiation of a larger number of species, e.g., the 39 species of indigenous North American bats for which cases of rabies have been reported [49]. The principle could also be extended to the analysis of the much larger number of bat species that inhabit tropical regions of the Americas, many of which harbor RV variants [50]; indeed an array that identifies many neotropical bat species from Guyana based on their barcodes [8] has been described previously [35]. Alternatively, the adaptation of this method to novel far less labour intensive microarray platforms could be beneficial. For example, use of a single instrument that fully integrates and automates a complex molecular assay from nucleic acid extraction and gene amplification to array-based detection would yield a highly user-friendly microarray platform with a much more streamlined workflow. An instrument that uses an inexpensive bio-contained disposable reaction cartridge to perform such a workflow has recently been developed and applied to the detection of viruses responsible for human sexually transmitted diseases [51]. A fully automated and integrated assay for molecular species identification that does not require user handling after sample addition may be more cost-effective and rapid than DNA sequencing for applications described in this study. Thus, future work will focus on the adaptation of the methods described in this report to such highly automated platforms. As a result genetic-based tools for bat species identification will become more accessible, not only for laboratory based analysis but also for studies of bats in the field where live specimens could be rapidly and unequivocally identified using wing-punches or hair buds. 4. Conclusions A new PCR that successfully amplified the COI gene of a reference collection of 152 out of 153 bats representing 14 different species of bats that have been submitted for rabies diagnosis in Canada was developed. Amplification of COI from a highly degraded sample required an alternate primer pair. A microarray-based method for identification of chiropteran specimens to the species level was also successfully developed and validated with 153 field samples. This chiropteran PCR and microarray assay would allow unequivocal identification to species of most, if not all, bat specimens diagnosed as rabid in Canada. This new test for species assignment of rabies positive bats can be used in conjunction with existing methods for rabies virus typing to provide scientific information into the host range and host association of various rabies virus variants in bat species in Canada. The associations made between bat species and rabies virus variants will improve knowledge of rabies epidemiology and provide much needed information to support ongoing surveillance of rabies and to assess the public health risk to Canadians. Further planned expansion of this barcoding microarray assay to include more probes for redundancy and allow differentiation of additional species, and adaptation to a fully automated platform would make this technology accessible to a wider group of potential users such as those involved in wildlife conservation. Acknowledgments Funding for this study was provided by Canadian Food Inspection Agency’s Technology Development Program. The authors would like to acknowledge Kim Burton-Hughes, Josephine Kush and Cody Buchanan for technical assistance and Kingsley Amoako and Zaheer Iqbal and three anonymous reviewers for suggestions regarding the manuscript. Conflict of Interest The authors declare no conflict of interest. ==== Refs References 1. Arnason U. Adegoke J.A. Bodin K. Born E.W. Esa Y.B. Gullberg A. Nilsson M. Short R.V. Xu X. Janke A. Mammalian mitogenomic relationships and the root of the eutherian tree Proc. Natl. Acad. Sci. USA 2002 99 8151 8156 10.1073/pnas.102164299 12034869 2. Frézal L. Leblois R. Four years of DNA barcoding: Current advances and prospects Infect. Genet. Evol. 2008 8 727 736 10.1016/j.meegid.2008.05.005 18573351 3. Hajibabaei M. Singer G.A.C. Hebert P.D.N. Hickey D.A. DNA barcoding: How it complements taxonomy, molecular phylogenetics and population genetics Trends Genet. 2007 23 167 172 10.1016/j.tig.2007.02.001 17316886 4. Tobe S.S. Kitchener A.C. Linacre A.M.T. Reconstructing mammalian phylogenies: A detailed comparison of the cytochrome b and cytochrome oxidase subunit I mitochondrial genes PLoS ONE 2010 5 e14156 10.1371/journal.pone.0014156 21152400 5. Hebert P.D.N. Cywinska A. Ball S.L. de Waard J.R. Biological identification through DNA barcodes Proc. R. Soc. Lond. B 2003 270 313 321 10.1098/rspb.2002.2218 6. Hebert P.D.N. Ratnasingham S. de Waard J.R. Barcoding animal life: Cytochrome c oxidase subunit 1 divergences among closely related species Proc. R. Soc. Lond. B 2003 270 S96 S99 10.1098/rsbl.2003.0025 7. Ratnasingham S. Hebert P.D.N. BOLD: The barcode of life data system Mol. Ecol. Notes 2007 7 355 364 18784790 8. Clare E.I. Lim B.K. Engstrom M.D. Eger J.I. Hebert P.D.N. DNA barcoding of Neotropical bats: Species identification and discovery within Guyana Mol. Ecol. Notes 2007 7 184 190 10.1111/j.1471-8286.2006.01657.x 9. Mayer F. Dietz C. Kiefer A. Molecular species identification boosts bat diversity Front. Zool. 2007 4 4 10.1186/1742-9994-4-4 17295921 10. Jones K.E. Bininda-Emonds O. Gittleman J. Bats, clocks, and rocks: Diversification patterns in chiroptera Evolution 2005 59 2243 2255 16405167 11. Teeling E.C. Springer M.S. Madsen O. Bates P. O’Brien S.J. Murphy W.J. A molecular phylogeny for bats illuminates biogeography and the fossil record Science 2005 307 580 584 10.1126/science.1105113 15681385 12. Tudge C. The Variety of Life: A Survey and a Celebration of All the Creatures that Have Ever Lived Oxford University Press Oxford, UK 2000 13. Calisher C.H. Childs J.E. Field H.E. Holmes K.V. Schountz T. Bats: Important reservoir hosts of emerging viruses Clin. Microbiol. Rev. 2006 19 531 545 10.1128/CMR.00017-06 16847084 14. Wibbelt G. Moore M.S. Schountz T. Voigt C.C. Emerging diseases in Chiroptera. Why bats? Biol. Lett. 2010 6 438 440 10.1098/rsbl.2010.0267 20427329 15. Banyard A.C. Hayman D. Johnson N. McElhinney L. Fooks A.R. Bats and lyssaviruses Adv. Virus Res. 2011 79 239 289 10.1016/B978-0-12-387040-7.00012-3 21601050 16. Wong S. Lau S. Woo P. Yuen K.Y. Bats as a continuing source of emerging infections in humans Rev. Med. Virol. 2007 17 67 91 10.1002/rmv.520 17042030 17. Kuzmin I.V. Rupprecht C.E. Bat rabies Rabies 2nd ed. Jackson A.C. Wunner W.H. Academic Press San Diego, CA, USA 2007 259 307 18. Hanlon C.A. Niezgoda M. Rupprecht C.E. Rabies in terrestrial animals Rabies 2nd ed. Jackson A.C. Wunner W.H. Academic Press San Diego, CA, USA 2007 201 258 19. Nadin-Davis S.A. Fehlner-Gardiner C. Lyssaviruses: Current trends Adv. Virus Res. 2008 71 207 250 10.1016/S0065-3527(08)00005-5 18585530 20. Nadin-Davis S.A. Huang W. Armstrong J. Casey G.A. Bahloul C. Tordo N. Wandeler A.I. Antigenic and genetic divergence of rabies viruses from bat species indigenous to Canada Virus Res. 2001 74 139 156 10.1016/S0168-1702(00)00259-8 11226582 21. De Serres G. Dallaire F. Côte M. Skowronski D.M. Bat rabies in the United States and Canada from 1950 through 2007: Human cases with and without bat contact Clin. Infect. Dis. 2008 46 1329 1337 10.1086/586745 18419432 22. Jackson A.C. Fenton M.B. Human rabies and bat bites Lancet 2001 357 10.1016/S0140-6736(00)04852-2 23. WHO Fact Sheet. “Rabies” Available online:http://www.who.int/mediacentre/factsheets/fs099/en/ (accessed on 21 May 2013) 24. Kuzmin I.V. Shi M. Orciari L.A. Yager P.A. Velasco-Villa A. Kuzmina N.A. Streiker D.G. Bergman D.L. Rupprecht C.E. Molecular inferences suggest multiple host shifts of rabies viruses from bats to mesocarnivores in Arizona during 2001–2009 PLoS Pathog. 2012 8 e1002786 10.1371/journal.ppat.1002786 22737076 25. Nagorsen D.W. Brigham R.M. Bats of British Columbia. Royal British Columbia Museum Handbook Series UBC Press Vancouver, BC, Canada 1993 26. Adams R.A. Bats of the Rocky Mountain West: Natural History, Ecology, and Conservation. Boulder University Press of Colorado Vancouver, BC, Canada 2003 27. National Audubon Society National Audubon Society Field Guide to North American Mammals Knopf Doubleday Publishing Group New York, NY, USA 1996 28. Environment Canada and Canadian Wildlife Federation Bats: Hinterland Who’s Who Available online:http://www.hww.ca/en/species/mammals/bats.html (accessed on 28 February 2013) 29. Nadin-Davis S.A. Guerrero E. Knowles M.K. Feng Y. DNA barcoding facilitates bat species identification for improved surveillance of bat-associated rabies across Canada Open J. Zool. 2012 5 27 37 10.2174/1874336601205010027 30. Carnieli P. Jr. de Oliveira F.W. Castilho J.G. Brandão P.E. Carrieri M.L. Kotait I. Species determination of Brazilian mammals implicated in the epidemiology of rabies based on the control region of mitochondrial DNA Braz. J. Infect. Dis. 2008 12 462 465 10.1590/S1413-86702008000600002 19287829 31. Harris S.L. Johnson N. Brookes S.M. Hutson A.M. Fooks A.R. Jones G. The application of genetic markers for EBLV surveillance in European bat species Dev. Biol. 2008 131 347 363 32. Schena M. Heller R.A. Theriault T.P. Konrad K. Lachenmeier E. Davis R.W. Microarrays: Biotechnology’s discovery platform for functional genomics Trends Biotech. 1998 16 301 306 10.1016/S0167-7799(98)01219-0 33. Garaizar J. Rementeria A. Porwollik S. DNA microarray technology: A new tool for the epidemiological typing of bacterial pathogens? FEMS Immunol. Med. Microbiol. 2006 47 178 189 10.1111/j.1574-695X.2006.00081.x 16831204 34. Pfunder M. Holzgang O. Frey J.E. Development of microarray-based diagnostics of voles and shrews for use in biodiversity monitoring studies, and evaluation of mitochondrial cytochrome oxidase I vs. cytochrome b as genetic markers Mol. Ecol. 2004 13 1277 1286 10.1111/j.1365-294X.2004.02126.x 15078463 35. Hajibabaei M. Singer G.A. Clare E.L. Hebert P.D. Design and applicability of DNA arrays and DNA barcodes in biodiversity monitoring BMC Biol. 2007 13 10.1186/1741-7007-5-24 36. Deblauwe I. de Witte J.C. de Deken G. de Deken R. Madder M. van Erk S. Hoza F.A. Lathouwers D. Geysen D. A new tool for the molecular identification of Culicoides species of the Obsoletus group: The glass slide microarray approach Med. Vet. Entomol. 2012 26 83 91 21973187 37. Kochzius M. Seidel C. Antoniou A. Botla S.K. Campo D. Cariani A. Vazquez E.G. Hauschild J. Hervet C. Hjörleifsdottir S. Identifying fishes through DNA barcodes and microarrays PLoS ONE 2010 5 e12620 10.1371/journal.pone.0012620 20838643 38. Chung I. Yoo H.S. Eah J. Yoon H. Jung J. Hwang S.Y. Kim C. A DNA microarray for identification of selected Korean birds based on mitochondrial Cytochrome c oxidase 1 gene sequences Mol. Cells 2010 30 295 301 20821060 39. Summerbell R.C. Lévesque C.A. Seifert K.A. Bovers M. Fell J.W. Diaz M.R. Boekhout T. de Hoog G.S. Stalpers J. Crous P.W. Microcoding: The second step in DNA barcoding Phil. Trans. Royal Soc. Lond. 2005 360 1897 1903 10.1098/rstb.2005.1721 40. Larkin M.A. Blackshields G. Brown N.P. Chenna R. McGettigan P.A. McWilliam H. Valentin F. Wallace I.M. Wilm A. Lopez R. Clustal W and Clustal X version 2.0 Bioinformatics 2007 23 2947 2948 10.1093/bioinformatics/btm404 17846036 41. Thompson J.D. Gibson T.J. Plewniak F. Jeanmougin F. Higgins D.G. The ClustalX windows interface: Flexible strategies for multiple sequence alignment aided by quality analysis tools Nucleic Acids Res. 1997 25 4876 4882 10.1093/nar/25.24.4876 9396791 42. OligoAnalyzer® Available online:https://www.idtdna.com/analyzer/Applications/OligoAnalyzer/ (accessed on 28 February 2013) 43. Desloire S. Moro C.V. Chauve C. Zenner L. Comparison of four methods of extracting DNA from D. gallinae (Acari: Dermanyssidae) Vet. Res. 2006 37 725 732 10.1051/vetres:2006031 16820136 44. Lung O. Fisher M. Beeston A. Hughes K.B. Clavijo A. Goolia M. Pasick J. Mauro W. Deregt D. Multiplex RT-PCR detection and microarray typing of vesicular disease viruses J. Virol. Methods 2011 175 236 245 10.1016/j.jviromet.2011.05.023 21620898 45. Meusnier I. Singer G.A. Landry J.F. Hickey D.A. Hebert P.D. Hajibabaei M. A universal DNA mini-barcode for biodiversity analysis BMC Genomics 2008 12 10.1186/1471-2164-9-214 46. Zinck J.M. Duffield D.A. Ormsbee P.C. Primers for identification and polymorphism assessment of Vespertilionid bats in the Pacific Northwest Mol. Ecol. Notes 2004 4 239 242 10.1111/j.1471-8286.2004.00629.x 47. Nadin-Davis S.A. Feng Y. Mousse D. Wandeler A.I. Aris-Brosou S. Spatial and temporal dynamics of rabies virus variants in big brown bat populations across Canada: Footprints of an emerging zoonosis Mol. Ecol. 2010 19 2120 2136 10.1111/j.1365-294X.2010.04630.x 20406385 48. Streicker D.G. Turmelle A.S. Vonhof M.J. Kuzmin I.V. McCracken G.F. Rupprecht C.E. Host phylogeny constrains cross-species emergence and establishment of rabies virus in bats Science 2010 329 676 679 10.1126/science.1188836 20689015 49. Constantine D.G. An updated list of rabies-infected bats in North America J. Wildl. Dis. 1979 15 347 349 480527 50. Sodré M.M. da Gama A.R. de Almeida M.F. Updated list of bat species positive for rabies in Brazil Rev. Inst. Med. Trop. São Paulo 2010 52 75 81 20464127 51. Spizz G. Young L. Yasmin R. Chen Z. Lee T. Mahoney D. Zhang X. Mouchka G. Thomas B. Honey W. Rheonix CARD® technology: An innovative and fully automated molecular diagnostic device Point Care 2012 11 42 51 22754401

PMC005xxxxxx/PMC5003477.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2020115microarrays-02-00115ArticleA Flexible Microarray Data Simulation Model Dembélé Doulaye Microarray Platform, IGBMC, CNRS-INSERM-UdS, 1 rue Laurent Fries, Parc d’Innovation, 67400 Illkirch, France; E-Mail: doulaye@igbmc.fr; Tel.: +33-388-653-52817 4 2013 6 2013 2 2 115 130 01 3 2013 07 4 2013 15 4 2013 © 2013 by the author; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Microarray technology allows monitoring of gene expression profiling at the genome level. This is useful in order to search for genes involved in a disease. The performances of the methods used to select interesting genes are most often judged after other analyzes (qPCR validation, search in databases...), which are also subject to error. A good evaluation of gene selection methods is possible with data whose characteristics are known, that is to say, synthetic data. We propose a model to simulate microarray data with similar characteristics to the data commonly produced by current platforms. The parameters used in this model are described to allow the user to generate data with varying characteristics. In order to show the flexibility of the proposed model, a commented example is given and illustrated. An R package is available for immediate use. microarray datasimulation modelR code ==== Body 1. Introduction Microarray is now a mature technology in the field of molecular biology for monitoring of gene expression profiling at the genome level. Using two-color microarray technology, the biological activities of two samples are compared on an array on which thousands of specific deoxyribonucleic (DNA) sequences are printed or synthesized in situ. Two fluorescent dyes (generally red and green) are used for labeling of the samples for hybridization. Then, after washing, scanning and quantification of images corresponding to the red and green fluorescence of the array, numerical values, or intensities are associated with each probe (gene). These values are normalized to correct for undesirable technical variations [1,2]. Using one-color microarray technology, each sample is hybridized on one array. One fluorescence color is used, and expression intensities are obtained for genes on the arrays used. These values should also be normalized [3]. For a given gene, a ratio of values from two samples is used as a measurement of its expression change. Throughout this paper, we use logarithm two scale to transform the values. Hence, we use log2 intensities and log2 ratios for sample intensities and their ratios. In general, log2 ratios are used for two-color microarray data, while log2 intensities are used for one-color microarray data. In a screening study, the goal is to select genes with differential expression from data whose characteristics are unknown. To study the performance of gene selection methods, we often use synthetic data, which can also serve in developing new methods. To properly play their role, synthetic data must resemble as closely as possible the real data they represent. This is achieved by simulating the physical phenomena through a model. The parameters of this model are approximations of the physical laws that govern the observed phenomenon, and a good knowledge of the physical laws is therefore required to obtain a good model. A complex model that takes into account several components of a phenomenon may become less flexible and may be difficult to modify in order to include new factors. In contrast, a simple model can be effective and easy to modify to take into account an unexpected situation. One way to generate synthetic data is to use real microarray values as seeds [4,5,6,7]. In [4], data were generated using a normal distribution and a real microarray dataset. This dataset was used to estimate some hyperparameters, and the percentage of the differentially expressed (DE) genes was fixed; see details in [4]. The simulation data obtained in [4] may be useful for statistical methods, but they differ from observed data, since the log2-intensities obtained vary between the unrealistic bounds 16 and 30. In [5], two samples are selected from the control and the test samples. The measurements associated with their genes are modified (exchange of values between the two samples…) in order to obtain two statistically undistinguishable samples. Finally, a given number of down- and up-regulated genes is used in this dataset. This procedure necessitates a real microarray dataset, which cannot be available. The procedure proposed in [6] is a modular system, including all steps of the microarray technology. This includes slide layout, hybridization, scanning and image processing. Many models already available in the literature are used with new ones in [6]. The system of [6] results in data as close as possible to real biological data, but is quite complex, and the large number of its parameter settings may discourage its use. In [7], two simulation methods are used. In the first, the number of DE genes and the number of levels of changes for these genes are fixed. Then, for each gene, a mean and a standard deviation are drawn from a uniform distribution. Finally, these data (mean and standard deviation) are used as normal distribution parameters to get expression values for the genes. In the second simulation method, the mean and standard deviation of test (μt, σt) and control (μc,σc) samples are estimated from observed real data and are used as normal distribution parameters to get expression values for all genes. There is no flexibility in the number of the DE genes of the method in [7]. A hierarchical model is used in [8,9], where the variance of each gene is simulated using two parameter settings and the χ2 distribution. This variance is then used to generate values from a normal distribution. DE genes are finally defined using levels compared to a threshold. The procedure used in [8,9] allows one to generate data with parameters derived from assumptions about real data. However, the percentage of the DE genes depends on a parameter that is difficult to control a priori. For some of the above methods, there is no distinction between the number of weakly expressed genes and those strongly expressed. The model proposed in [10] provides synthetic data fairly close to the characteristics of true data. In this model, the level of expression of a gene is obtained by the superposition of several components. These components allow one to define the DE genes and the overall level of variability in the data. However, it lacks flexibility in choosing the number of DE genes and the number of genes over- and under-regulated. In this paper, we are interested in generating synthetic microarray data associated with two biological conditions. These conditions correspond to comparison of wild-type samples versus knock-out ones, treated samples versus non-treated ones, etc. For these situations, we use the generic terms of control versus test samples. Two condition biological data can be obtained using either one-color or two-color microarray technology leading to log2 intensities or log2 ratios, respectively. We assume that the same reference sample was used for log2 ratio data. This paper is organized as follows. In the next section, we present the model used and describe its parameter settings. In Section 3, we present commented results obtained using our model. Conclusions are drawn in Section 4. 2. Methods We propose here a model that does not require knowledge of the physical laws governing gene expression to produce data with similar characteristics to the data commonly produced by current platforms. The user provides a subset of parameters that control the behavior of the data generated. We assume the following characteristics for microarray data:There are usually more genes with low intensities than genes with high intensities. One possible explanation for this observation is a non-response of all genes to given biological conditions. The log2-intensities of microarray data usually vary between zero and 20. Expression values are the results of 16- or 20-bit float per pixel images, representing probes on arrays. Under similar biological conditions, the expression level of a gene varies around an average value. Exceptionally different values will be due to technical problems. The total number of DE genes depends on the biological conditions used. The number of over-regulated and under-regulated genes may be different. The variability observed for weakly expressed genes is larger than that observed for highly expressed genes. This is mitigated with today’s scanners and tools. However, genes which are not really expressed still have non-zero values, and strongly expressed genes (saturated) receive a single value corresponding to the maximum level of quantification. In the model described below, the expression levels are log2 intensities. Ratios are derived from these values. 2.1. Model Used Given n genes, m1 and m2 control and test samples, respectively, we used the following model: (1) xi=ai+si+ni+ti where xi is a vector of values generated for gene, i, (i=1,…,n). These values come from four sources. The values of vector ai={aij,j=1,…,m1,m1+1,…,m1+m2}, are expression levels for samples. The values of vector si allows one to define DE genes. These values are zero for control samples and for genes that are not DE, while for DE genes, non-zero values are associated with test samples. The additive noise vector, ni, has values independent of gene expression levels. Values associated with technical problems are cast in vector ti, which has only a few non-zero components (samples) for some genes. The current implementation of our model does not include any technical problem terms, ti, but we intend to take into account this component in the future. To generate data using model (1) that satisfies the issues given above, we proceed as follows. The beta distribution is used to obtain n values varying between zero and one. Shape parameters (shape1 and shape2) of this distribution were chosen to produce more small values than higher ones (issue 1). The n values obtained are scaled to fit real data (issue 2), which vary between a lower bound (lb) and an upper bound (ub). The scaled values represent average expression levels for the n genes (issue 3). Optionally, real data can be used as seed at this step. We assume that intensities for each gene are uniformly distributed around its average level. The range of variation of values from this distribution is assumed to be dependent on the gene average level and is expressed as a percentage through a parameter, α=λ1e-λ1z¯i, where λ1 is a parameter setting and z¯i is the average level of gene, i (issue 5). This leads to vector ai. Two parameters are used for generating vector si, the percentage of DE genes (pde) and a setting determining the number of up- and down-regulated genes (sym). The first m1 values (control samples) of si are set to zero. For gene i, an integer is randomly drawn from the set {1,…,100} and is compared to 100pde to decide whether the corresponding gene is DE. If the answer is yes, a second integer is drawn from the set {1,…,100} and is compared to 100sym to decide up and down assignment. m2 normally distributed values are generated using a mean (μde) and a standard deviation (σde). These values (or their opposite) form the second part (test samples) of vector si for the up- (down-) regulated DE genes. Independent noise vector ni values are obtained using a normal distribution with zero mean and parameter (σn) for standard deviation. The details of this algorithm are given in Algorithm 1 and described in the following sections. 2.2. Model Parameters 2.2.1. Data Size: n This is the number of probes (genes) on the arrays. It is on the order of thousands, and the default setting is n=10,000. 2.2.2. Number of Samples: m1 and m2 These parameters are the number of control and test samples, respectively. Typically, the minimum number of samples per condition should be three to allow the use of statistical tests. The maximum value of these parameters rarely exceeds 100 (samples per biological condition); the default setting is seven for both. 2.2.3. Expression Level or Ratio: Ratio This parameter option generates log2 intensities or log2 ratios data. With ratio = 0 (default setting), log2 intensities data will be generated; otherwise, log2 ratios data are returned. Algorithm 1: Steps of the micorarray data generation model. Generate n values from a beta distribution using shape1=2 and shape2 as shape parameters, we obtain values z={zi,i=1,…,n} Transform values of z, such that they vary between settings lb and ub: z¯=lb+ub×z For each value of z¯, generate m1+m2+1 uniformly distributed values: ai+si={U((1-α)z¯i,(1+α)z¯i)}, where α=λ1e-λ1z¯i and λ1 is a user setting. Then, the first m1+1 values are used for a vector r1, and the last m2 values are used for a vector r2. (a) If gene i is differentially expressed (a sample ∈{1,…,100}<100pde) Set v1=r1 If probe i is up-regulated (a sample ∈{1,…,100}>100sym) Generate m2 normally distributed values and add them to r2: μde=μdemin+{E(λ2)} v2=r2+{N(μde,σde2)}, where λ2, μdemin and σde2 are settings Else Generate m2 normally distributed values and subtract them from r2: μde=μdemin+{E(λ2)} v2=r2-{N(μde,σde2)}, where λ2, μdemin and σde2 are settings Set: yi=ai+si=(v1v2) (b) Else Set: yi=ai+si=(r1r2) Now, we have a noise-free data matrix Y:={yi,i=1,…,n}. Do we need noisy data? (a) If the noise standard deviation parameter is positive (setting σn) N:={ni=N(0,σn),i=1,…,n} Add normally distributed values to data: Y¯=Y+N}, (b) Else Do not use noise: Y¯=Y Depending on user settings, log2-intensities or log2-ratios are returned (a) If ratios are required The first column of Y¯ is used as reference: X=Y¯[,2:m1+m2+1]-Y¯[,1] (b) Else log2-intensities are returned: X=Y¯[,2:m1+m2+1] 2.2.4. Beta Distribution Shape Parameters: Shape1 and Shape2 To obtain more small values than high ones, we use a beta density distribution. Based on many histogram plots using various shape parameter values, we propose to set parameter shape1 to two and allow the user to choose a value for the parameter shape2 in the interval [4,8]. The default setting for shape2 is four. 2.2.5. Log2 Intensities Variation Range: lb and ub These parameters specify the log2 intensities variation range. Quantification of microarray images is usually performed with 16- to 20-bit base, leading to 216 to 220 levels of gray. We suggest using values for parameters lb and ub from intervals [2,6] and [8,16], respectively. These values are typically observed for actual Affymetrix GeneChip© array data for gene expression profiling. Default settings are lb=4 and ub=14. Observed minimum and maximum expression values will not exactly match the settings because of the variations used and possible additive noise. 2.2.6. Percentage of DE Genes: Pde This parameter controls the number of differentially expressed genes the user would like to have in the data set. Its values are taken from the interval [0,1]. A value pde=0.02 (default setting) means that 2% of the n probes in the data set should be DE. 2.2.7. Number of Up- and Down-Regulated Genes: Sym This parameter results in nearly the same number of up- and down-regulated probes when sym = 0.5 (default setting). If the value of this parameter is less (greater) than 0.5, we will have more up- (down-) regulated probes in the data set. 2.2.8. Gene Average Level Variation Range: λ1 (Lambda1) We assume that the values of each probe are uniformly distributed around an average value. An exponential distribution is used, α=λ1e-λ1z¯i, where λ1 is a user setting for decreasing rate. Then parameter, α, controls the width of the uniform distribution and is expressed as a percentage of the average level. λ1 allows a high variability in weakly expressed genes and, at the same time, a low variability for strongly expressed genes. We use the default value, λ1=0.13, which allows an α≈10% for an average expression level equal to two and an α≈3.54% for an average expression level equal to 10. Increasing λ1 will lead to more variability for weakly expressed genes and a small variability for strongly expressed genes. Small values for λ1 (≈0.01) will lead to the same variability for all genes independently of their expression level. 2.2.9. Fold Change Variation Parameters: λ2, μdemin and σde (Lambda2, Muminde and Sdde) For a gene, the fold change is a shift of average expression levels between test and control samples. Assuming that (control and test) sample values of a given gene are uniformly distributed around an average level, the shift comes from a superposition of additional values to the test samples, see Figure 1. We assume a normal distribution for the shift values N(μde,σde). However, the same mean μde is not used for all DE genes. Hence, we use a minimum value (setting μdemin) and the exponential distribution to get μde=μdemin+{λ2e-λ2}, where λ2 is another setting. Default settings for these parameters are: μdemin=1.0, λ2=2 and σde=0.5. A higher λ2 value will lead to a small number of genes having a shift greater than μdemin; a small λ2 value leads to the opposite situation. Parameters μdemin and σde may be chosen using a one sample Student t-test analysis. The statistic of the shift value is zs=m2μdeminσde. The critical value for a significance level of 0.05 for zs is 1.96. Hence, parameters may be chosen using the relation σde≤m2μdemin1.96≈0.51m2μdemin. The choice of additive noise standard deviation, σn (see below), will modify this inequality. Figure 1 Variation of expression levels of a DE gene. The first seven points are for control samples, and the last seven points correspond to test samples. The average value of this gene in the control is eight and 9.5 for the test sample (blue horizontal lines). The green points correspond to noise free data, while the red are observed values for control and test samples. 2.2.10. Additive Noise Standard Deviation: σn (Sdn) This parameter represents a normal distribution standard deviation for additive noise. The default value is σn=0.4. A zero value for this parameter will lead to noise-free data. Too high σn values can lead to a number of DE genes different to those specified in pde. 2.2.11. Computer Random Generator Seed: Rseed This parameter is used for computer random number initialization. It will allow one to generate the same data at different times. The default value is 50. 3. Results and Discussion To evaluate the performance of the proposed model, we performed simulations, in which we studied the influence of different parameter settings. Their default values are: n=10,000, m1=m2=7, lb=4, ub=14, λ1=0.13, λ2=2, μdemin=1.0, σde=0.5, ratio=0, sdn=0.4, shape2=4, pde=0.02, sym=0.5). We performed 100 independent simulations by changing the initialization of the generator through parameter, rseed. For each simulation, we performed a Student t-test and selected genes with a p-value less than 0.006. Using the DE information, selected genes were split into two: true and false DE genes. The p-value threshold, 0.006, leads to an expected error (false discovery rate) equal to (0.006×n)/(pde×n)=30% for default settings. When studying one parameter, the others are set to their default value. 3.1. Parameter Pde We used four different values (1%, 2%, 5% and 10%). For these values, the theoretical numbers of DE genes are, respectively, 100, 200, 500 and 1,000. The boxplots in Figure 2 show the results obtained. The median numbers of true down- (up-) regulated genes are 39.5 (40), 75.5 (76.5), 194 (190) and 382 (386.5) for the above values of parameter pde, respectively. In comparison with the expected number of DE genes, the recovery powers of the Student t-test are 79.5%, 76%, 76.8% and 76.8%. Better power results can be obtained for this test by using smaller value for parameter σn. Panel C of Figure 2 shows the number of false DE genes obtained using the four values for parameter pde. The median numbers of the false DE genes are 49, 49, 48.5 and 45.5 for the above values of parameter pde, respectively. Figure 2 Boxplots of the number of down- and up-regulated genes (true DE, false DE) with four values of the parameter pde. 100 simulations were used for these results. 3.2. Parameter Sym We used the following three values: 0.3, 0.5 and 0.7. For these values, the expected numbers for pairs of down- and up-regulated genes are, respectively, (60,140), (100,100) and (140,60). Figure 3 shows the boxplot of results obtained. The median numbers of true down- and up-regulated genes observed are (46, 107), (75.5, 76.5) and (107, 46) for the above values of parameter sym, respectively. Hence, the recovery powers of the Student t-test are 76.5%, 76% and 76.5%. Figure 3 Boxplots of the number of down- and up-regulated genes (true DE) with three values of the parameter sym. 100 simulations were used for these results. The median number of false DE genes is 49 for the three values of parameter sym. 3.3. Parameter σn We used three values: 0.2, 0.4 and 0.6. Figure 4 shows the boxplots of the results obtained. The median numbers of down- and up-regulated genes observed are (89, 89), (75.5, 76.5) and (57, 58), leading to detection powers of 89%, 76% and 57.5%, respectively. The t-test detection power decreases when σn increases. Figure 4 Boxplots of the number of down- and up-regulated genes (true DE) with three values of the parameter σn. 100 simulations were used for these results. The median numbers of false DE genes are 54, 49 and 49 for the 3 values of parameter σn, respectively. 3.4. Parameters μdemin, σde, λ1, λ2 and Shape2 We performed simulations to examine the influence of these parameters. For each parameter, 100 simulations were used, and the results obtained are summarized in Table 1. Increasing the parameter μdemin setting introduces more change for the DE genes, while its decrease leads to the opposite effect. Parameter σde acts as noise. The effect of the modification of some parameters is investigated further in the MA plot representations described in the following paragraph. microarrays-02-00115-t001_Table 1Table 1 Number of down- and up-regulated detected genes using the Student t-test and various parameter settings. Parameters (down, up) power μdemin=2 (100, 100) 100% μdemin=0.5 (41, 40) 40.5% σde=0.2 (90.5, 90) 90% σde=0.4 (82, 83) 82.5% σde=0.6 (69, 69.5) 69% λ1=0.1,σn=0.4 (76, 6) 76% λ1=0.01,σn=0.4 (81, 81) 81% λ1=0.1,σn=0.2 (89, 88) 88.5% λ1=0.01,σn=0.2 (92, 92) 92% λ2=4, (66, 66) 66% λ2=0.5 (92, 91) 91.5% shape2=4 (75.5, 76.5) 76% shape2=6 (75, 76) 76.5% shape2=8 (76, 76.5) 76% 3.5. Volcano and MA Plots Using default settings, we performed one simulation. Then we computed the Student t-test p-value and fold change for all genes. These values were used in the volcano plot of Figure 5. Red circles represent genes having a p-value less than 0.01 and a fold change greater than two or less than 0.5. Intensity measurements of two samples can be used to create two new variables: M=Ix2-Ix1 and A=0.5(Ix2+Ix1), where Ix2 and Ix1 are log2 intensities of samples x2 and x1, respectively. A value, one (-1) for M means that the corresponding gene is up- (down-) regulated two-fold. A plot of M (log2 ratio) versus A (log2 intensities average) is denoted “MA plot" [11]. The MA plots in Figure 6 are obtained using either two control samples (panel A) or one control and one test sample (panel B). Additional MA plots in Figure 7, Figure 8, Figure 9 and Figure 10 showing the effect of some parameter settings. A small value for λ1 leads to a less dense cloud of points. A larger change is observed for higher values of λ2 than for smaller ones. The same applies to parameter, σde. Increasing the parameter shape2 value leads to a decrease of the dynamic range of the data. Figure 5 Volcano plot of data obtained. Figure 6 MA plot using (A) two control samples or (B) one control and one test sample data. Figure 7 MA plot with one control and one test sample, using (A) λ1=0.1,σn=0.2 or (B) λ1=0.01,σn=0.2. Figure 8 MA plot with one control and one test sample, using (A) λ2=4 or (B) λ2=0.5. Figure 9 MA plot with one control and one test sample, using (A) μdemin=2 or (B) μdemin=0.5. Figure 10 MA plot with one control and one test sample, using (A) shape2=4 or (B) shape2=8. 3.6. Discussion The choice of some setting parameters for the proposed model is easy and can be dictated by the experimental design. This applies to n, m1, m2 and ratio. Parameters lb and ub and sym and pde concern the dynamic range of variation of the data to generate and define the DE genes. The intervals indicated for lb and ub are those observed for data from common platforms. Parameters, shape2, lb and ub, have no effect if real microarray data are used as a seed. The number of DE genes (pde) and the proportion of under- and over-regulated genes (sym) is at the discretion of the user. The parameter, rseed, allows one to produce the same data at different times using the same computer. This parameter is also useful for generating test data for different analysis algorithms. Settings, λ1, λ2, μdemin, σde and σn, control the global behavior of the data generated. More precisely, λ1 allows one to make gene changes dependent on the average level of expression. λ2 introduces variation in the expression changes for DE genes. These changes are defined by μdemin and σde. Parameter, σde, also acts as noise. The parameter, σn, allows one to perturb the data generated. The example of Figure 4 shows that the data obtained can differ from those expected for large values of σn. The signal to noise ratio (defined as the ratio of standard deviations) for default settings is 6.25. This rises to 4.16 for σn=0.6. An interesting microarray study was reported in [12]. In that study, the same RNA samples were processed by many laboratories using three leading microarray platforms: Affymetrix (five labs), two-color cDNA (three labs) and two-color oligonucleotides (two labs). The results presented show a good agreement across-platforms in contrast to some results previously reported in the literature, see, for instance, references cited in [12]. The microarray data generated for the study described in [12] are available from the Gene Expression Omnibus [13] under accession number GSE2521. These data can be used as seed for our model, which can then be integrated into user friendly data analysis software, such as Partek Genomics Suite, GeneSpring GX, etc., for demonstration and/or teaching purposes. 3.7. R Code For immediate use of the proposed model, we provide an R code function madsim.R (MicroArray Data Simulation Model), which is deposed as a package on the Comprehensible R Archive Network (CRAN) server for download [14]. The outputs of this function are the data generated (xdata) and the indexes of DE genes (xid). Real data can be used as seed for each gene. An example of such data is available in the data folder of the package. Further explanations are available from the package’s help function. 4. Conclusions We proposed in this paper a simulation model of microarray data. This model is very flexible and allows one to generate data with similar characteristics to the data commonly produced by current platforms. We showed a commented example of its possible use. We considered the case of data from two biological conditions. This model can be extended to multiple biological conditions in different ways: (a) modify to take into account additional biological conditions and several levels for the parameters, μde and σde, and (b) use as is, then place data side by side. Acknowledgments This work was supported by the Centre National pour la Recherche Scientifique. We are grateful to Julie D. Thompson for critical reading of the manuscript and helpful comments. We thank anonymous reviewers for their helpful comments. ==== Refs References 1. Dabney A.R. Storey J.D. A new approach to intensity-dependent normalization of two-channel microarrays Biostatistic 2007 8 128 139 10.1093/biostatistics/kxj038 16636140 2. Fujita A. Sato J.R. de Oliveira Rodrigues L. Ferreira C.E. Sogoyar M.C. Evaluating different methods of microarrays data normalization BMC Bioinformatics 2006 7 10.1186/1471-2105-7-469 17059609 3. Irizarry R.A. Hobbs B. Collin F. Beazer-Barclay Y.D. Antonellis K.J. Scherf U. Speed T.P. Exploration, normalization and summaries of high-density oligonucleotide array probe level data Biostatistic 2003 4 249 264 10.1093/biostatistics/4.2.249 12925520 4. Lonnstedt I. Speed T. Repicated microarray data Stat. Sinica 2002 12 31 46 5. Martin D.E. Demougin P. Hall M.N. Bellis M. Rank difference analysis of microarrays RDAM, a novel approach to statistical analysis of microarray expression profiling data BMC Bioinformatics 2004 5 10.1186/1471-2105-5-148 6. Nykter M. Aho T. Ahdesmäki M. Ruusuvuori P. Lehmussola A. Yli-Harja O. Simulation of microarray data with realistic characteristics BMC Bioinformatics 2006 7 10.1186/1471-2105-7-349 16848902 7. Witten D.M. Tibshirani R. A Comparison of Fold-Change and the t-Test for Microarray Data Analysis Department of Statistics, Stanford University Stanford, CA, USA 2007 17 Available online: http://www-stat.stanford.edu/ tibs/ftp/FCTComparison.pdf (accessed on 1 March 2013) 8. Smyth G.K. Linear models and empirical bayes methods for assessing differential expression in microarray experiments Stat. Appl. Genet. Mol. Biol. 2004 3 Article 3 Available online: http://www.statsci.org/smyth/pubs/ebayes.pdf (accessed on 1 March 2013) 10.2202/1544-6115.1027 16646809 9. McCarthy D.J. Smyth G.K. Testing significance relative to a fold-change threshold is a TREAT Bioinformatics 2009 25 765 771 10.1093/bioinformatics/btp053 19176553 10. Kooperberg C.F. Aragaki A.D. Strand A. Olson J.M. Significance testing for small microarray experiments Stat. Med. 2005 24 2281 2298 10.1002/sim.2109 15889452 11. Dudoit S. Yang Y.H. Callow M.J. Speed T.P. Stistical methods for identifying differentially expressed genes in replicated cDNA microarray experiments Stat. Sinica 2002 12 111 139 12. Irizarry R.A. Warren D. Spencer F. Kim I.F. Biswal S. Frank B.C. Gabrielson E. Garcia J.G.N. Geoghegan J. Germino G. Multiple-laboratory comparison of microarray platforms Nat. Methods 2005 2 345 350 10.1038/nmeth756 15846361 13. Gene Expression Omnibus Available online: http://www.ncbi.nlm.nih.gov/geo/ (accessed on 1 March 2013) 14. The R Project for Statistical Computing Available online: http://www.r-project.org (accessed on 1 March 2013)

PMC005xxxxxx/PMC5003478.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2020097microarrays-02-00097ArticleExpanding the Diversity of Imaging-Based RNAi Screen Applications Using Cell Spot Microarrays Rantala Juha K. *Kwon Sunjong Korkola James Gray Joe W. Department of Biomedical Engineering and Knight Cancer Institute, Oregon Health and Science University, Portland, OR 97239, USA; E-Mails: kwons@ohsu.edu (S.K.); korkola@ohsu.edu (J.K.); grayjo@ohsu.edu (J.W.G.)* Author to whom correspondence should be addressed; E-Mail: rantala@ohsu.edu; Tel.: +1-503-494-3012; Fax: +1-503-494-3688.11 4 2013 6 2013 2 2 97 114 15 2 2013 02 4 2013 07 4 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Over the past decade, great strides have been made in identifying gene aberrations and deregulated pathways that are associated with specific disease states. These association studies guide experimental studies aimed at identifying the aberrant genes and networks that cause the disease states. This requires functional manipulation of these genes and networks in laboratory models of normal and diseased cells. One approach is to assess molecular and biological responses to high-throughput RNA interference (RNAi)-induced gene knockdown. These responses can be revealed by immunofluorescent staining for a molecular or cellular process of interest and quantified using fluorescence image analysis. These applications are typically performed in multiwell format, but are limited by high reagent costs and long plate processing times. These limitations can be mitigated by analyzing cells grown in cell spot microarray (CSMA) format. CSMAs are produced by growing cells on small (~200 μm diameter) spots with each spot carrying an siRNA with transfection reagent. The spacing between spots is only a few hundred micrometers, thus thousands of cell spots can be arranged on a single cell culture surface. These high-density cell cultures can be immunofluorescently stained with minimal reagent consumption and analyzed quickly using automated fluorescence microscopy platforms. This review covers basic aspects of imaging-based CSMA technology, describes a wide range of immunofluorescence assays that have already been implemented successfully for CSMA screening and suggests future directions for advanced RNAi screening experiments. RNA interferencehigh-throughput screeningcell spot microarraysimage cytometryquantitative immunofluorescence assays ==== Body 1. Introduction International genomic studies are revealing a growing number of genomic aberrations and aberrant pathways that are postulated to be important in a wide range of human diseases including cancer. The number of candidate “cancer genes” emerging from these efforts is especially daunting. However, the definitive functions of these aberrant genes and pathways must be established by experimental manipulation in laboratory models. Manipulation of gene expression levels using inhibitory RNAs (RNAi) is a key technique for this purpose. Large-scale nucleic acid synthesis techniques now enable convenient and low cost synthesis of thousands of RNAis so that assessment of the effects of manipulating the expression levels of thousands of genes is possible. Initial efforts in RNAi screening in mammalian cells were accomplished using automated strategies in which the effects of RNAi knockdown were assessed in multiwell format (usually 96- or 384-well plates). This required substantial laboratory instrumentation and automation, while the costs of RNAi reagent libraries were high. A strategy to miniaturize this process was first suggested by Sabatini et al. [1] and subsequently demonstrated with both siRNAs and shRNAs [2,3,4,5] before fully developed by Rantala et al. [6]. In this process, individual RNAi oligonucleotides are printed in ~200 μm diameter spots separated by a few hundred micrometers (Figure 1(a)). Each spot carries cell adherence promoting matrix proteins allowing spatially confined array patterning with cells growing only on the spots. Each spot can host up to a few hundred cells and contains a lipid transfection agent so that cells auto-transfect as they grow. This miniaturized platform provides an economical and robust alternative to multiwell screening systems for systematic assessment of gene function in vitro. In a typical experiment, siRNA-lipid microarrays are covered with adherent cells in a culture medium for reverse transfection-mediated uptake of the siRNA [6] or micro-RNA [7] reagents from the spatially confined array spots. siRNAs usually are arrayed in triplicate in order to enable assessment of experimental reproducibility. Spot densities of ~1,000/cm2 are routinely achieved on this platform so that responses to thousands of siRNAs can be robustly assessed in a single culture [6]. The well-less and miniature format of the CSMA platform allows cells to be immunofluorescently stained for specific molecular response endpoints (e.g., molecular events associated with proliferation, apoptosis, differentiation status, senescence, etc.) much as one would stain cells grown on a coverslip (Figure 1(b,c)). Alternately, cells can be genetically engineered to express fluorescent response reporter constructs for time-resolved analyses [6,8]. The quantitative assessment of the impact of high-throughput RNAi knockdown on immunofluorescently stained molecular features—the subject of the present article—has already been applied in several cell biological studies, and the potential of the approach is only beginning to be realized. Early applications include analysis of the impact of specific RNAi-induced knockdowns on cellular abundance of protein complexes [9], regulatory pathways [10,11], and changes in the spatial distribution of target proteins [12]. We describe here experiments using siRNAs as RNAi reagents, but the platform appears readily extensible to assessment of effects of shRNAs [13], miRNAs [7] and cDNAs [1]. The applications described herein illustrate how the CSMA platform can be used for efficient assessment of the roles that specific molecular entities or genomic aberrations play in many aspects of cancer pathophysiology. The ability to assess the impact of RNAi knockdown on endpoints such as differentiation, DNA repair activity, senescence and motility is a particular strength of the platform. The advantages of the platform relative to multiwell analysis approaches are the high analysis density which allows detailed analyses of responses to thousands of RNAis at low immunochemical reagent and cell culture cost and relatively high analysis speed. The disadvantages of the platform are the relatively small numbers of cells interrogated on each spot and the possibility that cells on one spot may be affected by a small amount of molecule secretion from cells on adjacent spots. Printing replicate spots to increase the number of cells analyzed can mitigate the former disadvantage and the latter can be mitigated by randomly positioning replicate spots across the culture surface so that each replicate is in close proximity to a different collection of other RNAi-perturbed cell spots. A typical experiment in which each RNAi is printed in triplicate provides analyses of sufficient cells to enable detection of RNAi effects that differ from the control by less than 10%. RNAi against genes like AURKB, CDK1, INCENP, KIF11 and PLK1 for which responses are well established are usually included as positive controls [6,8,14] to confirm that the RNAi transfection efficiency is high. This is especially important when working with cell types that have not been previously analyzed using the CSMA platform [6]. The advantages of assessing RNAi-induced changes using image cytometry following immunofluorescence staining—whether in CSMAs or in multiwell cultures—are substantial compared to strategies that assess bulk changes in cell number or metabolic activity (e.g., using the CellTiterGlow or MTT assays; [15]) or that identify RNAi effects following bulk transfection (e.g., by assessing loss of cells carrying specific RNAis). Specifically, imaging allows for the following: (a) Quantitative measurement of the cellular abundance of specific target molecules for which a fluorescence reporter can be developed. These studies take advantage of a growing number of antibodies, aptamers or other affinity ligands that bind with high affinity and specificity to proteins or post-translationally modified variants that comprise regulatory networks. The availability of fluorescent reporter constructs that level the expression levels of specific proteins further increases the information that can be obtained; (b) Assessment of the intracellular distributions of the proteins or organelles of interest. This enables assessment endpoints such as the number of discrete DNA repair foci (a measure of DNA repair activity), assessment of the fraction of cells incorporating EdU, mitochrondrial morphology and assessment of mitotic apparatus shape; (c) Analysis of molecular proximity through the use of Förster resonance energy transfer assays (FRET) or antibody-based proximity ligation assays (PLA); (d) Assessment of molecular response heterogeneity between the cells in a single cell spot—for example, induced by cell–cell proximity and/or transient differentiation; (e) Multiplex analysis of multiple molecular and biological response endpoints. For example, multicolor analysis allows assessment of how RNAi manipulation changes the relationship between molecular pathway components (e.g., PI3K and MAPK pathway activities), and biological endpoints such as EdU incorporation or cell cycle distribution, motility, differentiation status and cell death [7,10,11,16,17]; (f) Assessment of the impact of RNAi knockdown on responses to chemical, biological or microenvironmental perturbations. The platform is especially useful in identifying genes and pathways that influence responses to anticancer agents. Figure 1 (a) Cell spot microarrays (CSMAs) are produced by spotting siRNA samples mixed with transfection lipids and extra-cellular matrix proteins on a hydrophobic polystyrene surface in microplate sized vessels. This enables production of high density cell transfection microarrays with up to 1,000 siRNA samples/cm2. The slides are coated so that the cells adhere only to the spots containing the siRNAs. Transfer of the siRNAs to the cells occurs by reverse transfection during growth for a selected time, typically 48 to 72 h. (b) Multicolor image of immunofluorescently stained human kidney tumor cells following growth in CSMA format. The diameter of a cell spot is approximately 200 μm. (c) High resolution image of cells growing on one cell spot stained for DNA (blue), F-Actin (green) and beta-tubulin (red). In the following sections, we describe fundamental aspects of image cytometry-based RNAi screening applications on CSMAs and review some of the basic methods. We also discuss in detail the custom assays established in our laboratory for analysis of quantitative cancer cell phenotypes. Finally, we conclude with an assessment of future developments of imaging-based RNAi analyses. 2. Experimental Section 2.1. Cells and Cell Culture All cell lines grown on CSMAs were cultured according to the protocols recommended for the cell line. Primary kidney tumor cells were isolated from fresh patient surgical specimens obtained under an Institutional Review Board-approved protocol at Oregon Health and Science University (OHSU). These primary kidney cells, along with BT20, HCC1569, HCC1954, MDA-MB-468, U-2OS, KFr13, VCaP, and 22RV1 cells (ATCC, Manassos, VA, USA) were grown in RPMI-1640 (Gibco, Life Technologies, Grand Island, NY, USA) supplemented with 10% FBS, 10 µg/mL penicillin and streptomycin and 2 mM l-glutamine. HaCat, MFC7 and MDA-MB-231 cells were grown in DMEM (Gibco) supplemented with 10% FBS, 10 µg/mL penicillin and streptomycin and 2 mM L-glutamine. RWPE-1 (ATCC) cells were grown in Keratinocyte Serum Free Medium (K-SFM, Gibco) supplemented with 0.05 mg/mL BPE (bovine pituitary extract) and 5 ng/mL EGF. SKBR3 cells (ATCC) were grown in McCoy’s 5A medium (Gibco) supplemented with 10% FBS, 10 µg/mL penicillin and streptomycin, and 2 mM l-glutamine. 2.2. Preparation of Cell Spot Microarrays Transfections with siRNAs and cell culture on the CSMAs were carried out as described previously [6]. siRNAs against PLK1 and AURKB purchased from Qiagen as experimentally verified oligos were used for transfection validation experiments (PLK1 #A SI02223837, #B SI02223844; AURKB #A SI02622032, #B SI02622039). Briefly, the siRNAs and siLentFect (Bio-Rad) transfection reagent for array printing were prepared by mixing the lipid–siRNA samples with cold growth factor-reduced Matrigel (BD Biosciences, Bedford, MA, USA)—OptiMEM I (Gibco) solution resulting in final siRNA printing concentrations of 2.5 µM and 15% Matrigel. These solutions were printed as 200 µm diameter spots on the bottom of the wells of polystyrene microplates (typically 4 to 8 wells per plate, Nunc Brand, Roskilde, Denmark). The siRNA–Matrigel spots were allowed to polymerize for 30 min at room temperature and then stored at room temperature, desiccated and protected from light [6]. Arrays were stored for several weeks before use under these conditions. Approximately 2 × 106 cells in 4.5 mL of growth medium were added to each array well (4-well plates) and allowed to adhere at +37 °C for 5–15 min. Cells were dispersed with non-trypsin cell detachment reagent HyQtase (HyClone, South Logan, UT, USA) prior to seeding on the CSMAs since this dispersal method enabled rapid adhesion to the array spots. Non-adherent cells were washed off and 4.5 mL of fresh medium was added per array well. siRNA transfer to the adherent cells took place during growth periods ranging from 48 to 144 h prior to staining and imaging. 2.3. Antibody Staining Procedure Cells transfected with siRNAs during growth on CSMAs were immunofluorescently stained according to the following protocol. First the culture medium was aspirated carefully from each array well and the cells were fixed with 2% paraformaldehyde (Sigma-Aldrich, St. Louis, MO, USA) in PBS for 15 min at room temperature. Cells were then rinsed once briefly with 50 mM NH4Cl to quench any remainder of the paraformaldehyde fixative and washed 5 min with PBS. Cells were permeabilized with 0.3% Triton-X100 in PBS for 15 min at room temperature, washed once with PBS and blocked with 2% filtered BSA in PBS for 60 min at room temperature. After blocking, the arrays were washed 2 × 5 min with PBS, rinsed briefly with distilled H2O and air-dried. Array areas were inscribed with a hydrophobic border using a PAP-pen (Sigma-Aldrich) to reduce the amount of antibody used during staining. Cells were rinsed with 0.05% PBS-Tween 20 and stained with primary antibody in 2% BSA-PBS (100 μL per 20 × 20 mm array surface) for 1 h at room temperature or overnight at 4 °C. These arrays were washed 2 × 5 min with PBS and for 5 min with 0.05% PBS-Tween 20 and then stained with 100 μL of diluted Alexa fluorochrome-conjugated (Life Technologies) secondary antibodies. Secondary antibody incubation and parallel DAPI counterstaining was performed for 1 h at room temperature, followed by washing as described for primary antibodies. The stained arrays were then rinsed with distilled H2O, air-dried and stored for imaging. The cells were rehydrated for imaging by covering the arrays with PBS or by mounting under a coverslip using ProLong Gold anti-fade reagent (Life Technologies). 2.4. RNA Immuno-FISH Procedure Cells grown on CSMAs were fixed in 4% formaldehyde in PBS for 10 min, permeabilized with 70% ethanol at 4 °C for 1 h, and prehybridized with wash buffer (2× SSC, 10% formamide) at room temperature for 10 min. Cells were then incubated in parallel with cyclin D1 antisense-oligonucleotide probe-sets labeled with Cy-5 and anti-Ki67 antibody (1:300, Abcam, Cambridge, MA, USA) in hybridization solution (10% Dextran sulfate, 2× SSC, 10% formamide) at 37 °C overnight in a dark/humid chamber, and incubated with wash buffer at 37 °C for 30 min twice. Secondary antibody incubation (Alexa488-conjugated goat anti-rabbit antibody in the hybridization solution) was carried out at room temperature for 1 h, followed by incubation with wash buffer for 30 min, with DAPI for nuclear staining, and with 2× SSC. Cells on arrays were mounted in Prolong Gold anti-fade reagent (Life Technologies). Probe sets for CCND1 mRNA detection were designed using a Stellaris™ Probe Designer version 1.0 [18]. They were composed of 48 different 20 mer DNA oligonucleotides, each complementary to a different region of CCND1 mRNA, targeting sequences with 45% GC content, separating at least two bases between oligonucleotides. Images of 0.2 µm optical sections were acquired using Deltavision CoreDV Automated Widefield microscopy (Applied Precision™, 60× objective, NA = 1.42) with a Nikon Coolsnap ES2 HQ camera. These images were processed with deconvolution software to subtract blurred lights or to reassign them back to sources, and reconstructed into 3D image using IMARIS™ software (Bitplane, South Windsor, CT, USA). 2.5. Antibody-Based Proximity Ligation Assay for in situ Protein–Protein Interaction Analysis After transfection, cells grown on CSMA were fixed and stained according to the manufacturer instructions for the DuoLink II PLA kit (Olink, Uppsala, Sweden) with a minor change in PLA probe dilution. The primary antibodies were diluted 1:200 in 2% BSA-PBS and incubated overnight at 4 °C (ITGB1; Abcam 12G10, ITGA2; Millipore AB1936). The PLA probe antibodies were diluted 1:20 in 2% BSA-PBS supplemented with the PLA blocking concentrate and incubated for 2 h at 37 °C. The array surfaces were circumscribed with a hydrophobic border drawn with a PAP-pen (Sigma-Aldrich) to minimize antibody and PLA detection reagent consumption. Cells were counterstained for filamentous actin using fluorescently labeled phalloidin (Alexa488, Life Technologies) and DNA (DAPI). Images for each spot in the CSMAs were acquired automatically using an Olympus scan^R imager and image analysis software (Olympus-SIS, Münster, Germany) was used to quantify PLA levels in individual cells. 2.6. Western Blot Analysis Western blot analysis of total cell lysates prepared from cells transfected on CSMAs with 384 replicate spots of a single control or targeting siRNA were fractionated on SDS-polyacrylamide gels and transferred to nitrocellulose membranes (Whatman Inc., Kent, UK). The membranes were blocked against non-specific binding using 5% skim milk. Membranes were probed with primary antibodies (PLK1, Abcam; AURKB, Abcam) overnight at 4 °C. Equal loading was confirmed by probing the same filter with a specific antibody for β-tubulin or β-actin (1:5,000, Abcam). Signals were revealed with horseradish peroxidase-coupled secondary anti-mouse or anti-rabbit IgG antibodies (1:1,000; Sigma-Aldrich). 2.7. Imaging and Analysis Array imaging was performed using an Olympus scan^R integrated imager and image analysis suite (Olympus-SIS, Münster, Germany) equipped with a Hamamatsu ORCA-R2 CCD digital camera (Hamamatsu Photonics K.K., Tokyo, Japan). The Olympus scan^R system is an inverted microscope designed for fully automated image acquisition of biological samples in high-density sample platforms such as CSMA plus image analysis algorithms for feature quantification. Each spot on a CSMA was imaged individually with a 20× LUCPLFLN NA 0.40 objective using specific filter sets for DAPI, Alexa488, Alexa568 and Alexa647 dyes (Semrock, Inc., Rochester, NY, USA). The scan^R image analysis software suite was also used for quantitative analysis of image features. Analysis capabilities included cell/particle counting, protein expression analysis with immunofluorescence quantitation, subcellular particle quantitation assays, cell cycle analysis, and protein localization and co-localization assays. Image features quantified for cell populations using the scan^R software were further analyzed using FCS Express 3.0 software (De Novo Software, Los Angeles, CA, USA). The effects of each siRNA knockdown on each specific response endpoint were assessed by comparing image parameters (e.g., total fluorescence intensity per cell, fraction of cells incorporating EdU or number of segmented spots per cell) with comparable image parameters measured for cells transfected with a non-targeting scrambled control siRNA. 3. Results and Discussion 3.1. Image Cytometry of Cell Spot Microarrays Figure 2(a) illustrates the results obtained using an Olympus scan^R imager for analysis of cells grown in CSMA format. The scan^R software was used to define cell based on the extent of immunofluorescently stained cytokeratin and nuclear boundaries defined based on the extent of DAPI-stained DNA. Segmentation of secondary objects within the primary objects was then performed to define and quantify subcellular features. These measurements of subcellular and cellular shape features in adherent cells are the unique provenance of image cytometry since they cannot be readily assessed using flow cytometry. Figure 2(a), for example, shows the use of combinations of cell morphology parameters and cell differentiation markers to identify biologically distinct subpopulations within a cell culture environment that is often presumed to be homogeneous. In this example, BT-20 human breast cancer cells seeded on CSMAs were stained for basal cell lineage markers cytokeratin-8 and -14 and the sub-populations staining positive for one but not the other marker (94.26% KRT8+, 2.13% KRT14+; Figure 2(b)) were analyzed separately for nuclear size and cell cycle distribution. This analysis, which highlights the multiplexing capabilities of the image cytometry assays, indicated these two cell populations differed in both features. The KRT8-/KRT14+ cells within the parental population had consistently smaller, less circular nuclei and an increased fraction of cells in the G2-M-phase of the cell cycle (Figure 2(b)). Figure 2 Examples of quantitative image cytometry on CSMAs. (a) BT-20 breast cancer cells stained for DNA (blue), cytokeratin-8 (green) and -14 (red). Panels on the right show segmentation of the image according to DAPI and cytokeratin extent. (b) Identification of a subpopulation of KRT14 positive cells comprising 2% of the KRT8 positive parental BT-20 cell population (left panel). The panels on the right show that KRT14 positive cells had smaller, less circular nuclei and a different distribution of cells in the G1 and G1 phases of the cell cycle compared to KRT8 positive cells. (c) Assessment of the effects on cell cycle traverse of siRNA knockdown of AURKB and PLK1 in KFr13 ovarian cancer cells after 48 h. Panel c shows images of DAPI and EdU incorporation 48 h after growth on cell spots carrying a scrambled siRNA and siRNAs against PLK1 and AURKB. Insets show DNA distributions calculated from nuclear DAPI intensity measurements. These analyses show the decrease in proliferation and the increase in frequency of polylobed (siAURKB) or mitotic (siPLK1) cells induced by siRNA knockdown. The lower panels show levels of AURKB and PLK1 proteins after growth on CSMAs carrying siRNAs against AURKB and PLK1, respectively. Another example of the robustness of the image cytometry is its ability to assess cell cycle features. These analyses can be performed simply on the basis of a single DNA dye such as DAPI, PI or Hoechst in combination with nuclear size and shape measures. Measurements of the total nuclear fluorescence in proportion to the nuclear size of cells generates scatter plots of DNA content to nuclear area ratios that can be gated to accurately quantify percentages of cells in definite G1-, S- and G2-phases of cell cycle. The mean pixel intensity corresponding to chromatin condensation and the nuclear area ratio enables mitotic and late anaphase cells to be distinguished from G2 and G1 cells, respectively [9]. In addition, apoptotic cells can be separated from mitotic cells without use of any specific markers based on the size and circularity of the objects. Figure 2(c) shows the use of these features to assess responses to RNAi knockdown of well-established cell cycle regulating proteins, AURKB and PLK1. This type of cell cycle analysis may be performed directly from DNA measurements, leaving other available fluorescence channels available for assessment of additional markers of interest. As an example, assaying EdU (5-ethynyl-2'-deoxyuridine, Life Technologies) incorporation (Figure 2(c)) or detection of Histone-H3 phosphorylation [14] can be used to further fine-tune analyses of specified cell cycle phases. This ability to measure shape and quantity of multiple stained features in individual cells enabled by image cytometry allows coordinate analysis of cell cycle endpoints and the expression levels of the proteins that control cell cycle traverse. 3.2. Surrogate Markers for Analysis of Quantitative Cancer Cell Phenotypes High-throughput RNAi screening has become a major tool for identification of molecular functions and genomic aberrations that play a role in cancer pathophysiology and for identification of novel candidate therapeutic targets. These studies have been, and continue to be, important in categorizing general molecular networks crucial for cell survival and proliferation. However, assessment of cell viability provides little information about the detailed roles that the interrogated target genes may play in the other important aspects of cancer cell physiology. Decades of research have defined an atlas of cellular functions that go awry in cancer development and progression and manifest in the cancerous cells as changes of molecular and structural phenotypes. These key aberrant cellular features of cancer have been designated collectively as “hallmarks of cancer” [19]. Figure 3 (a) A gallery of representative images showing immunofluorescent staining patterns that can be analyzed as quantitative cancer cell phenotypes (qCP). These assays assess aspects of cell proliferation, differentiation, apoptosis, cell adhesion/motility, senescence and genomic integrity. Each marker is indicated on the images with text in the corresponding color. The complete list of the validated qCP detection reagents is listed in Table S1. (b) Assessment of changes in qCPs induced by siRNAs targeting genes known to produce change in specific cancer hallmarks. qCPs were measured after growth on targeting or control siRNAs for multiple cell lines. The cell lines, targeting siRNAs and assessed qCPs are indicated in the figure. We have now developed immunofluorescent staining and quantitative image analysis procedures for many of the key molecular features associated with the cancer hallmarks to facilitate use of RNAi screening for in-depth biological discovery in the context of cancer research. Measures of RNAi-induced changes in these features—designated herein as quantitative cancer phenotypes (qCPs)—enables accurate identification of aberrant genes or networks that are causal for many aspects of cancer. This approach assesses many more aspects of cancer physiology than are routinely assessed in screening strategies that are sensitive only to events that alter proliferation or immortalization. We are now applying this approach to systematic assessment of oncogenic events, signal transduction programs and associated deregulated gene networks that are postulated to contribute to the hallmarks of cancer by the Cancer Genome Atlas (TCGA) [20] and related international genomics efforts. Our current system measures qCPs that report on proliferation, apoptosis, senescence, differentiation, DNA replication and repair, and motility (Figure 3(a)). The cell proliferation qCP quantifies the fraction of cells incorporating a nucleotide analog (EdU or BrdU) or showing high level staining for Ki-67 [6]. The apoptosis qCP reports the level of staining for cleaved PARP [7] or members of the caspase protein family. The qCP for cell senescence reports the intensity of staining for trimethylation of histone 3 lysine 9 (H3K9me3), an established marker for heterochromatin formation. Epithelial-to-mesenchymal transition is one of the fundamental cellular phenotypes associated with the course of cancer pathogenesis. qCPs for differentiation status report the intensity of staining for differentiation associated proteins including Beta-Catenin, Cytokeratins-8/14/19, E-Cadherin, EpCAM, Fibronectin and Vimentin [21]. Loss of genomic integrity and aberrant DNA repair is characteristic of virtually all human cancers [22]. The qCP for DNA damage repair (DDR) reports the number of detect P53BP1 or gamma-H2Ax foci that are associated with DNA double-strand breaks [9,21]. Figure 3(a) shows a gallery of CSMA images of cells stained with antibodies that report on each of these cancer hallmarks. The complete description of the qCP marker antibodies that we now use for image-based analysis of qCPs is provided in Table S1. The utility of these assays for large-scale CSMA screens has been confirmed in our previous and on-going work [6,10,11,16,17]. Representative examples of siRNA-induced changes in qCPs are shown in Figure 3(b). Our panel of assays is not intended to be a definite list of assays for quantitative cancer phenotype assessment. Rather, it is a guiding collection setting the basis for future development of additional image cytometry assays for RNAi screening experiments that will contribute to the systematic assessment of the roles specific aberrations play in cancer pathophysiology. 3.3. Next-Generation RNAi Screen Readouts—Immuno-FISH on CSMAs The miniature scale of the CSMA platform also facilitates development of other molecular state assays such as Fluorescence in situ Hybridization (FISH) for genome number (DNA-FISH) [23], or abundance and cellular location of mRNA transcripts (RNA-FISH). These approaches rely on the in situ detection of specific DNA or RNA sequences in fixed cell samples using probe oligonucleotides that are complementary to the specific sequence(s) of interest. We tested the performance of RNA-FISH for quantification of RNA levels altered during RNAi screening by assessing CCND1 (Cyclin D1) mRNA levels. We use the Stellaris FISH method (Biosearch Technologies) in which detection probes are comprised of multiple short, fluorescently labeled oligonucleotides targeting one mRNA transcript [24]. The high signal amplification achieved in this procedure produces signals that can be readily detected and quantified during imaging. The Stellaris FISH technique does not require denaturation of the samples with heat so the staining procedure can be combined with other detection techniques such as immunofluorescent staining. Figure 4(a) shows a prototypic example in which we combined RNA-FISH detection of CCND1 transcripts with immunofluorescent staining for Ki-67. Dual-stained MDA-MB-468 breast cancer cells were imaged with the Olympus scan^R and a Deltavision CoreDV widefield microscope. These studies showed that CCND1 mRNA was located predominantly in the nuclei, and more specifically to nucleoli. We assessed signal intensities of the nuclear CCND1 mRNA and Ki-67 immunofluorescence staining in cells assigned to the G1-, S-, G2- and M-phases of the cell cycle phases based on DAPI fluorescence intensity and spatial distribution. In this example, nuclear staining for Ki-67 was highest in mitotic cells as previously shown [6] and nuclear CCND1 staining intensity was modestly higher in G2 cells in comparison to the other groups (Figure 4(b)). Figure 4 (a) Quantitative RNA immuno-FISH on CSMAs. MDAMB468 breast cancer cells grown on a CSMA were immunostained in for Ki-67 (green) and for CCND1 using RNA-FISH (red). Nuclei were counterstained with DAPI (blue). The left panel shows a scan^R image at 20× magnification. The right panel shows a high resolution widefield microscopic image of the same field. Red punctate dots with intense CCND1 mRNA staining in nuclei indicate localization of the CCND1 messenger to the nucleoli. Scale bar 5 μm. (b) Images of individual cells showing nuclear CCND1 RNA levels and Ki-67 staining (upper panel). The cells are organized according to cell cycle phase as determined from quantitative measurements of nuclear DAPI fluorescence. The lower panel shows quantitative analyses of the images in panel b. These analyses show that mitotic cells have the highest mean nuclear area signal for Ki-67 while cells in G2-phase have the highest nuclear intensity for CCND1 mRNA (lower right panels). Error bars indicate the variance present that resulted from analysis of 1,686 cells with 885 (52.5%) in G1, 173 (10.3) in S, 409 (24.3%) in G2 and 39 (2.3%) in M-phase (lower left panel). We anticipate that use of immuno-FISH as described coupled with high-throughput RNAi screens will enable coordinate analysis of the impact RNAi knockdown on both transcriptional and translational regulation of target genes and proteins. This will allow discrimination between RNAi effects that directly alter expression of the target genes and those that affect post-translational mechanisms regulating the protein functions. 3.4. Next-Generation RNAi Screen Readouts—Analysis of Endogenous Protein Interactions in situ Interactions between proteins enable and regulate a wide range of cellular processes. Protein–protein interactions typically are studied using methods such as mass spectrometry of immunoprecipitated protein complexes or via analysis in yeast two hybrid systems. These methods reveal a broad range of interacting proteins. However, they do not reveal where within cells these proteins interact nor do they assess the impact of RNAi knockdown on these interactions. This can only be accomplished using image-based analysis strategies. To this end, we have developed a strategy for in situ analysis of protein–protein interactions on cells grown on CSMAs using an antibody-based proximity ligation assay (PLA) [25]. The PLA method detects interacting target proteins in fixed cells by detecting the interactions between paired primary antibodies labeled with complementary oligonucleotides. The oligonucleotides interact to enable rolling circle amplification and oligonucleotide complex formation when the two antibodies are in close proximity (Figure 5(a)). The resulting bright fluorescent complexes can be imaged by standard fluorescence microscopy and the numbers of foci can be measured as a function of location within individual cells (Figure 5(a,b)). To date, PLA analyses has been extensively used to study protein interactions and protein co-localizations in the individual biological experiments and in a small-scale screening of chemical inhibitors of PDGF signaling [26]. The PLA technology can be adapted to high throughput RNAi manipulation of genes that (a) regulate protein–protein interactions; (b) participate in the interactions; or that (c) influence the localization of the interactions. However, it is prohibitively expensive when implemented in multiwell formats, due to the need for large quantities of the detection reagents including enzymes required for the DNA rolling circle amplification (Figure 5(a)). Implementation of PLA analysis in CSMA format allows assessment of the impact of the several thousand RNAi reactions using the same volume of PLA detection reagents that would be required to stain two individual 96-microplate wells [26]. This dramatically reduces reagent consumption, allowing use of PLA for large-scale functional genomics experiments. Figure 5 illustrates the application of the PLA technique to assessment of detection of activated heterodimeric integrin α2β1 cell-adhesion receptors in cells grown on CSMAs carrying siRNAs. This assay, described previously [10], enables detection of the ITGB1 sub-units with an active conformation in close proximity to ITGA2 alpha sub-units (Figure 5(a)). These results indicate that PLA staining on cell spot microarrays in combination with automated image cytometry can reliably detect PLA signals in single cells (Figure 5(b)) and therefore can be used for functional interrogation of regulators of endogenous protein interactions in situ. Analysis of protein proximity after PLA staining is accomplished by segmenting the individual PLA signals and quantifying the number of individual PLA signals per cell. In Figure 5(c), single cells were defined by segmenting according to the extent of F-Actin staining. This analysis enables cells to be classified according to the extent specific protein interactions (Figure 5(c)) or according to the spatial distribution of the protein complexes (Figure 5(b)). We anticipate that assessment of the impact of large-scale siRNA knockdown on specific protein interactions will facilitate elucidation of mechanisms that control protein–protein interactions and/or that regulate their spatial localization. Figure 5 (a) Principle of the PLA readout for analysis of active state heterodimeric integrin α2β1 receptors. Active integrin α2β1 heterodimers were recognized in pairs by rabbit monoclonal antibodies binding the alpha subunit ITGA2 and an active conformation-specific mouse monoclonal antibodies binding ITGB1 (clone 12G10). The primary antibodies were bound by species-specific probe antibodies, conjugated to PLA oligonucleotides. When in proximity, the oligonucleotides can be used as templates for the joining of two additional linear oligonucleotides into a DNA circle forming the template for rolling-circle amplification and fluorescence detection of interacting proteins in situ. (b) Illustration of the use of the PLA technique to assess DNA (blue), F-actin (green) and ITGB1-ITGA2 proximity (red) in individual cells. PLA signals were quantitated at the single signal level using automated image analysis object segmentation algorithms. Scale bar 10 µm. (c) CSMA detection of active α2β1-integrin in PC3 prostate cancer cells. Cells on spots were divided into groups according to the amount of PLA signals identified by signal segmentation (right panel) within the membrane boundaries of individual cells detected on basis of F-Actin staining. The colored rectangles surrounding each cell indicate the number of PLA signals associated with the given cell. 3.5. Summary The CSMA analysis platform represents a technological advance in RNAi screening. It enables assessment of the functional impact of large-scale gene knockdown in many different cell types quickly and at low reagent cost. The impact of siRNA knockdown on multiple molecular and cellular response endpoints can be revealed using immunofluorescent staining or fluorescence in situ hybridization or a combination of both. Analyses can be carried out for cells growing in diverse culture conditions and during treatment with therapeutic compounds or other siRNAs. Assays for >25 response endpoints already have been developed for the CSMA platform. Automated, image analysis allows images of cells on individual array spots to be acquired and analyzed the amount and spatial distribution of molecular features associated with several cancer hallmarks. We have demonstrated applications of the CSMA technology ranging from general cancer target gene discovery to detailed analyses of specific cellular processes including regulation of integrin cell-adhesion receptors [10,11]. The panel of assays compatible for screening using CSMA can be tailored to address many different biological questions. We expect the number of image cytometry endpoints to expand significantly during the coming years as genomics analyses of cancers and other disease types mature. 4. Future Directions The ability to assess changes induced by targeted RNAis using image cytometry assays enables assessment of the functional importance of a large number of disease-linked genomic aberrations and aberrant regulatory networks. Loss-of-function genetic screening has been used widely to assess the impact of RNAi knockdown on cell viability and/or immortalization. More recently, imaging-based screening strategies, coupled with immunofluorescent staining for cancer hallmarks has expanded the range of biological endpoints that can be assessed during RNAi screening. The ability to simultaneously assess the levels of specific proteins and RNA species, their spatial locations and the proximities of specific proteins, greatly increases the information that can be gained from RNAi screening experiments. This powerful form of high-throughput gene function assessment at cellular and subcellular resolution has already proven useful for functional annotation of normal and aberrant human genes. Several emerging imaging techniques such as automated confocal microscopy, super-resolution fluorescence microscopy and integrated light and electron microscopy will likely lead to further developments in the field of cell biological gene function assessment by enabling quantitative analysis of RNAi-induced changes in cellular function and structure. Understanding of human genetic and cell biological functions is increasing as the range of biomarkers for specific cellular processes increases. Multiplex analysis on CSMA carrying RNAis allows correlative analyses of the impact of gene perturbations on the molecular and cellular features they regulate in individual cells. In the future, RNAi screening experiments will likely shift from general cell biological discovery with established model cell lines to clinical applications where RNAi screening is performed directly with patient derived cells [6]. Acknowledgments The authors declare no competing interests. Work described here was partially funded by the Knight Cancer Institute, the NIH/NCI grant U54 CA112970 Integrative Cancer Biology Program (ICBP), NIH/NCI grant P50 CA58207 Bay Area Breast Cancer Translational Research Program (SPORE), European Commission Framework program 6 project ENLIGHT and European Commission Framework program 7 project AFFINOMICS. Appendix microarrays-02-00097-t001_Table 1Table S1 Surrogate markers for image cytometry-based quantitative cancer phenotypes. Marker Vendor Cat # Dilution Proliferation Ki-67 Abcam Ab15580 1:400 CDT1 Cell Signaling Tech. 8064 1:300 BrdU BD Pharmingen 555627 1:400 EdU (Click-iT A647 kit) Life Technologies C10356 10 µM Apoptosis Senescence Cleaved PARP Cell Signaling Tech. 9546S 1:400 Histone H3 (K9me3) Abcam Ab8898 1:300 DNA Damage Gamma-H2Ax Abcam Ab11174 1:300 P53BP1 Novus Biologicals NB100-904 1:300 CHEK2 (phospho T68) Abcam Ab38461 1:200 ATM (phospho S1981) Abcam Ab36810 1:300 TP53 (acetyl K381) Abcam Ab61241 1:300 Differentiation KRT8 Abcam Ab59400 1:300 KRT14 Abcam Ab7800 1:300 CTNNB1 Cell Signaling Tech. 8814 1:300 EpCAM Cell Signaling Tech. 2929 1:200 CDH1 (total) Cell Signaling Tech. 3195 1:200 CDH1 (extracellular epitope) Abcam Ab1416 1:200 Vimentin Cell Signaling Tech. 5741 1:300 Adhesion Matrix Active Integrin Beta-1 (12G10) Abcam Ab30394 1:200 Vinculin Abcam Ab18058 1:200 Fibronectin Epitomics 1574-1 1:300 Cytoskeleton Beta-Tubulin Santa Cruz Biotech. Sc-55529 1:300 F-Actin (phalloidin) Life Technologies A12379 1:100 ==== Refs References 1. Ziauddin J. Sabatini D.M. Microarrays of cells expressing defined cDNAs Nature 2001 411 107 110 10.1038/35075114 11333987 2. Mousses S. Caplen N.J. Cornelison R. Weaver D. Basik M. Hautaniemi S. Elkahloun A.G. Lotufo R.A. Choudary A. Dougherty E.R. RNAi microarray analysis in cultured mammalian cells Genome Res. 2003 13 2341 2347 10.1101/gr.1478703 14525932 3. Wheeler D.B. Carpenter A.E. Sabatini D.M. Cell microarrays and RNA interference chip away at gene function Nature Genet. 2005 37 25 30 15592469 4. Snijder B. Sacher R. Rämö P. Liberali P. Mench K. Wolfrum N. Burleigh L. Scott C.C. Verheije M.H. Mercer J. Single-cell analysis of population context advances RNAi screening at multiple levels Mol. Syst. Biol. 2012 8 10.1038/msb.2012.9 5. Hutchins J.R. Toyoda Y. Hegemann B. Poser I. Hériché J.K. Sykora M.M. Augsburg M. Hudecz O. Buschhorn B.A. Bulkescher J. Systematic analysis of human protein complexes identifies chromosome segregation proteins Science 2010 328 593 599 10.1126/science.1181348 20360068 6. Rantala J.K. Mäkelä R. Aaltola A.R. Laasola P. Mpindi J.P. Nees M. Saviranta P. Kallioniemi O. A cell spot microarray method for production of high density siRNA transfection microarrays BMC Genom. 2011 28 10.1186/1471-2164-12-162 7. Cekaite L. Rantala J.K. Bruun J. Guriby M. Agesen T.H. Danielsen S.A. Lind G.E. Nesbakken A. Kallioniemi O. Lothe R.A. Skotheim R.I. MiR-9, -31, and -182 deregulation promote proliferation and tumor cell survival in colon cancer Neoplasia 2012 14 868 879 23019418 8. Neumann B. Walter T. Hériché J.K. Bulkescher J. Erfle H. Conrad C. Rogers P. Poser I. Held M. Liebel U. Phenotypic profiling of the human genome by time-lapse microscopy reveals cell division genes Nature 2010 464 721 727 10.1038/nature08869 20360735 9. Doil C. Mailand N. Bekker-Jensen S. Menard P. Larsen D.H. Pepperkok R. Ellenberg J. Panier S. Durocher D. Bartek J. RNF168 binds and amplifies ubiquitin conjugates on damaged chromosomes to allow accumulation of repair proteins Cell 2009 136 435 446 10.1016/j.cell.2008.12.041 19203579 10. Rantala J.K. Pouwels J. Pellinen T. Laasola P. Potter C. Sundberg J.P. Kallioniemi O. Parsons M. Salmi M. Ivaska J. Sharpin is an endogenous inhibitor of β1-integrin activation Nat. Cell Biol. 2011 13 1315 1324 10.1038/ncb2340 21947080 11. Pellinen T. Rantala J.K. Arjonen A. Mpindi J.P. Kallioniemi O. Ivaska J. A functional genomic analysis of β1-integrin activity regulators J. Cell Sci. 2012 125 649 661 10.1242/jcs.090704 22389402 12. Winograd-Katz S.E. Itzkovitz S. Kam Z. Geiger B. Multiparametric analysis of focal adhesion formation by RNAi-mediated gene knockdown J. Cell Biol. 2009 186 423 436 10.1083/jcb.200901105 19667130 13. Bailey S.N. Ali S.M. Carpenter A.E. Higgins C.O. Sabatini D.M. Microarrays of lentiviruses for gene function screens in immortalized and primary cells Nat. Methods 2006 3 117 122 16432521 14. Moffat J. Grueneberg D.A. Yang X. Kim S.Y. Kloepfer A.M. Hinkle G. Piqani B. Eisenhaure T.M. Luo B. Grenier J.K. A lentiviral RNAi library for human and mouse genes applied to an arrayed viral high-content screen Cell 2006 124 1283 1298 10.1016/j.cell.2006.01.040 16564017 15. McGowan E.M. Alling N. Jackson E.A. Yagoub D. Haass N.K. Allen J.D. Martinello-Wilks R. Evaluation of cell cycle arrest in estrogen responsive MCF-7 breast cancer cells: Pitfalls of the MTS assay PLoS One 2011 6 e20623 10.1371/journal.pone.0020623 21673993 16. Björkman M. Östling P. Härmä V. Virtanen J. Mpindi J.P. Rantala J. Mirtti T. Vesterinen T. Lundin M. Sankila A. Systematic knockdown of epigenetic enzymes identifies a novel histone demethylase PHF8 overexpressed in prostate cancer with an impact on cell proliferation, migration and invasion Oncogene 2012 31 3444 3456 22120715 17. Rantala J.K. Edgren H. Lehtinen L. Wolf M. Kleivi K. Vollan H.K. Aaltola A.R. Laasola P. Kilpinen S. Saviranta P. Integrative functional genomics analysis of sustained polyploidy phenotypes in breast cancer cells identifies an oncogenic profile for GINS2 Neoplasia 2010 12 877 888 21082043 18. Stellaris™ Probe Designer Available online:http://www.biosearchtech.com/stellarisdesigner (accessed on 15 February 2013) 19. Hanahan D. Weinberg R.A. Hallmarks of cancer: The next generation Cell 2011 144 646 674 10.1016/j.cell.2011.02.013 21376230 20. Cancer Genome Atlas (TCGA) Available online:http://cancergenome.nih.gov/ (accessed on 15 February 2013) 21. Strauss R. Li Z.Y. Liu Y. Beyer I. Persson J. Sova P. Möller T. Pesonen S. Hemminki A. Hamerlik P. Analysis of epithelial and mesenchymal markers in ovarian cancer reveals phenotypic heterogeneity and plasticity PLoS One 2011 6 e16186 10.1371/journal.pone.0016186 21264259 22. Halazonetis T.D. Gorgoulis V.G. Bartek J. An oncogene-induced DNA damage model for cancer development Science 2008 319 1352 1355 10.1126/science.1140735 18323444 23. Kallioniemi A. Kallioniemi O.P. Waldman F.M. Chen L.C. Yu L.C. Fung Y.K. Smith H.S. Pinkel D. Gray J.W. Detection of retinoblastoma gene copy number in metaphase chromosomes and interphase nuclei by fluorescence in situ hybridization Cytogenet. Cell Genet. 1992 60 190 193 1354594 24. Raj A. van den Bogaard P. Rifkin S.A. van Oudenaarden A. Tyagi S. Imaging individual mRNA molecules using multiple singly labeled probes Nat. Methods 2008 5 877 879 18806792 25. Söderberg O. Gullberg M. Jarvius M. Ridderstråle K. Leuchowius K.J. Jarvius J. Wester K. Hydbring P. Bahram F. Larsson L.G. Landegren U. Direct observation of individual endogenous protein complexes in situ by proximity ligation Nat. Methods 2006 3 995 1000 17072308 26. Leuchowius K.J. Jarvius M. Wickström M. Rickardson L. Landegren U. Larsson R. Söderberg O. Fryknäs M. Jarvius J. High content screening for inhibitors of protein interactions and post-translational modifications in primary cells by proximity ligation Mol. Cell. Proteomics 2010 9 178 183 10.1074/mcp.M900331-MCP200 19864249

PMC005xxxxxx/PMC5003479.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2020063microarrays-02-00063ArticlePhenotypic MicroRNA Microarrays Kwon Yong-Jun Heo Jin Yeong Kim Hi Chul Kim Jin Yeop Liuzzi Michel Soloveva Veronica *Institut Pasteur Korea, IP-Korea, 696 Sampyeong-dong, Bundang-gu, Seongnam-si, Gyeonggi-do 463-400, Korea; E-Mails: yjkwon@ip-korea.org (Y.-J.K.); maple3644@ip-korea.org (J.Y.H.); island7boy@ip-korea.org (H.C.K.); jykim@ip-korea.org (J.Y.K.); michel.liuzzi@ip-korea.org (M.L.)* Author to whom correspondence should be addressed; E-Mail: Soloveva@ip-korea.org; Tel.: +82-31-8018-8008; Fax: +82-31-8018-8015.03 4 2013 6 2013 2 2 63 80 08 2 2013 20 3 2013 25 3 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Microarray technology has become a very popular approach in cases where multiple experiments need to be conducted repeatedly or done with a variety of samples. In our lab, we are applying our high density spots microarray approach to microscopy visualization of the effects of transiently introduced siRNA or cDNA on cellular morphology or phenotype. In this publication, we are discussing the possibility of using this micro-scale high throughput process to study the role of microRNAs in the biology of selected cellular models. After reverse-transfection of microRNAs and siRNA, the cellular phenotype generated by microRNAs regulated NF-κB expression comparably to the siRNA. The ability to print microRNA molecules for reverse transfection into cells is opening up the wide horizon for the phenotypic high content screening of microRNA libraries using cellular disease models. microRNAsiRNAphenotypic screen ==== Body 1. Introduction Two key events in biological sciences happened a decade ago around the year 2000: first, the sequence of the human genome was completed and published by Celera Co. and NIH, and second, the discovery of RNA interference was reported, where small double-stranded oligonucleotides mediate post-transcriptional gene silencing in a broad variety of organisms [1,2,3]. Since then, the new wave of technological applications has facilitated research and provided tools for basic research studies and drug discovery. Double-stranded RNA was found to mediate sequence-specific, post-transcriptional knock-down of the mRNA in cells and tissues of all origins of plants, insects and animals. RNA-mediated interference (RNAi) became useful in the analysis of functional genes and their roles in biological phenotypes in mammalian and particular human cells [3,4]. With the development of synthetic siRNAs and shRNAs, the RNA-interference methods were adopted for systematic and methodical screening approaches in studies of diverse biological processes, including mechanisms of disease pathogenesis [5,6,7]. The miRNAs (microRNA, miR) in contrast to the synthetic inhibitory RNAs, like siRNA and shRNA, are encoded by the genome and function as endogenous regulatory factors for both protein coding genes and non-coding RNAs. The expression of specific miRNAs in mammalian tissue was found to negatively correlate with the level of the mRNAs for genes whose 3'UTRs are targeted by miRNA [8]. Those and similar observations lead to the conclusion that miRNAs can control tissue- and disease-specific gene expression, mostly at the mRNA level. In brief, miRNA can induce two processes: the more common pathway involves promoting mRNA degradation by creating a perfect duplex and, thus, inducing duplex cleavage by the RNA-induced silencing complex (RISC) complex; the other process involves inhibiting mRNA translation by the formation of imperfect duplexes [8,9,10,11,12]. With the current knowledge of RNAi, the miRNA molecules can be engineered to mimic natural miRNAs and used to control the expression of genes of interest [13]. However, altogether, there are hundreds and thousands of natural miRNAs and synthetic miRNA mimics that could be involved in the regulation of at least 60% of genes in the human genome. The study of those molecules and of their function requires a robust methodology with a very good throughput. The hybridization techniques, like LNA or Q-PCR, were mostly used as a screening approach to search for miRNAs involved in the control of different cellular phenotypes or disease phenotypes of interest [14,15]. That allowed scientists to quantify overexpression of miRNAs in those cellular or tissue models, but did not reveal the function of each of the selected molecules. Recently, several screens were conducted to test miRNA collections for their specific effects on the expression of co-transfected reporter gene constructs [16,17,18]. However, the non-physiological overexpression of reporter genes cannot reflect correctly the complex mechanisms that are involved in regulation of post-transcriptional events. In this work, we are developing a phenotypic screening approach that would allow monitoring of the functional response of cells transfected with different RNAi molecules. High density spot microarrays became popular after 1995, when P. Brown with his group [19] created the first bimolecular printing solution for DNA microarrays. The microarray approach is useful when there is a need for repetitive analyses of different samples with a large collection of the probes, like polypeptides, mRNA (cDNAs), etc. [20,21]. The first application of microarrays that introduced genetic material into a monolayers of cultured cells was first suggested and demonstrated by Ziauddin, J. and Sabatini, M. in 2001 [22]. Even though this screen tested only 192 cDNA constructs, it clearly demonstrated the advantages of throughput and the flexibility of the microarray format for cell-based screening of a bio-molecule collection. The feasibility of using a phenotypic approach instead of a DNA or RNA hybridization method opened up a broad spectrum of applications for cellular assays. It also highlighted the conceptual idea that the cellular phenotypic response could be used in an automated high throughput mode with reproducible precision and accuracy, ensuring identification of new genes [23,24]. To develop the microarray application for over-expression of microRNAs, we selected the regulation of endogenous NF-κB expression in cancer cells. Nuclear factor kappa-light-chain-enhancer of activated B-cells (NF-κB) is a protein complex that controls DNA transcription and regulates diverse signaling processes in many animal cells. Incorrect regulation of NF-κB has been associated with cancer, inflammatory and autoimmune diseases, due to the control of cell proliferation, differentiation and survival [25,26]. There are five members in the NF-κB family. Proteins from class I NF-κB1 or p50 and NF-κB2 or p52 and class II RelA, RelB and c-Rel can form homo- or hetero-dimers to create complexes that are able to control gene transcription [27]. Due to the critical role of NF-κB in cellular functions, the regulation of its expression and activation happens at several different levels. One pathway controls the existing level of inactive NF-κB dimers by its association with the inhibitory protein complex (IkB). The degradation of IkB frees NF-κB for activation and transport into the nucleus. The other pathway is more complex and includes post-translational modifications of different members of the NF-κB family, as well as IkB and related signaling cascades (review Oeckinghaus and Ghosh 2009 [28]). MicroRNAs have been shown to be involved in cell differentiation, immune response and tumor development and metastasis progression [29,30,31,32]. It is not surprising that there have been several studies trying to establish connections between NF-κB signaling pathways and microRNA functions and, in particular, tumor development and progression [16,33,34,35,36]. Several miRNAs, like miR-146, miR-155, miR-181b, miR-21 and miR-301a, are involved in NF-κB activation, and at the same time, they play a significant role in tumorigenesis [31]. In line with bioinformatic predictions, the co-transfection of reporter genes and microRNA from the miR-520/373 family reduced expression of the reporter-gene associated with RelA 3'UTR [16,35,37]. In our work, we are using the regulation of NF-κB (RelA) expression as a model to develop a new microarray-based approach that can facilitate the use of phenotypic analysis of cellular responses in the search for new critical modulators within a collection or library of microRNAs or microRNA mimic molecules [13]. 2. Experimental Methods 2.1. Chemicals and Cell Culture All fine chemicals were purchased from Sigma-Aldrich. DRAQ5 (Cat#: DR50050) was from BioStatus (Shepshed, UK). The OTP (On-TARGET plus) version of siRNA duplexes—RelA siRNA smart pool, L-003533-00-0005, non-targeted siRNA #1, D-001810-01-05 and human miRIDIAN microRNA: has-miR-373 mimic, C-300680-03-0005, has-miR-520c-3p mimic C-300803-05-0005 and microRNA mimic negative control #1, CN-001000-01-05—were purchased from Dharmacon (Lafayette, CO, USA). Primary antibodies, NF-κB p65(# SC-109), were from Santa Cruz Biotechnology and fluorescent secondary antibodies, Alexa Fluor® 488 donkey anti-rabbit IgG (H+L), MOP-A-21206, from Molecular Probes/Invitrogen (Carlsbad, CA, USA). Hela cells (ATCC, Manassas, VA, USA) were cultivated in high-glucose Dulbecco’s Modified Eagle Medium (DMEM) (Invitrogen, Carlsbad, CA, USA) supplemented with glutamax, 110 mg/mL sodium pyruvate, 10% fetal calf serum (Gibco, Carlsbad, CA, USA) and 1% penicillin streptomycin (Invitrogen Carlsbad, CA, USA). For siRNA forward transfection, Hela cells were seeded at 1.5 × 104 cells/well in 96-well plates and cultured in growth medium without antibiotics for 16 h before transfection. For reverse transfection on arrays, cells were seeded at 2 × 106 cells/array and cultured in OptiMEM medium supplemented with 5% fetal calf serum (Gibco) and 1% penicillin streptomycin (Invitrogen) for 48 h. 2.2. miRNA and siRNA Transient Transfection Hela cells were trypsinized one day before transfection, diluted in fresh DMEM high glucose medium supplemented with 5% FBS without antibiotics and transferred to 96-well plates (Greiner, Washington, DC, USA). 5,000 Hela cells were seeded per well and cultivated for 16 h. Transient transfection of siRNAs was carried out using DharmaFECT 2 (Dharmacon, Thermofisher, West Lafayette, IN, USA). For each well, 9.9 µL of serum-free DMEM and 0.1 µL of DharmaFECT 2 were pre-incubated for 5 min at room temperature. At the same time, 5 µL of serum-free DMEM were mixed with 5 µL of each siRNA (1 µM) and also incubated for 5 min at room temperature. The two mixtures were combined and incubated for 20 min at room temperature for complex formation. After addition of 80 µL of complete DMEM medium to the mixture, the entire solution was added to the cells in each well, resulting in a final concentration of 50 nM for the siRNAs. After transfection, cells were incubated for 48 h to allow gene silencing to occur. 2.3. Phenotypic NF-κB Assay and Data Analysis To develop the NF-κB detection assay, cells were transfected with siRNAs for 48–72 h. After incubation, the cells were washed twice with phosphate-buffered saline (PBS), fixed for 10 min with 4% (w/v) paraformaldehyde in PBS, washed again with PBS and permeabilized with 0.1% (v/v) Triton X 100 in PBS for 10 min. Permeabilized cells were washed in PBS and incubated with a 1:200 dilution of anti-p65 antibody (Rel A, #SC-109, Santa Cruz, CA, USA) in 10% (v/v) goat serum PBS overnight at 4 °C. Cells were washed 3 times with PBS for 10 min on an orbital rotator and treated with Alexa 488 goat anti-rabbit secondary antibody (1:1,000) (Molecular Probes) for 60 min at room temperature. Cells were washed 3 times for 10 min with PBS on an orbital rotator before the addition of 5 μM DRAQ5 in PBS and incubated for 10 min at room temperature. The images of the cells on the plates were acquired with a 20× objective using an ImageXpress Ultra point scanning confocal microscope (Molecular Devices, Sunnyvale, CA, USA). Images were acquired at 488 nm (fluorescein isothiocyanate (FITC)) to detect RelA expression, and at 635 nm, to record nucleus staining with DRAQ5. Quantification of RelA silencing was performed using MetaXpress software (Molecular Devices). Cells were identified and counted using the nucleus mask. The intensity of the RelA signal was a threshold to exclude the population of cells expressing low levels of NF-κB. The data were normalized as a ratio of cells with an intensity of FITC/pixel above the threshold to the total amount of cells detected in the image at 635 nm. The ratio value of cells in wells with non-target control transfection was used as 100%. The numbers of cells that show the RelA signal higher than the threshold were converted in the % of the total cell number in that image or the percent of expressing cells. Statistical analysis of the data was done using GraphPad Prism. 2.4. RNA Extraction and Real-Time PCR Total RNA was extracted from Hela (up to 106) cells using Trizol reagent (Invitrogen). Target RNA was reverse transcribed using the Moloney murine leukemia virus (MMLV) reverse transcriptase enzyme (Promega). In the first step, 5 μM Oligo (dT) 16 was added to 0.5–1 μg of total RNA and annealed at 70 °C for 10 min. Then, 100 U MMLV reverse transcriptase was added in the presence of 50 mM Tris-HCl, pH 8.3, 75 mM KCl, 3 mM MgCl2 and 5 mM unlabeled deoxynucleotides (dNTPs) and incubated at 37 °C for 60 min. For each experiment, RT (reverse transcriptase)-minus controls were included to provide a negative control for subsequent PCR reactions. For PCR amplification, a maximum of 1 μL of cDNA was used per 50 μL PCR. Larger amounts were avoided, because they might significantly inhibit PCR amplification. To minimize variations in reverse transcriptase efficiency, all samples from a single experiment were reverse transcribed simultaneously. Real-time PCR was performed with the SYBR Premix Ex Taq II (TaKaRa Bio Inc., Shiga, Japan) and MJ Research PTC-200 Thermo Cycler (BioRad, Hercules, CA, USA). Amplification was done in a 20 µL final volume containing 1–5 μL of template cDNA, 10 μL of SYBR Premix Ex Taq II and 0.4 μM of each primer. The protocol included an initial denaturation step at 95 °C for 15 min, followed by 32 cycles of 95 °C for 20 s, 60–65 °C for 30 s and 72 °C for 30 s. After amplification, a melting curve was obtained by increasing temperatures from 65 °C to 95 °C with fluorescence detection at 0.2 °C intervals. The quantification of the target gene was performed using the cycle threshold (Ct) value in a PCR amplification curve by cluster analysis with variable cluster endpoints. Data were determined from duplicate assays. For normalization, the cell number in the specimen was determined from glyceraldehyde 3-phosphate dehydrogenase (GAPDH) gene quantification. 2.5. siRNA and miRNA Printing siRNA/miRNA transfection solution was prepared essentially as described [38,39,40,41]. Briefly, 2 μL of 20 μM siRNA was transferred into 384 well plates. 6 μL of Red mixture, which contains 20 μM RED siGLO (Thermofisher, West Lafayette, CO, USA) 2 μL of 0.8 M sucrose dissolved in OptiMEM media and 2 μL of RNASE free water were added. Then, lipofectamine 2000 was added and mixed thoroughly. Then, the mixture was incubated for 20 min at RT, and 5 μL of 0.3% (w/v) gelatin was added. The final transfection reagent mixture was dispensed or printed as 300 µm spots on a microscope glass slide (MAS slides, Mutsunami, Japan) using stealth pins SMP 9 (Telechem, Sunnyvale, CA, USA) and a high throughput microarray printer (Genomic Solutions, Ann Arbor, MI, USA) at 22–25 °C, 55–65% RH (room humidity) enclosed in a custom built clean chamber providing a sterile HEPA filtered atmosphere. For some experiments with siRNA, we also tested glass cover slips (#1 Marienfeld, Germany) coated with poly-L-lysine (Sigma, St. Louis, MO, USA). Arrays were stored in a desiccator with no significant alterations in performance from 1 week to 5 months post-printing. Five slides covered the genome and contained 16% of control siRNA. We printed the genomic library such that the whole library was composed of 5-arrays set with 3,888 siRNA spots per array, containing 4 different sequences for each gene of the 18,000 individual human open reading frames from the Dharmacon “On-TARGET plus siRNA” library. 2.6. Microarray-Based Phenotypic Assay and Data Analysis Cells were seeded at 2 × 106 cells on microarrays located in 4-well cell culture dishes (Nunc, Drive Rochester, NY, USA) and cultured for 48 h to allow the transient transfection to occur. Cells were fixed and stained with anti-RelA antibody, as described for the well-based assay (see Section 2.3). Confocal images were acquired 48 h post-transfection using ImageXpress Ultra (Molecular Devices), 10× objective lenses. Three channels were used for the readout of RelA expression at 488 nm (FITC), for the detection of siGLO-RED at 560 nm (Texas Red) and for the detection of nucleus staining with DRAQ 5 at 635 nm. Quantification of p65 silencing was performed using MetaXpress software (Molecular Devices). First, we extracted red miRNA spots based on spot intensity at 560 nm. Only those cells that were associated with that spot area were analyzed to measure RelA expression. The intensity of the RelA signal/pixel was thresholded to remove cells with a low intensity of RelA signal from the analysis. The results were normalized to the total amount of cells in the image and converted in the percent of expressing cells (or % of expression), as described for image analysis of the cells on plates (Section 2.3). 3. Results and Discussion 3.1. Phenotypic Assay of siRNA and miRNA Knock-down Effect in Cellular Model First, we established the functional assay that allows for the quantification of the cellular response to the transfected RNAi. As mentioned above, we chose to analyze the changes in the endogenous level of NF-κB protein after transfection with siRNA or miRNA. For the siRNA experiment, we selected OTP smart pool siRNA designed against the RelA member of the NF-κB family and, as a negative control, the OTP non-targeted siRNA #1. As any pooled siRNA, this sample consists of four different siRNA sequences against RelA to increase the silencing efficiency and decrease the chance of off-target effects. As was mentioned in the introduction, several miRNAs were found to control NF-κB expression. We selected has-miR-373 and has-miR-520c-3p mimic, as a miRNA that targets NF-κB, and microRNA mimic negative control #1 from the human miRIDIAN microRNA collection from Thermo scientific. The basic conditions that we are carrying out in the lab for direct transfection (see Experimental Methods) allow us to achieve 80% reduction of the mRNA copy number based on the RT-PCR analysis (Figure 1(a)). The same transfection protocol was used for image analysis of the cells in the treated wells. Figure 1(b,c) shows that the image analysis detecting anti-p65 (RelA) staining yielded a similar ~80% reduction of the RelA protein level after transfection of the anti-RelA siRNA. Next, the miR-520 and miR-373c were transiently transfected in Hela cells following exactly the same protocol as for siRNA. As was mentioned before, the siRNA was a pool of four different sequences targeting the same mRNA, but miRNA was represented only by one individual sequence and might be less efficient in silencing the target. Figure 2 shows that both microRNAs, targeting 3'UTR of ReLA, gave a very similar reduction effect on the level of mRNA as the pooled siRNA targeting the mRNA directly. Figure 1 Direct transfection of Hela cells with RelA siRNA. Hela cells were transfected with RelA targeted siRNA vs. scramble siRNA for 48 h. (a) The level of RelA mRNA was measured by qPCR after transfection of anti-p65 siRNA and incubation for 48 h. (b) Cells transfected with anti-p65 siRNA were stained with anti-RelA antibody (Green) and DRAQ5 for nuclei detection (white (nucleus)). The exposure for the representative cell images for this figure was increased to show the presence of the cells in the images where the signal would be too low. (c) Image analysis used MDS MetaMorph image analysis software. The total level of RelA signal was analyzed as described in Section 2.3. The resulted data are presented as % of expressing cells (RelA positive cells). Data are plotted as bar-graphs using GraphPad Prism software, the error-bar representing the standard deviation for n = 3 replicates. Figure 2 Comparison of NF-κB expression modulated by siRNA and miRNA-373c, miRNA-520 transfection. Hela cells were transfected with 20 nM of scramble siRNA and anti-RelA SiRNA, and the negative control (NC) for microRNA, miR-373c and miR-520 and incubated for 48 h before harvesting and analyzing the level of endogenous p65 by real-time PCR. We performed the titration microRNA to be sure that the concentration of miRNA was optimal for efficient knock-down. The siRNA concentration that produces 80% knock-down usually varied around 20–50 nM; the miRNA produced a similar silencing effect, already at 10 nM (Figure 3), and 80% of silencing was detected as low as 1 nM for miR-520. Figure 3 NF-κB expression was modulated by miRNA-373c and miRNA-520 transfection. Hela cells were transfected with negative control (NC), miR-373c and miR-520 and incubated for 48 h before fixation and image analysis. (a) Cells were stained with anti-RelA antibody (Green) and DRAQ5 for nuclei detection (white (nucleus)). As for the figure above, the exposure for cell images was increased to show the presence of the cells in the images where the signal would be too low. (b) Image analysis was carried out using MDS MetaXpress image analysis software. The RelA signal analysis is described in Section 2.3 and was presented as % of expressing cells (or % RelA-expressing cells). Data were plotted as bar-graphs using GraphPad Prism software, the error-bar representing the standard deviation for n = 3 replicates. This is the first phenotypic approach demonstrating the down regulation of endogenous RelA protein in Hela cells in response to transfected miR-520 and miR-373c. This is also a proof of concept that the same phenotypic imaging approach can be used to detect the effect of both siRNA and miRNA on the endogenous level of NF-κB expression in cultured cells. 3.2. RNA Interference Microarrays A robust high throughput approach is required to screen large libraries of RNAi designed for more than 18,000 individual genes of the human genome, as well as thousands of miRNAs or miRNA mimics designed by nature (or scientists) to control expression of those genes. Hybridization microarrays were mentioned above and currently used very broadly to study expression level in selected tissues or cells. We have adapted the high density spot microarrays for reverse-transfection of RNAi or cDNA into cells with microscopic monitoring of the changes in the selected phenotype or biomarkers. In our next experiments, we adapted the phenotypic analysis of RelA regulation by siRNA and micro-RNA to microarray-based reverse-transfection assay. 3.2.1. High-Density Spot Microarrays for Reverse Transfection The steps for array-based reverse-transfection (Figure 4) are different from those of the forward transfection in wells described in Section 2.2. Figure 4 The flow chart for reverse-transfection of RNAi into cell monolayers on microarrays. One of the main issues for reverse transfection is that the RNAi should be immobilized with a transfection reagent on the glass-slide in the form of a spot with a defined size and location. Once the cells are seeded on the top of the spots, RNAi should be transferred into cells attached right above the spot. To print uniform spots with a defined size, to reach high efficiency of transfection and to prevent cross contamination between spots, those are critical aspects in the process of the protocol optimization. The reverse transfection into cultured cell monolayers was fist demonstrated for cDNA constructs transfected into the Hek293 cell line by Ziauddin, J. and Sabatini in 2001 [21] and later was applied to the screening of a subset of siRNA libraries [22,23,38]. To adapt this method for experimental use, we applied multistep optimization to achieve accurate dispensing of the transfection mixture on the glass slides and efficient reverse transfection of the recombinant molecules. Several parameters needed to be considered simultaneously: A. Selection of the right coating of the glassware that would efficiently immobilize RNAi reagents in a spot, preventing cross-contamination between spots and allow cell adherence on the slide and on the spot after seeding cells on the slide. We used poly-L-lysine and/or the MAS coating of the microarray glass slides. The majority of adherent cell lines that were tested in the lab demonstrated proper attachment to those surfaces. The printed microarrays can be stored at a low humidity environment (~10%) up to 12 months without the significant change of the transfection efficiency of the spotted reagent. B. The transfection reagent mixture used for the arrays contains not only RNAi and the transfection reagent, but also labeled tracer RNA molecules for spot visualization (siGLO-RED in our case), gelatin and sucrose, which provide better immobilization of reagents after printing, as well as more efficient delivery of the RNA/transfection reagent complex into the cells. (see Table 1 and Section 2.2, Section 2.5). We optimized the composition of the reagent mixture that was used to print the OTP siRNA library (Table 1) and will discuss below the optimization of the reagent mixture for miRNA printing on glass slides. There is a big variety of commercially available transfection reagents that were developed to achieve good transfection efficiency with minimal cytotoxic effects. If there was a new cell line that needed to be transfected in microarray-based assays, we would have to test several of those reagents, like Lipofectamine 2000, RNAiMax and DharmaFECT, to obtain the best results. microarrays-02-00063-t001_Table 1Table 1 The composition of transfection reagent mixture for well-based or microarray-based transfection. Component of mixture for reverse transfection in wells on array 1 lipofectamine 0.1 μL 1 lipofectamine 3 μL DMEM 9.9 uL 2 SiRNA (1 μM) 5 μL (1 μM) 2 SiRNA (20 μM) 2 μL (20 μM) DMEM 5 μL 3 0.3 M Sucrose in OptiMEM 2 μL (0.3 M) RNAse free water 2 μL Mixed and incubate for 20 min Mixed and incubate for 20 min 3 OptiMEM 80 μL 4 siGLO (20 μM) 2 μL 5 0.2% gelatine 5 μL add 100 μL to cells in 1 well/96 well plate or 20 μL in 1 well/384 well plate Printed 3–6 μL/300 μm spot C. The size and quality of the spots that reflect the amount of reagent that will be delivered into cells must be consistent through all arrays. It depends not only on the size of the pins selected for printing, but also on the viscosity of the transfection mixture, time and depth of the pin immerging in the reagent. With our selection of the pin-size, MAS coating for slides and composition of the siRNA transfection mixture, as listed in Table 1, the size of the spots was about 300 µm, with a distance between spots of ~200 µm (Figure 5). Those dimensions allowed us to print 4,000 spots containing individual siRNAs on one 26 × 76 mm glass slide (Figure 6). Thus, the whole library of 18,000 human genome siRNAs occupies only five slides. This miniaturization allows us to run multiple replicates during the genomic screen to ensure high quality results [40,41]. Figure 5 Spotting of siRNA on microarrays. Mixture of siRNA, siGLO (red) and transfection reagents described in Table 1 were printed on PLL-coated slides. The red is the detection of siGLO-RED siRNA tracer at the 560 nm channel for visualization of the printed spots. Cells were stained with an anti-RelA/secondary antibody labeled with Alexa 488 (green), detected at 488 nm channel, and DRAQ5 (blue), measured at 635 nm for nuclei detection. (a,c) Containing non-targeting siRNA; (b) anti-RelA siRNA; the reduction of green immuno-staining signals indicates the decrease of NF-κB protein expression in cells. Figure 6 High density spot siRNA microarray. The tissue culture glass with siGLO siRNA tracer (red) spots visualized at 561 nm to assess the quality of the array printing. 3.2.2. Optimization of the Transfection Reaction for siRNA and miRNA on Microarrays The selected concentrations of sucrose and gelatin were also associated with the specific coating of the glassware used in this experiment. In order to immobilize miRNA spots for efficient delivery of miRNA into cells, both the gelatin and the sucrose are essential components in the miRNA transfection mixture. To optimize the miRNA reverse transfection efficiency, we tested gelatin and sucrose in a combination experiment. The high concentrations of both gelatin and sucrose were not optimal in spite of the better transfection efficiency, not only due to cross-contamination caused by the spread of the reagent form the spot area and inaccurate dispensing of the reagent, but because of the high viscosity of mixture. The results in Figure 7 show the matrix experiment designed to select the optimal concentrations of both reagents. The selected conditions, 100 mM sucrose and 0.09% gelatin, were confirmed in several experiments, not only with miR-373c, but also with miR-520 (data not shown). Figure 7 Optimization of the transfection reagent composition for miRNA array. (a) The immunostaining for RelA with Alexa 488 in Hela cells on small-scale siRNA microarrays containing miR-373 and non-targeted miRNA (NC, negative control) printed on MAS-coated glass slide with a matrix titration of different components of the transfection reagent mixture: sucrose from 50–125 mM (vertical) and gelatin from 0.03–0.25% (horizontal). (b) Annotation for the images of the cells representing three channels: siGLO-560 nm (Red), for spot localization; Nuclei-635nm (Blue), DRAQ5 signal; RelA-488 nm (Green). The overlay is the image created by merging all three images together. When cells are transfected with siRNA or miRNA, the concentration of those molecules needs to be minimal to avoid any cytotoxic effects, but sufficient to produce a >70% knock-down effect on the targeted protein. The concentration of the siRNA stock that needs to be added to the reagent mixture was 10–20 uM. The range of concentrations was tested for miR-373 (Figure 8(a)). As the inhibitory RNA requires time to mediate the degradation of targeted mRNA and to decrease the overall protein level detected in cells, in our experiments, siRNA usually takes 48 to 72 h after transfection to establish a silencing effect for proteins like RelA. That effect starts to disappear after 96 h post-transfection (data not shown). We also performed a kinetic study for miRNA transfection. Figure 8(b) demonstrates that the effect of the miRNA-373 diminishes slightly after 72 h. Thus, the 48 h incubation time appears to be optimal for both siRNA and miRNA. Figure 8 Optimization of the conditions for miRNA microarrays. (a) Test of different concentrations of miR-373 (373 miRNA) and non-targeted negative control in microarray. Data is plotted by GraphPad Prism, n = 4. (b) The time course for incubation, 48 and 72 h, of the cells on microarray. Data is plotted by GraphPad Prism; error bar: S.D., n = 4. All images of cells in experiments (a) and (b) were processed as described in Section 2.6. 3.3. Phenotypic Assay for Cellular Monolayer Microarrays with miRNA and siRNA As shown in Figure 1, Figure 3, both anti-RelA siRNA and miR-520/373 produced endogenous RelA protein knock-down in Hela cells. We applied the phenotypic assay using high content image analysis to the cells seeded and cultured on microarrays. The conditions for reverse transfection were optimized for both types of inhibitory RNA molecules. Cells were incubated on microarrays for 48 h to maximize the silencing of the RelA detected by imaging. Figure 9 shows that the knock-down effect produced by the siRNA and the miRNA is significant and comparable. However, the effect of miR-520 was slightly lower than expected, but still showed >70% silencing efficiency of the total RelA detectable in the cells. 3.4. Discussion The goal of this work was to adapt the previously used phenotypic siRNA microarray-based screening methodology [40,41] to screening of miRNAs, a new group of RNA interfering molecules. Our phenotypic approach is using analysis of cellular images acquired by high resolution confocal microscopy to assess the changes of biomarkers or proteins of interest, as well as other morphological changes in a variety of cells, including immortalized cell lines, primary cells or stem cells. The use of the endogenous cellular environment is a critical aspect that makes phenotypic assays more biologically relevant and better suited to study the mechanisms that are taking place in tissues and even in whole organs. The application of phenotypic screening could be used in a variety of projects to search for new regulatory (like miRNAs) or functional components that play a crucial role in different diseases, like tumorigenesis, liver, kidney, cardio- or autoimmune disorders [14,17,42,43]. Figure 9 miRNA high density spots microarray. (a) The array of miRNA (miR-373 and miR-520 and non-coding control) spots used for phenotypic assay and quantitative image analysis: each test (square) shows three images from three channels (red: spots; green: RelA signal; blue: nuclei) and one overlay/merge of those three images (see annotation for Figure 6(b)). (b) The bar-graph plot of the data from images normalized as described in methods and presenting the % of cells expressing a detectable level of NF-κB. The data shows the average % of cells expressing RelA, n = 4; error: S.D. In a Student’s t test, the difference between the control and each of the miRNAs is significant, with p < 0.0001. (c) The array of siRNA (anti-RelA and scramble non-targetd control) spots for phenotypic assay and quantitative image analysis. (d) The bar-graph plot of the data from images for siRNA spots. t test: p < 0.0001. In this work, we have developed the first phenotypic cell-based microarrays for screening of the miRNA or synthetic miRNA mimic collections. Phenotypic microarray screening (or PhenomicID™) allowed us to perform thousands of individual experiments under controlled conditions to analyze large RNAi collections. The uniformity of the conditions ensures the reproducibility and the statistical reliability of individual observations. This feature makes the microarray approach a reliable tool for complex phenotypic assays [40,41]. Another advantage of microarray-based techniques is the significant cost reduction, due to the miniaturization of the volumes of reagents used for array printing (pmoles of materials and pL of volumes) and for the detection assays. Of course, there is a primary investment in building up the printing capabilities in the laboratory that need to be considered. The third, but not less critical, advantage is the significant increase of the throughput that is only comparable to the use of 1,536 and 3,456-well plate formats. The microarrays characteristics allow very complex experimental designs with multiple replicates and a large variety of samples. We have been able to apply the microarray screening of siRNAs in combination with six different concentrations of a compound of interest, which altogether used 21 copies of the human genome siRNA library (results from the laboratory in preparation for publication). As we have demonstrated in this work, the phenotypic microarray technology fits perfectly well to analyze the functions of miRNA collections in which each molecule could potentially control hundreds of genes and produce a clear phenotypic effect on cellular functions critical in development and disease. For example, a large set of miRNAs are known to be under-expressed in human tumors compared to normal tissue [44]. That analysis was done using hybridization techniques. The introduction of the same panel of microRNA mimics into relevant cell lines via microarray-based transfection would potentially result in reduction of cell proliferation and initiation of differentiation or apoptotic phenotypes detected by phenotypic analysis [45]. Such cell line profiles could add new specific microRNAs to anti-oncogenic or anti-metastatic pathways. It could be even more interesting to test the anti-miRNA molecules as a potential suppressors of miRNAs overexpressed in some diseases, such as autoimmune deregulations in lupus models [46]. 4. Conclusions Regulation of the NF-κB complex transcription and activation is very critical for cellular functions and, overall, cell fate. In this work, phenotypic cellular imaging was used to assess the effect of miRNAs, in particular, miR-520/373, on the total level of RelA protein detectable in Hela cells. The combination of phenotypic assays with high-density spot microarrays will allow for the screening of miRNA libraries in order to identify new molecules involved in the control of NF-κB or other critical cellular factors. Acknowledgments This work was supported by the Bio and Medical Technology Development Program of the National Research Foundation (NRF) and the National Research foundation of Korea (NRF) grant funded by the Korean government (MEST) (No.2012-00011 and No. 2012M3A9B6055466), Gyeonggi-do and KISTI. ==== Refs References 1. Fire A. RNA-triggered gene silencing Trend Genet 1999 15 358 363 10.1016/S0168-9525(99)01818-1 2. Sharp P.A. RNA interference—2001 Genes Dev. 2001 15 485 490 10.1101/gad.880001 11238371 3. Elbashir S.M. Harborth J. Lendeckel W. Yalch A. Weber K. Tuschl T. Duplex of 21-nucleotide RNAs mediates RNA interference in cultured mammalian cells Nature 2001 411 494 498 11373684 4. Wianny F. Zrnicka-Goetz M. Specific interference with gene function by double-stranded RNA in early mouse development Nature Cell Biol. 2000 2 70 75 10.1038/35000016 10655585 5. Sharma S. Rao A. RNAi screening: Tips and techniques Nat. Immunol. 2009 10 799 804 10.1038/ni0809-799 19621037 6. Seyhan A.A. Ryan T.E. RNAi screening for the discovery of novel modulators of human diseases Curr. Pharmaceut. Biotechnol. 2010 11 735 756 10.2174/138920110792927766 7. Mohr S. Bakal C. Perrimon N. Genomic screening with RNAi: Results and challenges Ann. Rev. Biochem. 2010 79 37 64 10.1146/annurev-biochem-060408-092949 20367032 8. Sood P. Krek A. Zavolan M. Macino G. Rajewsky N. Cell-type-specific signatures of microRNAs on target mRNA expression Proc. Natl. Acad. Sci. USA 2006 103 2746 2751 16477010 9. Baek D. Villen J. Shin C. Camargo F.D. Gygi S.P. Bartel D.P. The impact of microRNAs on protein output Nature 2008 455 64 71 18668037 10. Guo H. Ingolia N.T. Mammalian microRNAs predominantly act to decrease target mRNA level Nature 2010 466 835 840 20703300 11. Pasquinelli A.E. Hunter S. Bracht J. MicroRNAs: A developing story Curr. Opin. Genet. Dev. 2005 15 200 205 10.1016/j.gde.2005.01.002 15797203 12. Janas M.M. Wang E. Love T. Harris A.S. Stevenson K. Semmelmann K. Shaffer J.M. Chen P.H. Novina C.D. Reduced expression of ribosomal proteins relieves micro-RNA-mediate repression Mol. Cell 2012 46 171 186 22541556 13. Wang Z. The guideline of the design and validation of miRNA mimics Methods Mol. Biol. 2011 676 211 223 10.1007/978-1-60761-863-8_15 20931400 14. Sokilde R. Barken K.B. Mouritzen P. Moller S. Litman T. MicroRNA expression analysis by LNA enhanced microarrays MicroRNA Profiling in Cancer Gusev Y. Pan Stanford Publisher Singapore 2010 15. Pandey P. Brors B. Srivastava P.K. Bott A. Boehn S.N.E. Groene H.J. Gretz N. Microarrays-based approach identifies microRNAs and their target functional patterns in polycystic kidney disease BMC Genomics 2008 10.1186/1471-2164-9-624 16. Keklikoglou I. Koerner C. Schmidt C. Zhang J.D. Heckmann D. Shavinskaya A. Allgayer H. Guckel B. Fehm T. Schneewewiss A. Sahin O. Wiemann S. Tschulena U. MicroRNA-520/373 family functions as a tumor suppressor in estrogen receptor negative breast cancer by targeting NF-κB and TGF-β signaling pathways Oncogene 2011 10.1038/onc.2011.571 17. Zhang Y. Fan K.J. Sun Q. Chen A.Z. Shen W.L. Zhao Z.H. Zheng X.F. Yang X. Functional screening for miRNAs targeting Smad4 identified miR-199a as a negative regulator of TGF-β signaling pathway Nucl. Acids Res. 2012 40 10.1093/nar/gks667 18. Eulalio A. Mank M. Ferro M.D. Zentilin L. Sinagra G. Zacchigna S. Giacca M. Functional screening identifies miRNAs inducing cardiac regeneration Nature 2012 492 376 381 23222520 19. Schena M. Shalon D. Davis R.W. Brown P.O. Quantitative monitoring of gene expression patterns with a complementary DNA microarray Science 1995 270 467 470 7569999 20. Cheung V.G. Morley M. Aguilar F. Massimi A. Kucherlapati R. Childs G. Making and Reading Microarrays Available online:http://www.rose-hulman.edu/~ahmed/making%20and%20reading%20cdna%20microarrays.pdf (accessed on 8 February 2013) 21. Barbulovic-Nad I. Lucente M. Sun Y. Zhang M. Wheeler A.R. Bussmann M. Bio-microarray fabrication techniques—A review Crit. Rev. Biotechnol. 2006 26 237 259 10.1080/07388550600978358 17095434 22. Ziauddin J. Sabatini D.M. Micro-array of cells expressing defined cDNAs Nature 2001 411 107 110 10.1038/35075114 11333987 23. Silva J.M. Mizuno H. Brady A. Lucito R. Hannon G.J. RNA interference microarrays: High-throughput loss-of-function genetics in mammalian cells PNAS 2004 101 6548 6552 15084744 24. Mousses S. Caplen N.J. Cornelison R. Weaver D. Basik M. Hautaniemi S. Elkahloun A.G. Lotufo R.A. Choudary A. Dougherty E.R. Suh E. Kallioniemi O. RNAi microarrays analysis in cultured mammalian cells Genome Res. 2007 10.1101/gr.1478703 25. Karin M. Greten F.R. NF-κB: Linking inflammation and immunity to cancer development and progression Nat. Rev. Immunol. 2005 5 739 759 26. Karin M. Nuclear factor-κB in cancer development and progression Nature 2006 441 431 436 16724054 27. Hoffmann A. Baltimore D. Circuitry of nuclear factor κB signaling Immunol. Rev. 2006 210 171 186 10.1111/j.0105-2896.2006.00375.x 16623771 28. Oeckinghaus A. Ghosh S. The NF-κB family of transcription factors and its regulation Cold Spring Harb. Perspct. Biol. 2009 10.1101/cshprespect.a000034 29. O’Donnell K.A. Wentzel E.A. c-Myc-regulated microRNAs modulate E2F1 expression Nature 2005 435 839 843 15944709 30. He L. Thomson J.M. Hemann M.T. Hernando-Monge E. Mu D. Goodson S. Powers S. Cordon-Cardo C. Lowe S.W. Hannon G.J. Hammond S.M. A microRNA polycistron as a potential human oncogene Nature 2005 435 828 833 10.1038/nature03552 15944707 31. Mraz M. Pospisilova S. Malinova K. Slapak I. Mayer J. MicroRNAs in chronic lymphocytic leukemia pathogenesis and disease subtypes Leuk. Lymphoma 2009 50 506 509 10.1080/10428190902763517 19347736 32. Cordes K. Srivastava D. MicroRNA regulation of cardiovascular development Circ. Res. 2009 10.1161/circresaha.108.192872 33. Ma X. Becker Buscaglia L.E Barker J.R. Li Y. MicroRNAs in NF-κB signaling J. Mol. Cell Biol. 2011 10.1093/jmcb/mjr007 34. Liu P. Wilson M.J. MiR-520c and miR373 upregulate MMP9 expression by targeting mTOR and SIRT1, and activate the Ras/Raf/MEK/Erk signaling pathway and NF-κB factor in Human fibrosarcoma cells J. Cell. Physiol. 2012 277 867 876 21898400 35. Wang L. Kang F. Shan B. Liu L. Sang M. Targeting NF-κB p65 with an artificial microRNA suppress growth of MDA-MB-231 human triple-negative breast cancer cell line Gene Ther. Mol. Biol. 2012 14 30 41 36. Huang S. Robinson J.B. Deguzman A. Bucana C.D. Fidler I.J. Blockade of nuclear factor-κB signaling inhibits angiogenesis and tumorigenecity of human ovarian cancer cells by suppressing expression of vascular endothelial growth factor and interleukin 8 Cancer Res. 2000 60 5334 5339 11034066 37. Huang Q. Gumireddy K. Schrier M. le Sage C. Nagel R. Nair S. Egan D.A. Li A. Huang G. Pure E. Agami R. The microRNAs miR-373 and miR-520c promote tumor invasion and metastasis Nat. Cell Biol. 2008 10 202 210 10.1038/ncb1681 18193036 38. Erfle H. Neumann B. Liebel U. Rogers P. Held M. Walter T. Ellenberg J. Pepperkok R. Reverse transfection on cell arreys for high content screening microscopy Nat. Protocol. 2007 2 392 399 39. Erfle H. Neumann B. Rogers P. Bulkescher J. Ellenberg J. Pepperkok R. Work flow for multiplexing siRNA assays by solid-pahse reverse trasnfection in multiwell plates J. Biomol. Screen. 2008 13 575 580 10.1177/1087057108320133 18599879 40. Genovesio A. Giardini M.A. Kwon Y.J. Dossin F.D.M. Choi S.Y. Kim N.Y. Kim H.C. Jung S.Y. Schenkman S. Almeida I.C. Emans N. Freitas-Junior L.H.F. Visual genome-wide RNAi screening to identify human hostfactors requered for Trypanosoma cruzi infection PLoS One 2011 6 e19733 10.1371/journal.pone.0019733 21625474 41. Genovesio A. Kwon Y.J. Windisch M.P. Kim N.Y. Choi S.Y. Kim H.C. Jung S. Mammano F. Perrin V. Boese A.S. Casartelli N. Swartz O. Nehrbass U. Emans N. Automated genome-wide visual profiling of cellular proteins involved in HIV infection J. Biomol. Screen. 2011 16 945 958 10.1177/1087057111415521 21841144 42. Ikeda S. Kong S.W. Lu J. Bisping E. Zhang H. Allen P.D. Golub T.R. Pieske B. Pu W.T. Altered microRNA expression in human heart disease Physiol. Genom. 2007 31 367 373 10.1152/physiolgenomics.00144.2007 43. Zhu S. Pan W. Sing X. Liu Y. Tang Y. Liang D. He D. Wang H. Liu W. Shi Y. Harley J.B. Shen N. Qian Y. The microRNA mir-23b suppress IL-17-associated autoimmune inflammation by Targeting TAB2, TAB3 and IKK-α Nat. Med. 2012 18 10.1038/nm.2815 44. Lu J. Getz G. Miska E.A. Alvarez-Saavedra E. Lamb J. Peck D. Sweet-Cordero A. Ebert B.L. Mak R.H. Ferrando A.A. Downing J.R. Jacks T. Horvitz H.R. Golub T.R. MicroRNA expression profiles classify human cancers Nature 2005 435 10.1038/nature03702 45. Tavazoie S.F. Alarcon C. Oskarsson T. Padua D. Wang Q. Bos P.D. Gerald W.L. Massague J. Endogenous human microRNAs that suppress breast cancer metastasis Nature 2008 451 147 152 18185580 46. Dai R. Zhang Y. Khan D. Heid B. Caudell D. Crasta O. Ahmed S.A. Identification of a common lupus disease-associated microRNA expression pattern in three different murine models of Lupus PLoS One 2010 5 e14302 10.1371/journal.pone.0014302 21170274

PMC005xxxxxx/PMC5003480.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2020051microarrays-02-00051ArticleGene Dosage Analysis in a Clinical Environment: Gene-Targeted Microarrays as the Platform-of-Choice Marquis-Nicholson Renate 1Prosser Debra 1Love Jennifer M. 1Love Donald R. 12*1 Diagnostic Genetics, LabPLUS, Auckland City Hospital, P.O. Box 110031, Auckland 1148, New Zealand; E-Mails: renate.mn@gmail.com (R.M.-N.); DProsser@adhb.govt.nz (D.P.); JLove@adhb.govt.nz (J.M.L.)2 School of Biological Sciences, University of Auckland, Private Bag 92019, Auckland 1142, New Zealand* Author to whom correspondence should be addressed; E-Mail: DonaldL@adhb.govt.nz; Tel.: +64-9-307-4949 (ext. 22013); Fax: +64-9-307-4939.27 3 2013 6 2013 2 2 51 62 06 2 2013 18 3 2013 20 3 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).The role of gene deletion and duplication in the aetiology of disease has become increasingly evident over the last decade. In addition to the classical deletion/duplication disorders diagnosed using molecular techniques, such as Duchenne Muscular Dystrophy and Charcot-Marie-Tooth Neuropathy Type 1A, the significance of partial or whole gene deletions in the pathogenesis of a large number single-gene disorders is becoming more apparent. A variety of dosage analysis methods are available to the diagnostic laboratory but the widespread application of many of these techniques is limited by the expense of the kits/reagents and restrictive targeting to a particular gene or portion of a gene. These limitations are particularly important in the context of a small diagnostic laboratory with modest sample throughput. We have developed a gene-targeted, custom-designed comparative genomic hybridisation (CGH) array that allows twelve clinical samples to be interrogated simultaneously for exonic deletions/duplications within any gene (or panel of genes) on the array. We report here on the use of the array in the analysis of a series of clinical samples processed by our laboratory over a twelve-month period. The array has proven itself to be robust, flexible and highly suited to the diagnostic environment. array comparative genomic hybridisation (aCGH)dosage analysistargeted microarraymolecular diagnosismaturity-onset diabetes of the young (MODY)familial phaeochromocytoma/paraganglioma syndromeCowden syndrome ==== Body 1. Introduction Large deletions and duplications have long been recognised as playing an important part in the aetiology of several disorders conventionally diagnosed using molecular techniques, such as Duchenne Muscular Dystrophy (DMD) and Charcot-Marie-Tooth Neuropathy Type 1A (CMT1A) [1,2]. In addition to these classical deletion/duplication disorders, the role of partial or whole gene deletions in the pathogenesis of a wide variety of single-gene disorders is becoming increasingly evident. A 2008 review of the entries in the online Human Gene Mutation Database showed that large deletions and duplications comprise 10% of the listed mutations [3], compared to 6% in 2003 [4]. This number will continue to rise as the increasingly widespread availability of cost-effective and robust analysis techniques enables more individuals to be subjected to dosage analysis on a routine basis. A variety of dosage analysis methods are available to the molecular diagnostic laboratory, including multiplex ligation-dependent probe amplification (MLPA) [5], quantitative real-time PCR (qPCR) [6] and customised fluorescence in situ hybridisation (FISH) [7]. Each of these methods, however, is relatively expensive and, in the case of MLPA and qPCR, is confined to a limited number of exons across a limited number of genes [8,9]. Low sample throughput in a small diagnostic laboratory prevents batching of samples if turn-around times are to be maintained, thereby further decreasing the cost-effectiveness of assays with such a limited scope. It can also be difficult to maintain staff competency across the full range of dosage assays required when sample numbers are modest. In order to address these limitations we have implemented the use of a bespoke Nimblegen 12 × 135 K CGH Array. This array targets a panel of genes chosen to complement the sequencing assays we offer in-house, as well as covering a number of genes (such as PMP22) for which partial or whole gene deletion/duplication is the predominant pathogenic mechanism. In addition to this gene-focused coverage, the array also provides low-density coverage of the entire human genome, which allows for carrier testing of genomic rearrangements that may have initially been detected by high density molecular karyotyping of a proband. We have previously reported on the validation of this custom-designed array and the cost-effectiveness of the method in a small diagnostic laboratory [10]. Here, we report on the use of the array in the routine investigation of a series of clinical samples that illustrate the suitability and flexibility of this approach for dosage analysis in a diagnostic environment. 2. Experimental Section 2.1. Patient Samples Peripheral blood EDTA samples from ninety-eight individuals were submitted over a twelve-month period to the Diagnostic Genetics department of LabPLUS, Auckland City Hospital, for molecular analysis of a range of genes (see Table 1). An archived Guthrie card, collected as part of routine newborn screening, was retrieved for one additional (deceased) patient. Analysis was requested principally for diagnostic purposes (eighty patients), with the remaining samples received for either carrier or predictive testing. Dosage analysis was performed as the primary assay for the PMP22 and DMD genes, as deletion/duplication is the predominant pathogenic mechanism in these genes [11,12,13]. Sequence analysis was performed first for the other genes, cascading to aCGH if no pathogenic mutations were found. microarrays-02-00051-t001_Table 1Table 1 Clinical samples analysed over a twelve-month period. Gene(s) of interest Number of patients Clinical indication Mode of inheritance Sample type APC 7 Familial adenomatous polyposis (FAP) Autosomal dominant Peripheral blood Dystrophin ( DMD) 7 Becker muscular dystrophy (BMD) X-linked Peripheral blood 17 Duchenne muscular dystrophy (DMD) Peripheral blood; Guthrie spot (1) 17 Carrier testing for BMD/DMD Peripheral blood Calcium-sensing receptor (CaSR) 1 Familial hypocalciuric hypercalcemia Autosomal dominant Peripheral blood E-cadherin (CDH1) 5 Familial gastric cancer Autosomal dominant Peripheral blood EPCAM 3 Familial colon cancer Autosomal dominant Peripheral blood HNF4α (MODY1), GCK (MODY2), HNF1α (MODY3), HNF1β (MODY5) 3 Maturity-onset diabetes of the young (MODY); 1 individual also with hepatic multiple adenomatosis Autosomal dominant Peripheral blood PMP22 19 Possible diagnosis of Charcot Marie Tooth Type 1A (CMT1A) Autosomal dominant Peripheral blood 7 Possible diagnosis of Hereditary Neuropathy with liability to Pressure Palsies (HNPP) Autosomal dominant Peripheral blood MSH2 2 Hereditary Non-Polyposis Colorectal Cancer (HNPCC) Autosomal dominant Peripheral blood PTEN 3 Cowden syndrome Autosomal dominant Peripheral blood RET proto-oncogene, SDHAF2, SDHB, SDHC, SDHD, TMEM127, VHL 6 Familial phaeochromocytoma/paraganglioma Autosomal dominant Peripheral blood 1 Predictive testing for familial paraganglioma VHL 1 Possible diagnosis of Von-Hippel-Lindau syndrome Autosomal dominant Peripheral blood 2.2. DNA Extraction Genomic DNA (gDNA) was extracted from peripheral blood EDTA samples using the Gentra Puregene DNA Extraction kit (Qiagen, Gaithersburg, MD, USA) and from the Guthrie card using the QIAmp DNA Miniblood Kit (Qiagen, Gaithersburg, MD, USA) as described by the manufacturer. 2.3. Dosage Analysis A Roche NimbleGen 12 × 135 K custom CGH Array was used for dosage analysis. This bespoke CGH array was designed to interrogate the coding regions of sixty-six genes of interest to our laboratory. Exonic probes overlapped by 25 bp in order to provide high-resolution detection of deletions or duplications within the coding regions of the genes of interest. Intronic probes were spaced on average every 175 bp. In addition to the targeted probes, approximately 75,000 “backbone” probes, with a mean probe interval of 45 kbp, were also included, providing low-density whole genome coverage. Two hundred and fifty nanograms of gDNA were processed according to the manufacturer’s instructions; NimbleGen Array User’s Guide: CGH and CNV Arrays v6.0 [14]. In brief, extracted gDNA from samples and Promega controls was denatured in the presence of a Cy3-(test) or Cy5-(control) labelled random primers and incubated with the Klenow fragment of DNA polymerase, together with dNTPs (5 mM of each dNTP), at 37 °C for 2 h. The reaction was terminated by the addition of 0.5 M EDTA (21.5 µL), prior to isopropanol precipitation and ethanol washing. Following DNA quantitation, the test and sex-matched control samples were combined in equimolar amounts and applied to one of the twelve arrays on the microarray slide. Hybridisation was carried out in a Roche NimbleGen Hybridisation Chamber (Madison, WI, USA) for a period of 48 h. Slides were washed and scanned using a NimbleGen MS200 Microarray Scanner (Madison, WI, USA Array image files (.tif) produced by the MS200 Data Collection Software were imported into DEVA v1.2.1 (Roche NimbleGen Inc., Madison, WI, USA) for analysis. Data was filtered using a log2ratio threshold of less than −0.4 over 6 probes for a deletion and greater than 0.4 over 15 probes for a duplication. All copy number changes meeting these thresholds were exported out of DEVA into a Microsoft Excel spreadsheet for further investigation. Each genomic region exhibiting a copy number change within one of the genes of interest was examined using the UCSC genome browser [15] to determine the location and significance of the change. Analysis of copy number changes was only performed for the gene(s) of interest; changes identified within other genes for which analysis had not been requested were not subjected to detailed examination. 3. Results and Discussion Dosage changes were detected in twenty-six of the eighty patients referred in for diagnostic testing, nine of the seventeen referred in for carrier testing, and in the one patient referred in for predictive testing (see Table 2). These changes are separated into disease/gene and are described in detail below. microarrays-02-00051-t002_Table 2Table 2 Mutations detected by array comparative genomic hybridisation (aCGH) analysis of clinical samples. Patient Gene(s) analysed Genotype Significance of result 1,2 DMD Hemizygous deletion of exons 45–47 (inclusive) In-frame deletion; consistent with BMD phenotype 3 Hemizygous deletion of exons 45–48 (inclusive) In-frame deletion; consistent with BMD phenotype 4 c.5199_5209del (p.Thr1734SerfsX10) Premature truncation of protein; consistent with DMD phenotype 5 Hemizygous deletion of exons 46–50 (inclusive) Out-of-frame deletion; consistent with DMD phenotype 6 Hemizygous duplication of exon 12 Out-of-frame duplication; consistent with DMD phenotype 7 Hemizygous duplication of exons 10–11 (inclusive) Out-of-frame duplication; consistent with DMD phenotype 8 Hemizygous deletion of exons 53–59 (inclusive) Out-of-frame deletion; consistent with DMD phenotype 9,10 Hemizygous duplication of exons 8–9 (inclusive) Out-of-frame duplication; consistent with DMD phenotype 11–19 Various (heterozygous deletion/duplication) Carrier of familial deletion/duplication 20 HNF1α Heterozygous deletion of exons 2–3 (inclusive) Consistent with clinical phenotype—adenomatosis and MODY3 21–29 PMP22 ~1.5 Mb heterozygous duplication encompassing PMP22 gene Consistent with CMT1A phenotype 30,31 Reciprocal deletion Consistent with HNPP phenotype 32,33 PTEN Heterozygous deletion of exon 2 Consistent with Cowden syndrome phenotype 34,35 SDHB Heterozygous deletion of exon 1 Consistent with clinical diagnosis of familial phaeo syndrome 36 Presence of familial deletion— appropriate surveillance/operative management required 3.1. PMP22 Gene Analysis—Charcot-Marie-Tooth Neuropathy Type 1A (CMT1A) and Hereditary Neuropathy with Liability to Pressure Palsies (HNPP) Nineteen patients were referred for CMT1A gene analysis and seven for HNPP. Of these, two were found to carry the classic 1.5 Mb deletion (HNPP) and nine carried the reciprocal duplication (CMT1A) at 17p11.2 (includes the PMP22 gene; see Figure 1) that is responsible for 80% of each of these disorders [11,12]. Figure 1 (a) DEVA software output showing copy number change (duplication; log2ratio: 0.4953) for probes localized to chr17: 14160052-15824662 (hg18 co-ordinates), encompassing the PMP22 gene; (b) UCSC genome browser graphic output of chr17: 14160052-15824662 (hg18 co-ordinates). 3.2. DMD Gene Analysis—Duchenne and Becker Muscular Dystrophy (DMD/BMD) More than 5,000 mutations have been identified in individuals with BMD or DMD [13,16]. These mutations are highly variable and run the full spectrum from deletion of the entire gene, to deletion or duplication of one or more exons, to small deletions or insertions, to single-base pair alterations. Deletions and duplications account for 60–70% of cases of DMD and 5–10% of cases of BMD [17]. For this reason, deletion/duplication analysis is the first-line diagnostic test for DMD/BMD, with sequence analysis performed if no dosage changes are found. As a general rule, mutations that alter the reading frame correlate with DMD, whereas those that preserve the reading frame are associated with BMD [16,18]. Twenty-four males with a clinical diagnosis of dystrophinopathy were referred for routine diagnostic testing. Array CGH analysis revealed a hemizygous deletion or duplication within the DMD gene in ten of these patients. Further assessment of each of these mutations was performed using a Reading-frame Checker [19]. In each case the predicted effect was consistent with the phenotype that was observed clinically. An intra-exonic deletion of six probes, the lower limit of size threshold for analysis, was identified within exon 37 in Patient 4. Sequence analysis of exon 37 confirmed a hemizygous deletion of 11 base pairs within the exon, c.5199_5209del (p.Thr1734SerfsX10). This frameshift mutation results in premature termination of translation and truncation of the protein and is therefore consistent with the clinical diagnosis of DMD. Molecular testing for BMD/DMD is not only useful to confirm the clinical diagnosis in affected males who are suspected to have a dystrophinopathy based on clinical signs and an elevated serum creatine kinase (CK) level, but identification of the causative mutation also informs genetic counselling for the family and allows carrier and prenatal testing to be performed as appropriate [20]. The familial mutation was identified in nine of the seventeen patients referred for DMD carrier testing during this twelve-month period. The absence of the familial mutation within female relatives within the extended family is reassuring, but lack of the familial mutation in the mother of an affected boy does not mean that the mutation is necessarily de novo as germline mosaicism remains a possibility. 3.3. PTEN Gene Analysis Three patients were referred for PTEN gene dosage analysis following a negative result on sequencing. Each of these patients had a probable clinical diagnosis of Cowden Syndrome, a multiple hamartoma syndrome that confers a high risk of benign and malignant tumours of the thyroid, breast, and endometrium [21]. Two of these three patients were found to carry a deletion encompassing exon 2 of the PTEN gene. Exonic or whole gene deletions are believed to be responsible for up to 10% of cases of Cowden syndrome [21,22]. The deletion of exon 2 is an out-of-frame deletion that alters the translational reading frame and results in premature truncation of the PTEN protein. It is extremely likely, therefore, to be the causative mutation in these cases. 3.4. Familial Paraganglioma/Phaeochromocytoma Syndrome Mutation Screening—SDHAF2, SHDB, SDHC, SDHD, VHL, RET Proto-Oncogene, and TMEM127 Gene Analysis The full familial paraganglioma/phaeochromocytoma gene panel (genes listed above) was analysed in six patients using both sequencing and aCGH. No pathogenic mutations were detected on sequence analysis in any of the genes for any of these patients. Array CGH revealed a deletion of exon 1 of the SDHB gene in two individuals. This deletion was later detected in the unaffected son of one of these patients. Heterozygous deletion of exon 1 of the SDHB gene has been reported in several unrelated families with hereditary phaeochromocytoma [23,24]. It has been proposed that the relatively high frequency of this deletion (three of the five instances of gross deletion listed in the online Human Gene Mutation Database) is due to a high density of Alu repeats within intron 1 of the SDHB gene [24]. 3.5. Maturity-Onset Diabetes of the Young (MODY) Mutation Screening—HNF4α, GCK, HNF1α, HNF1β Two patients were referred for sequence and deletion/duplication analysis of the full MODY gene panel offered at our laboratory (genes listed above). No pathogenic mutations were detected on either assay in these patients. Patient 20, however, was referred for HNF1α gene analysis only. He was a 38 years old man with a history of multiple hepatic adenomas, requiring surgical resection, and a diabetic profile suggestive of MODY type 3. Biallelic inactivation of HNF1α has been reported to be an important event in the occurrence of liver adenoma [25]; partial or whole gene deletions are responsible for approximately 3% of cases of MODY type 3 [26]. Histological investigation of Patient 20’s resected hepatic tissue showed not only the three large lesions that had previously been noted on imaging, but also several hundred micro-adenomas. No pathogenic mutations were detected on sequence analysis of the HNF1α gene, but aCGH revealed an heterozygous deletion of exons 2–3 (inclusive; see Figure 2). This deletion removes the main part of the B domain and a portion of the homeodomain of the HNF1α protein, resulting in destabilization [27]. Mutation analysis of the affected hepatic tissue was not performed, but it is expected that somatic inactivation of the second HNF1α gene allele would be evident. Figure 2 (a) DEVA software output showing copy number change (deletion; log2ratio: −0.5459) for probes localized to chr12:119906606-119915923 (hg18 co-ordinates), encompassing exons 2 and 3 of the HNF1α gene; (b) UCSC genome browser graphic output of chr12:119906606-119915923 (hg18 co-ordinates). 4. Conclusions Through the use of a gene-targeted CGH array for dosage analysis within the diagnostic environment, we have been able to confidently detect a spectrum of changes that would be invisible to sequence analysis: single exon, multiple exon and whole gene deletions/duplications. In addition, as a result of the high-density overlapping probes that tile the exons in our custom-designed array, we have found that large intra-exonic changes can also be detected (Patient 4 described above). The aCGH technique is robust and cost-effective, overcoming the problems associated with the use of expensive kits in the context of low sample throughput, and allowing for consolidation of previously separate gene-targeted dosage assays to a single validated technique. The cost-effectiveness is principally due to this ability to batch all samples received for deletion/duplication analysis, and to the fact that a separate assay does not need to be worked up for each gene, allowing analysis of a larger number of genes to be offered in-house and bringing more revenue into the laboratory. Furthermore, the aCGH process eliminates the risk of false positives that can occur as a result of polymorphisms under primer binding sites [20]. This risk is inherent in all PCR-based techniques, including the other dosage method most widely used by diagnostic laboratories, MLPA. To eliminate the occurrence of false positive results due to a one-off failure of hybridisation to a particular probe, each gene-focused probe on our custom-designed array is spotted in duplicate. In contrast to MLPA, aCGH allows the interrogation of intronic as well as exonic regions, allowing breakpoints to be mapped more accurately [20]. It can also be used to characterise some inversions and complex rearrangements, thereby offering a higher mutation detection rate than MLPA and other purely exon-focused dosage assays [28,29]. The disadvantages of the aCGH array approach described here are that it does not interrogate small-scale changes in deep intronic regions, nor rare and more complex rearrangements. These mutation events could, however, be detected by RNA analysis or whole genome sequencing. In the meantime, the combination of coding region sequence analysis and aCGH should detect the vast majority of pathogenic mutations known to be responsible for single gene disorders, thereby fulfilling the diagnostic needs of the clinical community. During the latter stages of our study, we were informed that Nimblegen had ceased production of arrays. As a consequence, readers are directed to an alternative company, Agilent Technologies, which offers custom microarray designs that might serve as a suitable substitute. Acknowledgments We acknowledge the contributions of Anthony Thrush and Elise Bal of Roche Diagnostics New Zealand Limited, and Ross Hewett of LabPLUS, in supporting the diagnostic initiative reported here. ==== Refs References 1. Kunkel L.M. Hejtmancik J.F. Caskey C.T. Speer A. Monaco A.P. Middlesworth W. Colletti C.A. Bertelson C. Muller U. Bresnan M. Analysis of deletions in DNA from patients with Becker and Duchenne muscular dystrophy Nature 1986 322 73 77 3014348 2. Roa B.B. Garcia C.A. Lupski J.R. Charcot-Marie-Tooth disease type 1A: Molecular mechanisms of gene dosage and point mutation underlying a common inherited peripheral neuropathy Int. J. Neurol. 1991–1992 25–26 97 107 3. Stenson P.D. Mort M. Ball E.V. Howells K. Phillips A.D. Thomas N.S. Cooper D.N. The human gene mutation database: 2008 update Genome Med. 2009 1 10.1186/gm13 4. Stenson P.D. Ball E.V. Mort M. Phillips A.D. Shiel J.A. Thomas N.S. Abeysinghe S. Krawczak M. Cooper D.N. Human gene mutation database (hgmd): 2003 update Hum. Mutat. 2003 21 577 581 10.1002/humu.10212 12754702 5. Eijk-Van Os P.G. Schouten J.P. Multiplex ligation-dependent probe amplification (MLPA) for the detection of copy number variation in genomic sequences Meth. Mol. Biol. 2011 688 97 126 10.1007/978-1-60761-947-5_8 6. Sieber O.M. Lamlum H. Crabtree M.D. Rowan A.J. Barclay E. Lipton L. Hodgson S. Thomas H.J. Neale K. Phillips R.K. Whole-gene APC deletions cause classical familial adenomatous polyposis, but not attenuated polyposis or “multiple” colorectal adenomas Proc. Natl. Acad. Sci. USA 2002 99 2954 2958 10.1073/pnas.042699199 11867715 7. Bendavid C. Kleta R. Long R. FISH diagnosis of the common 57-kb deletion in CTNS causing cystinosis Hum. Genet. 2004 115 510 514 10.1007/s00439-004-1170-2 15365816 8. Gouas L. Goumy C. Veronese L. Tchirkov A. Vago P. Gene dosage methods as diagnostic tools for the identification of chromosome abnormalities Pathol. Biol. 2008 56 345 353 10.1016/j.patbio.2008.03.010 18513889 9. Armour J.A.L. Barton D.E. Cockbuen D.J. Taylor G.R. The detection of large deletions or duplications in genomic DNA Hum. Mutat. 2002 20 325 337 10.1002/humu.10133 12402329 10. Marquis-Nicholson R. Doherty E. Thrush A. Love J.M. Lan C.-C. George A.M. Love D.R. Array-based identification of copy number changes: Simultaneous gene-focused and low resolution whole human genome analysis Sultan Qaboos Univ. Med. J. 2013 13 69 79 23573385 11. Bird T.D. Charcot-Marie-Tooth Neuropathy Type 1 Pagon R.A. Bird T.D. Dolan C.R. GeneReviews™ Seattle, WA, USA 1993 Available online:http://www.ncbi.nlm.nih.gov/books/ NBK1205/ (accessed on 1 February 2013) 12. Bird T.D. Hereditary Neuropathy with Liability to pressure Palsies Pagon R.A. Bird T.D. Dolan C.R. GeneReviews™ Seattle, WA, USA 1993 Available online:http://www.ncbi.nlm.nih.gov/books/NBK1392/ (accessed on 1 February 2013) 13. Flanigan K.M. Dunn D.M. von Niederhausern A. Soltanzadeh P. Gappmaier E. Howard M.T. Sampson J.B. Mendell J.R. Wall C. King W.M. Mutational spectrum of DMD mutations in dystrophinopathy patients: Application of modern diagnostic techniques to a large cohort Hum. Mutat. 2009 30 1657 1666 10.1002/humu.21114 19937601 14. Roche NimbleGen Available online:http://www.nimblegen.com (accessed on 25 March 2013) 15. UCSC Genome Browser Available online:http://genome.ucsc.edu (accessed on 25 March 2013) 16. Aartsma-Rus A. van Deutekom J.C. Fokkema I.F. van Ommen G.J. den Dunnen J.T. Duchenne muscular dystrophy mutation database: An overview of mutation types and paradoxical cases confirm the reading-frame rule Muscle Nerve 2006 34 135 144 10.1002/mus.20586 16770791 17. Beggs A.H. Koenig M. Boyce F.M. Kunkel L.M. Detection of 98% of DMD/BMD gene deletions by polymerase chain reaction Hum. Genetic. 1990 86 45 48 18. Monaco A.P. Bertelson C.J. Liechti-Gallati S. Moser H. Kunkel L.M. An explanation for the phenotypic differences between patients bearing partial deletions of the DMD locus Genomics 1988 2 90 95 3384440 19. DMD Exonic Deletions/Duplications Reading-Frame Checker 1.9 Available online:http://www.humgen.nl/scripts/DMD_frame.php (accessed on 25 March 2013) 20. Abbs S. Tuffery-Giraud S. Bakker E. Ferlini A. Sejersen T. Mueller C.R. Best practice guidelines on molecular diagnostics in Duchenne/Becker muscular dystrophies Neuromuscul. Disord. 2010 20 422 427 10.1016/j.nmd.2010.04.005 20466545 21. Eng C. PTEN Hamartoma Tumor Syndrome (PHTS) Pagon R.A. Bird T.D. Dolan C.R. GeneReviews™ Seattle, WA, USA 1993 Available online:http://www.ncbi.nlm.nih.gov/ books/NBK1488/ (accessed on 1 February 2013) 22. Chibon F. Primois C. Bressieux J.M. Lacombe D. Lok C. Mauriac L. Taieb A. Longy M. Contribution of PTEN large rearrangements in Cowden disease: A multiplex amplifiable probe hybridisation (MAPH) screening approach J. Med. Genet. 2008 45 657 665 10.1136/jmg.2008.058131 18456716 23. McWhinney S.R. Pilarski R.T. Forrester S.R. Schneider M.C. Sarquis M.M. Dias E.P. Eng C. Large germline deletions of mitochondrial complex II subunits SDHB and SDHD in hereditary paraganglioma J. Clin. Endocrinol. Metab. 2004 89 5694 5699 10.1210/jc.2004-0769 15531530 24. Cascon A. Landa I. Lopez-Jimenez E. Dieaz-Hernandez A. Buchta M. Montero-Conde C. Leskela S. Leandro-Garcia L.J. Leton R. Rodriquez-Antona C. Molecular characterisation of a common SDHB deletion in paraganglioma patients J. Med. Genet. 2008 45 233 238 18057081 25. Bluteau O. Jeannot E. Bioulac-Sage P. Marques J.M. Blanc J.F. Bui H. Beaudoin J.C. Franco D. Balabaud C. Laurent-Puig P. Zucman-Rossi J. Bi-allelic inactivation of TCF1 in hepatic adenomas Nat. Genet. 2002 32 312 315 10.1038/ng1001 12355088 26. Ellard S. Thomas K. Edghill E.L. Owens M. Ambye L. Cropper J. Little J. Strachan M. Stride A. Ersoy B. Partial and whole gene deletion mutations of the GCK and HNF1A genes in maturity-onset diabetes of the young Diabetologia 2007 50 2313 2317 10.1007/s00125-007-0798-6 17828387 27. Bach I. Pontoglio M. Yaniv M. Structure of the gene encoding hepatocytes nuclear factor 1 (HNF1) NAR 1992 20 4199 4204 10.1093/nar/20.16.4199 1354855 28. Del Gaudio D. Yang Y. Boggs B.A. Schmitt E.S. Lee J.A. Sahoo T. Pham H.T. Wiszniewska J. Chinault A.C. Beaudet A.L. Eng C.M. Molecular diagnosis of Duchenne/Becker muscular dystrophy: Enhanced detection of dystrophin gene rearrangements by oligonucleotide array-comparative genomic hybridization Hum. Mutat. 2008 29 1100 1107 10.1002/humu.20841 18752307 29. Bovolenta M. Neri M. Fini S. Fabris M. Trabanelli C. Venturoli A. Martoni E. Bassi E. Spitali P. Brioschi S. A novel custom high density-comparative genomic hybridization array detects common rearrangements as well as deep intronic mutations in dystrophinopathies BMC Genomics 2008 9 10.1186/1471-2164-9-572

PMC005xxxxxx/PMC5003481.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2020081microarrays-02-00081ReviewComparative Analyses of MicroRNA Microarrays during Cardiogenesis: Functional Perspectives Bonet Fernando 1Hernandez-Torres Francisco 1Esteban Franciso J. 2Aranega Amelia 1Franco Diego 1*1 Cardiovascular Research Group, Department of Experimental Biology, University of Jaén, Jaén 23071, Spain; E-Mails: fbonetmartinez@gmail.com (F.B.); fraheto@ujaen.es (F.H.-T.); aaranega@ujaen.es (A.A.)2 System Biology Group, Department of Experimental Biology, University of Jaén, Jaén 23071, Spain; E-Mail: festeban@ujaen.es* Author to whom correspondence should be addressed; E-Mail: dfranco@ujaen.es; Tel.: +34-953-212-763; Fax: +34-953-211-875.03 4 2013 6 2013 2 2 81 96 16 2 2013 14 3 2013 21 3 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).Cardiovascular development is a complex process in which several transcriptional pathways are operative, providing instructions to the developing cardiomyocytes, while coping with contraction and morphogenetic movements to shape the mature heart. The discovery of microRNAs has added a new layer of complexity to the molecular mechanisms governing the formation of the heart. Discrete genetic ablation of the microRNAs processing enzymes, such as Dicer and Drosha, has highlighted the functional roles of microRNAs during heart development. Importantly, selective deletion of a single microRNA, miR-1-2, results in an embryonic lethal phenotype in which both morphogenetic, as well as impaired conduction, phenotypes can be observed. In an effort to grasp the variability of microRNA expression during cardiac morphogenesis, we recently reported the dynamic expression profile during ventricular development, highlighting the importance of miR-27 on the regulation of a key cardiac transcription factor, Mef2c. In this review, we compare the microRNA expression profile in distinct models of cardiogenesis, such as ventricular chamber development, induced pluripotent stem cell (iPS)-derived cardiomyocytes and the aging heart. Importantly, out of 486 microRNAs assessed in the developing heart, 11% (55) displayed increased expression, many of which are also differentially expressed in distinct cardiogenetic experimental models, including iPS-derived cardiomyocytes. A review on the functional analyses of these differentially expressed microRNAs will be provided in the context of cardiac development, highlighting the resolution and power of microarrays analyses on the quest to decipher the most relevant microRNAs in the developing, aging and diseased heart. microRNAsmicroarrayscardiac developmentmeta-analyses ==== Body 1. Introduction Cardiac development is a complex process in which multiple cell types are integrated. Two distinct populations of cells contribute to the developing heart, a first heart field (FHF), which mainly leads to the left ventricle, and a second heart field (SHF), which contributes to the rest of the heart [1]. Commitment from the mesodermal lineage into the cardiomyocyte lineage seems to be equally achieved for the FHF and SHF precursors. Both FHF and SHF precursors are mainly dependent on Bmp and Fgf signaling, resulting on the activation of the core cardiac transcriptional machinery, i.e., Mef2, Gata, Nkx2.5 and Srf [2]. However, FHF cells are characterized by constitutive expression of Nkx2.5 with only a transient islet-1 expression, whereas SHF cells display strong and constitutive expression of Nkx2.5 and islet-1, providing the first insights into cardiac transcriptional heterogeneity [2]. As development of heart proceeds, distinct cardiomyocyte cell types are established, with distinct cellular, molecular and functional characteristics [3]. Transcriptional regulators, such as T-box, Hands and Pitx2 genes, participate on these processes [4]. In addition to cardiomyocyte formation, the heart develops two additional tissue layers, an inner endothelial lining (endocardium) and external epicardial layer (epicardium). Contribution to the formed heart from these two layers is complex; at discrete regions, the endocardium leads to an epithelial-to-mesenchymal transformation (EMT) configuring the endocardial cushions [5,6], whereas the epicardium displays also a similar EMT transformation, intermingling across the cardiomyocytes and contributing to the formation of the interstitial cells and the coronary vasculature [7,8]. Understanding of the transcriptional control of EMT in both the endocardium and epicardium is emerging, with key roles for Ets, Wt-1 and Snail/Slug transcription factors. In addition to the role of transcriptional regulation, a new layer of complexity is emerging, as non-coding RNAs are also capable of playing a pivotal role during cardiogenesis [9]. microRNAs are non-coding RNA with an average length of 22–24 nucleotides, which are capable of interacting with the 3' untranslated region of coding RNAs (mRNAs), leading to blockage of protein translation and/or mRNA degradation [10]. Understanding microRNA biogenesis has been achieved at a quick pace; however, knowledge about the tissue distribution and functional consequences remains more elusive. However, there is an increasing body of evidence that suggests a highly relevant role of microRNAs in multiple aspects of cardiac development and disease. 2. Experimental Section MicroRNA expression profiles were downloaded from Gene Expression Omnibus [11] GSE32935 (aging heart; [12] Exiqon Mouse microRNA v11.0) and GSE35672 (iPS-cardiomyocytes; [13], Illumina Human v2 MicroRNA expression beadchip datasets). microRNA microarray data from ventricular maturation (MirVana v2.0, miRBase version 8.0) were already available to us, as described by Chinchilla et al. [14]. To minimize technical (non-biological) variability among arrays, each data group was independently log2 transformed and then normalized using the quantiles normalization function implemented in the Bioconductor limma package [15], with default parameters run in R software [16], and finally, each probe was Z-scored (). MicroRNA expression data in ventricular development [14], also Z-scored, was identically processed as describe above, but with an initial k-nearest neighbor(KNN) imputation of the densitometry values <1 using the KNN algorithm implemented in the Bioconductor impute package [15] with default parameters. Hierarchical clustering (Euclidean distance and complete linkage), an unsupervised way of grouping samples based only on their gene expression similarities, was carried out using TM4 software suite [17]. 3. MicroRNAs in Cardiovascular Development Cardiac development is a complex process in which several cell types are involved. Morphogenetic and transcriptional regulation of gene expression during cardiogenesis has been extensively investigated over the last few decades, providing a good framework to understand the formation of the heart [4]. A novel layer of complexity has been gained by the discovery of microRNAs and their pivotal role in cardiogenesis. Several studies have provided evidence of differential expression of microRNAs during heart development, both during embryogenesis [14], as well as postnatal stages [18], supporting a pivotal roles of microRNA during heart formation. Functional evidence of the role of microRNAs in the developing heart was demonstrated by selective inhibition of Dicer in a tissue-restricted manner. Conditional ablation of Dicer using Nkx2.5Cre driver mice resulted in embryonic lethality, displaying pericardial edema and cardiac hypoplasia [19]. Inhibition of Dicer function using αMHC-Cre mice also resulted in cardiac developmental impairment, and newborns die from heart failure soon after birth [20]. Whereas these studies highlight the importance of microRNA biogenesis for heart development, the role of individual microRNAs in cardiac formation remains largely unknown. However, two examples highlight the importance of microRNAs during cardiogenesis, such as germ-line deletion of miR-1-2, which resulted in ventricular septal defects and early embryonic lethality in a subset of mice [19], and germline deletion of miR-126, which resulted in embryonic lethality due to vascular leakage [21]. Curiously, targeted deletion of other abundantly expressed microRNAs in the developing heart, such as miR-133 or miR-208, have resulted in completely viable mice, suggesting a modulatory rather than a fully determinant role of these microRNAs during cardiogenesis [22,23]. In an effort to grasp the variability of microRNA expression during cardiac morphogenesis, we recently reported the dynamic expression profile during ventricular development [14]. We demonstrate that most of the differentially expressed microRNAs during ventricular maturation (55/486; 11%) displayed increased expression levels, opening a new avenue to explore the role of these microRNAs during cardiogenesis. The rapid pace of deciphering the functional roles of microRNAs has provided a cardiovascular role for 24 of these microRNAs (24/55; ~45%) (Table 1), including multiple validated targets, as nicely compiled in TarBase database [24]. In addition, a large number of microRNA microarrays have been generated in distinct cardiovascular settings, such as human cardiogenesis, embryonic stem cell- and induced pluripotent stem cell (iPS)-derived cardiomyogenesis, as well as the aging and diseased heart, enabling to trace the putative role of these differentially expressed microRNAs in other cardiovascular contexts. Thus, within this review, we will update on the functional role of these previously described differentially expressed microRNAs, and we will explore the putative functional contribution in other cardiovascular settings. microarrays-02-00081-t001_Table 1Table 1 List of differentially expressed microRNAs during ventricular chamber development [14]. Several of them display a reported cardiovascular role (highlighted in the first column), and several validated targets have been reported, among which those with a cardiovascular role are highlighted. * MicroRNAs reported in humans (hsa-tagged), but not in mice. ** microRNAs reported in mice (mmu-tagged), but not in humans. Note that more that 85% (48/55) of the differentially expressed microRNAs are conserved in both species. iPS, pluripotent stem cell; na, not applicable. Differentially expressed microRNAs Cardio-vascular role Validated targets (cardiovascular role) Cardiac formation iPS cardio-myogenesis Cardiac aging References let-7a nras, kras, hmga2 up up down [24] let-7b up up na let-7c up up na let-7d up up na let-7i up up na miR-15b miR-15 dmtf1, c22orf5, bcl-2, Chek1 up down na [18,24] miR-17 miR-17 rbl2-p130, ncoa3, e2f1, adkcnna-p21, fog2 up na na [24,25,26] miR-23b miR-23 pou4f2, hes1, has2 up up na [24,27] miR-24 miR-24 notch1,mapk14, kiaa0152, dhfr, cdkn2a-p16, alk4, gata2, pak4, bcl2 up up na [24,28,29] miR-25 miR-25 na up down na miR-26a up up na miR-30a miR-30 ctgf, znf294, wnt5a, uap1, tnfrsf10b, tnfaip2, tmem87a, tmem59, tmem41b, tmed7, tmed3, tmed2, tmed10, tmco1, tloc1, ticam2, them4, sypl1, stx7, strn, slc9a3r2, slc9a3r2, slc7a11, slc7a1, slc4a7, slo4a10, slc38a2, slc38a1, slc12a4, sec23a, rpcd1, rbms1, rad23b, rab27b, ptrh1, ptprk, ptgfrn, prpf40a, ppp3r1, ppp3ca, ppp2r4, pgm1, ptprk, ptgfrn, prpf40a, ppp3r1, ppp3ca, ppp2r4, pgm1, pex11b, pafah1b2, pfha2, nufip2, nucb1, nt5e, nt5c3, npr3, np, ncl, napg, myo10, mpdu1, mllt11, mllt1, met, mbnl1, mat2a, lrrc8c, lmnb2, krthb5, kdelc2, jun, itga2, ifrd1, idh1, hnrpm, gpd2, gnai2, gilnt7, galnt1, fxr2, frg1, f2, elmod2, dock7, cpne8, chd1 up na na [30,31,32] miR-30d up up na miR-93 miR-93 na up down na miR-99a up up up miR-99b up up na miR-103 up down na miR-106a up down down miR-122a miR-122 trpr6, ndrg3, cd320, bckdk, aldoa, cck-8, caspase-3 up na na [24,33] miR-125a miR-125 Lin28, erbb2, erbb3, zfp385, tor2a, rhebl1, ppt2, mkk7, lin28, jub, entpd4, dus11, ddx19b, arid3b, arid3a, apln, abtb1 up na na [24] miR-125b up up na miR-126 miR-126 vam1 up down na [24] miR-130a up down na miR-130b up down na miR-133a miR-133 srf, ptp2, kcne1, hcn2, hcn4, erg, casp9, nfatc4 up up na [24,34,35,36] miR-133b up up na miR-140 up na na miR-143 miR-143 mapk7, mapk12, aduccine3 up up na [24,37] miR-145 miR-145 irs-1, flj21308, dab2 up up na [24,38] miR-181a up up na miR-181b up up na miR-183 up up na miR-190 up up na miR-191 up up na miR-198 * up up na miR-202 up up na miR-210 miR-210 efna3 up up na [24] miR-298 up na equal miR-320 up na na miR-322 ** miR-322 na up na up miR-324 up down na miR-351 ** up na equal miR-373 * up down na miR-422b up na na miR-453 up up na miR-455 up na na miR-494 miR-494 na up up down miR-494 up na na miR-503 up na down miR-513 ** up na na miR-517 ** up na na miR-518c up na na miR-546 * up na equal 4. A Meta-Analysis of MicroRNA Microarrays in Cardiogenesis and Cardiac Aging By using microRNA microarrays, we have recently reported both the most representative differentially expressed microRNAs during ventricular development, as well as the most abundantly expressed [14]. Most recently, using the complementary approach of deep-sequencing, Cao et al. [39] have provided similar findings in the developing heart. Interestingly, ten microRNAs were demonstrated to be abundantly expressed; miR-23b, miR-24, miR-23a, miR-375, miR-29a, miR-93, miR-21, miR-25, let-7b and miR-27b, in line with our previous findings. A meta-analyses approach of differentially expressed microRNAs during cardiogenesis comparing the developing mouse heart [14], the induced pluripotent stem cell-cardiomyogenesis [13] and the aging heart [12] reveals interesting patterns, as described below. From 55 microRNAs differentially expressed in the developing heart, 44 were found in the comparison to iPS-cardiogenesis and/or the aging heart. Since the aging heart experiments were performed in mice, only nine mmu-microRNA identities were similar, while in the human iPS-cardiomyogenesis, 35 equal hsa-microRNA identities were found. Comparison of the differentially expressed microRNAs in the developing heart [14] and aging heart [12] shows that those microRNAs with a progressive increase over time during heart development display arbitrary changes in the adult and aged heart (Figure 1), suggesting that these differentially expressed microRNAs during cardiogenesis are not overtly modified with aging. In line with these findings, only two (miR-494 and let-7a) among those nine microRNAs have been involved in cardiovascular biology, suggesting a rather subtle role in the adult cardiovascular system. In contrast, comparison of the differentially expressed microRNAs in the developing heart [14] and induced pluripotent stem cells-derived cardiomyocytes [13] revealed 35 shared microRNAs, among which 17 (17/35; 48%) display similar expression profiles during heart development and iPS-cardiogenesis, while seven (7/35; 20%) display an opposite pattern. The remaining 11 microRNAs display arbitrary trends (11/35; 31%). Importantly, 12 out of 17 (70%) with parallel expression profiles have been already involved in cardiovascular biology. Overall, these data nicely illustrate that on the one hand, similar microRNA profiles are observed during cardiac ventricular development and iPS-derived cardiomyogenesis, highlighting the parallelism between the in vivo and the in vitro system. On the other hand, the fact that the majority of the identified microRNAs in cardiac ventricular and iPS-derived cardiomyogenesis display increasing expression levels with maturation and the fact that a role in cardiovascular development has been established for a large number of them highlights the relevance of microRNA microarray comparison and reinforces the notion of pivotal role for these microRNAs during cardiac development. Furthermore, the extrapolation of these approaches to the diseased heart will further reinforce the significant mean of microarray comparison and will further identify key microRNAs with pivotal roles in cardiovascular development and disease. Figure 1 A meta-analyses of microRNA microarrays during in the developing and aging cardiogenesis. Panel A illustrates the heatmap of the differentially expressed microRNAs during mouse ventricular development, as described by Chinchilla et al. [14]. Panel B illustrates the heatmap of the comparative analyses of the differentially expressed microRNAs in the developing and aging mouse heart. Observe that while those differentially expressed during cardiogenesis display increasing expression trends, the young and adult heart display no significant trends. In contrast, comparative analyses of the differentially expressed microRNAs in the developing heart and induced pluripotent stem cell-derived cardiomyocytes, over a maturation period ranging incipient cardiomyocytes (day zero) to 120 days after differentiation, nicely show subset of microRNAs with similar trends (i.e., let7 cluster) or opposite trends (see, for example, miR-103 and miR-106b), as illustrated in this heatmap on panel C. Panels A'. B', C' and C'' are graphical illustrations of microRNA representative expression trends as depicted in heatmaps on panels A, B and C, respectively. Panel A' displays only a small representation of the full list of microRNAS represented in heatmap of panel A. More detailed information can be found in Chinchilla et al. [14]. Panel C' represents a subset of panel C microRNAs displaying increased expression levels in both ventricular maturation and iPS-cardiomyogenesis, while panel C'' represents a microRNA subset display increasing expression levels during ventricular maturation, but decreasing expression levels in iPS-cardiomyogenesis. 5. The Cardiovascular Role of Differentially Expressed MicroRNAs during Ventricular Development Over the last few years, we have witnessed an increasing interest on the contribution of microRNAs to cardiovascular development. Microarray analyses and deep-sequencing studies have increased our knowledge among the microRNA signatures of distinct cardiovascular tissues and conditions. For example, a large number of microRNAs that were initially reported to be upregulated within the developing heart by microarrays [14] have also been reported to be highly expressed in the developing heart by deep sequencing approaches [39]. Vacchi-Suzzi et al. [30] have further provided detailed information about the microRNA hallmarks of different cardiac structures, revealing an abundant expression of miR-125, miR-99 and miR-320 in valve tissues and of miR-133 and miR-30 in the myocardium, among those differentially expressed during mouse ventricular maturation [14]. However, the tissue distribution and the functional role of several other microRNAs, such as miR-320, miR-99, miR-125, miR-93 and miR-322 in the developing and/or diseased heart, remains to be elucidated, apart from being highly and dynamically expressed during cardiogenesis [14,30,39], while the function role of others, such as, for example, miR-23, miR-24, miR-83, miR-25, miR-27 and several members of the let-7 family members, is currently emerging, as stated below. As we have previously mentioned, the heart is a complex structure in which distinct tissue layers and structures can be delineated. Three distinct layers can be delimited, the endocardium, myocardium and epicardium, the myocardium being the most important functionally. In addition, the heart is composed of interstitial tissue, arterial and atrioventricular valves and its own system of blood perfusion, the coronary vasculature. Understanding of the impact of microRNA regulation in these tissues is progressively emerging. To date, the contribution of microRNAs to myocardium biology is emerging at a quick pace. The role of miR-133, together with miR-1, is one of the most extensively studied in both cardiac and skeletal muscle development and disease [40,41]. Chen et al. [42] describes the opposite roles of miR-1 and miR-133 in skeletal muscle proliferation and differentiation, despite being transcribed from the same polycistronic unit. Surprisingly, genetic deletion of miR-133a or miR-133b displays no cardiac phenotype [22,23], yet deregulation of miR-133 in distinct cardiovascular diseases has been extensively described, such as during myocardial infarction [43,44] and arrhythmias [45,46]. At the molecular level, miR-133 contributes to repression of cardiac hypertrophy by negatively regulating Nfatc4 signaling [34,35], controlling, thus, the metabolic status of cardiac myocytes [36], and it has been proposed as biomarker in the predicted regression of left ventricular hypertrophy after valve replacement [47]. Most importantly, a significant decrease of miR-133 was observed during zebrafish cardiac regeneration [48], and the experimental modulation of miR-133 by either over-expression or deletion demonstrates a pivotal role of this microRNA in cardiomyocyte proliferation during cardiac regeneration. Several other microRNAs, such as miR-143, miR-145 and the miR-17-92 cluster, have also been reported to play a role in cardiac muscle. Miyasaka et al. [49] elegantly demonstrate that cardiac development is modulated by hemodynamic inputs controlling miR-143 expression and, thus, cardiac morphogenesis in zebrafish. Knockdown of miR-143 elicits re-expression of retinoic acid signaling components leading to outflow tract and ventricular dysmorphogenesis. Further evidence on the role of miR-143 in heart development has been provided also in zebrafish by Deacon et al. [37], who nicely showed that miR-143 targets adducin3, and if impaired, abnormal growth and elongation of the ventricular chambers occurs, leading to decrease cardiac contractility and, eventually, collapse. Interestingly, in the human adult heart, deep-sequencing of the miRNA transcriptome in the left and right atrial chambers has revealed that miR-143 is the highest expressed microRNA in the atrial chambers [50], yet its functional role in the adult (normal and diseased) heart remains to be uncovered. miR-145 has been primarily implicated in smooth muscle cells, and particularly, it is highly upregulated within the lungs of both experimentally-induced pulmonary arterial hypertension, as well as in patients with idiopathic and hereditable pulmonary arterial hypertension [51]. While there is no evidence of a functional role of miR-145 in the developing cardiovascular development, impaired miR-145 expression has been reported in several cardiovascular diseased conditions, such as acute cardiac infarction and coronary artery disease [52,53]. A functional link between miR-145 and these diseased statuses has been recently reported by Li et al. [54], demonstrating that miR-145 protects against oxidative stress-induced apoptosis in cardiomyocytes and also regulating Wnt/beta-catenin signaling through targeting of Dab2 in the ischemic heart [38]. The precise contribution of miR-17 remains to be discerned, yet systemic deletion of the miR-17-92 cluster leads to lung hypoplasia and ventricular septal defects at birth [55]. It has been also demonstrated that in embryonic cardiomyocytes, miR-17-92 regulates Fog-2 and, thus, cardiomyocyte proliferation [25], while in neonatal cardiac progenitor cells, it regulates Rb12/p130 [26]. In addition to its role during cardiovascular development, miR-17-92 also plays a pivotal role in the adult heart, since it is differentially expressed in the aging heart [56], and overexpression of this microRNA cluster leads to cardiac hypertrophic cardiomyopathy and arrhythmias [57]. Yet to date, the most described functional phenotype of miR-17 relates to pulmonary hypertension [58,59] and extracellular matrix remodeling [60] in this context. In addition to the role reported for miR-133, miR-143, miR-145 and the miR-17-92, among those microRNAs differentially expressed during cardiac development, initial hints for miR-494 and miR-210 function in myocyte adaptation and survival during hypoxia/ischemia are also emerging, yet with limited understanding [61,62]. While the myocardium is the most relevant tissue layer in the pumping heart, the role of the interstitial cardiac tissue is progressively emerging, since, in fact, it is the most abundant cell type within the adult heart. In this context, several microRNAs, such as miR-15, miR-30 and miR-24, have been reported to play key functions in the cardiac fibroblasts, yet as previously mentioned, they also seem to provide pivotal roles in other cardiovascular tissues. miR-15 displays a key role governing cardiomyocyte cell cycle withdrawal and binucleation, by controlling Chek1 expression [63], while in cardiac fibroblasts, it modulates collagen deposition [64]. miR-30 display a differential expression in experimental pulmonary hypertension [51]. Within the heart, it is has been reported to play a role in extracellular matrix remodeling, by controlling Ctgf expression, in the context of ventricular hypertrophy [31], a role that might also be linked with downregulation of miR-30 in a model of induced atrial fibrillation with fibrosis and inflammation [46]. Wang et al. [65] reported that miR-24 is downregulated after myocardial infarction, correlating with increased extracellular matrix deposition. Forced in vivo miR-24 overexpression decreased fibrosis, by controlling furin expression, which in turn, regulated tgf-beta bioavailability. In addition to its role in the interstitial cardiac tissue, a role on the cardiac endothelium has also been demonstrated [28] in the context of myocardial infarction. These authors demonstrate that miR-24 impairs angiogenesis by controlling gata2 and pak4 expression and blocking miR-24 in vivo leads to decreased endothelial apoptosis, increased vascularization and preservation of cardiac function in a mouse experimental model of myocardial infarction. Importantly, a role for miR-24 in myocardial cells has also been reported. Li et al. [29] reported that ischemia increased miR-24 expression in cardiomyocytes, while forced miR-24 expression in cultured cardiomyocytes increased cell viability and apoptosis, and necrosis rates were reduced. Such effects seemed to be mediated by miR-24 mediated control of the pro-apoptotic gene, Blc2l11. Given these pivotal roles, therapeutic usage of miR-24, in combination with other miRs (miR-21 and miR-221), has provided increased survival and engraftment to cocktail-treated cardiac progenitor cells in a mouse model of coronary artery ligation [66], opening, thus, new therapeutic approaches to heal the diseased heart. Apart from microRNAs playing a role in cardiomyocytes and interstitial tissue, two microRNAs have been revealed to play an important role in valve and vascular development. miR-23 has been highlighted in the context of cardiac valve formation. Elegant studies in zebrafish demonstrate that miR-23 controls Has2 expression, therefore regulating the extracellular matrix remodeling during endocardial cushion formation [27]. Additional evidence on the role of miR-23 has been reported in the vascular context, as recently reviewed by Bang et al. [67]. Similarly, miR-126 is predominantly expressed in the vascular tissue [21]. Targeted deletion of miR-126 resulted in impaired angiogenesis and vascular integrity [21,68]. In addition, a role for miR-126 in valve development, controlled by VEGF, has been also reported [69]. Yet, more recently, a role of miR-126 has also been ascribed in the developing cardiomyocytes, being regulated by hypoxia and histone deacetylases (HDAC) inhibitors and, therefore, contributing to cardioprotection [70]. For other microRNAs, such as miR-122 and let-7, although there is unequivocal evidence of a determinant role during cardiovascular development, it remains to be elucidating at which tissue level they contribute. miR-122 has been reported to play a pivotal role in several endodermal-derived tissues [71,72,73,74], yet its role in the cardiovascular system is just emerging. Huang et al. [33] recently reported that miR-122 was upregulated in Pax8 null mutants, which display ventricular septal defects. Analyses of putative targets of miR-122 uncovered that cck-8 and caspase-3 are genuine targets, supporting the notion that Pax8-related cardiovascular defects are mediated by miR-122. Similarly, let-7 has been proposed to play a role in cardiac development [39] and differentiation [75,76], yet full mechanistic insights remain to be clarified. Similarly, abnormal expression of let-7 has been identified in a rat model of myocardial injury (doxorubicin treatment), yet it functional role remains unexplored [77]. 6. Conclusions & Perspectives We have reported herein a glimpse of the value that microRNA microarrays can provide to understand the role of microRNAs in cardiovascular development. A simple approach, comparison of the microRNA signature of the developing ventricular chambers, has highlighted that almost 50% of the differentially expressed microRNAs identified have been subsequently shown to play diverse functional roles in the cardiovascular system. Moreover, meta-analyses comparing two additional conditions revealed similar microRNA signatures in the developing cardiac chambers and the differentiating and maturing cardiomyocytes derived from induced pluripotent stem cells, which are not altered in the adult and aging heart. Furthermore, such a proof-of-principle microRNA microarray meta-analyses provide also novel hints, as decoding a subset of microRNAs that behave in the opposite pattern during in vitro (iPS-derived) and in vivo (chamber maturation) cardiogenesis, opening new avenues to dissect the functional role of these microRNAs in the cardiovascular setting. Moreover, future meta-analyses studies, including not only healthy conditions, but emerging microRNA signatures of diseased status, such as atrial fibrillation, cardiac hypertrophy or ischemia, will certainly increase our understanding of microRNA biology in the normal and diseased heart. In addition, microRNA microarrays also provided novel insights into the intricate biology of microRNA transcriptional regulation and putative target recognition, broadening, thus, the spectrum of their applicability. As we focus our attention on the developing heart, it is becoming clear that microRNAs play a pivotal role in cardiac muscle biology, the interstitial tissue and also valve development. Yet, one of the challenges that remains ahead of us is to fine-tune our understanding of microRNA function in distinct cardiovascular settings, starting from unravelling their tissue distribution and following with dissecting the role of microRNAs in the epicardium and coronary vasculature. Such endeavors will provide us novel understandings and tools for promising therapeutic usage. Acknowledgments This work is supported by grants from the Ministry of Science and Innovation of the Spanish Government (MICINN BFU2009-11566) and from the Junta de Andalucía Regional Council (CTS-1614) to DF. Conflict of interest The authors declare no conflict of interest. ==== Refs References 1. Kelly R.G. The second heart field Curr. Top. Dev. Biol. 2012 100 33 65 10.1016/B978-0-12-387786-4.00002-6 22449840 2. Prall O.W. Menon M.K. Solloway M.J. Watanabe Y Zaffran S. Bajolle F. Biben C. McBride J.J. Robertson B.R. Chaulet H. An Nkx2-5/Bmp2/Smad1 negative feedback loop controls heart progenitor specification and proliferation Cell 2007 128 947 959 17350578 3. De Castro M.P. Acosta L. Domínguez J.N. Aránega A. Franco D. Molecular diversity of the developing and adult myocardium: Implications for tissue targeting Curr. Drug. Targets Cardiovasc. Haematol. Disord. 2003 3 227 239 10.2174/1568006033481429 12871041 4. Chinchilla A. Franco D. Regulatory mechanisms of cardiac development and repair Cardiovasc. Hematol. Disord. Drug. Targets 2006 6 101 112 16787195 5. Kruithof B.P. Duim S.N. Moerkamp A.T. Goumans M.J. TGFβ and BMP signaling in cardiac cushion formation: Lessons from mice and chicken Differentiation 2012 84 89 102 10.1016/j.diff.2012.04.003 22656450 6. De Vlaming A. Sauls K. Hajdu Z. Visconti R.P. Mehesz A.N. Levine R.A. Slaugenhaupt S.A. Hagège A. Chester A.H. Markwald R.R. Atrioventricular valve development: New perspectives on an old theme Differentiation 2012 84 103 116 10.1016/j.diff.2012.04.001 22579502 7. Männer J. Pérez-Pomares J.M. Macías D. Muñoz-Chápuli R. The origin, formation and developmental significance of the epicardium: A review Cells Tissues Organs 2001 169 89 103 10.1159/000047867 11399849 8. Wessels A. Pérez-Pomares J.M. The epicardium and epicardially derived cells (EPDCs) as cardiac stem cells Anat. Rec. A Discov. Mol. Cell. Evol. Biol. 2004 276 43 57 10.1002/ar.a.10129 14699633 9. Liu N. Olson E.N. MicroRNA regulatory networks in cardiovascular development Dev. Cell 2010 18 510 525 10.1016/j.devcel.2010.03.010 20412767 10. Bauersachs J. Thum T. Biogenesis and regulation of cardiovascular microRNAs Circ. Res. 2011 109 334 347 10.1161/CIRCRESAHA.110.228676 21778437 11. Gene Expression Omnibus Available online:http://www.ncbi.nlm.nih.gov/geo/ (accessed on 16 February 2013) 12. Zhang X. Azhar G. Wei J.Y. The expression of microRNA and microRNA clusters in the aging heart PLoS One 2012 7 e34688 10.1371/journal.pone.0034688 22529925 13. Babiarz J.E. Ravon M. Sridhar S. Ravindran P. Swanson B. Bitter H. Weiser T. Chiao E. Certa U. Kolaja K.L. Determination of the human cardiomyocyte mRNA and miRNA differentiation network by fine-scale profiling Stem Cells Dev. 2012 21 1956 1965 10.1089/scd.2011.0357 22050602 14. Chinchilla A. Lozano E. Daimi H. Esteban F.J. Crist C. Aranega A.E. Franco D. MicroRNA profiling during mouse ventricular maturation: A role for miR-27 modulating Mef2c expression Cardiovasc. Res. 2011 89 98 108 20736237 15. Bioconductor Limma Package Available online:http://www.bioconductor.org (accessed on 16 February 2013) 16. R Software Available online:http://www.r-project.org (accessed on 16 February 2013) 17. TM4 Software Suite Available online:http://www.tm4.org (accessed on 16 February 2013) 18. Porrello E.R. Mahmoud A.I. Simpson E. Johnson B.A. Grinsfelder D. Canseco D. Mammen P.P. Rothermel B.A. Olson E.N. Sadek H.A. Regulation of neonatal and adult mammalian heart regeneration by the miR-15 family Proc. Natl. Acad. Sci. USA 2013 110 187 192 23248315 19. Zhao Y. Ransom J.F. Li A. Vedantham V. von Drehle M. Muth A.N. Tsuchihashi T. McManus M.T. Schwartz R.J. Srivastava D. Dysregulation of cardiogenesis, cardiac conduction, and cell cycle in mice lacking miRNA-1-2 Cell 2007 129 303 317 10.1016/j.cell.2007.03.030 17397913 20. Saxena A. Tabin C.J. miRNA-processing enzyme Dicer is necessary for cardiac outflow tract alignment and chamber septation Proc. Natl. Acad. Sci. USA 2010 107 87 91 10.1073/pnas.0912870107 20018673 21. Fish J.E. Santoro M.M. Morton S.U. Yu S. Yeh R.F. Wythe J.D. Ivey K.N. Bruneau B.G. Stainier D.Y. Srivastava D. miR-126 regulates angiogenic signaling and vascular integrity Dev. Cell 2008 15 272 284 10.1016/j.devcel.2008.07.008 18694566 22. Chen J.F. Mandel E.M. Thomson J.M. Wu Q. Callis T.E. Hammond S.M. Conlon F.L. Wang D.Z. The role of microRNA-1 and microRNA-133 in skeletal muscle proliferation and differentiation Nat. Genet. 2006 38 228 233 16380711 23. Liu N. Bezprozvannaya S. Williams A.H. Qi X. Richardson J.A. Bassel-Duby R. Olson E.N. microRNA-133a regulates cardiomyocyte proliferation and suppresses smooth muscle gene expression in the heart Genes Dev. 2008 22 3242 3254 10.1101/gad.1738708 19015276 24. Papadopoulos G.L. Reczko M. Simossis V.A. Sethupathy P. Hatzigeorgiou A.G. The database of experimentally supported targets: A functional update of TarBase Nucl. Acids. Res. 2009 37 D155 D158 10.1093/nar/gkn809 18957447 25. Xiang R. Lei H. Chen M. Li Q. Sun H. Ai J. Chen T. Wang H. Fang Y. Zhou Q. The miR-17-92 cluster regulates FOG-2 expression and inhibits proliferation of mouse embryonic cardiomyocytes Braz. J. Med. Biol. Res. 2012 45 131 138 10.1590/S0100-879X2012007500007 22267003 26. Sirish P. López J.E. Li N. Wong A. Timofeyev V. Young J.N. Majdi M. Li R.A. Chen H.S. Chiamvimonvat N. MicroRNA profiling predicts a variance in the proliferative potential of cardiac progenitor cells derived from neonatal and adult murine hearts J. Mol. Cell Cardiol. 2012 52 264 272 10.1016/j.yjmcc.2011.10.012 22062954 27. Lagendijk A.K. Goumans M.J. Burkhard S.B. Bakkers J. MicroRNA-23 restricts cardiac valve formation by inhibiting Has2 and extracellular hyaluronic acid production Circ. Res. 2011 109 649 657 10.1161/CIRCRESAHA.111.247635 21778427 28. Fiedler J. Jazbutyte V. Kirchmaier B.C. Gupta S.K. Lorenzen J. Hartmann D. Galuppo P. Kneitz S. Pena J.T. Sohn-Lee C. MicroRNA-24 regulates vascularity after myocardial infarction Circulation 2011 124 720 730 10.1161/CIRCULATIONAHA.111.039008 21788589 29. Li D.F. Tian J. Guo X. Huang L.M. Xu Y. Wang C.C. Wang J.F. Ren A.J. Yuan W.J. Lin L. Induction of microRNA-24 by HIF-1 protects against ischemic injury in rat cardiomyocytes Physiol. Res. 2013 61 555 565 23098654 30. Vacchi-Suzzi C. Hahne F. Scheubel P. Marcellin M. Dubost V. Westphal M. Boeglen C. Büchmann-Møller S. Cheung M.S. Cordier A. Heart structure-specific transcriptomic atlas reveals conserved microRNA-mRNA interactions PLoS One 2013 8 e52442 10.1371/journal.pone.0052442 23300973 31. Duisters R.F. Tijsen A.J. Schroen B. Leenders J.J. Lentink V. van der Made I. Herias V. van Leeuwen R.E. Schellings M.W. Barenbrug P. miR-133 and miR-30 regulate connective tissue growth factor: Implications for a role of microRNAs in myocardial matrix remodeling Circ. Res. 2009 104 170 178 10.1161/CIRCRESAHA.108.182535 19096030 32. Selbach M. Schwanhäusser B. Thierfelder N. Fang Z. Khanin R. Rajewsky N. Widespread changes in protein synthesis induced by microRNAs Nature 2008 455 58 63 18668040 33. Huang X. Huang F. Yang D. Dong F. Shi X. Wang H. Zhou X. Wang S. Dai S. Expression of microRNA-122 contributes to apoptosis in H9C2 myocytes J. Cell. Mol. Med. 2012 16 2637 2646 10.1111/j.1582-4934.2012.01577.x 22453009 34. Li Q. Lin X. Yang X. Chang J. NFATc4 is negatively regulated in miR-133a-mediated cardiomyocyte hypertrophic repression Am. J. Physiol. Heart Circ. Physiol. 2010 298 H1340 H1347 10.1152/ajpheart.00592.2009 20173049 35. Dong D.L. Chen C. Huo R. Wang N. Li Z. Tu Y.J. Hu J.T. Chu X. Huang W. Yang B.F. Reciprocal repression between microRNA-133 and calcineurin regulates cardiac hypertrophy: A novel mechanism for progressive cardiac hypertrophy Hypertension 2010 55 946 952 10.1161/HYPERTENSIONAHA.109.139519 20177001 36. Horie T. Ono K. Nishi H. Iwanaga Y. Nagao K. Kinoshita M. Kuwabara Y. Takanabe R. Hasegawa K. Kita T. MicroRNA-133 regulates the expression of GLUT4 by targeting KLF15 and is involved in metabolic control in cardiac myocytes Biochem. Biophys. Res. Commun. 2009 389 315 320 10.1016/j.bbrc.2009.08.136 19720047 37. Deacon D.C. Nevis K.R. Cashman T.J. Zhou Y. Zhao L. Washko D. Guner-Ataman B. Burns C.G. Burns C.E. The miR-143-adducin3 pathway is essential for cardiac chamber morphogenesis Development 2010 137 1887 1896 10.1242/dev.050526 20460367 38. Mayorga M.E. Penn M.S. miR-145 is differentially regulated by TGF-β1 and ischaemia and targets Disabled-2 expression and wnt/β-catenin activity J. Cell. Mol. Med. 2012 16 1106 1113 10.1111/j.1582-4934.2011.01385.x 21762377 39. Cao L. Kong L.P. Yu Z.B. Han S.P. Bai Y.F. Zhu J. Hu X. Zhu C. Zhu S. Guo X.R. MicroRNA expression profiling of the developing mouse heart Int. J. Mol. Med. 2012 30 1095 1104 22895573 40. Han M. Toli J. Abdellatif M. MicroRNAs in the cardiovascular system Curr. Opin. Cardiol. 2011 26 181 189 10.1097/HCO.0b013e328345983d 21464712 41. Malizia A.P. Wang D.Z. MicroRNAs in cardiomyocyte development WIREs Syst. Biol. Med. 2011 3 183 190 42. Chen J.F. Mandel E.M. Thomson J.M. Wu Q. Callis T.E. Hammond S.M. Conlon F.L. Wang D.Z. The role of microRNA-1 and microRNA-133 in skeletal muscle proliferation and differentiation Nat. Genet. 2006 38 228 233 10.1038/ng1725 16380711 43. Bostjancic E. Zidar N. Stajer D. Glavac D. MicroRNAs miR-1, miR-133a, miR-133b and miR-208 are dysregulated in human myocardial infarction Cardiology 2010 115 163 169 10.1159/000268088 20029200 44. Bostjancic E. Zidar N. Stajer D. Glavac D. MicroRNA miR-1 is up-regulated in remote myocardium in patients with myocardial infarction Folia Biol. (Praha) 2010 56 27 31 20163779 45. Belevych A.E. Sansom S.E. Terentyeva R. Ho H.T. Nishijima Y. Martin M.M. Jindal H.K. Rochira J.A. Kunitomo Y. Abdellatif M. MicroRNA-1 and -133 increase arrhythmogenesis in heart failure by dissociating phosphatase activity from RyR2 complex PLoS One 2011 6 e28324 10.1371/journal.pone.0028324 22163007 46. Li H. Li S. Yu B. Liu S. Expression of miR-133 and miR-30 in chronic atrial fibrillation in canines Mol. Med. Rep. 2012 5 1457 1460 22407060 47. Villar A.V. Merino D. Wenner M. Llano M. Cobo M. Montalvo C. García R. Martín-Durán R. Hurlé J.M. Hurlé M.A. Myocardial gene expression of microRNA-133a and myosin heavy and light chains, in conjunction with clinical parameters, predict regression of left ventricular hypertrophy after valve replacement in patients with aortic stenosis Heart 2011 97 1132 1137 21586423 48. Yin V.P. Lepilina A. Smith A. Poss K.D. Regulation of zebrafish heart regeneration by miR-133 Dev. Biol. 2012 365 319 327 10.1016/j.ydbio.2012.02.018 22374218 49. Miyasaka K.Y. Kida Y.S. Banjo T. Ueki Y. Nagayama K. Matsumoto T. Sato M. Ogura T. Heartbeat regulates cardiogenesis by suppressing retinoic acid signaling via expression of miR-143 Mech. Dev. 2011 128 18 28 10.1016/j.mod.2010.09.002 20869435 50. Hsu J. Hanna P. van Wagoner D.R. Barnard J. Serre D. Chung M.K. Smith J.D. Whole genome expression differences in human left and right atria ascertained by RNA sequencing Circ. Cardiovasc. Genet. 2012 5 327 335 10.1161/CIRCGENETICS.111.961631 22474228 51. Caruso P. MacLean M.R. Khanin R. McClure J. Soon E. Southgate M. MacDonald R.A. Greig J.A. Robertson K.E. Masson R. Dynamic changes in lung microRNA profiles during the development of pulmonary hypertension due to chronic hypoxia and monocrotaline Arterioscler. Thromb. Vasc. Biol. 2010 30 716 723 10.1161/ATVBAHA.109.202028 20110569 52. Meder B. Keller A. Vogel B. Haas J. Sedaghat-Hamedani F. Kayvanpour E. Just S. Borries A. Rudloff J. Leidinger P. MicroRNA signatures in total peripheral blood as novel biomarkers for acute myocardial infarction Basic Res. Cardiol. 2011 106 13 23 10.1007/s00395-010-0123-2 20886220 53. Fichtlscherer S. De Rosa S. Fox H. Schwietz T. Fischer A. Liebetrau C. Weber M. Hamm C.W. Röxe T. Müller-Ardogan M. Circulating microRNAs in patients with coronary artery disease Circ. Res. 2010 107 677 684 10.1161/CIRCRESAHA.109.215566 20595655 54. Li R. Yan G. Li Q. Sun H. Hu Y. Sun J. Xu B. MicroRNA-145 protects cardiomyocytes against hydrogen peroxide (H2 O2 )-induced apoptosis through targeting the mitochondria apoptotic pathway PLoS One 2012 7 e44907 10.1371/journal.pone.0044907 23028672 55. Ventura A. Young A.G. Winslow M.M. Lintault L. Meissner A. Erkeland S.J. Newman J. Bronson R.T. Crowley D. Stone J.R. Targeted deletion reveals essential and overlapping functions of the miR-17 through 92 family of miRNA clusters Cell 2008 132 875 886 18329372 56. Danielson L.S. Park D.S. Rotllan N. Chamorro-Jorganes A. Guijarro M.V. Fernandez-Hernando C. Fishman G.I. Phoon C.K. Hernando E. Cardiovascular dysregulation of miR-17–92 causes a lethal hypertrophic cardiomyopathy and arrhythmogenesis FASEB J. 2012 10.1096/fj.12-221994 57. Van Almen G.C. Verhesen W. van Leeuwen R.E. van de Vrie M. Eurlings C. Schellings M.W. Swinnen M. Cleutjens J.P. van Zandvoort M.A. Heymans S. MicroRNA-18 and microRNA-19 regulate CTGF and TSP-1 expression in age-related heart failure Aging Cell 2011 10 769 779 10.1111/j.1474-9726.2011.00714.x 21501375 58. Lee C. Mitsialis S.A. Aslam M. Vitali S.H. Vergadi E. Konstantinou G. Sdrimas K. Fernandez-Gonzalez A. Kourembanas S. Exosomes mediate the cytoprotective action of mesenchymal stromal cells on hypoxia-induced pulmonary hypertension Circulation 2012 126 2601 2611 10.1161/CIRCULATIONAHA.112.114173 23114789 59. Pullamsetti S.S. Doebele C. Fischer A. Savai R. Kojonazarov B. Dahal B.K. Ghofrani H.A. Weissmann N. Grimminger F. Bonauer A. Inhibition of microRNA-17 improves lung and heart function in experimental pulmonary hypertension Am. J. Respir. Crit. Care Med. 2012 185 409 419 22161164 60. Shan S.W. Lee D.Y. Deng Z. Shatseva T. Jeyapalan Z. Du W.W. Zhang Y. Xuan J.W. Yee S.P. Siragam V. MicroRNA MiR-17 retards tissue growth and represses fibronectin expression Nat. Cell Biol. 2009 11 1031 1038 19633662 61. Han M. Toli J. Abdellatif M. MicroRNAs in the cardiovascular system Curr. Opin. Cardiol. 2011 26 181 189 10.1097/HCO.0b013e328345983d 21464712 62. Mutharasan R.K. Nagpal V. Ichikawa Y. Ardehali H. microRNA-210 is upregulated in hypoxic cardiomyocytes through Akt- and p53-dependent pathways and exerts cytoprotective effects Am. J. Physiol. Heart Circ. Physiol. 2011 301 H1519 H1530 10.1152/ajpheart.01080.2010 21841015 63. Porrello E.R. Johnson B.A. Aurora A.B. Simpson E. Nam Y.J. Matkovich S.J. Dorn G.W. van Rooij E. Olson E.N. MiR-15 family regulates postnatal mitotic arrest of cardiomyocytes Circ. Res. 2011 109 670 679 10.1161/CIRCRESAHA.111.248880 21778430 64. Divakaran V. Adrogue J. Ishiyama M. Entman M.L. Haudek S. Sivasubramanian N. Mann D.L. Adaptive and maladptive effects of SMAD3 signaling in the adult heart after hemodynamic pressure overloading Circ. Heart Fail. 2009 2 633 642 10.1161/CIRCHEARTFAILURE.108.823070 19919989 65. Wang J. Huang W. Xu R. Nie Y. Cao X. Meng J. Xu X. Hu S. Zheng Z. MicroRNA-24 regulates cardiac fibrosis after myocardial infarction J. Cell Mol. Med. 2012 16 2150 2160 10.1111/j.1582-4934.2012.01523.x 22260784 66. Hu S. Huang M. Nguyen P.K. Gong Y. Li Z. Jia F. Lan F. Liu J. Nag D. Robbins R.C. Wu J.C. Novel microRNA prosurvival cocktail for improving engraftment and function of cardiac progenitor cell transplantation Circulation 2011 124 S27 S34 10.1161/CIRCULATIONAHA.111.017954 21911815 67. Fish J.E. Srivastava D. MicroRNAs: Opening a new vein in angiogenesis research Sci. Signal 2009 2 10.1126/scisignal.252pe1 68. Bang C. Fiedler J. Thum T. Cardiovascular importance of the microRNA-23/27/24 family Microcirculation 2012 19 208 214 10.1111/j.1549-8719.2011.00153.x 22136461 69. Stankunas K. Ma G.K. Kuhnert F.J. Kuo C.J. Chang C.P. VEGF signaling has distinct spatiotemporal roles during heart valve development Dev. Biol. 2010 347 325 336 10.1016/j.ydbio.2010.08.030 20816797 70. Shi H. Chen L. Wang H. Zhu S. Dong C. Webster K.A. Wei J. Synergistic induction of miR-126 by hypoxia and HDAC inhibitors in cardiac myocytes Biochem. Biophys. Res. Commun. 2013 430 827 832 10.1016/j.bbrc.2012.11.061 23201405 71. Tzur G. Levy A. Meiri E. Barad O. Spector Y. Bentwich Z. Mizrahi L. Katzenellenbogen M. Ben-Shushan E. Reubinoff B.E. Galun E. MicroRNA expression patterns and function in endodermal differentiation of human embryonic stem cells PLoS One 2008 3 e3726 10.1371/journal.pone.0003726 19015728 72. Stuckenholz C. Lu L. Thakur P. Kaminski N. Bahary N. FACS-assisted microarray profiling implicates novel genes and pathways in zebrafish gastrointestinal tract development Gastroenterology 2009 137 1321 1332 10.1053/j.gastro.2009.06.050 19563808 73. Xu H. He J.H. Xiao Z.D. Zhang Q.Q. Chen Y.Q. Zhou H. Qu L.H. Liver-enriched transcription factors regulate microRNA-122 that targets CUTL1 during liver development Hepatology 2010 52 1431 1442 10.1002/hep.23818 20842632 74. Laudadio I. Manfroid I. Achouri Y. Schmidt D. Wilson M.D. Cordi S. Thorrez L. Knoops L. Jacquemin P. Schuit F. A feedback loop between the liver-enriched transcription factor network and miR-122 controls hepatocyte differentiation Gastroenterology 2012 142 119 129 21920465 75. Wong S.S. Ritner C. Ramachandran S. Aurigui J. Pitt C. Chandra P. Ling V.B. Yabut O. Bernstein H.S. miR-125b promotes early germ layer specification through Lin28/let-7d and preferential differentiation of mesoderm in human embryonic stem cells PLoS One 2012 7 e36121 10.1371/journal.pone.0036121 22545159 76. Ahmed R.P. Haider H.K. Buccini S. Li L. Jiang S. Ashraf M. Reprogramming of skeletal myoblasts for induction of pluripotency for tumor-free cardiomyogenesis in the infarcted heart Circ. Res. 2011 109 60 70 10.1161/CIRCRESAHA.110.240010 21566212 77. Fu J. Peng C. Wang W. Jin H. Tang Q. Wei X. Let-7g is involved in doxorubicin induced myocardial injury Environ. Toxicol. Pharmacol. 2012 33 312 317 10.1016/j.etap.2011.12.023 22301161

PMC005xxxxxx/PMC5003482.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays2020131microarrays-02-00131ArticleEvaluation of Different Normalization and Analysis Procedures for Illumina Gene Expression Microarray Data Involving Small Changes Johnstone Daniel M. 1234*Riveros Carlos 15Heidari Moones 2Graham Ross M. 678Trinder Debbie 78Berretta Regina 145Olynyk John K. 91011Scott Rodney J. 1212Moscato Pablo 145Milward Elizabeth A. 121 Centre for Bioinformatics, Biomarker Discovery and Information-Based Medicine, The University of Newcastle, Callaghan, NSW 2308, Australia; E-Mails: carlos.riveros@newcastle.edu.au (C.R.); regina.berretta@newcastle.edu.au (R.B.); rodney.scott@newcastle.edu.au (R.J.S.); pablo.moscato@newcastle.edu.au (P.M.); liz.milward@newcastle.edu.au (E.A.M.) 2 School of Biomedical Sciences and Pharmacy, The University of Newcastle, Callaghan, NSW 2308, Australia; E-Mail: moones.heidari@uon.edu.au3 Discipline of Physiology and Bosch Institute, University of Sydney, Sydney, NSW 2006, Australia 4 Australian Research Council Centre of Excellence in Bioinformatics, Callaghan, NSW 2308, Australia 5 School of Electrical Engineering and Computer Science, the University of Newcastle, Callaghan, NSW 2308, Australia 6 School of Biomedical Sciences, CHIRI Biosciences Research Precinct, Faculty of Health Sciences, Curtin University, Bentley, WA 6102, Australia; E-Mail: rmgraham@curtin.edu.au7 School of Medicine and Pharmacology, University of Western Australia, Fremantle, WA 6160, Australia; E-Mail: debbie.trinder@uwa.edu.au8 Western Australian Institute for Medical Research, Perth, WA 6000, Australia9 Department of Gastroenterology, Fremantle Hospital, Fremantle, WA 6160, Australia; E-Mail: john.olynyk@health.wa.gov.au10 Curtin Health Innovation Research Institute, Curtin University, Bentley, WA 6102, Australia11 Institute for Immunology & Infectious Diseases, Murdoch University, Murdoch, WA 6153, Australia12 The Division of Molecular Medicine, Hunter Area Pathology Service, New Lambton, NSW 2305, Australia* Author to whom correspondence should be addressed; E-Mail: daniel.johnstone@sydney.edu.au; Tel.: +61-2-9351-5162; Fax: +61-2-9351-6470.21 5 2013 6 2013 2 2 131 152 25 3 2013 08 5 2013 10 5 2013 © 2013 by the authors; licensee MDPI, Basel, Switzerland.2013This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).While Illumina microarrays can be used successfully for detecting small gene expression changes due to their high degree of technical replicability, there is little information on how different normalization and differential expression analysis strategies affect outcomes. To evaluate this, we assessed concordance across gene lists generated by applying different combinations of normalization strategy and analytical approach to two Illumina datasets with modest expression changes. In addition to using traditional statistical approaches, we also tested an approach based on combinatorial optimization. We found that the choice of both normalization strategy and analytical approach considerably affected outcomes, in some cases leading to substantial differences in gene lists and subsequent pathway analysis results. Our findings suggest that important biological phenomena may be overlooked when there is a routine practice of using only one approach to investigate all microarray datasets. Analytical artefacts of this kind are likely to be especially relevant for datasets involving small fold changes, where inherent technical variation—if not adequately minimized by effective normalization—may overshadow true biological variation. This report provides some basic guidelines for optimizing outcomes when working with Illumina datasets involving small expression changes. gene expression microarraynormalizationIllumina ==== Body 1. Introduction Microarray studies have been particularly successful for identifying genes with large expression changes in conditions such as cancer. The challenge is to extend microarray technology into robust identification of smaller gene expression changes. This requires array platforms with a high degree of sensitivity and specificity and data analysis tools that generate accurate results. While increasing experimental group sizes can improve the detectability of subtle changes, one major challenge in microarray analysis is the detection of small, but “real”, expression changes in small datasets. The Illumina microarray platform has become one of the main platforms for “transcriptomic” studies. Each Illumina BeadChip array comprises randomly positioned silica beads, each containing hundreds of thousands of copies of a specific 50-nucleotide probe sequence. On average, each probe is replicated on at least 15 beads randomly distributed across each array. The large number of replicate beads minimizes artefacts that may arise due to intra-array location and other factors and provides a high degree of internal technical replication, facilitating generation of reliable raw data [1,2,3,4,5,6]. The technology has performed well in comparative studies of different platforms by the Microarray Quality Control (MAQC) consortium [7,8] and others [9,10,11], but such studies have not provided detailed comparisons of the performance of different data analysis tools. Various open source tools are available to analyse Illumina data, such as lumi [12], limma [13] and other Bioconductor packages [14], which use the R programming environment. Schmid and colleagues have compared different normalization methods available through the R environment and Illumina’s proprietary software, recommending particular approaches depending on the characteristics of a particular dataset [15]. However this study did not investigate how different differential expression analysis techniques or combinations of normalization strategy and differential expression analysis technique affect final outcomes—there is still little information available on this. In addition, as Bioconductor packages require knowledge of the R programming language, they are currently used primarily by researchers with stronger computing backgrounds and by more specialized research groups doing large quantities of array analysis. These approaches are less commonly used by researchers doing occasional array studies or performing downstream analyses of array data provided under contract by large facilities or by researchers with restricted computing expertise, as is the case for many graduates from biological disciplines. Most novice Illumina microarray users instead rely on established “black box” procedures developed by Illumina and other companies. Therefore, while the Illumina platform appears well-suited to working with datasets involving small expression changes, as described above, the effects of different computational approaches need to be investigated more closely. In this study, we have examined how different normalization and differential expression analysis tools may influence analyses of small, low fold-change datasets on this platform. Following initial scanning of BeadChips by Illumina’s BeadScan software, there are three phases of processing of scanned BeadChip data (bead level data): (1) Local background subtraction and averaging of probe replicates generating bead summary data; (2) Transformation and normalization; (3) Analysis of differential expression. The different data processing steps and associated issues are briefly reviewed below. 1.1. Generating Bead Summary Data Initial data pre-processing in the proprietary Illumina GenomeStudio (formerly BeadStudio) software provides users with bead summary data in the form of a single signal intensity value for each probe. This value is calculated by subtracting the local background from the signal intensity for each bead, then taking the mean of all beads containing a given probe. While the beadarray package available through R/Bioconductor allows the user to work with raw bead level data [16], these data impose considerable storage requirements and are not yet commonly utilized by novice microarray users. Furthermore, Dunning and colleagues investigated the effects on bead level data of the pre-processing summarization methods used by GenomeStudio and concluded that these are beneficial for reducing bias and robust determination of gene expression [17]. For these reasons, we have restricted the present investigation to bead summary data that have already been generated by pre-processing algorithms in GenomeStudio. 1.2. Transformation and Normalization Raw bead summary intensity values are usually normalized by one or more transforming functions. Reasons for normalizing can include forcing a normal data distribution or increasing comparability between probes, samples, chips, machines or platforms. Even small technical variations (e.g., cRNA loading on arrays, scanning and hybridization inconsistency) can sometimes cause considerable differences in signal intensities. The overarching aim of normalization is to reduce differences due to technical variation (false positives), while conserving true biological effects (i.e., maximizing true positives and minimizing false negatives). Prior to normalization, it is generally recommended that a correction step be performed to adjust for between-array differences in non-specific signal intensities (i.e., background correction). Using GenomeStudio, this correction involves subtracting the mean signal of negative control probes in a particular array from each bead summary value in that array. While recommended by Illumina, several groups suggest this particular method is flawed [15,17,18] and propose alternative correction approaches available through the Bioconductor project. Following background correction (or not), microarray data are generally normalized by one of several different approaches. Here, we have investigated all four optional normalization strategies in the GenomeStudio software: Average, Cubic Spline, Quantile and Rank Invariant, as well as the No Normalization option. Average involves normalization to the mean signal of each sample; Cubic Spline and Quantile apply different forms of quantile normalization to bead summary data [19,20]; Rank Invariant normalizes data based on values of probes that do not change their ranking across samples. In the first section of the study, we have compared the effects of the different GenomeStudio normalization strategies within each of three different analytical approaches. 1.3. Analysis of Differential Expression Following normalization, different analytical approaches are used to identify genes with altered expression between experimental conditions. The challenge for any analytical approach lies in reducing false positives (Type I or α errors), while avoiding false negatives (Type II or β errors). The use of a statistical p-value approach allows estimation of false positive error probability, which can be considerable when conducting large numbers of comparisons. Yet, conversely, the methods currently used to adjust for multiple comparisons [21] are often very conservative and may miss real changes. Adjustments of this kind may be most useful for identifying restricted groups of target genes (e.g., class prediction aimed at identifying biomarkers for diagnosis or prognosis). For studies aimed at identifying complete sets of target genes (e.g., class comparison or class discovery aimed at understanding biological mechanisms), accepting non-informative false positives may be less problematic than omitting informative genes. Consequently, minimizing false negatives by not applying a multiple testing correction has been recommended for such studies [22,23]. As our study has focused on approaches suitable for identifying complete sets of differentially expressed genes (class discovery), a multiple testing correction has not been applied to most analyses. In addition to exploring the effects of different normalization strategies, we have also assessed how outcomes are affected by applying each of three different analytical approaches to the normalized data. Two of the three approaches tested used statistical significance as the inclusion criteria: GenomeStudio differential expression and GeneSpring differential expression. GenomeStudio was assessed because it is the Illumina proprietary software. GeneSpring is a widely-used, commercially available application with a number of features not present in GenomeStudio, including additional statistical capabilities. The third analytical approach assessed was a Max Cover (α,β)-k Feature Set approach (Max Cover (α,β)-FS) [24,25,26,27]. Whereas the GenomeStudio and GeneSpring algorithms use the average magnitude and variance of the signal intensity, Max Cover (α,β)-FS considers primarily the distribution of the test and control replicates relative to one another and the ability of each probe to discriminate between replicates from different classes (e.g., different experimental conditions). It is not based upon fold-change cut-offs or the statistical significance of comparisons of mean expression measures. We analysed two comparable Illumina datasets with relatively small expression changes. These were from (i) heart and (ii) brain samples of biological replicates of mice fed a short-term high iron diet and control mice fed a normal diet. Short-term high iron diets cause only modest changes in heart gene expression [28], and our studies suggest changes in the brain are even smaller, possibly because the blood-brain barrier may help protect the brain from high systemic iron [29]. The study examines the effects of applying different normalization and expression analysis approaches to these datasets. 2. Experimental Section 2.1. Animals All animal work was approved by the Animal Ethics Committee of the University of Western Australia. Male mice of the AKR strain were fed either normal chow or a high-iron diet (normal chow supplemented with 2% carbonyl iron for three weeks prior to sacrifice). The high-iron regimen used results in significantly higher iron indices and iron loading in the liver [30]. At 10 weeks of age, mice were sacrificed under anaesthesia (50 mg/kg ketamine, 10 mg/kg xylazine), and blood was removed by transcardiac perfusion with isotonic saline. Heart and brain tissue was collected from biological replicates (n ≥ 4 per group), snap-frozen in liquid nitrogen and stored at −80 °C. 2.2. Microarray Experiments Total RNA was isolated using TRI Reagent (Ambion), purified and concentrated using the RNeasy MinElute Kit (Qiagen) and amplified with the Illumina TotalPrep RNA Amplification Kit (Ambion). Gene expression was assessed in biological replicates using Illumina Sentrix MouseRef-8 (v1.1) BeadChip arrays. BeadChips were scanned using Illumina BeadArray reader and BeadScan software. For each tissue, all sample preparation and microarray experimentation was done simultaneously using arrays from the same batch, in order to avoid any potential batch effects. Following quality control assessment of microarray data, one control heart RNA sample was flagged as an outlier and removed from further analysis. 2.3. Microarray Data Analysis 2.3.1. Normalization and Differential Expression Analysis Bead summary data were normalized separately for each dataset (heart, brain), using each of the four normalization procedures (Average, Cubic Spline, Quantile and Rank Invariant) available in GenomeStudio v2010.3 (Illumina). Non-normalized data were also examined. The algorithms and parameter settings used to assess differential gene expression were: (1) GenomeStudio v2010.3—The Illumina Custom algorithm in the GenomeStudio software assesses three components of variation (sequence specific biological variation, non-specific biological variation and technical error). Probes returning a p value < 0.05 in comparisons of the control and test classes were considered to be detecting differential expression. A more detailed description is given in the GenomeStudio Gene Expression Module User Guide [31]. (2) GeneSpring GX 11.0 Software—The usual default settings of the GeneSpring program apply further transformation and normalization steps; however, this can introduce substantial artefacts when applied to data already normalized by other approaches. For these reasons, these additional normalization steps were not applied. Differential expression was determined by an unpaired t test (p < 0.05). (3) Max Cover (α,β)-k Feature Set Approach—Max Cover (α,β)-FS is a multivariate method that selects a set of probes that, as a collective, can discriminate well between the experimental test and control groups [27]. This algorithm consists of a two-stage filter process. Firstly, Fayyad and Irani’s algorithm [32] is used to discretise the data. For each probe, the algorithm orders the samples based on signal intensity and converts continuous data to binary data based on different intensity thresholds. It then selects the threshold that minimizes the class-information entropy of the samples, creating a binary dataset and discards the probes that are not discriminative enough, according to the minimum description length principle (filtering) [27]. Secondly, the algorithm finds a solution for the Max Cover (α,β)-k Feature Set problem [24]. This is achieved by comparing, for each probe, all possible pairs of samples, whether controls or tests, in order to extract an optimal set (solution) of probes (“features”) with both strong inter-class differentiation and strong intra-class similarity [25,26,27]. This approach differs from statistical methods, such as GenomeStudio and GeneSpring, in that instead of only considering means and variance measures, it preserves information about the individual samples within each class. It also identifies solutions involving sets of probes. These solutions reflect interrelationships between different probes—information which is often lost when considering each probe individually. A subset of analyses were performed in which a multiple testing correction (Benjamini Hochberg False Discovery Rate) was applied to background-subtracted data normalized using the strategies described above and assessed for differential gene expression using GenomeStudio. In addition to investigating the effects of these normalization strategies and analytical approaches on background-subtracted data, we also investigated the effect of omitting background correction before normalizing data using the four available options (as well as No Normalization) and performing differential expression analysis in GenomeStudio. To compare these various approaches to those available through the Bioconductor project, bead summary data were exported from GenomeStudio and analysed with the Bioconductor packages limma [13] and lumi [12], using pipelines recommended by the tool creators. For limma, this involved invoking the neqc function (background correction using a normal-exponential convolution model, quantile normalization and log2 transformation) followed by replicate summarization by fitting a linear model and differential expression analysis using moderated t-statistics with empirical Bayes’ shrinkage of the sample variances [33]. For lumi, this involved background correction using bgAdjust, variance stabilizing transformation and robust spline normalization, followed by replicate summarization and differential expression analysis using the limma functions described above [12]. 2.3.2. Filtering of Non-Specific Probe Signals To avoid distortion of the results by noise, we removed probes returning signals that were highly likely to be due to non-specific background signal rather than specific probe-target hybridization. The specificity of individual probe signals was estimated using the detection p-value, which is the probability of seeing a certain signal level without probe-target hybridization [31]. All probes returning a detection p-value > 0.01 (1% false positive rate, as recommended by Illumina) in both the control group and the high iron group were eliminated from further analysis. As illustrated in Figure 1, this step was performed after normalization and differential expression analysis—the GenomeStudio software does not allow the removal of specific probes before normalization and analysis, as might be preferred. Figure 1 Flowchart illustrating the different normalization procedures and differential expression algorithms used. 2.3.3. Assessment of Probe Set Concordance Different combinations of normalization and analysis approaches were applied as detailed in the Results section. The degree of agreement of the resulting probe sets, henceforth termed “concordance”, was calculated as either a number or a percentage. In the first instance, the concordance of two probe sets generated by different normalization strategies or analytical approaches was defined as the number of overlapping probes between the two sets. In the second instance, the concordance was defined as the percentage of overlapping probes calculated against the total number of probes in each particular probe set. Comparable measures, notably number of overlapping genes (NOG) and percentage of overlapping genes (POG), have been used previously to assess outcome concordance [7,34]. In this study, we will be considering concordance in three separate contexts: (1) the concordance between the probe sets generated by the different normalization strategies; (2) the concordance between the probe sets generated by the various types of differential expression analysis approaches; and (3) the concordance between the pathways enriched within each probe set. 2.3.4. Summary of Analysis and Evaluation A schematic summarizing the different steps in normalization, differential expression analysis and subsequent filtering is given in Figure 1. 2.4. Pathway Analysis The Database for Annotation, Visualization and Integrated Discovery (DAVID [35]) was used to identify enriched pathways in select probe sets [36,37]. The full list of genes included on the array was used as the background list. DAVID organizes gene lists into pathways and identifies those that have an enrichment of differentially expressed genes relative to how many genes would be expected to fall into each pathway by chance alone. 3. Results 3.1. Comparison of Normalization Methods 3.1.1. Probe Set Generation For each of the two datasets (heart, brain), a total of 15 probe sets was generated. As summarized in Figure 1, these probe sets were generated by applying each of the four GenomeStudio normalization strategies (Average, Cubic Spline, Quantile, Rank Invariant) or the No Normalization option to background-corrected data, followed by each of the three analytical approaches (GenomeStudio, GeneSpring, Max Cover (α,β)-FS). These probe sets were then filtered to remove probes that returned a detection p-value above 0.01 in both conditions in order to eliminate probes at background levels. Both datasets showed generally small expression changes (<2-fold), with only 2.1% and 0.4% of changes being over 2-fold in the heart and brain datasets, respectively. Irrespective of the normalization strategy used, probe sets generated from the brain arrays contained a smaller number of probes than those from the heart arrays (Table 1, Figure 2), consistent with fewer gene expression changes in the brain. microarrays-02-00131-t001_Table 1Table 1 Concordance in probe sets generated by different normalization strategies. The data are presented as the means of the number of overlapping probes between each possible pairwise comparison of the five normalization strategies, with the means of the percentage overlaps for the same comparisons in parentheses. No Normalization Average Cubic Spline Quantile Rank Invariant Heart Dataset GenomeStudio 503 (88.2) 760 (80.4) 738 (83.3) 787 (78.8) 791 (74.5) GeneSpring 724 (73.6) 1,235 (78.0) 1,374 (78.1) 1,375 (78.3) 1,324 (77.3) Max Cover (α,β)-FS 781 (71.3) 1,181 (76.8) 1,282 (78.0) 1,278 (78.0) 1,231 (77.1) Brain Dataset GenomeStudio * 44 (82.4) 93 (70.2) 95 (56.9) 85 (67.2) GeneSpring * 134 (57.9) 248 (71.5) 248 (70.0) 209 (64.8) Max Cover (α,β)-FS * 190 (43.8) 402 (66.3) 401 (66.4) 320 (58.6) * Excluded from comparisons to avoid bias. 3.1.2. Effects of the Different Normalization Strategies on Probe Set Concordance In order to determine the influence of normalization on probe set concordance (defined in Section 2.3.3), for each particular analytical approach we performed pairwise comparisons between the different probe sets generated using each of the five normalization strategies. For example, the individual probe sets generated by GeneSpring for each of the five different normalization strategies were compared to each other, giving a total of 10 comparisons. This was also done for each of the other two analytical approaches (GenomeStudio, Max Cover (α,β)-FS), giving a total of 30 comparisons. Figure 2 Comparison of concordance between different analytical approaches for each normalization strategy. Concordance of probe sets generated by different analytical approaches was assessed for (a) heart array data and (b) brain array data. Numbers of fully or partially concordant or discordant probes are shown on the charts, with the total number of probes generated by each combination shown below. In general, irrespective of which analytical approach or dataset was used, the No Normalization strategy identified relatively small probe sets. In the case of the heart dataset, these were usually highly concordant with the sets identified by other normalization methods (Table 1, Table S1). This suggests that the omission of a normalization step yields fewer false positives, but at the cost of more false negatives, making it less effective than the other strategies for class comparison, although still of possible value for biomarker discovery. However, in the case of the brain dataset, the use of the No Normalization strategy gave extremely small probe sets for all analytical approaches, sometimes containing only a single probe. This grossly distorted the calculations of the concordance between the various normalization strategies for the brain dataset. Therefore, this strategy was not included in the concordance calculations for the brain dataset presented in Table 1. On average, all four normalization methods (i.e., Average, Cubic Spline, Quantile and Rank Invariant) gave comparable levels of concordance; however, the Average method produced smaller probe sets with a generally lower mean concordance in the brain dataset (Table 1). Similar trends were observed when a multiple testing correction was applied to GenomeStudio analysis of the heart dataset, with the No Normalization strategy producing far smaller probe sets than the four normalization methods. As for non-corrected data, concordance was high between the different normalization methods, with Average producing the smallest probe sets and Rank Invariant producing the largest (Table S2). When the multiple testing correction was applied to GenomeStudio analysis of the brain dataset, no probes were identified as having significantly altered expression, irrespective of which normalization strategy was used. As there have been questions raised in the literature over the suitability of the GenomeStudio background correction procedure [15,17,18], we generated probe lists from non-background corrected data using the five normalization strategies in combination with GenomeStudio differential expression analysis and repeated the pairwise comparisons described above. In almost all cases, omission of background correction gave rise to larger probe sets than those obtained when background correction was applied. Overall, percentage concordance between different normalization methods showed similar trends, whether data were background corrected or not (Table S3). Next, concordance was assessed across the different analytical approaches. 3.2. Comparison of Analytical Approaches 3.2.1. Definition of Concordance for Comparisons of Analytical Approaches For each particular normalization strategy (including the No Normalization strategy), we compared the concordance of the probe sets identified by each of the three different analytical approaches. (This is distinct from the concordance assessed by pairwise comparisons of normalization strategies for a single analytical approach, considered above). For each normalization strategy, a probe was classed as having “full concordance” if it was identified by all three analytical approaches, “partial concordance” if identified by two of the three approaches or “no concordance” if identified by only one approach. 3.2.2. Effects of the Different Analytical Approaches on Probe Set Concordance Figure 2 highlights the considerable differences in both numbers and proportions of identified probes that can occur with the various methods. However, some general conclusions can be drawn. For both datasets, the numbers of probes identified when using the No Normalization method were much lower than those identified when using each of the four normalization strategies. All four normalization strategies generally produced similar levels of concordance, again with the exception of the Average strategy, which produced a lower proportion and number of fully concordant probes in the brain dataset than other strategies (blue sectors, Figure 2). Of the other normalization strategies, overall Quantile performed most strongly when considered across both datasets, based on the percentage and number of fully concordant probes. When considering analytical approaches, GenomeStudio gave the highest proportion of full concordance (blue sectors, Figure 2). However, this was largely because this approach produced smaller probe sets. GeneSpring generally gave the highest proportion of combined full and partial concordance (blue and yellow sectors, respectively, Figure 2). Max Cover (α,β)-FS gave the largest probe sets and, therefore, the greatest number and proportion of discordant probes (red sectors, Figure 2). Some of these may be false positives, but others may be real changes missed by other approaches. This is assessed more fully in the pathway investigations discussed below. 3.3. Comparison with Bioconductor Packages To determine how the results obtained using these approaches compare with those obtained using more flexible, yet computationally-demanding, tools available through the Bioconductor project, data processing and analysis of the heart and brain datasets was undertaken using two Bioconductor tools designed for analysis of Illumina microarrays: lumi and limma. In the absence of a multiple testing correction, lumi and limma both generated probe sets that were larger than those generated by any other approach for the heart dataset, and only the Max Cover (α,β)-FS approach returned larger probe sets for the brain dataset. The probe sets generated by lumi and limma were highly concordant with one another (>90% for both heart and brain datasets). For the heart dataset, the lumi and limma approaches both identified more than 90% of the probes found by all analytical combinations involving Cubic Spline, Quantile and Rank Invariant, with GenomeStudio analyses showing the greatest percentage concordance, though possibly due to the smaller size of GenomeStudio probe sets (Table S4). Concordance was slightly lower for the brain dataset, particularly for combinations involving the Max Cover (α,β)-FS approach; however, this may simply reflect the large size of probe sets generated using this method, as described above. 3.4. Comparison of Pathway Analysis Outcomes 3.4.1. Definition of Concordance for Comparisons of Enriched Pathways We next conducted KEGG pathway enrichment analysis using DAVID for the 12 different gene sets generated for each dataset by using each of the four normalization strategies (Average, Cubic Spline, Quantile, Rank Invariant) with each of the three analytical approaches (GenomeStudio, GeneSpring, Max Cover (α,β)-FS). For each normalization strategy, we determined the number of concordant pathways across the different approaches, where “concordant” denotes pathways common to two or more approaches (Table 2). microarrays-02-00131-t002_Table 2Table 2 Comparison of outcomes from pathway enrichment analysis. Table displays the total number of pathways identified as enriched in gene lists generated using different combinations of normalization strategies and analytical approaches. Numbers of concordant pathways are shown in parentheses. Heart Dataset Average Cubic Spline Quantile Rank Invariant GenomeStudio 14 (12) 11 (8) 16 (10) 18 (11) GeneSpring 24 (22) 18 (16) 16 (13) 18 (17) Max Cover (α,β)-FS 18 (18) 20 (16) 19 (15) 19 (18) Brain Dataset Average Cubic Spline Quantile Rank Invariant GenomeStudio 0 (0) 2 (2) 3 (2) 3 (3) GeneSpring 2 (0) 2 (2) 2 (2) 3 (2) Max Cover (α,β)-FS 4 (0) 4 (2) 5 (2) 6 (3) 3.4.2. Effects of Different Normalization and Analytical Approaches on Pathway Analysis The pathways identified as enriched were strongly affected by both normalization strategy and analytical approach and also varied considerably between the two datasets. For all analytical approaches in both datasets, Rank Invariant normalization generally yielded both more pathways and more concordant pathways (Table 2). Unexpectedly, although (as described above) Max Cover (α,β)-FS generated probe sets with the most discordant probes (Figure 2), it generally yielded both more pathways and more concordant pathways than the other analytical approaches (Table 2). Of the other two approaches, GeneSpring identified more concordant pathways than GenomeStudio. 3.4.3. Probe Set Concordance and Outcomes of Pathway Analysis It was observed that approaches that generally show high probe set concordance can still fail to identify pathways of probable importance. One example was the “insulin signalling pathway”. Diabetes is one of the classical triad of symptoms seen at advanced stages of the human iron overload disorder hemochromatosis and iron overload arising due to various causes has been associated with insulin perturbations and type 2 diabetes [38,39]. Furthermore, the insulin signalling pathway has been observed to alter in association with oxidative stress and cell death in other mouse models of iron overload [40,41]. This pathway was identified as significantly enriched in the heart dataset when using all four normalization strategies in combination with the Max Cover (α,β)-FS approach (>1.9-fold enrichment, p < 0.01). In contrast, approaches that yielded relatively few discordant probes, such as Quantile or Rank Invariant in combination with GeneSpring, failed to identify this potentially important pathway as significantly enriched. Conversely, approaches that generally show high probe set discordance may sometimes identify pathways of potential importance not picked up by other approaches. For example, analysis of gene lists from the heart dataset generated using Average normalization with Max Cover (α,β)-FS, which had a relatively large number of discordant probes, identified the pathway “acute myeloid leukaemia” (2.6-fold enrichment, p = 0.009). This pathway was not detected by other approaches, yet is a potential true positive of probable clinical mechanistic relevance, since there is evidence for a relationship between acute myeloid leukaemia and gene mutations associated with hemochromatosis [42]. The Max Cover (α,β)-FS approach, therefore, was not only successful in identifying most of the concordant probes identified by the other analytical approaches, but also identified additional discordant probes of probable relevance. 4. Discussion This study demonstrates that, when expression changes are modest, the choice of normalization and analysis algorithms for Illumina microarray data can have a substantial effect on identification of altered genes and pathways. This may considerably influence decisions about which molecular systems are selected for further investigation and the direction of future research. The main findings are summarized here and discussed in detail below. - The No Normalization strategy may be poorly suited to discovery-driven research. - Background correction in GenomeStudio generally led to a reduction in the size of probe sets, but did not affect percentage concordance. - Of the four Illumina GenomeStudio normalization strategies, Cubic Spline, Quantile and Rank Invariant generally gave comparable outcomes for a particular analytical approach, although performance sometimes varied between the datasets. (Average did not perform as well, particularly in the brain dataset.) - Different analytical approaches (GenomeStudio, GeneSpring, Max Cover (α,β)-FS) often generated quite different probe sets that were enriched for different pathways, even when using the same normalization strategy. - Most combinations of normalization strategy and analytical approach compared favourably with the Bioconductor tools lumi and limma. The results showed that optimal combinations of normalization strategies and analytical approaches may vary considerably for different datasets in ways that are not always readily predictable. It was not possible to choose one combination that works best all the time. It is important to test combinations of different approaches to improve robustness and, wherever feasible, to validate outcomes by alternative methods. While a number of studies have evaluated the performance of the Illumina microarray platform compared to other platforms [7,8,9,10,11], there is little information on how the choice of different normalization and analysis approaches for Illumina data affects outcomes. One previous study investigated a range of different normalization strategies specifically using Illumina human microarray data [15], but incorporated various approaches only available through R/Bioconductor packages and did not assess the effects of different combinations of normalization strategy and analytical approach on pathway outcomes. Understanding the effects of using different approaches may be particularly important when analysing data involving subtle expression changes, where even minor differences in the scaling of raw data may lead to data adjustments that are comparable in size to the expression changes being investigated. This factor, combined with differences in the way that data are subsequently compared, could considerably influence the identification of “differentially expressed” genes. The findings suggest that some form of normalization should be applied, since the No Normalization strategy resulted in the generation of very small probe sets, as would be expected, since data not adjusted for technical variation are likely to show high variability. All four normalization strategies (i.e., Average, Cubic Spline, Quantile and Rank Invariant) performed well in most analyses. Except in the case of Cubic Spline and Quantile normalization, the high degree of concordance observed when using these methods is unlikely to be an artefact arising from similarities in the normalization procedures, as the various strategies use fundamentally different mathematical approaches. The variability in probe sets generated by different normalization strategies makes it difficult to recommend one that will invariably perform best for any analytical approach and any dataset. For optimal performance for discovery-driven research, we would suggest comparing all four normalization strategies for each new investigation. Similarly, it was shown that the same normalization strategy can give very different outcomes when used with different analytical approaches. The most accessible analysis software for Illumina users, the proprietary Illumina GenomeStudio, does well in that most of the probes it identified were concordant with the other methods investigated, including the Bioconductor tools lumi and limma. However, it typically generated substantially smaller probe sets than the other approaches and so may miss a considerable number of important genes in some datasets. GeneSpring generally identified a higher total proportion of fully and partially concordant probes than other approaches. Max Cover (α,β)-FS also generally identified high numbers of fully and partially concordant probes and in addition found further probes not identified by other approaches. While some of these additional probes may be false positives, some appear to represent real changes that help identify additional pathways of biological relevance. Max Cover (α,β)-FS has a very different mathematical basis from the analytical approaches based on statistical significance (GenomeStudio, GeneSpring). While this may decrease the numbers of fully concordant probes in comparisons of these approaches, those probes that are jointly identified by such very different methods are more likely to represent robust findings. Therefore, in addition to recommending that more than one normalization strategy be used, the use of more than one analytical approach, preferably not restricted solely to statistical testing, is also recommended. The findings also suggest that important pathways and processes may be overlooked if only one approach is used to analyse differential gene expression, further highlighting the need for using combinations of approaches. As there were often considerable differences between the findings for the two datasets, it is not possible to recommend a single combination of normalization strategy and analytical approach that will be optimal in all circumstances, particularly since the two datasets examined here were relatively similar (different tissues from the same model) and differences may be even greater for other datasets. Due to individual variability, there may be no “correct” approach—statistical methods may do better in some sample sets, in particular those with low variability, but may miss useful findings in others. The optimum combination of methods will also vary depending on whether the main aim is to minimize false positives, as required for class prediction aimed at biomarker discovery, or to maximize true positives and minimize false negatives, as required in class comparison or class discovery studies. The use of multiple approaches to identify robust changes differs from more conventional microarray analysis pipelines that utilize multiple testing corrections to avoid false positive findings; however, in this case, we believe it is appropriate. This point is particularly relevant since the GenomeStudio software does not allow the removal of low signal probes (representing non-expressed genes) prior to differential expression analysis, thereby increasing the burden of multiple testing. In addition, Max Cover (α,β)-FS appears to yield important findings of biological relevance; yet, as a non-statistical approach, it is not amenable to multiple testing correction. It would be unfortunate if this valuable complementary method were to be discarded solely on these grounds. Reference RNA that contains many transcripts of known concentration would be ideal for testing the ability of different approaches to identify true positives and true negatives. However, as far as we could determine, reference RNA of this type is not commercially available. Instead, experiments seeking to evaluate reproducibility across platforms or across processing and analysis approaches have relied on either titrations of two distinct RNA reference samples (e.g., universal RNA and brain RNA) [7] or “spike-in” experiments, where genes normally absent from the genome under investigation (e.g., bacterial or viral genes) are added at known concentrations [17,18]. While such experiments provide RNA pools where relative levels of certain transcripts are known a priori, they generally result in relatively large fold differences between samples. As our study specifically focused on datasets with small fold changes, it was not feasible to adopt a similar approach in our evaluation. Similarly, the small magnitude of most of the fold changes under investigation made it infeasible to test many results by quantitative reverse transcription PCR (qRT-PCR), which is often employed as a method for validating microarray findings. Other groups have reported that fold changes of less than 1.4 by microarray generally show poor correlation with qRT-PCR [43]. While we have used this technique previously to successfully validate some of the most robust findings in the brain dataset [29] and heart dataset (Johnstone et al., unpublished data), these specific changes exceeded the 1.4-fold threshold. Therefore, one important limitation of the study is that the accuracy of different outcomes could not be directly assessed and using concordance to estimate accuracy may not always give a true picture. While outside the scope of the present study, future research could compare microarray results obtained using different analytical approaches with other sensitive multiplex or transcriptome-wide technologies, such as other array platforms, RNA-seq, NanoString or Fluidigm. However, it is important to note that human and other technical errors will affect quantitative differential expression analysis by any method, and any comparison requires that the analysis methods for the comparison technology have been shown to be accurate for low fold changes. As far as we are aware, this has not yet been achieved. For example, RNA-seq is biased towards high expression transcripts, so the accuracy of differential expression determinations will vary depending on the expression levels of the transcript. Identifying probes as differentially expressed by two or three different methods and detecting enrichment of molecular pathways of strong biological relevance provides some assurance in the accuracy of the findings, as noted above. Also, the strong performance of particular approaches with respect to identifying concordant probes for two different datasets suggests a high degree of reliability in generating robust probe sets. Some of the issues addressed in this study may be circumvented by using larger replicate numbers or more sophisticated analytical algorithms. However, even when using high end software packages, consideration should still be given as to how different computational approaches affect study outcomes for different datasets [15]. Furthermore, many researchers lack the expertise to use tools such as lumi [12] or limma [13] or other Bioconductor packages, which require knowledge of the R programming language. For these reasons, it is important to understand and take into account the strengths and limitations of Illumina-recommended protocols, such as GenomeStudio and GeneSpring, for normalization and differential expression analysis. The findings should not be interpreted as implying that the Illumina platform and software give invalid or incorrect results. Probe sets identified by the GenomeStudio approach showed a high level of concordance with the other approaches, irrespective of the dataset and normalization strategies. However, our findings do indicate that outcomes can be further improved by using other analytical approaches. Most of the issues raised here are not unique to the Illumina platform. On other platforms, normalization and analysis methods can affect precision, sensitivity and other factors, and a method that is optimal in one context may be problematic in others [8,44]. The bead technology of Illumina arrays provides strong internal technical replication that is likely to be particularly important for detecting small expression changes. The platform successfully identified gene expression changes of high probable relevance in our study and appears likely to be appropriate for studies involving small expression changes, provided suitable normalization and analytical strategies are used. 5. Conclusions In conclusion, this study has identified a range of potential pitfalls in analysing low expression fold-change datasets and highlights the need for future studies using reference datasets of known positives. While these issues are particularly relevant for datasets where expression changes are expected to be modest, many similar considerations are likely to apply for datasets where most gene expression changes are large, since these will usually still also contain some genes of biological interest with small expression changes. Important effects may be overlooked if there is a habitual routine of using only one approach to investigate all array datasets in a laboratory or commercial testing service. The findings presented here provide guidelines to help researchers optimize outcomes when working with datasets involving small expression changes. Notably, it is proposed that microarray data should be routinely subjected to alternative normalization and analysis procedures and comparisons made between these to obtain more robust gene lists and pathway identifications. Acknowledgments This research was supported by the Australian National Health and Medical Research Council (NHMRC #572601), the Fremantle Hospital Medical Research Foundation (DT, RG, JO), the Hunter Medical Research Institute (RS, PM, EM) and the Australian Research Council Centre of Excellence in Bioinformatics (PM, DJ). JO was supported by an NHMRC Practitioner Fellowship, DT by an NHMRC Senior Research Fellowship, DJ by an NHMRC Early Career Fellowship. Conflict of Interest The authors declare no conflict of interest. Appendix Table S1 Pairwise comparisons of probe sets generated by different normalization strategies. Data are presented as the number of overlapping probes between each possible pairwise comparison of the five normalization strategies, with the percentage overlaps for the same comparisons in parentheses. Heart–GenomeStudio No Norm Average Cubic Spline Quantile Rank Invariant No Norm X 548 (58.0) 468 (52.8) 486 (48.6) 510 (48.0) Average 548 (96.1) X 775 (87.5) 845 (84.6) 872 (82.1) Cubic Spline 468 (82.1) 775 (82.0) X 873 (87.4) 836 (78.7) Quantile 486 (85.3) 845 (89.4) 873 (98.5) X 945 (89.0) Rank Invariant 510 (89.5) 872 (92.3) 836 (94.4) 945 (74.2) X Heart–GeneSpring No Norm Average Cubic Spline Quantile Rank Invariant No Norm X 821 (56.6) 696 (45.1) 694 (45.1) 685 (45.5) Average 821 (83.4) X 1,241 (80.5) 1,241 (80.6) 1,224 (81.3) Cubic Spline 696 (70.7) 1,241 (85.5) X 1,509 (98.1) 1,373 (91.2) Quantile 694 (70.5) 1,241 (85.5) 1,509 (97.9) X 1,374 (91.2) Rank Invariant 685 (69.6) 1,224 (84.4) 1,373 (89.0) 1,374 (89.3) X Heart–Max Cover (α,β)-FS No Norm Average Cubic Spline Quantile Rank Invariant No Norm X 870 (56.6) 759 (46.2) 752 (45.9) 742 (46.5) Average 870 (79.5) X 1,297 (78.9) 1,288 (78.6) 1,268 (79.5) Cubic Spline 759 (69.3) 1,297 (84.4) X 1,616 (98.6) 1,456 (91.3) Quantile 752 (68.7) 1,288 (83.8) 1,616 (98.3) X 1,456 (91.3) Rank Invariant 742 (67.8) 1,268 (82.5) 1,456 (88.6) 1,456 (88.8) X Brain–GenomeStudio Average Cubic Spline Quantile Rank Invariant Average X 44 (33.1) 44 (26.3) 43 (33.9) Cubic Spline 44 (83.0) X 132 (79.0) 104 (81.9) Quantile 44 (83.0) 132 (99.2) X 109 (85.8) Rank Invariant 43 (81.1) 104 (78.2) 109 (65.3) X Brain–GeneSpring Average Cubic Spline Quantile Rank Invariant Average X 145 (41.8) 145 (40.8) 111 (34.4) Cubic Spline 145 (62.8) X 341 (96.1) 258 (79.9) Quantile 145 (62.8) 341 (98.3) X 259 (80.2) Rank Invariant 111 (48.1) 258 (74.4) 259 (73.0) X Brain–Max Cover (α,β)-FS Average Average Cubic Spline Quantile Rank Invariant Cubic Spline X 213 (35.1) 209 (34.6) 149 (27.2) Quantile 213 (49.0) X 588 (97.4) 406 (74.2) Rank Invariant 209 (48.0) 588 (96.9) X 406 (74.2) Average 149 (34.3) 406 (66.9) 406 (67.2) X Table S2 Pairwise comparisons of probe sets generated by different normalization strategies, with multiple testing correction. Data are presented as the number of overlapping probes between each possible pairwise comparison of the five normalization strategies, with the percentage overlaps for the same comparisons in parentheses. Heart–GenomeStudio No Norm Average Cubic Spline Quantile Rank Invariant No Norm X 17 (34.0) 16 (28.1) 16 (26.2) 17 (21.5) Average 17 (100) X 47 (82.5) 48 (78.7) 49 (62.0) Cubic Spline 16 (94.1) 47 (94.0) X 57 (93.4) 57 (72.2) Quantile 16 (94.1) 48 (96.0) 57 (100) X 60 (75.9) Rank Invariant 17 (100) 49 (98.0) 57 (100) 60 (98.4) X Table S3 Pairwise comparisons of probe sets generated by different normalization strategies, with no background correction. Data are presented as the number of overlapping probes between each possible pairwise comparison of the five normalization strategies, with the percentage overlaps for the same comparisons in parentheses. Heart–GenomeStudio No Norm Average Cubic Spline Quantile Rank Invariant No Norm X 689 (50.6) 621 (48.6) 626 (47.9) 648 (43.2) Average 689 (92.7) X 1,146 (89.7) 1,183 (90.4) 1,242 (82.9) Cubic Spline 621 (83.6) 1,146 (80.2) X 1,249 (95.5) 1,185 (79.1) Quantile 626 (84.3) 1,183 (86.9) 1,249 (97.7) X 1,225 (81.7) Rank Invariant 648 (87.2) 1,242 (91.3) 1,185 (92.7) 1,225 (93.7) X Brain–GenomeStudio Average Cubic Spline Quantile Rank Invariant Average X 61 (33.0) 61 (30.3) 56 (45.2) Cubic Spline 61 (82.4) X 181 (90.0) 113 (91.1) Quantile 61 (82.4) 181 (97.8) X 114 (91.9) Rank Invariant 56 (75.7) 113 (61.1) 114 (56.7) X Table S4 Comparison of probe sets generated by different combinations of the normalization strategy and analytical approach, with probe sets generated by the Bioconductor packages, lumi and limma. Heart Dataset vs. Lumi (2,239 probes) vs. Limma (2,107 probes) Number Concordant Number Discordant % Concord Number Concordant Number Discordant % Concord GenomeStudio No Norm 551 19 96.7 535 35 93.9 Average 935 10 98.9 922 23 97.6 Cubic Spline 884 2 99.8 876 10 98.9 Quantile 997 2 99.8 989 10 99.0 Rank Invariant 1,060 2 99.8 1,051 11 99.0 GeneSpring No Norm 828 156 84.1 820 164 83.3 Average 1,371 80 94.5 1,366 85 94.1 Cubic Spline 1,512 30 98.1 1,508 34 97.8 Quantile 1,507 32 97.9 1,507 32 97.9 Rank Invariant 1,460 46 96.9 1,458 48 96.8 Max Cover (α,β)-FS No Norm 900 195 82.2 885 210 80.8 Average 1,382 155 89.9 1,365 172 88.8 Cubic Spline 1,532 112 93.2 1,522 122 92.6 Quantile 1,530 109 93.3 1,517 122 92.6 Rank Invariant 1,480 115 92.8 1,464 131 91.8 Brain Dataset vs. Lumi (488 probes) vs. Limma (420 probes) Number Concordant Number Discordant % Concord Number Concordant Number Discordant % Concord GenomeStudio No Norm 1 0 100 1 0 100 Average 47 6 88.7 43 10 81.1 Cubic Spline 128 5 96.2 116 17 87.2 Quantile 157 10 94.0 142 25 85.0 Rank Invariant 118 9 92.9 107 20 84.3 GeneSpring No Norm 1 3 25.0 1 3 25.0 Average 161 70 69.7 151 80 65.4 Cubic Spline 313 34 90.2 309 38 89.0 Quantile 316 39 89.0 311 44 87.6 Rank Invariant 271 52 83.9 261 62 80.8 Max Cover (α,β)-FS No Norm 1 11 8.3 1 11 8.3 Average 168 267 38.6 160 275 36.8 Cubic Spline 298 309 49.1 280 327 46.1 Quantile 299 305 49.5 283 321 46.9 Rank Invariant 249 298 45.5 240 307 43.9 ==== Refs References 1. Michael K.L. Taylor L.C. Schultz S.L. Walt D.R. Randomly ordered addressable high-density optical sensor arrays Anal. Chem. 1998 70 1242 1248 10.1021/ac971343r 9553489 2. Oliphant A. Barker D.L. Stuelpnagel J.R. Chee M.S. BeadArray technology: Enabling an accurate, cost-effective approach to high-throughput genotyping Biotechniques 2002 56–58 60 61 3. Fan J.B. Yeakley J.M. Bibikova M. Chudin E. Wickham E. Chen J. Doucet D. Rigault P. Zhang B. Shen R. A versatile assay for high-throughput gene expression profiling on universal array matrices Genome Res. 2004 14 878 885 10.1101/gr.2167504 15123585 4. Gunderson K.L. Kruglyak S. Graige M.S. Garcia F. Kermani B.G. Zhao C. Che D. Dickinson T. Wickham E. Bierle J. Decoding randomly ordered DNA arrays Genome Res. 2004 14 870 877 10.1101/gr.2255804 15078854 5. Kuhn K. Baker S.C. Chudin E. Lieu M.H. Oeser S. Bennett H. Rigault P. Barker D. McDaniel T.K. Chee M.S. A novel, high-performance random array platform for quantitative gene expression profiling Genome Res. 2004 14 2347 2356 10.1101/gr.2739104 15520296 6. Stokes T.H. Han X. Moffitt R.A. Wang M.D. Extending Microarray Quality Control and Analysis Algorithms to Illumina Chip Platform Proceedings of the IEEE 29th Annual International Conference Lyon, France 22–26 August 2007 4637 4640 7. Shi L. Reid L.H. Jones W.D. Shippy R. Warrington J.A. Baker S.C. Collins P.J. de Longueville F. Kawasaki E.S. Lee K.Y. The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements Nat. Biotechnol. 2006 24 1151 1161 16964229 8. Shippy R. Fulmer-Smentek S. Jensen R.V. Jones W.D. Wolber P.K. Johnson C.D. Pine P.S. Boysen C. Guo X. Chudin E. Using RNA sample titrations to assess microarray platform performance and normalization techniques Nat. Biotechnol. 2006 24 1123 1131 16964226 9. Chen J.J. Hsueh H.M. Delongchamp R.R. Lin C.J. Tsai C.A. Reproducibility of microarray data: A further analysis of Microarray Quality Control (MAQC) data BMC Bioinform. 2007 8 412 10.1186/1471-2105-8-412 10. Maouche S. Poirier O. Godefroy T. Olaso R. Gut I. Collet J.P. Montalescot G. Cambien F. Performance comparison of two microarray platforms to assess differential gene expression in human monocyte and macrophage cells BMC Genomics 2008 9 302 10.1186/1471-2164-9-302 18578872 11. Asare A.L. Gao Z. Carey V.J. Wang R. Seyfert-Margolis V. Power enhancement via multivariate outlier testing with gene expression arrays Bioinformatics 2009 25 48 53 19015138 12. Du P. Kibbe W.A. Lin S.M. Lumi: A pipeline for processing Illumina microarray Bioinformatics 2008 24 1547 1548 18467348 13. Smyth G.K. Linear models and empirical bayes methods for assessing differential expression in microarray experiments Stat. Appl. Genet. Mol. Biol. 2004 3 10.2202/1544-6115.1027 14. Bioconductor Available online:http://www.bioconductor.org (accessed on 13 May 2013) 15. Schmid R. Baum P. Ittrich C. Fundel-Clemens K. Huber W. Brors B. Eils R. Weith A. Mennerich D. Quast K. Comparison of normalization methods for Illumina BeadChip HumanHT-12 v3 BMC Genomics 2010 11 349 10.1186/1471-2164-11-349 20525181 16. Dunning M.J. Smith M.L. Ritchie M.E. Tavare S. Beadarray: R classes and methods for Illumina bead-based data Bioinformatics 2007 23 2183 2184 17586828 17. Dunning M.J. Barbosa-Morais N.L. Lynch A.G. Tavare S. Ritchie M.E. Statistical issues in the analysis of Illumina data BMC Bioinform. 2008 9 85 10.1186/1471-2105-9-85 18. Dunning M.J. Ritchie M.E. Barbosa-Morais N.L. Tavare S. Lynch A.G. Spike-in validation of an Illumina-specific variance-stabilizing transformation BMC Res. Notes 2008 18 10.1186/1756-0500-1-18 18710543 19. Workman C. Jensen L.J. Jarmer H. Berka R. Gautier L. Nielser H.B. Saxild H.H. Nielsen C. Brunak S. Knudsen S. A new non-linear normalization method for reducing variability in DNA microarray experiments Genome Biol. 2002 3 10.1186/gb-2002-3-9-research0048 20. Bolstad B.M. Irizarry R.A. Astrand M. Speed T.P. A comparison of normalization methods for high density oligonucleotide array data based on variance and bias Bioinformatics 2003 19 185 193 12538238 21. Reiner A. Yekutieli D. Benjamini Y. Identifying differentially expressed genes using false discovery rate controlling procedures Bioinformatics 2003 19 368 375 12584122 22. Rothman K.J. No adjustments are needed for multiple comparisons Epidemiology 1990 1 43 46 10.1097/00001648-199001000-00010 2081237 23. Bender R. Lange S. Adjusting for multiple testing—When and how? J. Clin. Epidemiol. 2001 54 343 349 10.1016/S0895-4356(00)00314-0 11297884 24. Cotta C. Sloper C. Moscato P. Evolutionary search of thresholds for robust feature set selection: Application to the analysis of microarray data Applications of Evolutionary Computing Raidl G.R. Springer Berlin, Germany 2004 21 30 25. Cotta C. Langston M.A. Moscato P. Combinatorial and algorithmic issues for microarray analysis Handbook of Approximation Algorithms and Metaheuristics Gonzalez T.F. Chapman & Hall/CRC London, UK 2007 74:1 74:14 26. Gomez Ravetti M. Moscato P. Identification of a 5-protein biomarker molecular signature for predicting Alzheimer’s disease PLoS One 2008 3 e3111 10.1371/journal.pone.0003111 18769539 27. Berretta R. Costa W. Moscato P. Combinatorial optimization models for finding genetic signatures from gene expression datasets Methods Mol. Biol. 2008 453 363 377 10.1007/978-1-60327-429-6_19 18712314 28. Rodriguez A. Hilvo M. Kytomaki L. Fleming R.E. Britton R.S. Bacon B.R. Parkkila S. Effects of iron loading on muscle: Genome-wide mRNA expression profiling in the mouse BMC Genomics 2007 8 379 10.1186/1471-2164-8-379 17949489 29. Johnstone D. Milward E.A. Genome-wide microarray analysis of brain gene expression in mice on a short-term high iron diet Neurochem. Int. 2010 56 856 863 10.1016/j.neuint.2010.03.015 20350576 30. Drake S.F. Morgan E.H. Herbison C.E. Delima R. Graham R.M. Chua A.C. Leedman P.J. Fleming R.E. Bacon B.R. Olynyk J.K. Iron absorption and hepatic iron uptake are increased in a transferrin receptor 2 (Y245X) mutant mouse model of hemochromatosis type 3 Am. J. Physiol. Gastrointest. Liver Physiol. 2007 292 G323 G328 16935854 31. Illumina (2008) GenomeStudio Gene Expression Module v1.0 User Guide Available online:http://support.illumina.com/documents/MyIllumina/c94519f7-9348-4308-a32f-b66ff3959e99/GenomeStudio_GX_Module_v1.0_UG_11319121_RevA.pdf (accessed on 15 May 2013) 32. Fayyad U.M. Irani K.B. Multi-Interval Discretization of Continuous-Valued Attributes for Classification Learning Proceedings of the 13th International Joint Conference on Artificial Intelligence Chambery, France 28 August–3 September 1993 Bajcsw R. Morgan Kaufmann San Francisco, CA, USA 1993 1022 1029 33. Ritchie M.E. Dunning M.J. Smith M.L. Shi W. Lynch A.G. BeadArray expression analysis using bioconductor PLoS Comput. Biol. 2011 7 e1002276 10.1371/journal.pcbi.1002276 22144879 34. Barbacioru C.C. Wang Y. Canales R.D. Sun Y.A. Keys D.N. Chan F. Poulter K.A. Samaha R.R. Effect of various normalization methods on applied biosystems expression array system data BMC Bioinform. 2006 7 533 10.1186/1471-2105-7-533 35. DAVID: Functional Annotation Result Summary Available online:http://david.abcc.ncifcrf.gov/ (accessed on 13 May 2013) 36. Dennis G. Jr. Sherman B.T. Hosack D.A. Yang J. Gao W. Lane H.C. Lempicki R.A. DAVID: Database for annotation, visualization, and integrated discovery Genome Biol. 2003 4 10.1186/gb-2003-4-9-r60 37. Huang D.W. Sherman B.T. Lempicki R.A. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources Nat. Protoc. 2009 4 44 57 19131956 38. Swaminathan S. Fonseca V.A. Alam M.G. Shah S.V. The role of iron in diabetes and its complications Diabetes Care 2007 30 1926 1933 10.2337/dc06-2625 17429063 39. Rajpathak S.N. Crandall J.P. Wylie-Rosett J. Kabat G.C. Rohan T.E. Hu F.B. The role of iron in type 2 diabetes in humans Biochim. Biophys. Acta 2009 1790 671 681 10.1016/j.bbagen.2008.04.005 18501198 40. Cooksey R.C. Jouihan H.A. Ajioka R.S. Hazel M.W. Jones D.L. Kushner J.P. McClain D.A. Oxidative stress, beta-cell apoptosis, and decreased insulin secretory capacity in mouse models of hemochromatosis Endocrinology 2004 145 5305 5312 15308612 41. Huang J. Gabrielsen J.S. Cooksey R.C. Luo B. Boros L.G. Jones D.L. Jouihan H.A. Soesanto Y. Knecht L. Hazel M.W. Increased glucose disposal and AMP-dependent kinase signaling in a mouse model of hemochromatosis J. Biol. Chem. 2007 282 37501 37507 17971451 42. Viola A. Pagano L. Laudati D. D’Elia R. D’Amico M.R. Ammirabile M. Palmieri S. Prossomariti L. Ferrara F. HFE gene mutations in patients with acute leukemia Leuk Lymphoma 2006 47 2331 2334 17107905 43. Morey J.S. Ryan J.C. van Dolah F.M. Microarray validation: Factors influencing correlation between oligonucleotide microarrays and real-time PCR Biol. Proced. Online 2006 8 175 193 10.1251/bpo126 17242735 44. Tefferi A. Bolander M.E. Ansell S.M. Wieben E.D. Spelsberg T.C. Primer on medical genomics. Part III: Microarray experiments and data analysis Mayo Clin. Proc. 2002 77 927 940 12233926

PMC005xxxxxx/PMC5003483.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020007microarrays-05-00007ArticleA Mismatch EndoNuclease Array-Based Methodology (MENA) for Identifying Known SNPs or Novel Point Mutations Comeron Josep M. 12*Reed Jordan 1Christie Matthew 1Jacobs Julia S. 1Dierdorff Jason 1Eberl Daniel F. 12Manak J. Robert 123*Alekseyev Yuriy Academic EditorLiu Gang Academic Editor1 Department of Biology, University of Iowa, Iowa City, IA 52242, USA; jordanreed.uiowa@gmail.com (J.R.); matthew.c.christie@gmail.com (M.C.); julie-jacobs@uiowa.edu (J.S.J.); jason-dierdorff@uiowa.edu (J.D.); daniel-eberl@uiowa.edu (D.F.E.)2 Graduate Program in Genetics, University of Iowa, Iowa City, IA 52242, USA3 Department of Pediatrics, University of Iowa, Iowa City, IA, 52242, USA* Correspondence: josep-comeron@uiowa.edu (J.M.C.); john-manak@uiowa.edu (J.R.M.); Tel.: + 319-335-0180 (J.R.M.); Fax: 319-335-1069 (J.R.M.)05 4 2016 6 2016 5 2 708 2 2016 30 3 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Accurate and rapid identification or confirmation of single nucleotide polymorphisms (SNPs), point mutations and other human genomic variation facilitates understanding the genetic basis of disease. We have developed a new methodology (called MENA (Mismatch EndoNuclease Array)) pairing DNA mismatch endonuclease enzymology with tiling microarray hybridization in order to genotype both known point mutations (such as SNPs) as well as identify previously undiscovered point mutations and small indels. We show that our assay can rapidly genotype known SNPs in a human genomic DNA sample with 99% accuracy, in addition to identifying novel point mutations and small indels with a false discovery rate as low as 10%. Our technology provides a platform for a variety of applications, including: (1) genotyping known SNPs as well as confirming newly discovered SNPs from whole genome sequencing analyses; (2) identifying novel point mutations and indels in any genomic region from any organism for which genome sequence information is available; and (3) screening panels of genes associated with particular diseases and disorders in patient samples to identify causative mutations. As a proof of principle for using MENA to discover novel mutations, we report identification of a novel allele of the beethoven (btv) gene in Drosophila, which encodes a ciliary cytoplasmic dynein motor protein important for auditory mechanosensation. microarraymismatchendonucleaseSNP detectiongenetic variationdisease mutation ==== Body 1. Introduction Current methodologies to identify known single nucleotide polymorphisms (SNPs) include microarray analysis (either based on differential hybridization strategies or hybridization coupled with DNA polymerase activity, as in the Affymetrix (Santa Clara, CA, USA) and Illumina (San Diego, CA, USA) platforms, respectively), Taqman® assays (Roche Diagnostics, Indianapolis, IN, USA), MassARRAY®/iPLEX® (Agena Bioscience, San Diego, CA, USA) single-base extension after amplification, mismatch endonuclease-based detection followed by suitable separation methods, or DNA sequencing (either directed or whole-genome, as in medical resequencing or next generation sequencing strategies) [1,2,3,4,5,6,7,8,9]. However, the aforementioned microarray-based, Taqman®, MassARRAY® and mismatch endonuclease strategies are only able to interrogate SNPs whose identity is already known. Moreover, most of these methods are not easily scalable to query thousands of SNPs at a time with high accuracy. For instance, the mismatch endonuclease-based detection approach involves PCR to amplify target DNA fragments from both mutant and wild-type reference DNA, enzyme treatment to cleave hybrid heteroduplexes, and detection of potential variants using conventional gel electrophoresis or high-performance liquid chromatography (HPLC). On the other hand, direct sequencing of large genomic regions can identify novel SNPs (and other genomic variation) but oftentimes this strategy is time-consuming and can be expensive and inefficient when genotyping known variants. We thus wanted to develop an economic methodology that could effectively identify both types of mutations (novel and previously discovered). Therefore, we combined the enzymatic efficiency and specificity of the mismatch endonuclease strategy with a tiling microarray hybridization strategy to produce a platform, which we call Mismatch EndoNuclease Array (MENA). In particular, we focused on the Surveyor® endonuclease CEL II which is part of the CEL nuclease family derived from celery. CEL II specifically targets heteroduplex DNA (mismatched base pairs and small indels) and produces a double-strand cleavage at the mismatch site [8,10,11,12,13]. The general MENA strategy (see Figure 1) relies on: (1) hybridizing a Cy3-labeled genomic DNA sample to a tiling microarray that interrogates one or more large genomic regions of interest in a sequenced genome; (2) scanning all features on the array to quantify hybridized DNA samples; (3) treating the array with a mismatch endonuclease (Surveyor®) that scans for and cleaves mismatches in heteroduplex DNA, reducing the length of heteroduplex formed by genomic fragments and array features when mismatches exist; and (4) re-scanning the array to obtain signal intensities post-treatment. Intensities of the post-digestion features are then compared with the intensities of the pre-digestion features. Probes with hybridized DNA including mismatches (SNPs and/or small indels) are expected to show a specific drop in post-treatment signal relative to pre-treatment signal that is more significant than for probes with no mismatches. It is important to note that traditional SNP array methods rely on a subtle reduction in hybridization strength (and thus signal) between DNA fragments from the sample and array probes when mismatches are present relative to perfect match. In contrast, MENA is expected to generate much more pronounced reductions in signal because genomic fragments are end-labeled and Surveyor® treatment will shorten the heteroduplex via double-strand cleavage (acting on the labeled DNA sample as well as on the array feature or probe) and thus liberate signal from the array. Array features exhibiting the greatest relative drop in signal can validate the presence of a known SNP or, alternatively, identify a new candidate genetic variant that can be earmarked for validation using PCR and Sanger sequencing. Here, we report 99% accuracy of the MENA platform in calling previously identified SNPs, as well as successful identification of novel, unknown genetic variation such as single nucleotide variants and indels. Using MENA, we also identify a novel mutation in a dynein motor protein gene required for hearing in Drosophila melanogaster, a mutation that was previously missed using standard Sanger sequencing. MENA is thus an effective alternative methodology for mapping unknown mutations localized to a genomic interval. Finally, given the rise in whole genome sequencing efforts, oftentimes with low sequence read coverage, we envision that this platform could play a significant role in SNP confirmation for these studies. 2. Materials and Methods 2.1. Array Designs All designs were devised for the Roche NimbleGen 385K microarray. Design files (ndf and pos) for each array are available on request. Design strategies and genomic resources are articulated in the Results section. 2.2. Protocol for Labeling DNA, MENA Array Processing One microgram of genomic DNA was labeled with Cy3-coupled random primers as described in the NimbleGen Arrays User’s Guide-CGH Analysis (version 5.1) using the labeling protocol for 385K feature arrays. Sixty-four micrograms of labeled DNA was lyophilized, followed by resuspension in the hybridization buffer including alignment oligo. Sixteen microliters was loaded into the array mixer (the chamber device that is affixed to the array in which hybridization of the labeled DNA occurs), and the sample was hybridized for up to 3 days. The array was washed in 42 °C Wash Buffer 1 for 2 min, followed by Wash buffer 2 for 1 min, and wash buffer 3 for 15 s. The array was then dried for 15–20 s on an ArrayIt slide dryer. Arrays were scanned on either the Roche NimbleGen MS 200 microarray scanner set to auto gain, or the Molecular Devices Axon 2.5 uM resolution microarray scanner, and the PMT value was recorded. Next, a new array mixer was placed on the dried, scanned array, and the array was placed on the NimbleGen Hybridization System set to 42 °C. Eighteen microliters of the Surveyor® nuclease master mix (see below) was then added to the mixer, and the array was mixed for up to 40 min. The array was then rewashed at 42 °C in Wash Buffers 1 to 3 days as described above, and dried on the ArrayIt slide dryer. The array was rescanned with the same PMT setting as the first scan. Array data was then processed in NimbleScan as per the NimbleGen Arrays User’s Guide. Surveyor® nuclease reaction master mix was prepared, as following, using the relevant components of the Surveyor® Mutation Detection Kit for Standard Gel Electrophoresis from Transgenomic, Inc. (now sold and registered by Integrated DNA Technologies (IDT), Inc., Coralville, IA, USA): 2.2 μL Surveyor® Nuclease S, 2.2 μL Surveyor® Enhancer S, 15.6 μL Surveyor® Reaction Buffer (25 mM·Tris-HCl, pH 9.0; 50 mM·KCl; 10 mM·MgCl2). For further dilutions of Surveyor®, the Dilution Buffer used was as follows: 50 mM·Tris-HCl (pH 7.5), 100 mM·KCl, 0.01% Triton X-100, 10 uM·ZnCl2, 5% Glycerol. 2.3. Drosophila Strains Flies carrying the l(2)k07109 chromosome from the Kiss collection [14] contain a lethal PlacW insertion at polytene location 25F1-2, as well as an incomplete PlacW element that is w- at the Fas3 gene locus in polytene section 36E. This chromosome fails to complement other btv alleles and deficiencies that define the btv locus [15]. We named this allele btv2. To maintain a non-lethal stock of btv2, we crossed the l(2)k07109 chromosome to a deficiency, Df(2L)cact255rv64, leaving the btv2 mutation hemizygous and the remainder of the chromosome freely recombining. We presume that over time, the deficiency has been lost, with the btv2 allele becoming homozygous. Because the background chromosome on which the Kiss lines were generated is not available, we used a different insertion line from the collection, l(2)k12913 containing a PlacW insertion in the rempA locus [16], as a proxy for the genetic background. To recover this reference DNA, we used flies of genotype rempAk12913/Df(2L)cact255rv64. 3. Results To test the feasibility of our strategy, we first designed a 385K Roche NimbleGen microarray that interrogates 128 human genome SNPs across 41 genes (see Supplemental Table S1). In this original design, the SNPs are interrogated by all four bases at the SNP position using four sets of oligonucleotides. Moreover, the SNP position region is tiled at 1 bp resolution, with oligonucleotides designed to position the SNP at every possible position along the oligo, plus 5 extra “buffer” probes on either side of each SNP that do not contain the SNP position. Thus, for a 30 m probe, 5 + 30 + 5 probes per strand and per SNP nucleotide are required for this design strategy, with 5 + 40 + 5 probes per strand per SNP nucleotide required for 40 mer probes, etc. Additionally, both forward and reverse strands are interrogated for each SNP region. This strategy was employed four different ways, using oligonucleotides that were 30, 40, 50 and 60 nucleotides long. Our initial experiments demonstrated that only 60 m probes were able to perform as anticipated under the experimental conditions used (see Experimental Section), so we redesigned the array using only 60 m, and in addition to once again employing the overall design strategy outlined above, we made sure to include four replicates for each probe set in order to test whether increased replication might have an effect on call accuracy. Our MENA array, therefore, was designed to employ three out of the four probe sets (with either an A, G, C or T interrogating a particular SNP, using multiple replicates) to expose a mismatch, with only one set (with multiple replicates) being complementary to the SNP, thus protecting it from the Surveyor® endonuclease. This strategy allowed us to test the efficacy of combining Surveyor® and microarrays to correctly genotype specific nucleotide positions under a wide variety of conditions and different base pair mismatches. We also investigated the genotyping accuracy when different numbers of partially overlapping (tiling) probes are used to infer pre- vs. post-digestion signal in order to maximize sensitivity and reliability. For example, the 128 SNPs were further split into three groups of SNPs of similar size, one that was analyzed at 1 bp resolution, a second that was analyzed at 2 bp resolution, and a third that was analyzed at 3 bp resolution. Based on knowledge generated in our original set of arrays, all probes were designed to be 60 bp (see below). We performed several initial experiments using a previously genotyped CEPH (1362-02) human genomic DNA sample (Supplementary Table S1). We labeled the DNA with Cy3 using the standard Roche NimbleGen labeling protocol (see Experimental Section), and we varied both amount of DNA hybed, length of time hybed, and amount of Surveyor® endonuclease used (Table 1). We then compared the pre- and post-digestion signals from the arrays to determine whether we could observe evidence of cleavage, and whether it was discriminate and showed specificity to heteroduplexes with mismatches. Once the pre and post digestion probe signal intensities were determined via the microarray scanner, we used an algorithm which we developed in house to reveal evidence of cleavage and assess whether MENA generated the correct SNP calls (accurate genotyping). We estimated the difference in relative signal due to Surveyor® treatment by measuring the signal pre- and post-treatment for each of the four possible probes, where all nucleotides of the probe are the same except for a single site where each one contains a different nucleotide. Importantly, we normalized these changes to allow for differences in signal intensity among probes as well as overall reduction in signal after treatment and washing. Our measure of pre- and post-treatment change in signal intensity (“Relative Signal Change” or RSC) for each nucleotide (RSCi with i = A, C, G, or T) is expected to be positive for probes with a nucleotide with perfect match to the sample while all other probes with mismatched nucleotide are expected to generate negative RSC values, with the sum of all four RSCi equal to zero (see two examples in Figure 2). In detail, (1) RSCi(i=A,C,G,T)=RSipre−RSipost= (Sipre−S¯pre)−(Sipost−S¯post) where pre- and post- indicate signal intensity pre- and post-treatment with Surveyor®, respectively, measured in Log2 units. Si indicates the signal for probes with nucleotide i, and S¯ is the average signal for all four probes. Note that Si would represent the average signal among all replicate probes in an array. The probe with the most positive RSCi is therefore the probe corresponding to the perfect match and informs our genotyping at the interrogated nucleotide site. We classify the genotyping of a SNP as correct or accurate when the highest positive value among the four RSCi corresponds to the nucleotide with perfect match and correlates with the known SNP in the sample. Note that if the region is tiled densely enough, RSCi can be estimated from multiple adjacent probes, all of which interrogate the same SNP thus adding robustness to the study (see below). Our analysis of several experimental and microarray conditions (Table 1) allowed us to discriminate cleavage and obtain high genotyping accuracy when using probes 60 nucleotides long. For 60-m probes, most experimental conditions allowed detection of a robust decrease in signal (negative RSC) for three out of the four sets of SNP-interrogating oligonucleotides while the fourth set showed a positive RSC, indicating that this latter set of probes contain a perfect match to the SNP present in the sample. Figure 2 shows representative results when genotyping two specific nucleotide sites using the MENA approach and 60-m probes. For these reference SNPs located in the BACH2 gene (rs9451298 and rs9359876), probes interrogating the correct nucleotide (T for rs9451298 and C for rs9359876) show a positive RSC while the probes interrogating the other three nucleotides show negative RSC values. The analysis of many (up to 27) different combinations of experimental conditions allowed us to obtain a general set of guidelines to obtain high genotyping accuracy when using MENA. First, we determined that 2.2 μL of the Surveyor® nuclease was optimal for achieving the maximal accuracy of calls (up to 99%) under a varied combination of amounts of DNA and digestion times (Table 1 and Figure 3). The use of limited (1.1 μL) Surveyor® gave less accurate results particularly when digestion times were also reduced. Not surprisingly, the use of higher amounts of Surveyor® (4.4, or 5.5 μL) required the reduction of digestion time (20 min instead of 40 min) to maintain high genotyping accuracy, likely to reduce unspecific cleavage. Either 40 or 50 min of digestion were able to achieve 99% accuracy when keeping all other parameters constant, including 64 μg of DNA hybridized, three days of hybridization and 2.2 μL of Surveyor®, suggesting that the enzyme digestion period could be somewhat flexible. Additionally, as little as 8 μg of amplified, labeled DNA (see Methods) could be used per hybridization to achieve this accuracy, using the same hybridization and digestion time, as well as Surveyor amount. Since labeling of the DNA usually results in an amplification of up to 70 fold, this suggests that as little as ~100 ng of starting genomic DNA might be sufficient to perform our analysis. Figure 3 shows genotyping accuracy for known SNPs under 14 different experimental conditions, varying the amount of DNA hybed, concentration of the Surveyor® endonuclease, and the time of digestion (Table 1). As shown, we obtained accuracies of 99% using the RSC algorithm described above for a number of conditions, thus indicating that MENA is not only highly accurate in genotyping SNPs, but also that the experimental conditions are fairly robust. This study also allowed us to draw additional conclusions about array design when using MENA (Figure 4). We observed that interrogating both strands always increases genotyping accuracy relative to doubling the number of probes for a single strand. We also observed that high genotyping accuracy can be obtained with low tiling overlap for probes: 95% accuracy is obtained with only five overlapping probes (double strand and four replicates). Finally, we were able to investigate the accuracy in detecting different nucleotide mismatches. We observed that accurate genotyping was lowest (albeit still very high) for probes interrogating “T” variants (with accurate genotyping of 96%). Having shown that MENA is highly efficient for detecting known SNPs, we next wanted to determine whether we could identify novel genomic variation. To this end, in our second array design we had also included the entire ~64 kb region of the human IRF6 gene (including the intergenic space 5′ and 3′ to the gene; Chr1: 208,021,000–208,085,000, hg18), tiling both strands at 2 bp resolution, with two replicates of both strands. In this first attempt to investigate whether MENA could detect candidate novel SNPs in the IRF6 region, we used the 1362-1 CEPH sample and employed MENA under optimal conditions that were identified as indicated above. In this case, probes showing the largest signal reduction (signal change, SC) are candidates for harboring differences relative to the reference (array) genome. Note that for this study we did not include all four possible nucleotides at all possible sites across the entire IRF6 region. Instead, we applied a sliding-window approach using probes that match the reference genome to describe SC along the complete region analyzed, and focused on the probes showing the most extreme decay in signal (most negative SC). As a first approximation, we used a conservative FDR of 0.1% as threshold to identify putative probes harboring genetic variants in the sample analyzed. When using the 1362-1 CEPH sample, we identified 18 genomic regions that had signal difference signatures consistent with the presence of a genetic change relative to the reference sequence. We designed primers to PCR the regions from the CEPH sample and performed Sanger sequencing in order to confirm the presence of novel mutations. We were able to identify 13 SNP variants (12 being heterozygous and one being homozygous; Supplemental Table S2) and indels in 10 of the interrogated regions. It is important to note that the identification of known SNPs is further validation that these variants are in fact real. Notably, we were able to identify not only homozygous single nucleotide mutations but also heterozygous mutations as well as indels. Finally, we wanted to determine whether we could identify an uncharacterized mutation from a model organism, Drosophila melanogaster. Previously, Eberl and colleagues identified a gene critical for fly hearing called beethoven (btv) [17,18]. Several mutations in this gene were identified; however, one allele that failed to complement the other btv alleles (btv2) was never characterized, as initial Sanger sequencing efforts covering several exonic regions based on early annotations failed to identify a likely mutation. The btv2 mutation was discovered as a second-site lesion on a chromosome with a partial P-element insertion in the fasciclin III gene (Fas3) near btv. We thus considered the possibility that the btv2 lesion resulted from a “hit-and-run” event during hybrid dysgenesis in the recovery of the k07109 chromosome. To apply MENA we thus generated a tiling array design (OID34126) in which we tiled a portion of the Drosophila second chromosome (dm3, 2L: 17,933,592-18,006,679) using 60 m probes with 3 bp spacing (both strands, three total replicates per strand) and performed our MENA assay on DNA isolated from flies harboring the btv2 allele (see Materials and Methods for a more detailed description) as well as flies that were wild-type for the btv locus. Note that, in this case, we used MENA with two samples (mutant and wild-type) and we compared the reduction in signal (SC) after Surveyor® treatment for btv mutant samples (SCbtv2) relative to the reduction in signal for the wild-type (SCcontrol) sample. This approach, we believe, is superior to the one applied for IRF6 because can capture better differences among probes after Surveyor® treatment. Probes showing the strongest relative reduction in the btv2 sample (see Figure 5) were thus candidates for harboring uncharacterized SNPs or small indel variants. Candidate SNP regions were then targeted for PCR and Sanger sequencing (Figure 5; red squares indicate the 1% of probes with strongest reduction relative to controls, and the black square is the probe amongst this 1% that led to identification of the relevant mutation), focusing on missense, nonsense or frameshift variants. Upon PCR amplification and sequencing of the candidate regions, a single base pair deletion mutation was confirmed in exon 22 (transcript btv-RD), at coordinate 17966613, which, on the minus strand encoding btv, begins a string of 5 T nucleotides that, in btv2, leave only four Ts. This indel results in a frameshift that causes early stop codons, thus substantially shortening the amino acid sequence of the cytoplasmic dynein protein (Figure 6). Thus, our MENA analysis succeeded in identifying a candidate novel variant that we experimentally confirmed to be a novel single base deletion mutation. 4. Discussion We have developed a mutation detection system called MENA (Mismatch EndoNuclease Array) in which we couple mismatch endonuclease enzymology with tiling DNA microarrays. We believe this system will be particularly well-suited for verification of SNP and SNV calls for genomes sequenced at low coverage depths, whether they be from model organisms, commercially relevant organisms such as livestock or domesticated animals, or humans. Importantly, number of genomes sequenced is oftentimes considered more essential than achieving maximum accuracy in calling variants for each individual genome, as is often the case in large-scale human sequencing efforts. Although such a sequencing strategy allows the sequencing of more genomes, both low and high sequence coverage can invariably lead to inaccurate base calling (the latter due to systematic sources of error [19], particularly in species with large genomes and few novel variants, such as humans. Having an accurate and independent genomic strategy for calling the suspected variants will allow confirmation of their identity. The current Agilent 8-plex slide, which contains eight independent microarrays, for instance, would allow confirming more than 5000 variants per sample for less than $200, with the advantage that each array can be custom designed to include novel SNPs/SNVs and genotype eight different genomic samples per slide. A second utility for MENA is identification of novel variants of interest in a genomic interval which can be tiled on a microarray. Such a strategy led us to identify a loss-of-function variant in the beethoven (btv) gene of Drosophila melanogaster, which is required for fly hearing [17,18]. Importantly, Sanger sequencing of a large genomic interval does not allow quick identification of relevant mutations, and the time required to design appropriate PCR primers, not to mention perform the sequencing, may be prohibitive depending on the size of the interval of interest. With MENA, regions of a genome that are up to approximately 50–100 kp can be assessed for variants in a matter of a few days once the arrays are synthesized. As shown, moreover, MENA can be applied to detect variants relative to a reference genome or to compare mutant and wild-type samples, and both approaches have allowed us to detect novel variants. A third utility for MENA is the screening of panels of genes associated with a particular disease in order to quickly identify known or novel gene variants likely underlying the disease for a particular patient. This strategy could be particularly useful for diseases or disorders associated with large numbers of genes such as metabolic disorders. Moreover, since the newest array platforms contain large numbers of features (approximately one million features for the Agilent platform), a single array design could be used to interrogate disease genes from a number of different disorders, thus minimizing the number of array designs needed. After employing MENA, the researcher would only need to focus on analysis of the relevant genes for his or her disorder of interest. We chose to use Roche NimbleGen microarrays during the planning phases of this study due to their long oligos, which provide increased sensitivity and specificity over short oligonucleotide microarrays. Although Roche NimbleGen no longer produces microarrays, Agilent microarrays have equivalent long oligonucleotides (similarly built base by base off the array surface) and oligo density. Thus, it is likely that our strategy will work for the Agilent platform as well. MENA, we propose, can be a rigorous, user-friendly and efficient methodology to simultaneously validate and test the presence of multiple known SNPs and/or detect novel variants using long-oligo array-based hybridization assays. 5. Conclusions We have developed an accurate, efficient low-cost mutation detection methodology (Mismatch EndoNuclease Array, or MENA) that pairs DNA mismatch endonuclease enzymology with tiling microarray hybridization to detect novel or known genomic variation at the level of point mutations or indels, provided the genome or genomic region of interest has been sequenced. Importantly, MENA is an independent genomics methodology that can be used to confirm variants identified by next generation sequencing methodologies, especially in cases where sequencing coverage is low. Since correlating genomic variation with disease is a primary focus of the biomedical research community, new platforms that can aid in correctly characterizing such variation will be in demand. Acknowledgments The authors would like to thank Aline Petrin and Jeffrey C. Murray for providing the genotyping information for the Centre d’Etude du Polymorphisme Humain (CEPH) sample used in this study, as well as helpful discussions. This project was supported by the following University of Iowa internal grants to Josep M. Comeron and J. Robert Manak: Carver Collaborative Pilot Grant, Institute for Clinical and Translational Science (ICTS) Translational Grant, Grow Iowa Values Fund Seed Grant. The project was also supported by an National Institutes of Health grant to Daniel F. Eberl (R01 DC04848). Supplementary Materials The following are available online at www.mdpi.com/2076-3905/5/2/7/s1. Click here for additional data file. Author Contributions Josep M. Comeron and J. Robert Manak conceived and designed the experiments; Jordan Reed, Matthew Christie, Julia S. Jacobs, and Jason Dierdorff performed the experiments; Josep M. Comeron, Daniel F. Eberl, and J. Robert Manak analyzed the data; Daniel F. Eberl contributed reagents/materials/analysis tools; and Josep M. Comeron and J. Robert Manak wrote the paper. Conflicts of Interest The authors declare no conflict of interest. Figure 1 General Mismatch EndoNuclease Array (MENA) strategy. Red stars indicate probes with mismatches to the hybridized DNA. Lightning bolts represent mismatches where the endonuclease cleaves the duplex DNA. Figure 2 Relative signal change (RSC) between pre- and post-treatment arrays with Surveyor. Nucleotides (X-axis) indicate the nucleotides interrogated by the probes (not the nucleotides in the probe). In the sample analyzed, rs9451298 and rs9359876 have “T” and “C”, respectively (forward strand). See text for a detailed description of the RSC measure. Red arrows indicate identity of the SNP. Figure 3 Summary of genotyping accuracy. Genotyping accuracy of MENA under different experimental conditions. Results shown after combining all homozygous SNPs for each condition. See Table 1 for detailed experimental conditions. Figure 4 Genotyping accuracy with different number of probes. Percentage of genotyping accuracy based different number of probes interrogating a given SNP. Results shown after combining all homozygous SNPs under experimental conditions 405722 (see Table 1 and Figure 3). Because probes are 60 nucleotides long, analyses based on probes tiled every three nucleotides use the combined information of 20 probes, analyses based on probes tiled every six nucleotides use information of 10 probes, etc. Figure 5 Comparison of signal change (SC) using MENA between mutant and control samples. SC indicates the reduction in signal after Surveyor treatment in controls and btv2 samples. In red, probes showing the strongest (1%) reduction in signal in mutant btv2 sample relative to controls and thus putative genetic variants. In black, the probe corresponding to a single bp deletion in exon 22 of the btv gene that results in a frameshift that causes early stop codons. In blue, all remaining probes that were interrogated with MENA. SC shown in Log2 units. Figure 6 Identification of a novel allele of the beethoven (btv) gene in Drosophila. Results of PCR and Sanger sequencing around the candidate mutation detected by MENA confirm a single bp (“A”) deletion in btv2 relative to a wild type (WT) sequence (A); this deletion maps to exon 22 of btv (red rectangle in B; red arrow in C), which is located in chromosome arm 2L (position 17966613-17966617); and (D) the consequences of the “A” deletion causing a frameshift and early stop codons. Green letters represent the altered amino acids that are translated as a result of the frameshift, which ends with a stop codon indicated by the red x. microarrays-05-00007-t001_Table 1Table 1 Experimental conditions to study genotyping accuracy using MENA. Slide Number Amount of DNA (μg) Amount of Surveyor (μL) Time of Surveyor (min) 405722 64 2.2 50 405442 64 2.2 40 406241 16 2.2 40 406228 8 2.2 40 405703 64 4.4 20 406247 32 2.2 40 405711 64 5.5 20 411044 64 4.4 40 410888 64 2.2 30 406295 64 2.2 20 406306 64 1.1 50 410527 64 3.3 20 406246 64 1.1 40 406205 64 1.1 20 All MENA experiments allowed three days of hybridization. See text for details. ==== Refs References 1. De la Vega F.M. Lazaruk K.D. Rhodes M.D. Wenz M.H. Assessment of two flexible and compatible SNP genotyping platforms: Taqman SNP genotyping assays and the snplex genotyping system Mutat. Res. 2005 573 111 135 10.1016/j.mrfmmm.2005.01.008 15829242 2. Fan J.B. Chee M.S. Gunderson K.L. Highly parallel genomic assays Nat. Rev. Genet. 2006 7 632 644 10.1038/nrg1901 16847463 3. Gresham D. Dunham M.J. Botstein D. Comparing whole genomes using DNA microarrays Nat. Rev. Genet. 2008 9 291 302 10.1038/nrg2335 18347592 4. Hoheisel J.D. Microarray technology: Beyond transcript profiling and genotype analysis Nat. Rev. Genet. 2006 7 200 210 10.1038/nrg1809 16485019 5. LaFramboise T. Single nucleotide polymorphism arrays: A decade of biological, computational and technological advances Nucleic Acids Res. 2009 37 4181 4193 10.1093/nar/gkp552 19570852 6. Trevino V. Falciani F. Barrera-Saldana H.A. DNA microarrays: A powerful genomic tool for biomedical and clinical research Mol. Med. 2007 13 527 541 10.2119/2006-00107.Trevino 17660860 7. van Dijk E.L. Auger H. Jaszczyszyn Y. Thermes C. Ten years of next-generation sequencing technology Trends Genet. TIG 2014 30 418 426 10.1016/j.tig.2014.07.001 25108476 8. Qiu P. Shandilya H. D‘Alessio J.M. O‘Connor K. Durocher J. Gerard G.F. Mutation detection using surveyor nuclease BioTechniques 2004 36 702 707 15088388 9. Olcaydu D. Harutyunyan A. Jager R. Berg T. Gisslinger B. Pabinger I. Gisslinger H. Kralovics R. A common JAK2 haplotype confers susceptibility to myeloproliferative neoplasms Nat. Genet. 2009 41 450 454 10.1038/ng.341 19287385 10. Bannwarth S. Procaccio V. Paquis-Flucklinger V. Surveyor nuclease: A new strategy for a rapid identification of heteroplasmic mitochondrial DNA mutations in patients with respiratory chain defects Hum. Mutat. 2005 25 575 582 10.1002/humu.20177 15880407 11. Oleykowski C.A. Bronson Mullins C.R. Godwin A.K. Yeung A.T. Mutation detection using a novel plant endonuclease Nucleic Acids Res. 1998 26 4597 4602 10.1093/nar/26.20.4597 9753726 12. Till B.J. Reynolds S.H. Greene E.A. Codomo C.A. Enns L.C. Johnson J.E. Burtner C. Odden A.R. Young K. Taylor N.E. Large-scale discovery of induced point mutations with high-throughput tilling Genome Res. 2003 13 524 530 10.1101/gr.977903 12618384 13. Yang B. Wen X. Kodali N.S. Oleykowski C.A. Miller C.G. Kulinski J. Besack D. Yeung J.A. Kowalski D. Yeung A.T. Purification, cloning, and characterization of the CEL I nuclease Biochemistry 2000 39 3533 3541 10.1021/bi992376z 10736152 14. Spradling A.C. Stern D. Beaton A. Rhem E.J. Laverty T. Mozden N. Misra S. Rubin G.M. The Berkeley Drosophila Genome Project gene disruption project: Single P-element insertions mutating 25% of vital Drosophila genes Genet. 1999 153 135 177 15. Mancebo R. Zhou X. Shillinglaw W. Henzel W. Macdonald P.M. BSF binds specifically to the bicoid mRNA 3′ untranslated region and contributes to stabilization of bicoid mRNA Mol. Cell. Biol. 2001 21 3462 3471 10.1128/MCB.21.10.3462-3471.2001 11313472 16. Lee E. Sivan-Loukianova E. Eberl D.F. Kernan M.J. An IFT-A protein is required to delimit functionally distinct zones in mechanosensory CILIA Curr. Biol. CB 2008 18 1899 1906 10.1016/j.cub.2008.11.020 19097904 17. Eberl D.F. Hardy R.W. Kernan M.J. Genetically similar transduction mechanisms for touch and hearing in Drosophila J. Neurosci. 2000 20 5981 5988 10934246 18. Eberl D.F. Duyk G.M. Perrimon N. A genetic screen for mutations that disrupt an auditory response in Drosophila melanogaster Proc. Natl. Acad. Sci. USA 1997 94 14837 14842 10.1073/pnas.94.26.14837 9405700 19. Wall J.D. Tang L.F. Zerbe B. Kvale M.N. Kwok P.Y. Schaefer C. Risch N. Estimating genotype error rates from high-coverage next-generation sequence data Genome Res. 2014 24 1734 1739 10.1101/gr.168393.113 25304867

PMC005xxxxxx/PMC5003484.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020008microarrays-05-00008ArticleTime-Resolved Study of Nanoparticle Induced Apoptosis Using Microfabricated Single Cell Arrays Röttgermann Peter J. F. 1Dawson Kenneth A. 2Rädler Joachim O. 1*Erfle Holger Academic Editor1 Faculty of Physics and Center for NanoSciene (CeNS), Ludwig-Maximilians-Universität, Geschwister-Scholl-Platz 1, 80539 Munich, Germany; peter.roettgermann@physik.lmu.de2 Centre for BioNano Interactions, School of Chemistry and Chemical Biology, University College Dublin, Belfield, Dublin 4, Ireland; Kenneth.A.Dawson@cbni.ucd.ie* Correspondence: raedler@lmu.de; Tel.: +49-89-2180-2438; Fax: +49-89-2180-318215 4 2016 6 2016 5 2 803 3 2016 07 4 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Cell fate decisions like apoptosis are heterogeneously implemented within a cell population and, consequently, the population response is recognized as sum of many individual dynamic events. Here, we report on the use of micro-patterned single-cell arrays for real-time tracking of nanoparticle-induced (NP) cell death in sets of thousands of cells in parallel. Annexin (pSIVA) and propidium iodide (PI), two fluorescent indicators of apoptosis, are simultaneously monitored after exposure to functionalized polystyrene (PS−NH2) nanobeads as a model system. We find that the distribution of Annexin onset times shifts to later times and broadens as a function of decreasing NP dose. We discuss the mean time-to-death as a function of dose, and show how the EC50 value depends both on dose and time of measurement. In addition, the correlations between the early and late apoptotic markers indicate a systematic shift from apoptotic towards necrotic cell death during the course of the experiment. Thus, our work demonstrates the potential of array-based single cell cytometry for kinetic analysis of signaling cascades in a high-throughput format. single cell arraycytotoxicitynanoparticledose-responsecorrelation studiesmicropatterning ==== Body 1. Introduction The interactions of nanoparticles (NPs) with cells remain poorly understood and this has raised concerns about potential cytotoxicity and environmental risks [1,2]. In recent years, many in-vitro and in-vivo studies have probed the safety and biocompatibility of NPs. Evidence for cytotoxicity was found in particular cases of NPs, depending on the cell line and test conditions used [3,4,5,6]. The majority of studies uses population-based toxicity assays, such as colorimetric assays for cell viability [7,8] and DNA fragmentation assays [9], or techniques with single-cell sensitivity, such as flow cytometry [10,11], image cytometry [12], or fluorescence microscopy [3], but data are taken at limited number of specific time points. It has recently been noted that cell-to-cell variations, which are averaged out in populations measurements but are revealed in single cell analysis, have non-genetic origins and provide important information on noise in apoptosis regulating circuitry [13,14]. Naturally occurring fluctuations in the levels of regulatory proteins can lead to ”fractional killing” and subpopulations of very sensitive or robust cells [14,15,16]. Moreover, time-lapse microscopy allows for fully time-resolved studies, in which every cell is tracked over time via brightfield and fluorescence microscopy [17,18,19]. These studies can directly assess the heterogeneous dynamic response of individual cells. It has become clear that, in NP toxicity studies, the precise experimental conditions have a crucial bearing on the results, and great care is required in the preparation and administration of NPs. Depending on the biological media chosen, NPs may be coated with a protein corona that further facilitates their entry into cells and determines their effect on cells [20,21]. However, we still know little about the biochemical pathways that are affected by NPs and how NPs eventually induce cell death. In order to understand the internal signaling processes and discriminate between various pathways that lead to cell death, it is crucial to measure cellular responses to NPs at the single-cell level using quantitative readouts. Typical cell death markers used in microscopy are Annexin V and propidium iodide (PI). The Annexin V-based marker pSIVA shows increased fluorescence when bound to phosphatidylserine (PhS), and hence indicates the externalization of plasma-membrane PhS induced by activation of the caspase-dependent pathway. The impermeable dye PI stains the nucleus only when the integrity of the cell membrane is lost, and this can be related to the late stage of apoptosis, the so-called secondary necrosis [22,23]. The use of cells captured on microfluidic- [24] or micro-patterned cell arrays offers a route towards high-throughput analysis. We recently introduced micro-patterned substrates for time-resolved measurements on regularly arrayed cells, and showed that cells self-organize onto fibronectin-coated sites surrounded by boundaries passivated by treatment with poly-l-lysine- polyethylene glycol [25,26]. Here, we perform NP toxicity studies on single cell arrays which yield time-resolved data at single-cell resolution. For a first proof of concept, we choose hepato carcinoma derived HuH7 liver cells exposed to PS−NH2 NPs, because liver cells are relevant in bio-accumulation and frequently used in toxicity studies. The timing of the onset of activation of two fluorescent markers — pSIVA, indicating the early apoptotic events and PI, the late stage of apoptosis or necrosis — was measured and the corresponding distribution function was analyzed as a function of dose. We show that the dynamics of NP-induced apoptosis is dose dependent, and relate the time-to-death value τ50 to the effective EC50 value. Furthermore, we find that the timing and signal intensities of the different apoptosis events are correlated. 2. Results 2.1. Highly Parallel Assessment of the Kinetics of Cell Death Figure 1 shows the workflow for data acquisition in time-resolved single-cell measurements. Microstructured surfaces for the preparation of single cell arrays are fabricated by selective plasma-induced patterning on either dishes or 8 well slides. Cells are seeded onto the microstructures and are left for five hours to self-arrange and adhere to the fibronectin adhesion sites. Cells are then exposed to different doses (0.1 to 100 μg · mL−1) of PS−NH2 NPs, and incubated together with the fluorescent markers pSIVA-IANBD (an Annexin B12 derivative) and PI. No further washing step is applied. NPs were obtained from the QualityNano project and were characterized by transmission electron microscopy (see Supplementary 1, Figure S1). The patterned cell samples are then scanned at 10-min intervals under physiological conditions for up to 48 h, using an automated microscope. Typically, 500 single cell traces of fluorescence intensities are extracted from image stacks using grid-based read-out software. Figure 2 shows brightfield (left) and fluorescence (right) microscopy images taken at selected time-points (0, 7, 14, 21 and 28 h, see movie in Supplementary 2). Green fluorescing cells indicate the binding of pSIVA to the cell surface at an early stage of apoptosis. During the the first 14 h, few cells show green fluorescence, and fluorescence emission begins in earnest only after 20 h. A similar pattern is observed for the red PI signal, albeit with a somewhat later onset (see below). In the time series and in Movie S2, the subsequent onset times of the green and red signal is clearly visible. The fading of the green signal towards the end of the measurement (28 h) is due to bleaching of the IANBD fluorophore. Fluorescence intensities integrated over the adhesion sites (indicated as squares in Figure 2) for each single cell can be extracted from the image stacks, and Figure 3 shows the time courses of single-cell intensities as a function of the NP exposure time. The plasma membrane transition of PhS in individual cells is readily detectable from the time course of the pSIVA signal (Figure 3A). Loss of plasma membrane integrity is indicated by a sharp increase of the PI signal (Figure 3B). The times of onset of these changes in individual cells can be clearly seen for pSIVA as well as for PI. For an NP dose of 10 μg · mL−1, the first apoptotic events are seen after 3 h and most cell deaths occur about 20 h after administration of NPs. Although not discernible in Figure 3B, in any given cell the PI signal appears about 70 min on average after the onset of the pSIVA signal (see below). The pSIVA time courses typically reveal a sudden, sharp increase followed by an exponential decrease. The latter behavior is attributable to the bleaching of the fluorophores. Some pSIVA traces show a second peak after the onset. Morphological changes such as formation of apoptotic bodies can lead to an increase in area and hence cause a higher mean fluorescence signal. In contrast, the decrease in the PI signal after the onset of cell death is related to non-specific staining of RNA, which lowers the intensity of PI fluorescence [27]. Most notably the peak intensities of both pSIVA and PI time courses vary quite considerably. The peak intensity distribution are plotted in the Supplementary Figure S3A, and we will discuss this cell-to-cell variability later on. In the following we focus on the onset-times of the pSIVA and PI signals. In our automated data analysis these values were determined as the times at which the signal first exceeds a level equivalent to 25% of the maximum background intensity. Cells which were dead from the beginning (less than 5%) are excluded from analysis. Also cell sites showing multiple occupation were excluded. Multiple occupation occurs either by the attachment of two cells on the same site or by cell division. The exclusion of dividing cells could lead to a bias in the cell death population as cells in a late cell cycle are neglected. However, the percentage of dividing cells over 40 hours is relatively low, about 15% in the case of cell arrays without exposure to NPs. Addition of NPs even decreases the ratio to 10% for a dose of 0.1 μg · mL−1 and below 1% for higher doses. This decreased cell division rate is due to cell-cycle arrest [28]. Hence, as the fraction of dividing cells is low, the restriction of our analysis to pure single cell events should not adversely bias the assessment of cell death timing. 2.2. Time Dependent Dose Response Function The distribution of times of apoptosis provides a measure for the dynamic response of a cell population. We studied the evolution of the onset distribution with increasing NP dosage (from 0.1 μg · mL−1 to 100 μg · mL−1). In Figure 4A, normalized frequency distributions of onset times of the PI signal are plotted against time for various doses. It can be seen that with increasing NP dose, cell death onset shifts to earlier times. The same behavior is found in the corresponding time distribution for the pSIVA signal (data not shown), which also exhibits the same characteristic broadness in the distribution. The distributions are fitted by log-normal distributions. A log-normal distribution describes a process which is the product of many independent processes and is found (in our case as well as in literature) to fit the cell death distribution better than Gaussians [29]. For the lowest dose (0.1 μg · mL−1), the mean onset time is 27.8 ± 0.4 h, and for the highest dose it is shifted to the markedly earlier time of 8.9 ± 0.3 h. In addition, the distributions become narrower as the dosage is increased. The fraction of cells that dies is given by the cumulative sum. Figure 4B shows the cumulative sum against time and the fits to the integrated log-normal function. The time-course of the fraction of dead cells highlights the dependence of the dynamic response on dosage. At the highest dose almost all cells are dead after 24 h, whereas only 20% of those exposed to the lowest dose have died by this time. We evaluate this ”dynamic dose response” by plotting the time of 50% cell death, τ50, against dosage (Figure 4E). A drop in time-to-death with increasing dose can clearly be seen. The variance, σ, i.e., the width of the log-normal distribution, in contrast, increases with dose (Figure 4F). In order to retrieve standard EC50 values from the log-normal distributions shown in Figure 4A,B, we replotted the data as a whole as a function of dose rather than time (see Figure 4C). In this way, we obtain dose-response curves for various time points, i.e., effective end-points. We find that with increasingly late end-point the dose-response curves shift to lower dose values as depicted in Figure 4C. As a consequence, the EC50 value, i.e., the dose that result in 50% cell death, appears to depend on the time point of examination (Figure 4D). After a sufficiently long time, the EC50 values asymptotically approach a constant value. In our case, sufficiently long means 20 h, as is clearly seen in Figure 4D. However, it is not obvious a priori at which time point a reliable EC50 value can be obtained. In order to test if the heterogeneous response is caused by the NPs negative controls were performed (see Supplementary S4, Figure S3A,D). The negative control for the time-lapse measurement as well as a viability assay showed a high viability (95%) after four days. Further, in order to show the applicability of the single cell assay to other toxic agents, a dose response measurement for the anti-cancer drug staurosporine was performed (see Supplementary 4, Figure S3A–C). 2.3. Two-Parameter Correlation of Cell Death Next, the degree of correlation between onset times of the early and late apoptotic markers was examined. In Figure 5A, the interval Δt between the onset of the pSIVA signal tpSIVA and that of the PI signal tPI is plotted against the time of cell death, which we again define as tPI. The scatter plot shows that there is a tendency for Δt to decrease with time of death. In fact, cells that die earlier, i.e., in the time window between 5 and 10 h exhibit a delay time of about 100 to 300 min while late events of cell death show almost no delays. The correlation is underlined by a principal component analysis (PCA), which shows a tilted major axis, as indicated by the grey ellipse in the scatter plot. The correlation between pSIVA and PI onset times is also well brought out by the Pearson’s correlation coefficient (for details, see data analysis in Experimental Section). The Pearson’s correlation in this case is 0.96, and hence clearly indicates a strong correlation at the scale from zero (no correlation) to one (full correlation). Extending this concept to all measurable parameters of the apoptotic time trace we also investigated the cross-correlations of the onset time ton with maximal intensity Imax, and duration of onset tduration (i.e., the time of increase from background to maximum signal for both the pSIVA and PI fluorescence). The correlation matrix is shown in Figure 5B for a dose of 10 μg · mL−1. Interestingly, the maximal intensity Imax(pSIVA) correlates negatively with the onset of apoptosis (−0.6/−0.56). This possibly indicates a connection with cell size or caspase activity, or it may be related to fluorescence bleaching. Indeed, the heterogeneity of the pSIVA fluorescence intensity depends on the amount of PhS exposed, and hence depends on the level of activation of caspase3 [30,31]. Also the maximal intensity Imax (PI) is negatively correlated, albeit weakly, with the onset of apoptosis ton (−0.23, −0.2). Likewise Imax (PI) is weakly correlated with Imax(pSIVA) (0.34). Both pSIVA and PI signals depend on cell size, which is in turn cell-cycle dependent. However, there is also substantial intrinsic cell-to-cell variation in the size of the nucleus (see Supplementary Figure S2B) [32,33]. For completeness, we also show the correlation of the duration times of the onset. The duration tduration(PI) is negatively correlated with the onset of apoptosis ton (−0.46/−0.43). tduration(PI) moderately correlates with Imax of pSIVA (0.47). Cells that die late undergo more rapid loss of membrane integrity, which leads to a faster increase in the PI signal. pSIVA tduration does not, or only weakly, correlates with other parameters (0.14, 0.3, −0.02, and 0.03). 3. Discussion In this work, we demonstrate the advantages of time-resolved analysis of single cell arrays on microstructured substrates for studies of the dynamics of NP-induced cell death. The individual cell traces permit virtually continuous monitoring of apoptotic events indicated by the markers pSIVA and PI. The wealth of information provided by this approach, in comparison to standard population-based toxicity tests, is obvious. The technique has allowed us to demonstrate that the time distribution function of apoptosis onset is dose dependent, and to introduce the τ50 value as a measure for fast or slow toxic efficacy. Conversely, the EC50 value is seen to depend on the time-point of measurement. The dependence of the EC50 value has been discussed previously [34]. It has also been noted that a time-dependent EC50 becomes exceedingly problematic in the low-dose regime, when toxicity occurs on time scales that exceed the duration of experimental observations [35,36]. In this case it can be argued that the τ50 value is the more useful measure. In the conventional approach, the evaluation of an EC50 value for weakly toxic nanomaterials is complicated not only by the choice of end-point, but also involves the risk of under- or over-estimating the toxic effect. In contrast to highly time-resolved measurements of many cells, our methodology allows us to extrapolate from the behavior of the relevant measured curve in order to get a first indication of toxicity risk in in-vitro systems. A second major advantage of single cell data with high time resolution lies in the fact that successive events can be correlated in time. As described earlier, single-cell time-courses yield distinct onset times showing that the pSIVA signal precedes the PI signal. It is generally accepted that a difference in onset time between the two events is indicative of apoptosis-induced cell death. Hence in our case, as we clearly observe a pSIVA-PI time difference — at least in the early phase of NP exposure — cells death is induced via the apoptosis pathway. However, as the experiment proceeds, the delay between the onset of pSIVA and PI activation becomes shorter. Short delay times are interpreted as a signature of necrotic or necroptotic cell death. In our set-up, the number of necrotic cells rises with increasing duration of the experiment. This time-dependent upsurge in the incidence of necrosis in toxicity assays has been reported before, and has two possible causes. First, failure to eliminate apoptotic bodies, due to the lack of phagocytes in-vitro, promotes the transition from apoptotic to necrotic cell death [22,23]. Secondly, cells that undergo apoptosis affect neighboring cells towards cell death [37]. In addition, accumulation of NPs in cells over time, with a concomitant decrease in cell fitness, might contribute the increase of necrosis. At all events, the time-resolved single-cell approach quantifies the necrotic shift in the cell culture death with unprecedented directness. The utility of the single-cell strategy for exploring temporal correlations between fluorescent signals can also be extended by employing multiple markers indicative for distinct cellular events. As shown by the work of Wang et al. multiple markers can be used to elucidate the pathway induced by NPs [21]. 4. Experimental Section 4.1. Cell Culture HuH7 cells were cultured in RPMI supplemented with 2 mM l-glutamine (c-c-pro) and 10% fetal calf serum. Cells were grown to 70%–80% confluence, trypsinized and centrifuged at 1000 rcf for 3 min. Cell pellets were re-suspended in either cell medium or, for the experiments in Leibovitz’s L15 medium with GlutaMAX (Gibco), 10% fetal calf serum and 1 mM calcium chloride. Cells were stained with polarity Sensor for Viability and Apoptosis (pSIVA-IANBD, Novus Biologicals) [38,39] (the stock solution was diluted 1:50) and 1 μM propidium iodide (Novus Biologicals). Cell dyes were added without any further washing step. As a positive control for cell death staurosporine (Sigma Aldrich) was used at doses between 0.5 and 25 μM. 4.2. NP Preparation and Characterization Core size and shape of the PS−NH2 NPs (provided by QualityNano, UCD) were determined with a transmission electron microscope (TEM). NPs were adsorbed onto a Formvar/carbon film-coated grid and were observed with a Jeol 1011 TEM. Sizes of the NPs were determined with ImageJ and further analyzed in MATLAB (see Supplementary 1 Figure S1). Further characterization was carried out in the context of the QualityNano project [10,40]. Prior to use, NPs were vortexed for 1 min to obtain maximal dispersity, although aggregates of >1 μm were still present. Cells were exposed either to PS−NH2 (0.1 μg · mL−1, 1 μg · mL−1, 10 μg · mL−1, and 100 μg · mL−1, corresponding to a surface concentration of 14 ng · cm−2 to 14 μg · cm−2). NPs were kept in solution for the duration of the measurement. 4.3. Micropatterning The micro-structured surfaces were produced by selective oxygen plasma treatment (Femto Diener, 40 W for 3 min) on a topas substrate (μ-dishes ibidi GmbH) with subsequent passivation using either μ-dishes or 8 well slides. Selectivity was achieved using a polydimethylsiloxane (PDMS) stamp (cast from a master produced by photolithography) as a mask. The parts exposed to plasma were passivated with 1 mg · mL−1 PLL(20k)-g(3.5)-PEG(2k) (SuSoS AG) in aqueous buffer (10 mM HEPES pH 7.4 and 150 mM NaCl). The remaining sectors were rendered cell adherent by exposure to 50 μg · mL−1 fibronectin (YoProteins) for 1 h. The samples were thoroughly rinsed with PBS and stored in cell medium at room temperature prior to seeding (8000 cells per 8 well and 50,000 per μ-dish). 4.4. Time-Resolved Fluorescence Microscopy Images were taken with an inverted Nikon Ti Eclipse microscope with phase-contrast and a fluorescent lamp with multiple filter sets (GFP, propidium iodide). Samples were kept at a constant temperature of 37 °C with an ibidi heating system (ibidi GmbH). Pictures were recorded every 10 min over a period of between 24 and 48 h. 4.5. Image and Data Analysis Raw images were pre-processed in ImageJ. For the image analysis on the cell grid and a background correction the in-house plug-in Microwell Analysis was used. Multiple occupied sites, determined at the end of the measurement, were filtered out. The time of onset of apoptosis/secondary necrosis was defined as the point at which the pSIVA/PI signal first exceed the threshold of 25% of the basal background . This definition was in good agreement with manual tracking. The duration of the onset time was determined as the interval between the first local maximum of the smoothed curve and the crossing of the threshold. The dose-response of the log-normal distribution of dead cells dc(t) was fitted by dc(t)=A*explog(t)−μ(2σ)2 and the fraction of dead cells F(t), the cumulative sum of the distribution, by F(t)=0.5*1+erflog(t)−μ(2σ2)2 where A denotes amplitude, t the time, σ the standard deviation, and μ the mean. The dose-response curves for the fraction of dead cells F(dose) in Figure 4C were fitted by F(dose)=y0+−y01+(EC50/dose)n with rate n, the basal rate y0, and the EC50 value at which 50% of cells are dead. Onset times of the two markers were analyzed, and the subsequent correlation analysis was carried out, in MATLAB. For the principal component analysis, the eigenvectors were calculated for the data set of tPI and Δt. In Figure 5A, the plotted axis of the ellipse correspond to those eigenvectors with lengths of two standard deviations σ. The ellipse is centered on the mean. The correlation coefficient was defined by Pearson’s correlation : ρ(X,Y)=Cov(X,Y)σ(X)σ(Y) with the covariance Cov(X,Y) of the two parameters X and Y and their related standard deviation σ. 5. Conclusions In summary, we provide the first kinetic analysis of NP-induced cell death based on single cell arrays. The resulting time-resolved measurements enable us to measure the onset distribution function of NP-induced apoptosis and a differentiated time-dependent picture of the NP dose-response relationship emerges. Future use of micro-arrays for high-throughput acquisition of single cell time-lapse data will pave the way to kinetic investigations of cell death in in-vitro and could potentially uncover aspects of the action NPs trapped within cells, thus allowing for the detailed characterization of nanomaterial pathways using single cell cross-correlation. Hence automated microscopy combined with micro-array technology might evolve into a potent platform for nanomaterial characterization and beneficially complement toxicity risk assessment. Acknowledgments We are grateful to Christian Meggle for writing the image analysis plug-in in ImageJ. Financial support from the FP7 EU grants NanoMILE and NanoTransKinetics, and the Excellence Cluster ’Nanosystems Initiative Munich (NIM)’ is gratefully acknowledged. Supplementary Materials The following are available online at http://www.mdpi.com/2076-3905/5/2/8/s1. Click here for additional data file. Author Contributions Peter J. F. Röttgermann and Joachim O. Rädler conceived and designed the experiments; Peter J. F. Röttgermann performed the experiments; Peter J. F. Röttgermann analyzed the data; Kenneth A. Dawson contributed reagents/materials/analysis tools; Peter J. F. Röttgermann and Joachim O. Rädler wrote the paper. Conflicts of Interest The authors declare no conflict of interest. Figure 1 Protein-coated arrays are generated following plasma-induced patterning of μ-dishes. Cells distribute themselves on the patterned sites of an array after seeding. They are then exposed to NPs and monitored for activation of fluorescent markers of cell death for up to 72 h using an automated fluorescence microscope. The ease of automated image processing on the cell lattice enables high-throughput analysis of the kinetics of single-cell fluorescence. Scale bar: 500 μm (close-ups 5× and 25× magnified, respectively). Figure 2 Time series of apoptotic events induced by PS−NH2-NPs monitored by brightfield illumination and fluorescence imaging at 0, 7, 14, 21 and 28 h after seeding. The graininess seen in the brightfield images is attributed to sedimentation of nanoparticle (NP) aggregates. In the fluorescence images, green staining (pSIVA-IANBD) indicates exposure of phosphatidylserine (PhS) on the outer leaf of the plasma membrane bilayer, the red nuclear staining (PI) indicates subsequent loss of plasma membrane integrity. The heterogeneity in the times of onset of apoptosis can be clearly seen. Square lattices are drawn for better visualization. Scale bar: 50 μm. Figure 3 Representative time traces of fluorescent signals monitored in the polarity Sensor for Viability and Apoptosis (pSIVA) (A) and propidium iodide (PI) (B) channels. A few typical traces are highlighted in green and red, respectively. The first apoptotic events are observed after 3 h whereas the majority of induced cell deaths occur after 20 h. The decrease in the pSIVA signal at late times can be attributed to bleaching of the fluorophores. Fluorescent signals are background corrected. Figure 4 (A) Normalized frequency distributions of times of cell death (onset times of the PI signal) are plotted against time for the indicated NP doses. Note that the distributions shift to earlier time points and get narrower with increasing NP dose. The distributions are fitted to log-normal functions (red curve); (B) The cumulative fraction of dead cells is plotted against time for NP doses of 0.1 μg · mL−1 (light gray), 1 μg · mL−1 (medium gray), 10 μg · mL−1 (dark gray), and 100 μg · mL−1 (black). At the lowest dose 20% of the cells die within 24 h, whereas at the highest dose all cells (100%) are dead by this point. The cumulative distributions are fitted to log-normal functions (red curve); (C) Standard dose-response curves with fraction of dead cells which can be extracted from the distribution of (B) for several different time points between 10 and 24 h (black to light gray). Data are fitted to dose-response functions (dashed lines). The EC50 values shift towards lower dose with increasing late time endpoints (black arrow); (D) The EC50 values are plotted in logarithmic scale against the endpoints. At an end-point of 18 h, the EC50 value approaches a constant value; (E, F) Time points τ50 (E) and a rate-dependent σ (F) for the cumulative distributions are plotted against dose. Both values exhibit dose-response behavior Figure 5 (A) The interval separating early-stage apoptosis (only pSIVA) from secondary necrosis (pSIVA+PI) varies depending on the onset time of apoptosis. This time difference Δt is plotted against the onset time. The grey ellipse represents a 2σ interval of a principal component analysis (PCA). Three representative time traces of apoptotic cells are depicted for better visualization: Cells that die at later time-points transit faster into secondary necrosis; (B) Different parameters of apoptotic cells are extracted from the time traces and displayed as a correlation matrix: onset times ton, maximum intensity Imax and duration of onset for pSIVA and PI. Maximal pSIVA intensity Imax, and duration (spread) of onset time of PI tduration are negatively correlated with the onset of apoptosis. ==== Refs References 1. Colvin V.L. The potential environmental impact of engineered nanomaterials Nat. Biotech. 2003 21 1166 1170 10.1038/nbt875 14520401 2. Nel A. Xia T. Madler L. Li N. Toxic potential of materials at the nanolevel Science 2006 311 622 627 10.1126/science.1114397 16456071 3. Anguissola S. Garry D. Salvati A. O’Brien P.J. Dawson K.A. High content analysis provides mechanistic insights on the pathways of toxicity induced by amine-modified polystyrene nanoparticles PLoS ONE 2014 9 8 10.1371/journal.pone.0108025 25238162 4. Elsaesser A. Howard C.V. Toxicology of nanoparticles Adv. Drug Deliv. Rev. 2012 64 129 137 10.1016/j.addr.2011.09.001 21925220 5. Love S.A. Maurer-Jones M.A. Thompson J.W. Lin Y.S. Haynes C.L. Assessing nanoparticle toxicity Annu. Rev. Anal. Chem. 2012 5 181 205 10.1146/annurev-anchem-062011-143134 22524221 6. Lu X. Liu Y. Kong X. Lobie P.E. Chen C. Zhu T. Nanotoxicity: A growing need for study in the endocrine system Small 2013 9 1654 1671 10.1002/smll.201201517 23401134 7. Fonseca C. Simões S. Gaspar R. Paclitaxel-loaded PLGA nanoparticles: preparation, physicochemical characterization and in vitro anti-tumoral activity J. Control. Release 2002 83 273 286 10.1016/S0168-3659(02)00212-2 12363453 8. Kommareddy S. Amiji M. Preparation and evaluation of thiol-modified gelatin nanoparticles for intracellular DNA delivery in response to glutathione Bioconjugate Chem. 2005 16 1423 1432 10.1021/bc050146t 16287238 9. Jose G.P. Santra S. Mandal S.K. Sengupta T.K. Singlet oxygen mediated DNA degradation by copper nanoparticles: potential towards cytotoxic effect on cancer cells J. Nanobiotechnology 2011 9 9 10.1186/1477-3155-9-9 21439072 10. Bexiga M.G. Varela J.A. Wang F. Fenaroli F. Salvati A. Lynch I. Simpson J.C. Dawson K.A. Cationic nanoparticles induce caspase 3-, 7- and 9-mediated cytotoxicity in a human astrocytoma cell line Nanotoxicology 2011 5 557 567 10.3109/17435390.2010.539713 21142842 11. Wlodkowic D. Skommer J. Darzynkiewicz Z. Flow cytometry-based apoptosis detection Methods Mol. Biol. 2009 559 19 32 19609746 12. Summers H.D. Rees P. Holton M.D. Rowan Brown M. Chappell S.C. Smith P.J. Errington R.J. Statistical analysis of nanoparticle dosing in a dynamic cellular system Nat. Nano 2011 6 170 174 10.1038/nnano.2010.277 21258333 13. Xia X. Owen M.S. Lee R.E. Gaudet S. Cell-to-cell variability in cell death: can systems biology help us make sense of it all? Cell Death Differ. 2014 5 e1261 10.1038/cddis.2014.199 24874733 14. Spencer S.L. Gaudet S. Albeck J.G. Burke J.M. Sorger P.K. Non-genetic origins of cell-to-cell variability in TRAIL-induced apoptosis Nature 2009 459 428 432 10.1038/nature08012 19363473 15. Love K.R. Bagh S. Choi J. Love J.C. Microtools for single-cell analysis in biopharmaceutical development and manufacturing Trends Biotechnol. 2013 31 280 286 10.1016/j.tibtech.2013.03.001 23582471 16. Ware M.J. Godin B. Singh N. Majithia R. Shamsudeen S. Serda R.E. Meissner K.E. Rees P. Summers H.D. Analysis of the Influence of Cell Heterogeneity on Nanoparticle Dose Response ACS Nano 2014 8 6693 6700 10.1021/nn502356f 24923782 17. Aftab O. Nazir M. Fryknas M. Hammerling U. Larsson R. Gustafsson M.G. Label free high throughput screening for apoptosis inducing chemicals using time-lapse microscopy signal processing Apoptosis 2014 19 1411 1418 10.1007/s10495-014-1009-9 24923770 18. Albeck J.G. Burke J.M. Aldridge B.B. Zhang M. Lauffenburger D.A. Sorger P.K. Quantitative analysis of pathways controlling extrinsic apoptosis in single cells Mol. Cell 2008 30 11 25 10.1016/j.molcel.2008.02.012 18406323 19. Forrester H.B. Albright N. Ling C.C. Dewey W.C. Computerized video time-lapse analysis of apoptosis of REC:Myc cells X-irradiated in different phases of the cell cycle Radiat. Res. 2000 154 625 639 10.1667/0033-7587(2000)154[0625:CVTLAO]2.0.CO;2 11096419 20. Milani S. Baldelli Bombelli F. Pitek A.S. Dawson K.A. Rädler J. Reversible versus irreversible binding of transferrin to polystyrene nanoparticles: Soft and hard corona ACS Nano 2012 6 2532 2541 10.1021/nn204951s 22356488 21. Wang F. Yu L. Monopoli M.P. Sandin P. Mahon E. Salvati A. Dawson K.A. The biomolecular corona is retained during nanoparticle uptake and protects the cells from the damage induced by cationic nanoparticles until degraded in the lysosomes Nanomedicine 2013 9 1159 1168 10.1016/j.nano.2013.04.010 23660460 22. Silva M.T. Secondary necrosis: The natural outcome of the complete apoptotic program FEBS Lett. 2010 584 4491 4499 10.1016/j.febslet.2010.10.046 20974143 23. Vanden Berghe T. Grootjans S. Goossens V. Dondelinger Y. Krysko D.V. Takahashi N. Vandenabeele P. Determination of apoptotic and necrotic cell death in vitro and in vivo Methods 2013 61 117 129 10.1016/j.ymeth.2013.02.011 23473780 24. Wlodkowic D. Faley S. Zagnoni M. Wikswo J.P. Cooper J.M. Microfluidic single-cell array cytometry for the analysis of tumor apoptosis Anal. Chem. 2009 81 5517 5523 10.1021/ac9008463 19514700 25. Röttgermann P.J. Hertrich S. Berts I. Albert M. Segerer F.J. Moulin J.F. Nickel B. Rädler J.O. Cell motility on polyethylene glycol block copolymers correlates to fibronectin surface adsorption Macromol. Biosci. 2014 14 1755 1763 10.1002/mabi.201400246 25204968 26. Röttgermann P.J.F. Piera Alberola A. Rädler J.O. Cellular Self-Organization on Micro-Structured Surfaces Soft Matter 2014 10 2397 2404 10.1039/c3sm52419a 24623049 27. Waring M.J. Complex formation between ethidium bromide and nucleic acids J. Mol. Biol. 1965 13 269 282 10.1016/S0022-2836(65)80096-1 5859041 28. Kim J.A. Åberg C. de Cárcer G. Malumbres M. Salvati A. Dawson K.A. Low dose of amino-modified nanoparticles induces cell cycle arrest ACS Nano 2013 7 7483 7494 10.1021/nn403126e 23941353 29. Furusawa C. Suzuki T. Kashiwagi A. Yomo T. Kaneko K. Ubiquity of log-normal distributions in intra-cellular reaction dynamics Biophysics 2005 1 25 31 10.2142/biophysics.1.25 30. Frasch S.C. Henson P.M. Kailey J.M. Richter D.A. Janes M.S. Fadok V.A. Bratton D.L. Regulation of phospholipid scramblase activity during apoptosis and cell activation by protein kinase Cδ J. Biol. Chem. 2000 275 23065 23073 10.1074/jbc.M003116200 10770950 31. Lee S.H. Meng X.W. Flatten K.S. Loegering D.A. Kaufmann S.H. Phosphatidylserine exposure during apoptosis reflects bidirectional trafficking between plasma membrane and cytoplasm Cell Death Differ. 2013 20 64 76 10.1038/cdd.2012.93 22858544 32. Edens L.J. White K.H. Jevtic P. Li X. Levy D.L. Nuclear size regulation: from single cells to development and disease Trends Cell Biol. 2013 23 151 159 10.1016/j.tcb.2012.11.004 23277088 33. Huber M.D. Gerace L. The size-wise nucleus: Nuclear volume control in eukaryotes J. Cell Biol. 2007 179 583 584 10.1083/jcb.200710156 17998404 34. Krippendorff B.F. Neuhaus R. Lienau P. Reichel A. Huisinga W. Mechanism-based inhibition: Deriving K(I) and k(inact) directly from time-dependent IC(50) values J. Biomol. Screen 2009 14 913 923 10.1177/1087057109336751 19675314 35. Vandenberg L.N. Colborn T. Hayes T.B. Heindel J.J. Jacobs D.R.J. Lee D.H. Shioda T. Soto A.M. vom Saal F.S. Welshons W.V. Hormones and endocrine-disrupting chemicals: Low-dose effects and nonmonotonic dose responses Endocr Rev. 2012 33 378 455 10.1210/er.2011-1050 22419778 36. Yedjou C. Moore P. Tchounwou P. Dose- and time-dependent response of human leukemia (HL-60) cells to arsenic trioxide treatment Int. J. Environ. Res. Public Health 2006 3 136 140 10.3390/ijerph2006030017 16823087 37. Gregory C.D. Pound J.D. Devitt A. Wilson-Jones M. Ray P. Murray R.J. Inhibitory effects of persistent apoptotic cells on monoclonal antibody production in vitro: Simple removal of non-viable cells improves antibody productivity by hybridoma cells in culture mAbs 2009 1 370 376 10.4161/mabs.1.4.9124 20068393 38. Kim Y.E. Chen J. Chan J.R. Langen R. Engineering a polarity-sensitive biosensor for time-lapse imaging of apoptotic processes and degeneration Nat. Meth. 2010 7 67 73 10.1038/nmeth.1405 19966809 39. Kim Y.E. Chen J. Langen R. Chan J.R. Monitoring apoptosis and neuronal degeneration by real-time detection of phosphatidylserine externalization using a polarity-sensitive indicator of viability and apoptosis Nat. Protoc. 2010 5 1396 1405 10.1038/nprot.2010.101 20671723 40. Wang F. Bexiga M.G. Anguissola S. Boya P. Simpson J.C. Salvati A. Dawson K.A. Time resolved study of cell death mechanisms induced by amine-modified polystyrene nanoparticles Nanoscale 2013 5 10868 10876 10.1039/c3nr03249c 24108393

PMC005xxxxxx/PMC5003485.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020009microarrays-05-00009ReviewMicroarray-Based Comparative Genomic and Transcriptome Analysis of Borrelia burgdorferi Iyer Radha Schwartz Ira *Loewy Zvi Academic EditorDepartment of Microbiology and Immunology, New York Medical College, School of Medicine, Valhalla, NY 10595, USA; radha_iyer@nymc.edu* Correspondence: schwartz@nymc.edu; Tel.: +1-914-594-4658; Fax: +1-914-594-417616 4 2016 6 2016 5 2 907 3 2016 11 4 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Borrelia burgdorferi, the spirochetal agent of Lyme disease, is maintained in nature in a cycle involving a tick vector and a mammalian host. Adaptation to the diverse conditions of temperature, pH, oxygen tension and nutrient availability in these two environments requires the precise orchestration of gene expression. Over 25 microarray analyses relating to B. burgdorferi genomics and transcriptomics have been published. The majority of these studies has explored the global transcriptome under a variety of conditions and has contributed substantially to the current understanding of B. burgdorferi transcriptional regulation. In this review, we present a summary of these studies with particular focus on those that helped define the roles of transcriptional regulators in modulating gene expression in the tick and mammalian milieus. By performing comparative analysis of results derived from the published microarray expression profiling studies, we identified composite gene lists comprising differentially expressed genes in these two environments. Further, we explored the overlap between the regulatory circuits that function during the tick and mammalian phases of the enzootic cycle. Taken together, the data indicate that there is interplay among the distinct signaling pathways that function in feeding ticks and during adaptation to growth in the mammal. microarrayBorrelia burgdorferitranscriptomeLyme diseasetranscriptional regulators ==== Body 1. Introduction The spirochete Borrelia burgdorferi is the causative agent of Lyme disease, the most commonly reported arthropod-borne disease in the United States [1,2,3]. B. burgdorferi is maintained in a natural enzootic cycle involving small mammals and a tick vector of the Ixodes species [3,4]. These two diverse host environments vary with respect to temperature, pH, oxygen tension and nutrients [5]. In order to adapt to growth in its mammalian and tick hosts, the spirochete must profoundly alter its gene expression in response to these environmental cues. B. burgdorferi has a unique genome organization; the genome of strain B31 (the type strain) is comprised of a linear chromosome of 910,724 bp and 21 linear and circular plasmids totaling an additional 610,694 bp [6,7]. Elucidation of the complete genome sequence enabled production of whole genome arrays. Since 2002, more than 25 microarray-based studies on the comparative genomic structure and transcriptome of B. burgdorferi have been published. Here, we review the contribution of microarray technology to our understanding of B. burgdorferi biology with particular emphasis on the variation in gene expression under different environmental conditions. B. burgdorferi Microarray Methodology To date, all B. burgdorferi genome arrays have been designed based on the genome sequence of strain B31. Initially, whole genome arrays were constructed with PCR products of >1600 putative B. burgdorferi open reading frames (ORFs) and were spotted on either glass slides or nylon membranes [8,9,10]. Subsequently, 70-mer oligonucleotides spotted on glass slides were employed in order to improve the reliability of hybridization signal intensity [11,12]. In addition, several groups have employed other custom glass slide or chip designs representing the complete genome or smaller sub-arrays of selected ORFs [13,14,15]. Table 1 contains a listing of all published B. burgdorferi microarray studies and provides the array types employed. In general, the steps for studying global gene expression changes in the B. burgdorferi transcriptome are as follows: RNA isolation, generation of labeled cDNA, array hybridization and scanning, data acquisition and analysis. In initial studies with nylon membrane arrays, cDNA was radioactively labeled with 33P; a detailed protocol is described in Ojaimi et al. [9]. Subsequently, a variety of high density microarray designs were developed and used fluorescently labeled DNA or cDNA for hybridization. Data acquisition, normalization and statistical analysis were particular to each type of microarray employed and details can be found in the respective references in Table 1. All published microarray data were deposited either in Array Express or NCBI’s Gene Expression Omnibus (GEO) databases. In addition, the results of virtually all published microarray studies reported to date have been validated by quantitative reverse transcription polymerase chain reaction (qRT-PCR). 2. Comparative Genomic Studies Liang et al. [13], constructed a sub-array comprised of PCR products for 137 putative lipoproteins in order to study lipoprotein gene content across three B. burgdorferi sensu lato genospecies that cause Lyme disease in humans. There was extensive conservation of chromosomally-encoded lipoprotein gene content among all strains tested. By contrast, lipoproteins encoded on the plasmid portion of the genome were substantially less conserved [13]. This pattern was confirmed by Terekhova et al. [11], who employed whole genome microarrays to perform comparative genome hybridization of seventeen B. burgdorferi isolates, including clinical isolates with varying capacity for hematogenous dissemination in mice or humans. This revealed that chromosomal genes are more highly conserved among the isolates than are plasmid genes. The linear chromosome and plasmids lp54 and cp26 are the most conserved genomic elements among all isolates studied, which implies that they may encode functions required for bacterial viability. The most substantial variation was found among the linear plasmid portion of the genome; this variability was the result of presence/absence of entire plasmids, deletions or nucleotide sequence divergence. Zhong and Barbour [16] used B. burgdorferi whole genome membrane arrays to study the similarity in gene content between B. burgdorferi and B. hermsii, a relapsing fever spirochete. They demonstrated that B. hermsii genomic DNA cross-hybridized with 81% of B. burgdorferi chromosomal genes and 46% of plasmid ORFs. They were also able to demonstrate the expression of 642 genes with similarity to B. burgdorferi ORFs in the blood of B. hermsii-infected mice [16]. Taken together, these microarray studies demonstrated that there is relatively little variation in the chromosomal portion of the B. burgdorferi genome, but much greater variation in plasmid content and sequence. Genomic sequencing of multiple B. burgdorferi isolates subsequently validated these findings [34,35,36]. 3. Global Transcriptome Studies The principal application of microarray technology to B. burgdorferi has been for global transcriptome analysis. These studies have informed our understanding of regulation of B. burgdorferi gene expression under different environmental conditions and, most importantly, elucidation of the roles for several transcriptional regulators in this process. As noted, in nature B. burgdorferi cycle through two distinct environments—tick vector and mammalian host. The limited number of organisms present in infected ticks or mammals constrains robust global transcriptome analysis from in vivo material. Initial transcriptome studies employed in vitro cultivation of B. burgdorferi in BSK medium under conditions thought to mimic either the tick or mammalian environments as surrogates for the in vivo state. Although subsequent studies demonstrated that global transcriptome analyses of in vitro-cultivated organisms do not fully reflect the in vivo state [12], these initial studies provided valuable insights into B. burgdorferi gene regulation. 3.1. Response to Temperature To identify temperature-responsive genes, Ojaimi et al. [10] compared gene expression of B. burgdorferi cells grown at 23 or 35 °C (to mimic the tick or mammalian environment, respectively). 215 genes were differentially expressed at the two temperatures; with 133 showing greater expression at 35 °C relative to 23 °C. Interestingly, 136/215 (63%) temperature-responsive genes were encoded on plasmids. Of particular note, are linear plasmid lp54 and the circular cp32 plasmids; 45% of the putative ORFs encoded on lp54 exhibited temperature-regulated expression and >20% of cp32-encoded ORFs responded to temperature shift. Transcripts known to have elevated levels during mammalian infection (e.g., those for outer surface protein C (OspC), decorin binding proteins A/B (DbpAB) and the alternative sigma factor RpoS) displayed elevated expression at 35 °C. Similarly, genes subsequently shown to be more highly expressed during the tick phase (glycerol uptake and utilization operon (glpFKD) and chbC, encoding a component of the chitobiose transporter) had significantly elevated transcript levels at 23 °C [10]. Revel et al. [8] carried out a similar analysis, but also varied the pH of the growth medium so as to mimic the environment in the unfed tick (23 °C/pH 7.5) and fed tick states (37 °C/pH 6.8). A total of 94 genes were differentially expressed between the two temperatures; 79 had higher expression at 37 °C. Among transcripts elevated at the higher temperature were those for OspC and DbpA/B, as expected. In addition, transcripts for chemotaxis and motility functions and the OppA components of the oligopeptide ABC transporter were also elevated under the “fed tick” condition. Fifteen transcripts encoded on lp54 were differentially expressed, consistent with the findings of Ojaimi et al. [10]. Interestingly, there was only limited concordance between the Ojaimi and Revel data sets. This is likely the result of methodological differences between the two studies, including different array types (membrane array vs. glass slide microarray), pH of the BSK growth medium, slightly different temperatures (35 °C vs. 37 °C) and differences in the data analysis approaches. 3.2. Transcriptome of B. burgdorferi in the Host-Adapted State The paucibacillary nature of B. burgdorferi infection in mammals led Akins et al. [37] to develop an alternative approach for isolating spirochetes in the host-adapted state. The method involves cultivating B. burgdorferi in BSK medium contained within dialysis membrane chambers (DMCs) implanted in a rat peritoneal cavity [37]. Numerous studies have demonstrated that gene expression of B. burgdorferi cultivated in DMCs is markedly different from that observed for spirochetes cultivated in vitro in the same medium at 37 °C [38,39,40]. Three microarray studies have been published in which transcriptome comparisons between B. burgdorferi cultivated in vitro at 37 °C and in DMCs were reported. In Revel et al. [8], 66 genes showed altered expression between these two conditions; only 6/66 exhibited higher expression in DMCs. Surprisingly, expression of some recognized mammalian phase genes such as ospC and dbpA/B was not induced. In a subsequent study by Brooks et al. [18], a total of 125 transcripts were differentially expressed between B. burgdorferi cultivated at 37 °C in vitro and DMCs—58 transcripts were induced (including ospC) and 67 were repressed. Among the latter, only three were chromosomally-encoded and the vast majority encode putative proteins of unknown function [18]. Interestingly, there was less than 10% overlap between the Revel and Brooks’ datasets, likely the result of methodological differences between the two studies. Caimano et al. [12], also performed whole transcriptome analysis of B. burgdorferi grown at 37 °C and in DMCs. Their study was designed to identify the regulon controlled by RpoS and a direct comparison of wild-type B. burgdorferi at the two conditions was not provided. However, the results clearly demonstrate that gene expression differs substantially between in vitro- and DMC-cultivated organisms. Taken together, these findings demonstrate that temperature alone does not elicit the distinctive mammalian host modulation of B. burgdorferi gene expression, but in addition requires mammalian host-specific signals. As an alternative to DMC cultivation, Tokarz et al. [19], examined the combined effect of temperature and blood so as to simulate the environmental changes B. burgdorferi encounter as they transit from tick vector to a mammalian host. Spirochetes were incubated in the presence or absence of 6% human blood for 48 h and transcriptomes were compared. A total of 154 transcripts were differentially expressed in the presence of blood (75 induced and 79 repressed relative to no addition of blood); greater than two-thirds of the regulated transcripts are plasmid-encoded. Among the induced transcripts were those for OspC and DbpA, as expected, transcripts encoding for chemotaxis and motility functions and for two transcriptional regulators, RpoS and BosR. Given the induction of RpoS by incubation with human blood, it was of interest to compare the list of differentially expressed genes during blood co-incubation to that of RpoS-regulated genes [12]. Thirty-nine genes were found in common; 36/39 are activated by RpoS during in vitro cultivation at 37 °C and during co-incubation with blood. Further, 40 genes that were differentially expressed in the presence of human blood were also RpoS-regulated during growth in DMCs (28 induced, 12 repressed). This analysis indicates that RpoS is induced during the nymphal blood meal and controls a regulon required for tick-to-mammal transmission and during mammalian infection and is supported by additional microarray studies discussed below. Livengood et al. [14], performed global transcriptome analysis of B. burgdorferi following a 20-h incubation with human neuroglial cells. A total of 72 B. burgdorferi transcripts were differentially expressed in the neuroglial cells relative to in vitro-cultivated spirochetes; the levels of 58 were induced and 14 were repressed. 63/72 differentially expressed genes are located on either the chromosome or plasmid lp54. Numerous genes involved in motility/chemotaxis were induced in neuroglial cells, as was ospC. Among the transcripts with decreased expression was glpK, which has been shown in other studies to be repressed in the mammalian environment [41,33]. 3.3. Transcriptome of B. burgdorferi in the Tick Vector Iyer et al. [33], characterized and compared the transcriptional profiles of B. burgdorferi during acquisition (fed larvae), transmission (fed nymphs) and in a mammalian host-like environment (DMCs). This analysis required the introduction of a pre-amplification step prior to array hybridization in order to enrich for B. burgdorferi RNA [33]. A core transcriptome consisting of 397 genes was expressed under all experimental conditions and is likely required for spirochetal survival in nature. The three in vivo transcriptomes differ substantially among each other, as well as to that obtained from organisms cultivated in vitro at 37 °C indicating that spirochetes respond to a variety of host-specific signals. Among the key findings were the differential expression of genes encoding lipoproteins, transporters and enzymes in several metabolic pathways including the oxidative branch of the pentose phosphate pathway, glycerophospholipid biosynthesis and isoprenoid biosynthesis. Alterations in gene expression for chemotaxis/motility proteins were also noted suggesting that the chemotaxis/motility apparatus may vary in the tick and mammalian environments. This was the first report describing B. burgdorferi global gene expression profiles from in vivo samples containing limited copies of pathogen. The findings provide the necessary transcriptional framework for delineating B. burgdorferi regulatory pathways that operate throughout the enzootic cycle. 4. Transcriptional Regulation As already noted, B. burgdorferi must alter its gene expression program in order to adapt to growth in either the tick or mammalian environments. This adaptation is mediated by several transcriptional regulators including RpoS, Rrp1, BosR and RelBbu [4,5]. RpoS, an alternative sigma factor, controls a regulon whose members are important for transmission of B. burgdorferi from tick vector to mammalian host and/or during mammalian infection. Expression of RpoS is controlled by a signaling cascade involving Rrp2, a response regulator, and RpoN, a second alternative sigma factor [4,5]. Another signaling pathway comprised of Hk1 and Rrp1 promotes the synthesis of cyclic di-GMP and expression of c-di-GMP-dependent genes; evidence indicates that genes comprising this regulon are required for spirochetal survival in ticks [27,42]. In addition, BosR and RelBbu have been shown to control expression of substantial numbers of genes [22,28,43]. Microarray analyses have informed much of our current understanding of transcriptional regulation in B. burgdorferi. Comparative transcriptome studies employing regulatory mutants have been particularly helpful in defining the regulons controlled by various transcriptional regulators. In this section, we review these studies and also provide a secondary analysis by merging the statistically significant gene lists from the processed data reported in comparisons of wild type and mutant transcriptomes for components of the Rrp2-RpoN-RpoS and Hk1-Rrp1 regulatory cascades. In addition, we also included transcriptome data for BosR in these analyses. 4.1. Rrp2-RpoN-RpoS Regulatory Cascade Caimano et al. [12], performed a comparative microarray analysis of B. burgdorferi strain 297 wild-type and an rpoS mutant cultivated either in vitro following temperature-shift to 37 °C or within DMCs. The expression of 110 genes was affected by the absence of RpoS during in vitro growth; all had higher expression in the wild type strain implying that their transcription is at least partially dependent on RpoS. No transcripts were found to be significantly repressed. 137 genes had altered expression in spirochetes cultivated under mammalian host-like conditions (i.e., in DMCs); 103 transcripts had significantly elevated levels in wild type relative to the RpoS mutant and 44 of these were also higher in vitro. Importantly, in contrast to in vitro-grown spirochetes, 34 genes had higher expression in mutant B. burgdorferi cultivated in DMCs demonstrating that host-specific signals are required for RpoS-dependent repression. Significantly, a number of genes in this group (ospA, bba62, glp operon) have been shown to have higher transcript levels in ticks [40,33]. Norgard and co-workers generated individual mutants in rrp2, rpoN and rpoS in a strain 297 background and compared gene expression between wild-type and mutant strains in vitro [24]. They identified 98 genes that were regulated in common by either Rrp2, RpoN or RpoS; 97 exhibited higher expression in wild type and only one (bba62) had lower expression. The substantial overlap between genes regulated by RpoS and RpoN provides evidence that the two alternative sigma factors form a congruous pathway and that RpoN regulates B. burgdorferi gene expression through RpoS [44,45]. Importantly, two-thirds (68/98) of the genes were similarly regulated by RpoS in the study by Caimano et al. [12]. It is noteworthy that transcription of an additional 106 genes was affected in the rrp2 mutant. This implies that Rrp2 controls expression of a regulon unrelated to the RpoS response. Two additional publications reported on the RpoS, RpoN and Rrp2 regulons. Boardman et al. [23] generated an rrp2 mutant in an infectious strain B31 background. They observed 144 genes with altered expression in the mutant. Due to strain variation, the overlap among the two Rrp2 gene sets was only 42%. Fisher et al. [25], studied the RpoS and RpoN regulons using mutants in each of these alternative sigma factors. Curiously, there is <20% concordance between these datasets and those of Caimano et al. [12], Ouyang et al. [24] and Boardman et al. [23] probably the result of methodological differences. To generate a composite picture of the Rrp2-RpoN-RpoS regulatory circuit, we merged the datasets from Caimano et al. [12], Ouyang et al. [24], and Boardman et al. [23]. The composite gene list contained 395 genes. This gene set (referred as rrp2-RpoN-RpoS) is provided in Table S1 and was used for further analysis as described below. 4.2. Borrelia Oxidative Stress Regulator (BosR) Borrelia oxidative stress regulator (BosR; BB0647) was initially thought to mediate the oxidative stress response in B. burgdorferi [46,47]. Subsequently, it was shown to be required for transcription of RpoS [22,48,49]. Two groups have generated bosR mutants and performed microarray-based comparative transcriptome analyses. Hyde et al. [21], employed a BosR point mutant that was sensitive to oxidative stress and an insertional disruption of this point mutant that restored resistance to oxidative stress. Due to the unusual nature of the strains used (both comparison strains were mutants and no wild-type strain was included), this study is not considered further. Ouyang et al. [22], inactivated bosR in an infectious B31 strain background. They found the BosR regulon to encompass 199 genes, 137 of which were induced. These induced genes included rpoS and ospC, as was expected, and nearly two-thirds (87/137) were also part of the RpoS regulon described by these investigators [24]. The genes regulated by the Rrp2-RpoN-RpoS cascade were compared to those in the BosR regulon as shown in Figure 1. The Venn diagram reveals a substantial overlap of Rrp2-RpoN-RpoS activated genes to those activated by BosR (125/137; 91%). This finding supports the accumulating evidence that both RpoS and BosR are required for modulation of gene expression during mammalian infection [22,49,50,51]. For further analysis (see below), the Rrp2-RpoN-RpoS and BosR regulated genes were merged to generate an Rrp2-RpoN-RpoS-BosR regulon consisting of 451 genes. The additional 56 genes that are regulated by BosR only are given in Table S2. This gene set should represent B. burgdorferi genes that are differentially regulated during late nymphal tick feeding, tick-to-mammal transmission and during mammalian infection. 4.3. Hk1-Rrp1 Regulatory Circuit Three groups have reported the construction of rrp1 mutants and studied their gene expression profiles compared to those of the parent strain. Rogers et al. [26], generated an rrp1 mutant in a non-infectious strain background that also lacked many plasmids. They identified 140 transcripts with altered abundance, 131 of which had higher expression in the wild-type than the mutant. These included products of the glp operon, bba74 and spoVG which have been shown to be repressed by RpoS in DMCs and expressed at higher levels in ticks [12,33,40]. Subsequently, He et al. [27] and Caimano et al. [42] employed mutants that were generated in an infectious strain B31 background; both mutants can infect mice but cannot survive in ticks. He et al. [27] found that 120 genes had altered expression; 99 had higher transcript levels in the wild type strain. Although Caimano et al. [42], used RNA-seq, these two datasets were merged to generate a composite rrp1 regulon. The regulon contained a total of 297 genes (222 induced and 75 repressed) (Table S3). 4.4. Interaction between the Rrp2-RpoN-RpoS-BosR and Hk1-Rrp1 Regulatory Circuits Regulation of differential gene expression during the enzootic cycle is mediated primarily by RpoS and Rrp1. RpoS is responsible for modulating gene expression during spirochetal transmission from the tick vector to the mammalian host and during mammalian infection; BosR also plays a role in these processes. Rrp1 mediates changes in tick phase gene expression and regulates protective responses during the tick blood meal [4,27,42]. The interplay of the two regulatory circuits controlling RpoS and Rrp1 activity is thus critical to the adaptation and survival of B. burgdorferi in the vector and host milieus. The overlap between these two regulatory cascades has not been well characterized. In order to gain further insight into the linkage between these pathways, we compared the genes comprising the Rrp2-RpoN-RpoS-BosR regulon (Tables S1 and S2) with those comprising the Hk1-Rrp1 regulon (Table S3). The resulting Venn diagram is presented in Figure 2. 140 genes were regulated by both pathways; 83 genes were activated by both Hk1-Rrp1 and Rrp2-RpoN-RpoS-BosR and three genes were commonly repressed by both pathways. In addition, 42 genes were activated by Hk1-Rrp1 but repressed by Rrp2-RpoN-RpoS-BosR and 12 genes were induced by the RpoS pathway but repressed by Rrp1. Interestingly, the 42 genes induced by Rrp1 and repressed by RpoS include the glp operon, spoVG, bba62 and bba74. These genes are known to have significantly higher expression in ticks than in the mammalian host [33,40,41,42]. These findings are consistent with the model that Rrp1 controls a regulon whose members are required during the tick phase of the B. burgdorferi life cycle [4,42]. 5. Conclusions and Prospects for the Future Microarray studies have contributed significantly to the current understanding of B. burgdorferi genome content and transcriptional regulation. Delineation of differential gene expression patterns throughout the enzootic cycle and characterization of the regulons controlled by various transcriptional regulators mediating these processes have provided roadmaps for more detailed mechanistic investigations. The limitations of microarray analyses include the inability to detect low copy transcripts and small RNAs, restriction of gene/transcript detection to only those genes represented on the microarray and failure to recognize post-transcriptional processing events. NextGen sequencing methodologies are not subject to these limitations and will ultimately replace microarray approaches for comparative genomic and transcriptomic investigations. Acknowledgments The studies conducted in the authors’ laboratory were supported by NIH grant AI45801. Supplementary Materials The following are available online at http://www.mdpi.com/2076-3905/5/2/9/s1, Table S1: B. burgdorferi genes regulated by Rrp2-RpoN-RpoS as identified by whole genome microarray analysis; Table S2: B. burgdorferi genes regulated by BosR only as identified by whole genome microarray analysis; Table S3: B. burgdorferi genes regulated by Hk1-Rrp1 as identified by global transcriptome analysis. Click here for additional data file. Author Contributions Radha Iyer and Ira Schwartz contributed equally to the literature review, data analysis and writing of this review. Conflicts of Interest The authors declare no conflict of interest. Figure 1 Overlap between the Rrp2-RpoN-RpoS and BosR regulons. Numbers indicate the count of genes that show statistically significant induction (up arrow) or repression (down arrow). Asterisk indicates genes that are induced in one regulon and repressed in the other. Figure 2 Interplay between the Rrp2-RpoN-RpoS-BosR and Hk1-Rrp1 regulons. Numbers indicate the count of genes that show statistically significant induction (up arrow) or repression (down arrow). Asterisk indicates genes that are induced in one regulon and repressed in the other. microarrays-05-00009-t001_Table 1Table 1 Published studies utilizing B. burgdorferi microarrays. Experimental Condition Strain Microarray Type Reference Comparative genomics B31 Glass slide Liang et al., Infect. Immun., 2002 [13] Comparative genomics B31 membrane Zhong & Barbour, Mol. Microbiol., 2004 [16] Comparative genomics B31 70 m oligo glass slide Terekhova et al., J. Bacteriol., 2006 [11] Temperature response B31 Membrane Ojaimi et al., Infect. Immun., 2003 [10] Strain transcriptome comparison B31 Membrane Ojaimi et al., Infect. Immun., 2005 [17] In vitro and host-adapted (DMC) B31 Glass slide Revel et al., PNAS, 2002 [8] In vitro and host-adapted (DMC) B31 Membrane Brooks et al., Infect. Immun., 2003 [18] Blood co-incubation B31 Membrane Tokarz et al., Infect. Immun., 2004 [19] Monoclonal OspB antibody co-cultivation B31 Membrane Anderton et al., Infect. Immun., 2004 [20] Neuroglial cell co-incubation B31 Affymetrix slide Livengood et al., Infect. Immun., 2008 [14] RpoS regulon 297 70 m oligo glass slide Caimano et al., Mol. Microbiol., 2007 [12] BosR regulon B31 Membrane array Hyde et al., Microbiology, 2006 [21] BosR regulon B31 70 m oligo glass slide Ouyang et al., Mol. Microbiol., 2009 [22] Rrp2 regulon B31 70 m oligo glass slide Boardman et al., Infect. Immun., 2008 [23] Rrp2/RpoN/RpoS regulon 297 70 m oligo glass slide Ouyang et al., Microbiology, 2008 [24] RpoN/RpoS regulon B31 70 m oligo glass slide Fisher et al., PNAS, 2005 [25] Rrp1 regulon B31 70 m oligo glass slide Rogers et al., Mol. Microbiol., 2009 [26] Rrp1 regulon B31 70 m oligo glass slide He et al., PLoS Pathog., 2011 [27] RelBbu regulon 297 70 m oligo glass slide Bugrysheva et al., PLoS ONE, 2015 [28] BadR regulon B31 Nimblegen Miller et al., Mol. Microbiol., 2013 [15] HrpA regulon B31 Nimblegen Salman-Dilgimen et al., PLoS Pathog., 2013 [29] Non-human primate tissues N40, JD1 Glass slide Narasimhan et al., PNAS, 2003 [30] Fed Ticks N40 Glass slide Narasimhan et al., J. Bacteriol., 2002 [31] Mouse tissues B31 70 m oligo glass slide Pal et al., J. Infect. Dis., 2008 [32] Tick feeding stages and host-adapted (DMC) B31 70 m oligo glass slide Iyer et al., Mol. Microbiol., 2015 [33] ==== Refs References 1. Steere A.C. Grodzicki R.L. Kornblatt A.N. Craft J.E. Barbour A.G. Burgdorfer W. Schmid G.P. Johnson E. Malawista S.E. The spirochetal etiology of Lyme disease N. Engl. J. Med. 1983 308 733 740 10.1056/NEJM198303313081301 6828118 2. Benach J.L. Bosler E.M. Hanrahan J.P. Coleman J.L. Habicht G.S. Bast T.F. Cameron D.J. Ziegler J.L. Barbour A.G. Burgdorfer W. Spirochetes isolated from the blood of two patients with Lyme disease N. Engl. J. Med. 1983 308 740 742 10.1056/NEJM198303313081302 6828119 3. Mead P.S. Epidemiology of Lyme disease Infect. Dis. Clin. N. Am. 2015 29 187 210 10.1016/j.idc.2015.02.010 25999219 4. Radolf J.D. Caimano M.J. Stevenson B. Hu L.T. Of ticks, mice and men: Understanding the dual-host lifestyle of Lyme disease spirochaetes Nat. Rev. Microbiol. 2012 10 87 99 10.1038/nrmicro2714 22230951 5. Samuels D.S. Gene regulation in Borrelia burgdorferi Annu. Rev. Microbiol. 2011 65 479 499 10.1146/annurev.micro.112408.134040 21801026 6. Fraser C.M. Casjens S. Huang W.M. Sutton G.G. Clayton R. Lathigra R. White O. Ketchum K.A. Dodson R. Hickey E.K. Genomic sequence of a Lyme disease spirochaete, Borrelia burgdorferi Nature 1997 390 580 586 10.1038/37551 9403685 7. Casjens S. Palmer N. van Vugt R. Huang W.M. Stevenson B. Rosa P. Lathigra R. Sutton G. Peterson J. Dodson R.J. A bacterial genome in flux: The twelve linear and nine circular extrachromosomal DNAs in an infectious isolate of the Lyme disease spirochete Borrelia burgdorferi Mol. Microbiol. 2000 35 490 516 10.1046/j.1365-2958.2000.01698.x 10672174 8. Revel A.T. Talaat A.M. Norgard M.V. DNA microarray analysis of differential gene expression in Borrelia burgdorferi , the Lyme disease spirochete Proc. Natl. Acad. Sci. USA 2002 99 1562 1567 10.1073/pnas.032667699 11830671 9. Ojaimi C. Brooks C. Akins D. Casjens S. Rosa P. Elias A. Barbour A. Jasinskas A. Benach J. Katonah L. Borrelia burgdorferi gene expression profiling with membrane-based arrays Methods Enzymol. 2002 358 165 177 12474386 10. Ojaimi C. Brooks C. Casjens S. Rosa P. Elias A. Barbour A. Jasinskas A. Benach J. Katona L. Radolf J. Profiling of temperature-induced changes in Borrelia burgdorferi gene expression by using whole genome arrays Infect. Immun. 2003 71 1689 1705 10.1128/IAI.71.4.1689-1705.2003 12654782 11. Terekhova D. Iyer R. Wormser G.P. Schwartz I. Comparative genome hybridization reveals substantial variation among clinical isolates of Borrelia burgdorferi sensu stricto with different pathogenic properties J. Bacteriol. 2006 188 6124 6134 10.1128/JB.00459-06 16923879 12. Caimano M.J. Iyer R. Eggers C.H. Gonzalez C. Morton E.A. Gilbert M.A. Schwartz I. Radolf J.D. Analysis of the RpoS regulon in Borrelia burgdorferi in response to mammalian host signals provides insight into RpoS function during the enzootic cycle Mol. Microbiol. 2007 65 1193 1217 10.1111/j.1365-2958.2007.05860.x 17645733 13. Liang F.T. Nelson F.K. Fikrig E. DNA microarray assessment of putative Borrelia burgdorferi lipoprotein genes Infect. Immun. 2002 70 3300 3303 10.1128/IAI.70.6.3300-3303.2002 12011030 14. Livengood J.A. Schmit V.L. Gilmore R.D. Jr. Global transcriptome analysis of Borrelia burgdorferi during association with human neuroglial cells Infect. Immun. 2008 76 298 307 10.1128/IAI.00866-07 17984208 15. Miller C.L. Karna S.L. Seshu J. Borrelia host adaptation regulator (BadR) regulates rpoS to modulate host adaptation and virulence factors in Borrelia burgdorferi Mol. Microbiol. 2013 88 105 124 10.1111/mmi.12171 23387366 16. Zhong J. Barbour A.G. Cross-species hybridization of a Borrelia burgdorferi DNA array reveals infection- and culture-associated genes of the unsequenced genome of the relapsing fever agent Borrelia hermsii Mol. Microbiol. 2004 51 729 748 10.1046/j.1365-2958.2003.03849.x 14731275 17. Ojaimi C. Mulay V. Liveris D. Iyer R. Schwartz I. Comparative transcriptional profiling of Borrelia burgdorferi clinical isolates differing in capacities for hematogenous dissemination Infect. Immun. 2005 73 6791 6802 10.1128/IAI.73.10.6791-6802.2005 16177357 18. Brooks C.S. Hefty P.S. Jolliff S.E. Akins D.R. Global analysis of Borrelia burgdorferi genes regulated by mammalian host-specific signals Infect. Immun. 2003 71 3371 3383 10.1128/IAI.71.6.3371-3383.2003 12761121 19. Tokarz R. Anderton J.M. Katona L.I. Benach J.L. Combined effects of blood and temperature shift on Borrelia burgdorferi gene expression as determined by whole genome DNA array Infect. Immun. 2004 72 5419 5432 10.1128/IAI.72.9.5419-5432.2004 15322040 20. Anderton J.M. Tokarz R. Thill C.D. Kuhlow C.J. Brooks C.S. Akins D.R. Katona L.I. Benach J.L. Whole-genome DNA array analysis of the response of Borrelia burgdorferi to a bactericidal monoclonal antibody Infect. Immun. 2004 72 2035 2044 10.1128/IAI.72.4.2035-2044.2004 15039324 21. Hyde J.A. Seshu J. Skare J.T. Transcriptional profiling of Borrelia burgdorferi containing a unique bos R allele identifies a putative oxidative stress regulon Microbiology 2006 152 2599 2609 10.1099/mic.0.28996-0 16946255 22. Ouyang Z. Kumar M. Kariu T. Haq S. Goldberg M. Pal U. Norgard M.V. BosR (BB0647) governs virulence expression in Borrelia burgdorferi Mol. Microbiol. 2009 74 1331 1343 10.1111/j.1365-2958.2009.06945.x 19889086 23. Boardman B.K. He M. Ouyang Z. Xu H. Pang X. Yang X.F. Essential role of the response regulator Rrp2 in the infectious cycle of Borrelia burgdorferi Infect. Immun. 2008 76 3844 3853 10.1128/IAI.00467-08 18573895 24. Ouyang Z. Blevins J.S. Norgard M.V. Transcriptional interplay among the regulators Rrp2, RpoN and RpoS in Borrelia burgdorferi Microbiology 2008 154 2641 2658 10.1099/mic.0.2008/019992-0 18757798 25. Fisher M.A. Grimm D. Henion A.K. Elias A.F. Stewart P.E. Rosa P.A. Gherardini F.C. Borrelia burgdorferi σ54 is required for mammalian infection and vector transmission but not for tick colonization Proc. Natl. Acad. Sci. USA 2005 102 5162 5167 10.1073/pnas.0408536102 15743918 26. Rogers E.A. Terekhova D. Zhang H.M. Hovis K.M. Schwartz I. Marconi R.T. Rrp1, a cyclic-di-GMP-producing response regulator, is an important regulator of Borrelia burgdorferi core cellular functions Mol. Microbiol. 2009 71 1551 1573 10.1111/j.1365-2958.2009.06621.x 19210621 27. He M. Ouyang Z. Troxell B. Xu H. Moh A. Fu X.-Y. Piesman J. Norgard M.V. Gomelsky M. Yang X.F. Cyclic di-GMP is essential for the survival of Borrelia burgdorferi in ticks PLoS Pathog. 2011 7 9 10.1371/journal.ppat.1002133 21738477 28. Bugrysheva J.V. Pappas C.J. Terekhova D.A. Iyer R. Godfrey H.P. Schwartz I. Cabello F.C. Characterization of the RelBbu regulon in Borrelia burgdorferi reveals modulation of glycerol metabolism by (p)ppGpp PLoS ONE 2015 10 9 10.1371/journal.pone.0118063 25688856 29. Salman-Dilgimen A. Hardy P.O. Radolf J.D. Caimano M.J. Chaconas G. HrpA, an RNA helicase involved in RNA processing, is required for mouse infectivity and tick transmission of the Lyme disease spirochete PLoS Pathog. 2013 9 9 10.1371/journal.ppat.1003841 24367266 30. Narasimhan S. Caimano M.J. Liang F.T. Santiago F. Laskowski M. Philipp M.T. Pachner A.R. Radolf J.D. Fikrig E. Borrelia burgdorferi transcriptome in the central nervous system of non-human primates Proc. Natl. Acad. Sci. USA 2003 100 15953 15958 10.1073/pnas.2432412100 14671329 31. Narasimhan S. Santiago F. Koski R.A. Brei B. Anderson J.F. Fish D. Fikrig E. Examination of the Borrelia burgdorferi transcriptome in Ixodes scapularis during feeding J. Bacteriol. 2002 184 3122 3125 10.1128/JB.184.11.3122-3125.2002 12003955 32. Pal U. Dai J. Li X. Neelakanta G. Luo P. Kumar M. Wang P. Yang X. Anderson J.F. Fikrig E. A differential role for BB0365 in the persistence of Borrelia burgdorferi in mice and ticks J. Infect. Dis. 2008 197 148 155 10.1086/523764 18171298 33. Iyer R. Caimano M.J. Luthra A. Axline D. Jr. Corona A. Iacobas D.A. Radolf J.D. Schwartz I. Stage-specific global alterations in the transcriptomes of Lyme disease spirochetes during tick feeding and following mammalian host adaptation Mol. Microbiol. 2015 95 509 538 10.1111/mmi.12882 25425211 34. Schutzer S.E. Fraser-Liggett C.M. Casjens S.R. Qiu W.G. Dunn J.J. Mongodin E.F. Luft B.J. Whole-genome sequences of thirteen isolates of Borrelia burgdorferi J. Bacteriol. 2011 193 1018 1020 10.1128/JB.01158-10 20935092 35. Casjens S.R. Mongodin E.F. Qiu W.G. Luft B.J. Schutzer S.E. Gilcrease E.B. Huang W.M. Vujadinovic M. Aron J.K. Vargas L.C. Genome stability of Lyme disease spirochetes: Comparative genomics of Borrelia burgdorferi plasmids PLoS ONE 2012 7 9 10.1371/journal.pone.0033280 22432010 36. Mongodin E.F. Casjens S.R. Bruno J.F. Xu Y. Drabek E.F. Riley D.R. Cantarel B.L. Pagan P.E. Hernandez Y.A. Vargas L.C. Inter- and intra-specific pan-genomes of Borrelia burgdorferi sensu lato: Genome stability and adaptive radiation BMC Genom. 2013 14 10.1186/1471-2164-14-693 24112474 37. Akins D.R. Bourell K.W. Caimano M.J. Norgard M.V. Radolf J.D. A new animal model for studying Lyme disease spirochetes in a mammalian host-adapted state J. Clin. Investig. 1998 101 2240 2250 10.1172/JCI2325 9593780 38. Yang X. Popova T.G. Hagman K.E. Wikel S.K. Schoeler G.B. Caimano M.J. Radolf J.D. Norgard M.V. Identification, characterization, and expression of three new members of the Borrelia burgdorferi mlp (2.9) lipoprotein gene family Infect. Immun. 1999 67 6008 6018 10531261 39. Caimano M.J. Eggers C.H. Gonzalez C.A. Radolf J.D. Alternate sigma factor RpoS is required for the in vivo -specific repression of Borrelia burgdorferi plasmid lp54-borne osp A and lp6.6 genes J. Bacteriol. 2005 187 7845 7852 10.1128/JB.187.22.7845-7852.2005 16267308 40. Mulay V.B. Caimano M.J. Iyer R. Dunham-Ems S. Liveris D. Petzke M.M. Schwartz I. Radolf J.D. Borrelia burgdorferi bba74 is expressed exclusively during tick feeding and is regulated by both arthropod- and mammalian host-specific signals J. Bacteriol. 2009 191 2783 2794 10.1128/JB.01802-08 19218390 41. Pappas C.J. Iyer R. Petze M.M. Caimano M.J. Radolf J.D. Schwartz I. Borrelia burgdorferi requires glycerol for maximum fitness during the tick phase of the enzootic cycle PLoS Pathog. 2011 7 9 10.1371/journal.ppat.1002102 21750672 42. Caimano M.J. Dunham-Ems S. Allard A.M. Cassera M.B. Kenedy M. Radolf J.D. Cyclic di-GMP modulates gene expression in Lyme disease spirochetes at the tick-mammal interface to promote spirochete survival during the blood meal and tick-to-mammal transmission Infect. Immun. 2015 83 3043 3060 10.1128/IAI.00315-15 25987708 43. Drecktrah D. Lybecker M. Popitsch N. Rescheneder P. Hall L.S. Samuels D.S. The Borrelia burgdorferi RelA/SpoT homolog and stringent response regulate survival in the tick vector and global gene expression during starvation PLoS Pathog. 2015 11 9 10.1371/journal.ppat.1005160 26371761 44. Hubner A. Yang X. Nolen D.M. Popova T.G. Cabello F.C. Norgard M.V. Expression of Borrelia burgdorferi OspC and DbpA is controlled by a RpoN-RpoS regulatory pathway Proc. Natl. Acad. Sci. USA 2001 98 12724 12729 10.1073/pnas.231442498 11675503 45. Smith A.H. Blevins J.S. Bachlani G.N. Yang X.F. Norgard M.V. Evidence that RpoS (σ s ) in Borrelia burgdorferi is controlled directly by RpoN (σ 54 / σ n ) J. Bacteriol. 2007 189 2139 2144 10.1128/JB.01653-06 17158681 46. Boylan J.A. Posey J.E. Gherardini F.C. Borrelia oxidative stress response regulator, BosR, a distinctive Zn-dependent transcriptional activator Proc. Natl. Acad. Sci. USA 2003 100 11684 11689 10.1073/pnas.2032956100 12975527 47. Katona L.I. Tokarz R. Kuhlow C.J. Benach J. Benach J.L. The fur homologue in Borrelia burgdorferi J. Bacteriol. 2004 186 6443 6456 10.1128/JB.186.19.6443-6456.2004 15375125 48. Hyde J.A. Shaw D.K. Smith III R. Trzeciakowski J.P. Skare J.T. The BosR regulatory protein of Borrelia burgdorferi interfaces with the RpoS regulatory pathway and modulates both the oxidative stress response and pathogenic properties of the Lyme disease spirochete Mol. Microbiol. 2009 74 1344 1355 10.1111/j.1365-2958.2009.06951.x 19906179 49. Samuels D.S. Radolf J.D. Who is the BosR around here anyway? Mol. Microbiol. 2009 74 1295 1299 10.1111/j.1365-2958.2009.06971.x 19943896 50. Ouyang Z. Deka R.K. Norgard M.V. BosR (BB0647) controls the RpoN-RpoS regulatory pathway and virulence expression in Borrelia burgdorferi by a novel DNA-binding mechanism PLoS Pathog. 2011 7 9 10.1371/journal.ppat.1001272 21347346 51. Wang P. Dadhwal P. Cheng Z. Zianni M.R. Rikihisa Y. Liang F.T. Li X. Borrelia burgdorferi oxidative stress regulator BosR directly represses lipoproteins primarily expressed in the tick during mammalian infection Mol. Microbiol. 2013 89 1140 1153 10.1111/mmi.12337 23869590

PMC005xxxxxx/PMC5003486.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020010microarrays-05-00010ArticleStromal Activation by Tumor Cells: An in Vitro Study in Breast Cancer Merlino Giuseppe 1Miodini Patrizia 1Paolini Biagio 2Carcangiu Maria Luisa 2Gennaro Massimiliano 3Dugo Matteo 4Daidone Maria Grazia 1Cappelletti Vera 1*Negrini Massimo Academic Editor1 Department of Experimental Oncology and Molecular Medicine, Fondazione IRCCS Istituto Nazionale dei Tumori, Via Amadeo 42, 20133 Milan , Italy; peppem1985@gmail.com (G.M.); patrizia.miodini@istitutotumori.mi.it (P.M.); mariagrazia.daidone@istitutotumori.mi.it (M.G.D.)2 Department of Pathology, Fondazione IRCCS Istituto Nazionale dei Tumori, Via Venezian 1, 20133 Milan , Italy; biagio.paolini@istitutotumori.mi.it (B.P.); marialuisa.carcangiu@istitutotumori.mi.it (M.L.C.)3 Department of Surgery, Fondazione IRCCS Istituto Nazionale dei Tumori, Via Venezian 1, 20133 Milan , Italy; massimiliano.gennaro@istitutotumori.mi.it4 Functional Genomics Core Facility, Department of Experimental Oncology and Molecular Medicine, Fondazione IRCCS Istituto Nazionale dei Tumori, Via Amadeo 42, 20133Milan, Italy; matteo.dugo@istitutotumori.mi.it* Correspondence: vera.cappelletti@istitutotumori.mi.it; Tel.: +39-02-2390-2700; Fax: +39-02-2390-276418 5 2016 6 2016 5 2 1031 3 2016 10 5 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Background: The tumor microenvironment participates in the regulation of tumor progression and influences treatment sensitivity. In breast cancer, it also may play a role in determining the fate of non-invasive lesions such as ductal carcinoma in situ (DCIS), a non-obligate precursor of invasive diseases, which is aggressively treated despite its indolent nature in many patients since no biomarkers are available to predict the progression of DCIS to invasive disease. In vitro models of stromal activation by breast tumor cells might provide clues as to specific stromal genes crucial for the transition from DCIS to invasive disease. Methods: normal human dermal fibroblasts (NHDF) were treated under serum-free conditions with cell culture media conditioned by breast cancer cell lines (SkBr3, MDA-MB-468, T47D) for 72 h and subjected to gene expression profiling with Illumina platform. Results: TGM2, coding for a tissue transglutaminase, was identified as candidate gene for stromal activation. In public transcriptomic datasets of invasive breast tumors TGM2 expression proved to provide prognostic information. Conversely, its role as an early biosensor of tumor invasiveness needs to be further investigated by in situ analyses. Conclusion: Stromal TGM2 might probably be associated with precancerous evolution at earlier stages compared to DCIS. breast cancerstroma-tumor interactionstromal activationfibroblastsbiomarkersductal carcinoma in situin-vitro modelsgene expression profiles ==== Body 1. Introduction Tumors arise within a microenvironment, which cannot be considered a simple bystander since it plays an active role in the acquisition of the malignant traits of the diseases to such an extent, that it may even be regarded as a therapeutic target [1,2,3,4,5,6]. The tumor microenvironment consists of the extracellular matrix (ECM) and various cells of hematopoietic and mesenchymal origin, such as cells from the lymphoid lineage (T cells, B cells and natural killer (NK) cells) or the myeloid lineage such as macrophages, neutrophils and myeloid-derived suppressor cells. Mesenchymal cells such as fibroblasts, myofibroblasts and adipocytes also share the microenvironment with the tumor cells. Current molecular subtyping and prognostic assessment of breast tumors is based on molecular features of cancer cells, but numerous stromal signatures with clinical validity have also been defined [7,8]. The current paradigm indeed envisages a strict connection between evolution of epithelial and stromal compartments, which is resulting into a symbiotic relation when tumor cells co-opt stromal cells to activate tumor promoting programs such as acquirement of invasive traits [9,10], or in a restrictive interaction when the microenvironment offers a barrier to tumor development [11]. The latter is particularly evident considering the frequently observed local immunological response mediated by infiltrating immune and inflammatory cells. Infiltrating lymphocytes and lately immune signatures do indeed represent tools predicting better prognosis and major sensitivity to chemotherapy especially in triple negative (TNBC) and in Her2 breast tumor [12,13]. Although they are the result of a co-evolution with the tumor, changes in the stromal compartment occurring in response to the development and progression of a tumor share to a certain extent traits that are common among different tumor types. Common and distinct mechanisms shared by activated fibroblasts have been recently reviewed in different malignant contexts such as breast, prostate and lung tumors [14,15]. Cancer associated fibroblasts (CAFs), although still eluding a clear-cut definition, often represent the most abundant cells in the breast tumor microenvironment, and behave as activated fibroblasts producing ECM components such as collagens, proteoglycans, cytokines, proteases and growth factors [15]. In breast cancer, DCIS has been recognized as a non-obligate precursor of invasive disease. In fact, a variable proportion (25%–50%, mainly depending on histological grade) of patients diagnosed with DCIS progress to invasive disease. This raises many unanswered questions as to the biological reasons for progression and constitutes a clinical dilemma for the management of DCIS patients that consequently often undergo over-treatment [16]. Invasive ipsilateral recurrence are the result of host- and tumor-specific factors, and tools such as the early developed Van Nuys prognostic index [17] or other later developed nomograms [18] are trying to define accurate predictions. Nonetheless, the field is still awaiting biologically-derived biomarkers able to refine prediction. In this study, we hypothesized that studying in vitro the transcriptional activation of normal fibroblasts in the presence of a tumor might inform on tumor-subtype specific alterations occurring in the stromal compartment of clinical tumors offering hints on biomarkers reflecting the early sensing of the stroma in the presence of the incipient tumor. We therefore developed an in vitro approach to model the early activation of fibroblasts. Our approach was focused on paracrine-mediated signals as it included the treatment of normal, dermal-derived fibroblast with the secretome (tumor conditioned medium) obtained from three distinct breast cancer cell lines, representing the luminal, the HER2-enriched and the triple negative molecular subtypes. Genes showing perturbations at the transcriptional level were considered as possible candidate biomarkers recapitulating an early activation of the stromal compartment. The hypothesis was challenged in the context of clinical DCIS. We identified TGM2 as a possible stromal early sensor of the malignancy and its role should be tested in earlier stages of malignancy development. 2. Material and Methods 2.1. Cell Lines Human breast cell lines (BCCLs), SkBr3, MDA-MB-468, T47D, purchased from the American Type Culture Collection, were cultured in McCoy’s 5A (SkBr3) or Dulbecco’s modified Eagle’s medium (MDA-MB-468, T47D) (Lonza, Slough, UK) supplemented with 5% (T47D) or 10% fetal bovine serum (Lonza). The human normal fibroblast (NF) cell line NHDF (normal human dermal fibroblasts), derived from human normal derma (Lonza) was cultured in Fibroblast Basal Medium (FBM) supplemented with Fibroblast Growth Medium-2 (FGM-2) Bullet kit (Lonza), containing 2% fetal bovine serum (FBS), 0.1% Insulin, 0.1% gentamicin, amphotericin GA 1000, 0.1% fibroblast growth factor (FGF). All cell lines were cultured at 37 °C in 95% humidified air in the presence of 5% CO2 and authenticated with short tandem repeat DNA profiling analysis by the Functional Genomic Unit of the Department of Experimental Oncology at Fondazione IRCCS Istituto Nazionale Tumori of Milano (INT). 2.2. Conditioned Medium Collection Conditioned media (c.m.) were collected from BCCLs (SkBr3, MDA-MB-468, T47D) separately plated (1.65 × 106 cells) in F25 cm2 flasks, in DMEM F/12 5% FBS/FBM 2% FBS (1:1) thereafter referred as MIX medium. After cell attachment, the medium was replaced with 7 mL of serum-free MIX medium and collected at 72 h. After collection, media were clarified by centrifugation (1400× g for 3 min). C.m. produced by SkBr3, MDA-MB-468, T47D were used for the treatment (72 h) of NHDF (NAF) plated in 6-wells plate at a density of 750,000 cells. 2.3. Flow Cytometry Analysis Fibroblasts cell suspension was washed and incubated in staining solution containing bovine serum albumin (BSA) 1% and EDTA 2 mM with specific antibodies used at appropriate dilution as indicated by datasheet. The following antibodies, directed against extracellular antigens, were employed: FITC anti-human CD90 (cat.# 11-0909-41, eBiosciences, Hatfield, UK) at a 1:50 dilution, PE anti-human CD105 (cat.# FAB10971P, R&D Systems Inc, Minneapolis, MB, USA) at a 1:10 dilution, PE anti-human CD166 (cat.#559263, BD Biosciences, San Jose, CA, USA ) at a 1:10 dilution CD73 (cat.#550257, BD Biosciences) at a 1:10 dilution. For detection of the intracellular expression of α- Smooth Muscle Actin, cells were fixed with paraformaldehyde 4% and permeabilized with a solution containing 0.5% saponin, 0.1% BSA in phosphate buffered saline. The anti-human α-Smooth Muscle Actin antibody (R&D Systems, Minneapolis, MN, USA) was used at 1:50 dilution in the permeabilization solution. 2.4. RNA Extraction and Microarray Hybridization Total RNA was extracted from treated NHDF cells using Qiazol (Qiagen, Valencia, CA, USA) reagent followed by a clean-up treatment with the RNeasy Mini kit (Qiagen) according to manufacturer’s recommendations to remove contaminating genomic DNA. RNA integrity and purity were assessed by Bioanalyzer (Agilent Technologies, Waldbronn, Germany) and concentration was evaluated using a NanoDrop 2000c spectrophotometer (Thermo Scientific, Waltham, MA, USA). RNA samples were processed for microarray hybridization by INT Functional Genomics core facility. Briefly, 300 ng of total RNA were reverse-transcribed, labeled with biotin and amplified overnight with the Illumina RNA TotalPrep Amplification kit (Ambion, Austin, TX, USA) according to manufacturer’s instructions. One µg of biotinylated cRNA was mixed with the Hyb E1 hybridizatioin buffer containing 37.5% (w/w) formamide and then hybridized to Illumina HumanHT-12v4 Expression BeadChip (Illumina, Inc., San Diego, CA, USA) at 58 °C for 18 h. Arrays were washed with manufacturer’s E1BC solution (Illumina Inc, San Diego, CA, USA), stained with 1 µg/mL Cy3-streptavidine (GE Healthcare, Buckinghamshire, UK) and scanned with Illumina BeadArray Reader. Illumina BeadScan software was used for image acquisition and recovery of primary signals. 2.5. qPCR Expression levels for TGM2, IL6 and TGFB, were evaluated by qPCR with TaqMan Fast Universal PCR Master Mix assay (Thermo Scientific) using GAPDH as housekeeping gene. Following primers were obtained from Applied Biosystems (Foster City, CA, USA): TGM2, assay Hs00190278_m1; IL6 assay Hs00985639_m1; TGFB, assay Hs00998133_m1; GAPDH, Hs00266705_g1). cDNA was generated from 400 ng of RNA. Reverse transcription was run for 10 min at 25 °C, followed by 60 min at 42 °C and 5 min at 85 °C with the High-Capacity cDNA Reverse Transcription kit (Thermo Scientific) in a total volume of 20 µL, according to the manufacturer’s instructions. Data were computed with the ΔΔCt method [19]. 2.6. Protein Studies For evaluation of secreted TGM2 protein, 7.5 × 105 NHDFs were seeded on 6-wells and grown for 72 h with BCCL-derived c.m. or control medium. At the end of the treatment, dishes were washed twice with PBS and treated with a hypotonic buffer (NH4OH 20 mM) for 20 min. After two more washes with PBS, extracellular matrix proteins deposed on the plastic dish were directly recovered in loading buffer with the help of a scraper and analyzed for TG2 presence by Western blotting using the mouse monoclonal anti-TG2 antibody (CUB 7402, cat.#ab2386, Abcam, Cambridge, UK). TGM2 protein levels in cell-derived extracellular matrix were normalized respect to the number of seeded cells grown in parallel to the cells used for extracellular matrix recovering. For evaluation of intracellular TGM2 protein, cells were lysed in Laemmli sample buffer containing 5% β-mercaptoethanol and boiled for 3 min. Aliquots containing 80 µg of total cell proteins were fractionated on 12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to nitrocellulose membranes. Membranes were blocked in 3% nonfat milk in Tris-buffered saline for 20 min at room temperature and then incubated overnight at 4 °C with mouse monoclonal anti-TG2 antibody. After washing in Tris-buffered saline containing 0.1% Tween 20, the filters were incubated with peroxidase anti-rabbit immunoglobulin G, and specific complexes were revealed by chemiluminescence according to the Amersham enhanced chemiluminescence Western blotting Detection kit (GE Healthcare, Buckinghamshire, UK). TGM2 protein levels were densitometrically analyzed and normalized for ß-actin protein expression evaluated on the same gel. 2.7. Data Analysis and Statistics 2.7.1. Microarray Data Analysis The Illumina BeadStudio software version 3.8 was used to retrieve microarray raw data and the lumi [20] Bioconductor package was employed for data pre-processing. After quality control, robust spline normalization was applied to log2-transformed data and probes with a detection p-value < 0.01 in at least one sample were selected. For genes represented by multiple probes we selected the probe detected in the highest number of samples (according to detection p-value < 0.01). In case of ties, the probe with the highest interquartile range was chosen. Raw and processed data were deposited to the Gene Expression Omnibus data repository [21] with accession number GSE80035. The limma [22] Bioconductor package (R version 2.15.2) was used for differential expression analysis. The Benjamini-Hochberg false discovery rate (FDR) method was applied to adjust for multiple testing. Genes with an FDR < 0.0001 and a log2 fold change < −1 or > 1 were selected as differentially expressed (DE). Gene Set Enrichment Analysis (GSEA) [23] was run in pre-ranked mode using pre-ranked gene lists according to the t statistic obtained from differential expression analysis with limma. A total of 1843 gene sets, including canonical pathways (c2) and gene ontology (c5) collections from MSigDB database (http://software.broadinstitute.org/gsea/msigdb) were tested. Enrichment was considered significant at FDR < 0.05. 2.7.2. Statistical and Survival Analyses Differences in two-group comparisons of continuous variables were assessed using two-tailed unpaired or paired Student’s t-test, as appropriate. For comparisons involving more than two groups we applied ANOVA followed by Tukey’s post-hoc test. Association between TGM2 expressions with outcome was evaluated using the Kaplan-Meier method. Samples included in each dataset were classified in TGM2-high or TGM2-low according to whether the expression of TGM2 was greater than its median expression. Disease-specific survival (DSS) was the main end point in the METABRIC cohort while distant metastasis free survival (DMFS) was the endpoint in the combined dataset. Survival differences were evaluated using the log-rank test. All observations were censored at 10 years of follow-up. A p-value < 0.05 was considered for statistical significance. 2.8. Case Series Samples collected at our Institution were used for immunohistochemical detection of TG2. Ten patients with histologically confirmed DCIS, median age 60 years (range 40–66) were included in the present study. The DCIS median size was 7 mm (range 5–20 mm); six tumors were defined as grade-3, three as grade-2 and one was grade-1. Four patients had estrogen receptor-positive tumors by immunohistochemistry and six were defined as human epidermal growth factor 2 (HER)-2 positive by immunohistochemistry and FISH analysis. A written informed consent signed by each patient authorized the use of material left over from the diagnosis for research purposes. Correlations with clinical data were investigated in publicly available datasets and in a small case series from our Institution. TGM2 expression in stromal tissue adjacent to normal or tumor breast cells was evaluated in two publicly available datasets whose normalized data were downloaded from GEO with accession numbers GSE14548 [24] and GSE26910 [25]. Survival analysis for TGM2 was performed using METABRIC [26] and a combined dataset derived from publicly available gene expression data retrieved from GEO (GSE2034, GSE2990, GSE5327, and GSE11121). For the combined dataset, gene expression data were merged similarly as described in Callari et al. [13]. 2.9. Immunohistochemistry Four-µm thick formalin-fixed paraffin-embedded tissue sections were used for the immunohistochemistry (IHC) procedure. After drying overnight at 37 °C, deparaffinization with xylene/ethanol and rehydration were performed. The sections were treated at 95 °C in citrate buffer for antigen retrieval, cooled and washed in TBS. Slides were incubated with an anti-TGM2 protein primary antibody (CUB 7402, abcam ab2386) overnight at 4 °C using the Envision Dual Link System-HRP DAB+ (cat.# k5007, Dako Italia, Cernusco sul Naviglio, IT) for detection and counterstained with Mayer’s haematoxylin. 3. Results The normal dermal fibroblast cell line NHDF was used for modeling the fibroblast component of the stroma. NHDFs have been previously characterized by FACS analysis with respect to the expression of superficial mesenchymal markers: CD90, CD105, CD166, CD73 confirming their mesenchymal phenotype. The CD90 marker identified two cell populations, 39.1% CD90− cells and 60.9% CD90+ cells. For the others mesenchymal markers we observed: 97.2% CD105+ cells, 87.4% CD166+ cells and 100% CD73+. Activation of the NHDF to CAF-like fibroblasts induced by treatment with BCCL conditioned medium, was checked by analyzing the expression of intracellular α- smooth muscle actin (α-SMA). Only SkBr3 increased two-fold the expression of α-SMA (8.77% vs. 18.2%). However, the activation status of NHDF cells upon treatment with BCCL-derived conditioned media is supported by our previous study [14], where we showed that a core of stromal genes and pathways reported as modulated in our in vitro model is shared with other tissue-derived and activated fibroblast from breast lesions, but also from lung and prostate cancers. 3.1. In Vitro Modeling of Tumor-Stroma Interactions BCCL-derived conditioned medium and the NF cell line NHDF were used to build an in vitro model mimicking stromal activation by human tumors. To such a purpose NHDF cells were treated with the conditioned media obtained from three different BCCLs under serum-free conditions, with the aim of recapitulating tumor cell secretome effects exerted in the immediate proximity of stromal cells. The three BCCLs lines were chosen to represent luminal (T47D), HER2-enriched (SkBr3) and triple negative (MDA-MB-468) cell lines. To infer gene expression alterations in the stroma induced by factors released by proximal tumor cells we compared the transcriptomic profiles of NHDF treated with conditioned medium from tumor cell lines and their respective control cells, grown in serum-free medium. Using a highly stringent significance threshold (FDR < 0.0001 and log2 fold change > 1 or < −1) we identified 80, 177 and 106 differentially expressed (DE) genes after stimulation by the triple negative MDA-MB-468 cell line, the HER2-enriched and the luminal cell line, respectively (Figure 1A–C). In all cases, upon stimulation with the tumor-derived secretome, the number of gene up-regulated in the stroma was superior to the number of down-regulated genes as can be seen from the volcano plots reported in Figure 1. To get insight into the biological processes altered upon stimulation with tumor conditioned media we performed GSEA and we identified several gene sets significantly (FDR < 0.05) positively or negatively associated to BCCL-promoted NF stimulation. To facilitate biological interpretation of the obtained results, significantly enriched gene sets were grouped into macro-categories related to specific pathways or to distinct biological functions. Results are summarized in Figure 2A. These results show that when focusing on the macro-categories there were no differences in the biological effects exerted on the NF transcriptome as a function of the specific tumor cell line employed for stimulation (Figure 2A). However, at the single gene level there were differences in gene modulations across treated samples (Figure 2B). Moreover, given the specific type of experimental model, i.e. focused on paracrine effects, the stimulations were mainly observed for signaling pathways including TNF, EGF, IL-6 and TGFβ and for targets of transcription factors activated by intracellular signaling (STAT3 and NFkB). In keeping with the role of fibroblasts in the deposition of the extracellular matrix, a positive enrichment in NF conditioned cells of genes involved in ECM was also observed. Only in the case of apoptosis we observed a negative enrichment, suggesting a suppression of apoptotic pathways in the stroma in the presence of tumor foci. Since as reported above, the activations observed in the NF transcription program did not appear to depend on the molecular subtype of the BCCLs used for stimulation, we focused our attention on the genes commonly modulated by all three BCCLs. The lists of DE genes from each of the three comparisons (Supplementary File 1) were intersected to identify shared and private modulated genes. We identified a core of 28 genes commonly up-regulated (Figure 3) and a core set of 16 genes commonly down-regulated (Supplementary Figure S1). Twenty-eight genes were commonly up-regulated and are listed in the right part of the Figure 3. Among the 28 up-regulated genes shared by the three comparisons we focused our attention on TGM2. TGM2 gene codes for a tissue transglutaminase (TG2) that can crosslink ECM components, stabilizing the matrix and promoting cell attachment, cell motility and a general remodeling of the matrix [27]. Besides, TG2 is also considered a component of cell and tissue defense mechanisms in response to cell damage [28], making us hypothesize that it might represent an early sensor of transformation in a context where ECM plays a crucial role due to its activation in stimulated NHDF. To re-enforce the interpretation of results derived from microarray experiments, a technical validation was carried out on residual RNA derived from the same experiment for TGM2, IL6 and TGFB. Figure 4 reports the log2-transformed fold change values derived from microarray gene expression data for three identified genes (panel A) and the −ΔΔCt values obtained from qPCR measurements (panel B). As expected, qPCR yielded larger gene expression differences compared to microarray data: however, the direction of the gene modulations observed by the two technical approaches was consistent as was the gene expression pattern. Despite TGFB was not included in the lists of DE genes from the microarray experiment it was included in the technical validation as the TGFβ pathway was still among the positive enriched pathways. The qPCR data confirmed the poor modulation of the gene itself. The in vitro stroma activation model was further characterized with respect to TGM2. To such a purpose, cells were seeded again and treated as described for the gene expression experiments using BCCL-derived conditioned media. After a 72-h treatment cell lysates were obtained for evaluation by Western blotting of TG2, the protein encoded by the TGM2 gene (Figure 5). The TG2 protein was not expressed in our BCCLs, but was instead present in NHDF both at baseline ad after treatment with BCCL-derived conditioned media. TG2 expression was slightly up-regulated even by treatment with control culture media as well as by treatment with the conditioned media. We therefore reasoned that if, despite the up-regulation observed for the TGM2 gene upon stimulation by tumor cells, no regulation was observed in the NHDF intracellular levels, the TG2 protein might be not retained within the cell, but immediately secreted in keeping with its role in ECM remodeling. In fact, as shown in Figure 5, neither untreated NHDF, nor BCLL line did depose any detectable amounts of TG2 on the culture dish, whereas stimulation with media conditioned by BCCL and not by control media, caused a massive increase in TG2 deposition. Finally, we excluded the presence of TG2 in the BCCL supernatants (data not shown). 3.2. Clinical Role of TGM2 To gain insight into the clinical role of TGM2 we first evaluated its expression in two publicly available gene expression datasets of stromal and tumor cells laser-capture microdissected from breast cancer clinical specimens. In GSE14548 dataset, including a total of 34 stromal samples referring to stroma adjacent to normal mammary cells, DCIS or invasive ductal carcinoma (IDC) we found TGM2 was consistently over-expressed in tumor-associated stroma with respect to stroma associated with normal mammary cells (Figure S2A). In particular, TGM2 expression levels were increased in the stroma associated with in situ tumor lesions, suggesting that this gene may act as an early stromal sensor of malignant transformation, also in consideration of the slight decrease in IDC. In GSE26910 dataset, composed of 6 matched tumor and normal stroma laser-microdissected from breast tumors, we observed an up-regulation of TGM2 in the tumor-associated stroma (Figure S2B), again suggesting that TGM2 expression could be associated with the presence of a stimulation exerted by tumor cells similar to the one reproduced in our in vitro model. We next sought to assess the association between TGM2 expression and clinical outcome. Unfortunately, no publicly available gene expression datasets of breast cancer stromal samples with associated clinical data were available to test the prognostic value of stromal TGM2 expression. Therefore, we could only evaluate the clinical relevance of TGM2 in datasets derived from non-microdissected invasive tumors. When the 475 untreated lymph-node negative patients from the METABRIC dataset were stratified using the TGM2 expression median value as cutoff, no differences in DSS were observed by survival analysis both considering all patients or by separately analyzing the patients subdivided according to their tumor molecular subtype. Kaplan-Meyer curves are reported in Supplementary Figure S3. In contrast, in the combined dataset including stage I breast cancers from untreated patients, where the absence of treatment and the information on DMFS allowed testing pure prognostic effects, stratification by TGM2 expression revealed a statistically significant longer DMFS in patients with primary tumors expressing higher TGM2 levels (log-rank p = 0.027, n = 611). The impact of TGM2 expression on DMFS was even stronger in the subset of triple negative tumors (log-rank p = 0.024, n = 130) whereas it was negligible in patients bearing HER2-enriched (log-rank p = 0.41, n = 95) or luminal tumors (log-rank p = 0.51, n = 386). Kaplan-Meier curves are reported in Figure 6. As already stated all the above data refer to non-microdissected tumors and, despite the increased percentage of tumor cells with respect to stromal compartment in the analyzed samples, the possibility that the contribution of stroma may impact on the data cannot be excluded. Considering our hypothesis that TG2 might represent an early sensor of tumor-induced modification in the stroma, we evaluated TG2 expression at protein level by IHC separately in stroma and in the tumor compartments of 5 patients diagnosed with DCIS and who later developed in invasive tumor matched with 5 women of similar age and diagnosed with DCIS lesions of comparable grade and size, albeit free of invasive relapse for a similar follow up duration. An example of TG2 staining in the stromal and epithelial compartments is shown in Supplementary Figure S4. Remarkably, all sections showed positivity for endothelial cell. Forty percent of sections showed a moderate-to-strong cytoplasmic positivity in neoplastic cells and a moderate-to-strong positivity of fibroblast cells in peritumoral stroma. In 20% of the cases, neoplastic cells of DCIS were negative for TG2, but peritumoral fibroblast showed cytoplasmic positivity. Among women known to have progressed to an invasive lesion after DCIS diagnosis, the TG2 expression in the tumor cells was trendily (p = 0.077, two-tailed Student’s t test) associated with a longer DFS (77 ± 15.5 months) compared to women with TG2 negative DCIS (46.3 ± 11 months). No differences were instead observed stratifying the women by stromal expression of TG2. Similarly, when looking at patients defined as controls, who represent a similar subset of women with comparable DCIS lesions, but who have not developed an invasive lesion at a similar observation time, stromal and tumoral did not associate as expected with any differences in follow-up time. This suggests that a high tumoral rather than stromal TG2 expression might play a role in predicting subsequent invasive evolution. In fact, when considering only women whose tumors did in fact undergo a progression towards the invasive phenotype, all women with a late progression after 5 years (3/3) were positive in their tumor cells, whereas those with an early progression (<5 years) had TG2 negative in situ lesions (2/2). No such association could be observed in the stromal compartment. Importantly, grading was not associated with early or late progression to invasive phenotype. 4. Discussion In the last two decades, the role of the microenvironment as one of the crucial hallmarks in cancer has been increasingly recognized, while a raising amount of evidences has been accumulated on the actual participation of normal cells in the acquisition of cancer hallmarks [29]. In the mammary gland the role of the microenvironment is important both in the development of the organ and in its neoplastic transformation [30]. Fibroblasts in particular, are known to play a major role on the stage of the microenvironment [31], and we are understanding that not all fibroblasts act in the same way [32]. The heterogeneous nature of breast cancer both at molecular and clinical level, although extensively unraveled [33], still poses more questions than answers since robust biomarkers of proven clinical utility are still lacking. In such a context the characterization of the tumor microenvironment through gene expression profiling proved to inform on the clinical outcome, albeit no biomarkers of clinical usefulness have been so far developed (reviewed in [7]). So far, to the best of our knowledge, the search for biomarkers of stromal activation with translational purposes has rarely followed an approach based on the generation of in vitro models. We therefore tried to address such a gap by developing a model for breast cancer. The approach is very similar to the one already previously used in our laboratory for modeling tumor cell activation by stromal cells, and which proved to have a translational value as it generated a molecular subtype-specific gene signature associated with increased risk of developing distant metastases in women with early breast tumors (manuscript submitted). The biological reliability of the present model is supported by the satisfactory overlap between both the genes and the biological categories identified as modulated in our model and activated fibroblasts derived from similar and different contexts (including both ex vivo and in vitro samples) such as lung and prostate tumors [14]. This finding gives an indirect support to the reliability of the candidate genes obtained by our approach. When specific genes involved in fibroblasts activation were examined, the biological categories related to intracellular signaling activation and the genes herein were among the strongest candidates for becoming stromal activation biomarkers. They were however, excluded from our screen for candidate biomarkers, since signaling activation (although being definitely expected given the paracrine type of stimulation) represents a category at the crossroad of many pathways. In contrast, the increased expression of the TGM2 gene represented an interesting possible candidate. This gene encodes for a tissue transglutaminase 2, a multifunctional protein with many enzymatic activities spanning from transglutaminase, GTPase/ATPase, protein disulfide isomerase, and protein kinase, but also entangled with non-enzymatic functions due to its interaction with multiple cellular proteins. Finally, TG2, the protein encoded by TGM2 is involved in modulation and deposition of extracellular matrix and is up-regulated in inflammation and wound [28,34]. Since TGM2 with its multifunctional role could represent an interesting candidate, after confirming the up-regulation of the gene with an orthogonal assay, we evaluated whether in our activated fibroblasts its up-modulation affected the intracellular or the secreted encoded protein or both, and confirmed that treatment with tumor cell-derived conditioned medium was responsible for an increase in the secreted protein. In fact, we could observe a massive increase in TG2 protein levels among the protein deposed on the cell culture dishes and corresponding to the ECM released by fibroblasts. This latter finding further increased our interest in this protein since it represents a tumor-stimulated stromal target directly affecting the ECM and with a potential, thanks to its enzymatic activity, to modify the EMT texture eventually affecting tumor cell migration. At difference with the data reported by Park et al. [35] and Tchou et al. [36], we did not observe molecular subtype–specific gene modulations in our activated fibroblasts. This could be an indication of the fact that our model captures the very early alterations induced in fibroblast in response to the activating tumor secretome, whereas modulations observed in clinical samples are the result of a long-time crosstalk between fibroblasts and tumor cells. Being convinced, based on our in vitro experiments, of the potential of TGM2 as candidate biomarker acting as an early sensor of stromal activation, we preceded with evaluating its clinical role. This was initially done in two public dataset of invasive tumors, an approach presenting limitations that we were aware of. The first limitation relates to the fact that we had identified TGM2 as a potential early marker, but were instead looking at it role in invasive tumors, which did not at all represent the situation of a naïve normal stroma undergoing a stimulation derived from the very early steps of malignant transformation. The second limitation relates to the nature of the clinical sample itself, which is not representative of the stroma, but is a non-pure, but still tumor-cell enriched rather than stroma-cell enriched tissue specimen. In fact, this evaluation resulted in a rather unexpected result with TGM2 expression being associated with a longer DMFS in the general population, a finding which, was probably driven by the minority of basal-like tumors where the ‘protective’ effect of a high TGM2 expression seemed to be stronger. We therefore turned our attention to the clinical setting of DCIS representing a non-obligate precursor of invasive tumors and where, as explained in the introduction, the need for a biomarker of invasive evolution is very strong. This time we could overcome the limitation linked to mixed samples as by using IHC we separately evaluated the stromal and tumor cells expression of TG2. In this context, the outcome considered for evaluating the role of TG2 expression was the time to invasive evolution after the diagnosis and the surgical removal of an in situ lesion. The sample size was quite limiting, but we still could confirm a protective role for the tumor cell-associated TGM2 (albeit for the protein and not the gene) similar to the one reported in infiltrating tumors. Conversely, stromal TG2 expression was not associated with an earlier progression to invasive disease, a finding that could appear to be in contrast with our hypothesis, although a possible explanation can be given as explained below. Where the limited sample size does not help in supporting strong and definitive conclusions, the obtained results might however suggest that the role of stromal TG2 should be better evaluated in earlier precursor lesions such as atypic ductal hyperplasia, atypical apocrine hyperplasia, microglandular adenosis [37]. DCIS, besides representing a real clinical dilemma, is an already established tumor that even retains its molecular subtype from in situ to invasive disease. This is supported by pathological considerations, but also by gene expression studies [24]. In fact, in their seminal paper Ma et al. [24] profiled for gene expression 14 patient-matched samples from normal epithelium, normal stroma and tumor including both the invasive and the in situ lesions. In the epithelium gene expression changes occurred only in the transition from normal cells to DCIS, and similarly in the stroma, the very great majority of gene expression modulations were again occurring in the transition from normal tissue to DCIS. This would therefore provide a possible explanation for the failed validation of the stromal TGM2 role as early sensor of transition to an invasive phenotype, since DCIS would already share the same traits of the invasive lesion and would not represent the ideal model for early stromal activation. On the other hand, the outcome that we have considered (invasive tumor occurring before or after 5 years) for our evaluations of the clinical role of TG2 might not be the best surrogate for a true early stromal reaction. A literature study [38] reports data on tumor and stromal TG2 expression in 44 patients with in situ breast lesions in combination with HJURP and HIF. In addition, in this larger case series no correlation was found between neither tumor nor stromal TG2 expression and relapse (both in situ and invasive) occurring within 12 months from surgery in the ipsilateral breast. However, similarly to our study the authors report higher TG2 expression in tumor cells in non-relapsed compared to relapsed patients. Therefore, we feel that TGM2 with its potential role as a modifier of the ECM texture represents an extremely interesting biomarker, but that the clinical validation should probably address earlier pre-neoplastic lesions rather than DCIS. Accordingly, TGM2 might represent a diagnostic rather than prognostic biomarker. Despite the failed clinical validation, these results are still biologically relevant as supported by the above mentioned overlapping of our up-regulated genes with other genes obtained from clinically derived and from in vitro models of cancer activated fibroblasts, and would therefore deserve a validation in a different clinical context. Acknowledgments We wish to thank the core facilities of Fondazione IRCCS Istituto Nazionale dei Tumori Department of Experimental Oncology, and in specifically the Functional Genomics Core Facility for microarray hybridization and the Histochemistry Core Facility for immunohistochemical determination. We are indebted to Francesca D’Aiuto for assistance with bioinformatic analyses. We thank Francesca Andriani for carrying out cell staining and FACS analysis. The study was supported by grants from: the Associazione Italiana per la Ricerca sul Cancro (AIRC), grant number 10611 to M.G.D and AIRC 5X1000 project “EDERA” Tumor microenvironment-related changes as new tools for early detection and assessment of high-risk diseases (ED12162); the Italian Ministry of Health (Special Project on Female Cancers) and from the European Commission under the 7th Framework Program, grant agreement # 260791 Eurocan Platform. The study used data generated by the Molecular Taxonomy of Breast Cancer International Consortium, with the support of Cancer Research UK and of British Columbia Cancer Agency Branch. Supplementary Materials The following are available online at http://www.mdpi.com/2076-3905/5/2/10/s1. Click here for additional data file. Author Contributions Vera Cappelletti and Giuseppe Merlino conceived and designed the experiments; Giuseppe Merlino and Patrizia Miodini performed the experiments; Matteo Dugo, Giuseppe Merlino, Maria Grazia Daidone and Vera Cappelletti analyzed the data; Maria Luisa Carcangiu and Biagio Paolini identified the DCIS cases and scored the immunostained slides; Vera Cappelletti and Maria Grazia Daidone wrote the manuscript; Massimiliano Gennaro was responsible for collection of clinical data on patients diagnosed with DCIS. Conflicts of Interest The authors declare no conflict of interest. Figure 1 Volcano plots showing differentially expressed gene derived from Class comparison analysis NHDH treated with serum-free control medium versus the NHDF treated with serum-free cultured medium conditioned for 72 h by MDA-MB-468 (panel A: NHDF c.m. MDA-MB-231 vs. NHDF in DMEM), SkBr3 (panel B: NHDF c.m. SkBr3 vs. NHDF in McCOYS’s ) and T47D (panel C: NHDF c.m. T47D vs. NHDF in DMEM) cells. Dashed vertical lines delimit genes (dots) with log2 fold changes > 1, or < −1 (x-axis) and with high (false discovery rate FDR< 0.0001, dashed horizontal line) statistical significance (−log10 of p-value, y-axis), respectively. Black dots represent genes with fold changes and statistical significance values below the defined thresholds. Genes up-regulated above the defined cutoffs are reported as red dots, whereas down-regulated genes are reported as green dots. Figure 2 Biological interpretation of gene expression modulations in activated fibroblasts. (A) Selected macro biological categories found to be specifically enriched by GSEA analysis after treatment of NHDF cells with conditioned media derived from SkBr3, T47D and MDA-MB-468 cells. Positively-enriched categories are reported in red, while negatively-ones are reported in green. Color intensity reflects FDR values as shown in the inset; (B) Heat map of genes found to be differentially expressed (DE) across treated NHDF with an FDR < 0.0001. Figure 3 Pattern of genes up-regulated in fibroblast upon stimulation by breast cancer cells. Eulero-Venn diagrams highlighting the numbers of common and exclusive DE genes detected as up-regulated in samples derived from NHDF cells treated with serum-free medium conditioned by MDA-MB-468, by SkBr3 and by T47D breast cancer cell lines compared to their respective controls. The complete list of common up-regulated genes is also reported. Figure 4 Relative expression of TGM2, IL6 and TGFB in NHDF cells treated with breast cancer cell line-derived conditioned media. (A) Fold-change values (log2-tranformed) derived from class comparison analyses (treated versus control HNDF) of microarray data are reported in the Table; (B) Relative TGM2, IL6 and TGFB expression values of HNDF treated with medium conditioned by breast cancer cell lines versus control media are reported. Bars represent mean –ΔΔCt values (treated Ct value—control Ct value) of three independent biological triplicates ±SD. Statistical significance of differences between genes compared to controls were evaluated by two-tailed unpaired Student’s t-test. Figure 5 Intracellular and secreted TG2. (A) Western blotting results for NHDF intracellular TG2 expression upon stimulation with control medium or medium conditioned by breast cancer cell lines. The histogram reports relative quantification data by densitometry with respect to β-actin; (B) Western blotting results for TG2 deposited by NHDF on culture dishes upon stimulation with control or tumor cell-conditioned media. The histogram reports densitometric quantification in arbitrary units after normalization with respect to the total number of cells grown in the dish. Figure 6 TGM2 and distant metastasis-free survival. Kaplan-Meier analysis for the association of TGM2 expression, stratified by median value, in tumors from patients in the combined dataset with distant metastasis-free survival (DMFS). Survival differences were evaluated by log-rank test. ==== Refs References 1. Egeblad M. Nakasone E.S. Werb Z. Tumors as organs: Complex tissues that interface with the entire organism Dev. Cell 2010 18 884 901 10.1016/j.devcel.2010.05.012 20627072 2. Pietras K. Ostman A. Hallmarks of cancer: Interactions with the tumor stroma Exp. Cell Res. 2010 316 1324 1331 10.1016/j.yexcr.2010.02.045 20211171 3. Polyak K. Haviv I. Campbell I.G. Co-evolution of tumor cells and their microenvironment Trends Genet. 2009 25 30 38 10.1016/j.tig.2008.10.012 19054589 4. Hanahan D. Coussens L.M. Accessories to the crime: Functions of cells recruited to the tumor microenvironment Cancer Cell 2012 21 309 322 10.1016/j.ccr.2012.02.022 22439926 5. Quail D.F. Joyce J.A. Microenvironmental regulation of tumor progression and metastasis Nat. Med. 2013 19 1423 1437 10.1038/nm.3394 24202395 6. Erickson A.C. Barcellos-Hoff M.H. The not-so innocent bystander: The microenvironment as a therapeutic target in cancer Expert Opin. Ther. Targets 2003 7 71 88 10.1517/14728222.7.1.71 12556204 7. Giussani M. Merlino G. Cappelletti V. Tagliabue E. Daidone M.G. Tumor-extracellular matrix interactions: Identification of tools associated with breast cancer progression Semin Cancer Biol. 2015 35 3 10 10.1016/j.semcancer.2015.09.012 26416466 8. Conklin M.W. Keely P.J. Why the stroma matters in breast cancer: Insights into breast cancer patient outcomes through the examination of stromal biomarkers Cell Adh. Migr. 2012 6 249 260 10.4161/cam.20567 22568982 9. Zou W. Regulatory T cells, tumour immunity and immunotherapy Nat. Rev. Immunol. 2006 6 295 307 10.1038/nri1806 16557261 10. Pollard J.W. Tumour-educated macrophages promote tumour progression and metastasis Nat. Rev. Cancer 2004 4 71 78 10.1038/nrc1256 14708027 11. Dushyanthen S. Beavis P.A. Savas P. Teo Z.L. Zhou C. Mansour M. Darcy P.K. Loi S. Relevance of tumor-infiltrating lymphocytes in breast cancer BMC Med. 2015 13 10 10.1186/s12916-015-0431-3 25598008 12. Karn T. Pusztai L. Rody A. Holtrich U. Becker S. The Influence of Host Factors on the Prognosis of Breast Cancer: Stroma and Immune Cell Components as Cancer Biomarkers Curr. Cancer Drug Targets 2015 15 652 664 10.2174/156800961508151001101209 26452382 13. Callari M. Cappelletti V. D’Aiuto F. Musella V. Lembo A. Petel F. Karn T. Iwamoto T. Provero P. Daidone M.G. Subtype-Specific Metagene-Based Prediction of Outcome after Neoadjuvant and Adjuvant Treatment in Breast Cancer Clin. Cancer Res. 2016 22 337 345 10.1158/1078-0432.CCR-15-0757 26423797 14. Gandellini P. Andriani F. Merlino G. D’Aiuto F. Roz L. Callari M. Complexity in the tumour microenvironment: Cancer associated fibroblast gene expression patterns identify both common and unique features of tumour-stroma crosstalk across cancer types Semin Cancer Biol. 2015 35 96 106 10.1016/j.semcancer.2015.08.008 26320408 15. Qiao A. Gu F. Guo X. Zhang X. Fu L. Breast cancer-associated fibroblasts: Their roles in tumor initiation, progression and clinical applications Front. Med. 2016 10 33 40 10.1007/s11684-016-0431-5 26791754 16. Benson J.R. Wishart G.C. Predictors of recurrence for ductal carcinoma in situ after breast-conserving surgery Lancet Oncol. 2013 9 e348 e357 10.1016/S1470-2045(13)70135-9 23896274 17. Silverstein M.J. Poller D.N. Waisman J.R. Colburn W.J. Barth A. Gierson E.D. Lewinsky B. Gamagami P. Slamon D.J. Prognostic classification of breast ductal carcinoma-in situ Lancet 1995 345 1154 1157 7723550 18. Rudloff U. Jacks L.M. Goldberg J.I. Wynveen C.A. Brogi E. Patil S. Van Zee K.J. Nomogram for predicting the risk of local recurrence after breast-conserving surgery for ductal carcinoma in situ J. Clin. Oncol. 2010 28 3762 3769 10.1200/JCO.2009.26.8847 20625132 19. Livak K.J. Schmittgen T.D. Analysis of relative gene expression data using real-time quantitative PCR and the 2(−ΔΔC t) Method Methods 2001 25 402 408 10.1006/meth.2001.1262 11846609 20. Du P. Kibbe W.A. Lin S.M. Lumi: A pipeline for processing Illumina microarray Bioinformatics 2008 24 1547 1548 10.1093/bioinformatics/btn224 18467348 21. Barrett T. Troup D.B. Wilhite S.E. Ledoux P. Evangelista C. Kim I.F. Tomashevsky M. Marshall K.A. Phillippy K.H. Sherman P.M. NCBI GEO: Archive for functional genomics data sets—10 years on Nucleic Acids Res. 2011 39 D1005 D1010 10.1093/nar/gkq1184 21097893 22. Smyth G.K. Linear models and empirical bayes methods for assessing differential expression in microarray experiments Stat. Appl. Genet. Mol. Biol. 2004 3 1 25 10.2202/1544-6115.1027 16646809 23. Subramanian A. Tamayo P. Mootha V.K. Mukherjee S. Ebert B.L. Gillette M.A. Paulovich A. Pomeroy S.L. Golub T.R. Lande E.S. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles Proc. Natl. Acad. Sci. USA 2005 102 15545 15550 10.1073/pnas.0506580102 16199517 24. Ma X.J. Dahiya S. Richardson E. Erlander M. Sgroi D.C. Gene expression profiling of the tumor microenvironment during breast cancer progression Breast Cancer Res. 2009 11 R7 10.1186/bcr2222 19187537 25. Planche A. Bacac M. Provero P. Fusco C. Delorenzi M. Stehle J.C. Stamenkovic I. Identification of prognostic molecular features in the reactive stroma of human breast and prostate cancer PLoS ONE 2011 6 10 10.1371/journal.pone.0018640 21611158 26. Dvinge H. Git A. Graf S. Salmon-Divon M. Curtis C. Sottoriva A. Zhao Y. Hirst M. Armisen J. Miska E.A. The shaping and functional consequences of the microRNA landscape in breast cancer Nature 2013 97 378 382 10.1038/nature12108 23644459 27. Nurminskaya M.V. Belkin A.M. Cellular functions of tissue transglutaminase Int. Rev. Cell Mol. Biol. 2012 294 1 97 22364871 28. Agnihotri N. Kumar S. Mehta K. Tissue transglutaminase as a central mediator in inflammation-induced progression of breast cancer Breast Cancer Res. 2013 25 202 10.1186/bcr3371 23673317 29. Hanahan D. Weinberg R.A. Hallmarks of Cancer: The Next Generation Cell 2011 144 646 674 10.1016/j.cell.2011.02.013 21376230 30. Polyak K. Kalluri R. The role of the microenvironment in mammary gland development and cancer Cold Spring Harb. Perspect Biol. 2010 2 a003244 10.1101/cshperspect.a003244 20591988 31. Kalluri R. Zeisberg M. Fibroblasts in cancer Nat. Rev. Cancer 2006 6 392 401 10.1038/nrc1877 16572188 32. Sugimoto H. Mundel T.M. Kieran M.W. Kalluri R. Identification of fibroblast heterogeneity in the tumor microenvironment Cancer Biol. Ther. 2006 5 1640 1646 10.4161/cbt.5.12.3354 17106243 33. Prat A. Ellis M.J. Perou C.M. Practical implications of gene-expression-based assays for breast oncologists Nat. Rev. Clin. Oncol. 2011 9 48 57 10.1038/nrclinonc.2011.178 22143140 34. Mehta K. Kumar A. Kim H.I. Transglutaminase 2: A multi-tasking protein in the complex circuitry of inflammation and cancer Biochem. Pharmacol. 2010 80 1921 1929 10.1016/j.bcp.2010.06.029 20599779 35. Park S.Y. Kim H.M. Koo J.S. Differential expression of cancer-associated fibroblast-related proteins according to molecular subtype and stromal histology in breast cancer Breast Cancer Res. Treat. 2015 149 727 741 10.1007/s10549-015-3291-9 25667103 36. Tchou J. Kossenkov A.V. Chang L. Satija C. Herlyn M. Showe L.C. Puré E. Human breast cancer associated fibroblasts exhibit subtype specific gene expression profiles BMC Med. Genomics 2012 5 10.1186/1755-8794-5-39 22954256 37. Lopez-Garcia M.A. Geyer F.C. Lacroix-Triki M. Marchiò C. Reis-Filho J.S. Breast Cancer precursors revisitated: Molecular features and progression pathways Histopathology 2010 57 171 192 10.1111/j.1365-2559.2010.03568.x 20500230 38. Bravaccini S. Tumedei M.M. Scarpi E. Zoli W. Rengucci C. Serra L. Curcio A. Buggi F. Folli S. Rocca A. New biomarkers to predict the evolution of in situ breast cancers Biomed. Res. Int. 2014 2014 159765 10.1155/2014/159765 25243117

PMC005xxxxxx/PMC5003487.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020011microarrays-05-00011ReviewLiving Cell Microarrays: An Overview of Concepts Jonczyk Rebecca †Kurth Tracy †Lavrentieva Antonina †Walter Johanna-Gabriela Scheper Thomas Stahl Frank *Erfle Holger Academic EditorInstitute of Technical Chemistry, Leibniz University of Hannover, Callinstr. 5, Hannover 30167, Germany; rjonczyk@iftc.uni-hannover.de (R.J.); kurth@iftc.uni-hannover.de (T.K.); lavrentieva@iftc.uni-hannover.de (A.L.); walter@iftc.uni-hannover.de (J.-G.W.); scheper@iftc.uni-hannover.de (T.S.)* Correspondence: stahl@iftc.uni-hannover.de; Tel.: +49-511-762-2968; Fax: +49-511-762-3004† These authors contributed equally to this work. 26 5 2016 6 2016 5 2 1130 3 2016 11 5 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Living cell microarrays are a highly efficient cellular screening system. Due to the low number of cells required per spot, cell microarrays enable the use of primary and stem cells and provide resolution close to the single-cell level. Apart from a variety of conventional static designs, microfluidic microarray systems have also been established. An alternative format is a microarray consisting of three-dimensional cell constructs ranging from cell spheroids to cells encapsulated in hydrogel. These systems provide an in vivo-like microenvironment and are preferably used for the investigation of cellular physiology, cytotoxicity, and drug screening. Thus, many different high-tech microarray platforms are currently available. Disadvantages of many systems include their high cost, the requirement of specialized equipment for their manufacture, and the poor comparability of results between different platforms. In this article, we provide an overview of static, microfluidic, and 3D cell microarrays. In addition, we describe a simple method for the printing of living cell microarrays on modified microscope glass slides using standard DNA microarray equipment available in most laboratories. Applications in research and diagnostics are discussed, e.g., the selective and sensitive detection of biomarkers. Finally, we highlight current limitations and the future prospects of living cell microarrays. living cell microarrays (LCMAs)transfected cell microarraysaffinity-based arrayscell attachment and stimulation3D cell constructsmicrofluidic devicesorgan-on-a-chip (OOC) ==== Body 1. Introduction Microarrays became standard tools for gene expression analysis in the 1990s. They combine high-throughput screening and miniaturization on the same platform, and were adapted to many kinds of biomolecules/biological samples like proteins, antibodies, small molecules, cell lysates, fixed tissues, living cells, and organoids. In a microarray, probes are arranged in rows and columns on a solid slide surface in a highly reproducible pattern, in order to allow easy and precise identification. All microarray formats share a common principle: well-defined probes are fixed onto a surface at defined positions, and a mixture of multiple components is applied. Recognition events are detected by fluorescence or by an electrochemical signal. There are sophisticated developments towards micro-electromechanical systems (MEMS) and lab-on-a-chip systems, but for standard microarray experiments performed on modified glass slides, the major equipment consists of a spotter and a scanner. To transfer a standard microarray experiment to living cells, the microarray surface should primarily be suitable for the attachment, proliferation, and differentiation of living cells. Furthermore, the storage of printed living cell microarrays must be optimized. Apart from these challenges, living cell microarrays can be produced, processed, and analyzed with standard microarray equipment. Two- and three-dimensional cell-based microarrays on modified slides are the focus of this paper. Additionally, we discuss recently developed specialized platforms, such as microfluidic devices, as well as lab-on-a-chip/organ-on-a-chip (OOC) applications. An overview about the applications of living cell microarrays (LCMAs) is provided in Scheme 1. 2. Literature Review Section 2.1. Historical Background Initially, microarray technology was developed to investigate gene regulation via DNA microarrays [1]. In this format, 105 short oligonucleotides are immobilized on a surface-modified slide at defined locations (spots) to serve as capture molecules (probes). Here, each spot represents one gene. Immobilization, ranging from a few dozen genes up to the immobilization of the whole genome of an organism on a single slide, results in a high-throughput analyzing method. Two different fluorescently labeled DNA- or reverse-transcribed mRNA samples, which are isolated from the control and diseased or experimental tissues or cells, function as targets. These molecules bind and hybridize to their corresponding capture molecules on the microarray. The fluorescence intensity of each spot is quantified and is proportional to the amount of fluorescent targets, which hybridized on the microarray. Therefore, the determined ratio of the different fluorescent DNA-targets represents the relative abundance of DNA or mRNA in the investigated cells [2,3]. DNA microarrays are used in clinical diagnostics for the profiling of gene expression in tissues and cells, the characterization of diseases, and their corresponding genetic risk factors and for the identification of biomarkers for cancer treatment. Furthermore, with the help of these microarrays, gene polymorphisms, as well as the gene expression of markers involved in drug-metabolism or toxicology can be studied [4,5]. To permit accurate probe detection, special fluorescence scanners for glass objective slides were established, which provide high sensitivities and a good resolution of the defined circular spots. Since all samples on the microarray are treated under identical conditions and are incubated with the same amounts of analyzing reagents, the hundreds or thousands of results of one slide are highly comparable to each other. At the same time, only short processing times and low material amounts are required. The density of information is controlled by the number of spots, while the reliability of detection depends on the distance between spots [6]. Protein microarrays were historically established to compensate for the fact that there can be discrepancies between mRNA and protein levels detected in cells at a given time [7,8]. The first protein microarrays used antibodies; later, platforms also employed aptamers (short single-stranded oligonucleotides) as capture molecules on the microarray surface [9,10,11]. In an alternative setup, the target proteins can be directly printed onto the microarray surface, where they are detected by mobile capture molecules [11]. Molecules can be immobilized on the microarray surface by the use of computer-controlled spotters, either via direct contact of a pin to the microarray’s surface or by using a contact-free automated pipette [12,13]. 2.2. Tissue and Cell Microarrays To improve the direct phenotyping of cells from tissues, high-throughput tissue microarrays were developed [14] and optimized in the late 1990s [15]. Previously, tissue sections were used for this purpose, which had to be individually examined in a manual manner. This technique requires high amounts of antibodies and reagents, as well as large quantities of limited tissue probes. By the optimization of tissue microarrays (TMAs), the number of examinations performable with one single biopsy probe increased from 100 to 500,000 tests [16]. Thus, TMAs became the method-of-choice for protein and cells staining in tissues or for the validation of cDNA microarray results in experiments seeking the identification of new biomarkers and therapeutic target proteins [16,17]. A variety of analytic methods can be used to analyze the TMAs, e.g., immunohistochemistry (protein detection via enzyme-linked antibodies), immunofluorescence staining (protein detection via fluorescent-labeled antibodies), and fluorescence in-situ hybridization (determination of the copy number of genes via fluorescent-labeled DNA-fragments complementary to a gene) [16,18]. The recently published review of Vogel gives an overview of the milestones in TMA preparation from 1986 to 2014 [19]. In general, a hollow needle takes tissue core samples out of a donor tissue block, which is fixated in formalin and embedded in paraffin. The tissue core samples are then placed in an empty acceptor block at defined positions. Further tissue core samples from other donor blocks are transferred to the acceptor block. The acceptor blocks prepared in this manner are cut into sections by a microtome, after which the sections are placed on glass microscope slides [16]. In a single investigation, a variety of up to 10,000 tissue core samples can be simultaneously stained and analyzed under identical conditions. Two or three core samples from the same donor block per TMA ensure representative results of the biopsy probe [20]. Each tissue microarray can be arranged individually, in order to investigate a specific experimental question, resulting in TMAs containing, e.g., tumors of the same type in different stages of the disease [18,21]. In addition to tissue samples, well-defined and standardized controls consisting of native/healthy tissues or cell lines are used to enable a quantitative comparison of microarray experiments between different laboratories and dates [16,18]. The malignant transformation, differentiation, and other cellular processes of adherent cell lines are well-known and thoroughly described in the literature. As a result, adherent cell lines serve as an ideal control in tissue microarrays [22,23,24]. Pure cell microarrays (CMAs) are used for the easy identification of controls for immunostaining (positive or negative control) and for assay optimization by replacing expensive tissue probes. Protein expression profiles of whole cells, the effects of drug treatments, or other stimuli, as well as the effects of gene silencing experiments, were identified in 2005 using this method [25,26,27]. For this purpose, the cells were stimulated, fixed in formalin or paraformaldehyde, and then embedded in paraffin, agar, or low-melting agarose. These cell blocks serve as donor blocks for the creation of microarrays comparable to TMAs [26,27,28,29]. La Spada and coworkers [30] reported a preservation of elongated cell morphology in the prepared CMAs after fixation and scraping of induced pluripotent stem cells (iPSC) differentiated into neuronal lineage. Furthermore, they described an easier detection of the protein markers, as well as better image analysis, and thus a reduction of misinterpretation of the immunofluorescence staining of cell microarrays [30]. Stimulated and fixed cells can be directly transferred to microarrays utilizing a contact nanoprinter [31,32]. The procedure of fixation and embedding in paraffin influences the quality of cells and tissues, as well as the reproducibility of results, depending on the fixation time and antigen recovering protocols. Thus, the in-situ analysis of DNA, RNA, or proteins can lead to incorrect results [33,34]. In order to prevent this, frozen cell and tissue microarrays were developed [35,36,37]. Freshly frozen cells and tissues, however, lose their structure, resulting in severe alterations in cell morphology [35]. 2.3. Living Cell Microarrays Several research groups have established microarrays of living mammalian or prokaryotic cells over the last few years. In 2001, Ziauddin & Sabatini [38] laid down the basis for living cell microarrays. They printed DNA at defined locations on a microarray. Mammalian adherent cells grew on the printed area and took up this DNA. Thus, spots of localized transfection were created, which led to the rapid discovery of gene functions and the identification of drug targets, as well as gene products [38]. Further developments of the first transfected cell microarrays (presented in Section 2.3.2) led to the dissemination and application of this high-throughput screening platform to several research areas. Angres and the working group of Belkin gave detailed insights into the first steps of the evolution of whole-cell arrays [39,40,41]. In contrast to the working group of Belkin, which specializes in the development of biosensor arrays consisting of genetically tailored microbial cells, our review will focus exclusively on mammalian cells. All microarrays using living cells instead of purified cellular components are utilized to monitor complex, functional, and vital cellular responses. Living cell microarrays (LCMAs) were applied to characterize cell–cell interactions, cell interactions with their microenvironment, and reactions to applied stimuli, as well as to gain insight into molecular cellular mechanisms [42]. With the help of LCMAs, cells can be easily characterized with regard to their surface molecules. Cell activation caused by external stimuli can be detected in terms of intracellular signaling, inducing transcription as well as translation of genes, cell differentiation and proliferation, or cell death [40,43]. In this manner, toxic, genotoxic or other effects of biologically active molecules can be determined faster. Thus, at the beginning of a drug development process, a huge number of drug candidates and their in-situ enzyme-generated metabolites can be screened, resulting in a more efficient development of drugs [43,44,45]. Effects on gene or protein expression can be investigated after the extraction of DNA, RNA, or proteins from the cells [46,47]. However, such a miniaturized format can represent a challenge for the isolation of sufficient amounts of analytes to be studied. The isolation of DNA and RNA from small biomass quantities was successfully performed by Mutiu et al. [48]. The nucleic acids can be isolated from the whole cell population or after separation of individual living cells using high-speed cell sorting. Combining flow cytometric analysis and sorting of living cells with transcriptome analysis helps to relate molecular regulation processes within cellular subpopulations with the dynamics of the whole cell population [49]. A non-destructive observation of cell responses in real-time is easily feasible. Therefore, the applications range from basic cell biology studies to sophisticated drug testing procedures. Responses of the LCMAs are visualized in two different ways [43,50,51]: either label-free by electrochemical detection methods, such as surface plasmon resonance (SPR), differential interference contrast (DIC) microscopy, electric cell-substrate impedance sensors (ECIS), and magnetic resonance imaging (MRI), or by optical techniques using fluorescent/bioluminescent probes for specifically staining cellular targets. In the latter techniques, detection is performed by a fluorescence microscope or by a high-resolution fluorescence scanner [43,50,51]. Multi-electrode arrays (MEAs) can be additionally utilized as an electrochemical detection method. Moreover, they are a powerful tool for studying the electrophysiological effects of stimuli to single cells in neuroscience and cardiology [52,53,54]. For more information concerning microbial and mammalian cell biosensors, we would like to refer to a review article dealing with the construction of cell microarrays for biosensing purposes [55]. Furthermore, there are two recently published books focusing on the devices, cells, and applications of whole cell sensing systems [56,57]. The preparation of a LCMA requires a stable attachment of cells or cell-binding biomolecules to the microarray surface, as well as preventive measures for avoiding cross-contamination. Microarray surfaces should not interfere with cultivated cells in order to ensure meaningful results of stimulation experiments [42,58]. In addition, the microarray surface has to be chemically and physically stable, preventing cross-reactivity with medium ingredients or changes during the sterilization procedure [59]. The preparation strategies for LCMAs can be divided into two different techniques. In the first method, the cell suspension is positioned at defined locations on the microarray [60], which is further described in Section 2.3.1. In the second method, the microarray is modified to allow cells exposed to the surface to adhere only at defined spots [61,62,63]. Different applications of both techniques are introduced in Section 2.3.3, Section 2.3.4 and Section 2.3.5. Angres [58] introduced these two main techniques for LCMA preparation with the termini “genuine” and “substrate-based cell arrays”. In addition, she subdivided the first type into arrays of dielectrophoretic-positioned and printed cells. The latter category contained different types of microarrayed biomolecules (antibodies, peptides, glycans, proteins, and lipid membranes), as well as transfected cell microarrays. In our review, we will track Angres’ categories roughly, while presenting additional aspects ranging from cell-specific ligands and adsorption to entrapment or encapsulation [58]. New developments, such as three-dimensional (3D) cell constructs are described in Section 2.3.5, which were applied to LCMAs to create in vivo-like structures for better and more reliable models. The next evolutionary step of this microarray type was the recent combination of microfluidics and 3D cell constructions to generate lab-on-a-chip platforms. These specialized systems are described in Section 2.4, and a figurative overview of some of these systems is given in Figure 1. 2.3.1. Microarraying of Cells Cell microarrays are generated by a variety of cell patterning techniques including printing, photolithography, soft-lithography, and dielectric, microfluidic devices [64,65,66]. Here, we will describe cell printing either by contact or contact-free inkjet printing. As contact printing can damage the cells mechanically, this cell-patterning method is commonly used for fixed cells [31]. In contrast, contact-free devices do not directly interact with the cell sample and are thus feasible for the printing of living cells. Cell suspensions have been printed utilizing a contact-free printer in few studies only. Although the shear stress applied to cells in a droplet of liquid during the printing process should be considerable, in general, high cell viabilities have been observed [60,67,68]. In the first reported study, a modified computer printer was used for cell patterning [67]. A variety of cell types, as well as isolated single-cells in a droplet of few picoliters can be printed at defined locations of a single microarray via inkjet technology [69,70]. The application of recently established piezoelectric non-contact nanoprinters provided us with the possibility to transfer very small amounts of living cells (1200 cells and fewer) to microscope slides and to subsequently cultivate and monitor the applied cells online [60]. Here, one cell per droplet (0.4 nL) was printed. We used an incubation chamber to divide the microarray into 16 separate wells, in which the cells grew in clearly defined circular spots in a manner comparable to standard conditions. Since inkjet printing technology offers the possibility of printing cells on different surface topographies, cells can be transferred to slides coated with specific proteins, to topographic structures for enhanced attachment, or to microtiter plates, if preferred [60]. By increasing the viscosity of the printing solution, it is possible to prevent settling and aggregation of cells before and during the printing process. This leads to a standardized and reproducible dosage of cells. The addition of glycerol or sugars is the simplest method to increase viscosity in this case [31]. More complex systems, known as bio-inks, use the cross-linking reaction of alginate and calcium ions or fibrin and thrombin, to rapidly form a gel [50,71]. These bio-inks are not suitable for every cell type. Therefore, Ferris et al. [72] recently applied a bio-ink based on gellan gum and surfactants in culture medium to print different cell types. All of these bio-ink systems are suitable for the preparation of multiplex cell microarrays and single-cell microarrays [73,74]. The main advantages are the spatial separation of cell types and the 3D microenvironment provided for the cells. Proteins, such as collagen or fibronectin, are often added for enhanced cell-matrix interaction [75]. Thus, even single cells have contact to the matrix in all three dimensions. Bio-ink systems are also the basis for layer-by-layer printing of 3D structures [50]. Recently, state-of-the-art articles were published describing several methods used in 3D bioprinting technology [76,77]. 2.3.2. Transfected Cell Microarrays In 2001, Ziauddin & Sabatini [38] printed different sets of complementary DNAs, which were cloned in vectors and solved in aqueous gelatin, onto a glass slide. The aim of the experiment was to investigate a variety of cell reactions and changes in cellular behavior as a result of the activation of genes. The DNA spots were dried and treated with a lipid transfection reagent, resulting in a lipid-DNA complex. Afterwards, the microarray was covered with adherently growing HEK293T cells in medium. The oligonucleotides were incorporated into the cells via reverse transfection. The transiently transfected cells on the microarray express the genes and can be fixed, analyzed (by in-situ hybridization, immunofluorescence, or autoradiography), and visualized, e.g., by a laser fluorescence scanner or by a fluorescence microscope [38]. One spot on the microarray consisted of 30–500 transfected living cells. Since each spot was 120–250 µm in diameter, a spot density of up to 10,000 spots per standard glass slide was feasible [78]. Total expression levels directly depended on the applied amount of plasmid DNA, whereby only transfectable cells, such as HEK293T cells, could be used. HeLa and A-549 cells compared unfavorably, which was related to poorer transfection efficiencies [38]. By optimization of the transfection protocol for each cell type, the transfection efficiencies were improved, and a variety of cell types have been investigated successfully to date [78]. An alternative composition of the printed gel solution consisted of oligonucleotides, gelatin, and the lipid transfection reagent [79,80,81]. The addition of sucrose enhanced the storage stability of the printed microarrays up to 15 months. Fibronectin or other proteins reduced cross-contamination of transfected cells due to enhanced cell adherence and minimized migration. Furthermore, transfection efficacy was increased [79,80,81]. Thus, even primary cells were successfully transfected in an efficient and nontoxic procedure [80]. Detailed protocols were described by Erfle and coworkers [82]. Since the complete slide surface is covered with a cell carpet, a successful gene transfer into the cells could be easily observed by the additional expression of a reporter gene, e.g., green fluorescent protein [38]. This procedure results in fluorescent cells at the spots of oligonucleotides and non-fluorescent cells attached between the oligonucleotide spots [38,42,83,84,85]. Other research groups used GFP-transfected cells for creating a kind of grid between the non-fluorescent cells used in their studies [38,86]. Co-transfections of more than one gene into the same cell were achievable [38,79,87]. Transfected cell microarrays were suitable for the rapid and systematic high-throughput screening for genes that cause a desired phenotype or encode the products of interest. Ziauddin & Sabatini [38] demonstrated the proof-of-principle by the evaluation of binding specificities of a dopamine antagonist to a cell membrane protein, as well as by the detection of a pharmacologically relevant target of an immunosuppressive drug [38]. Furthermore, screening of transfected cell microarrays resulted in the identification of several proteins, which are involved in apoptosis, cell adhesion, and the kinase-signaling pathway [38,86,87]. Transfected cell microarrays were also utilized to analyze the effects of human herpesvirus-8 on a cellular transcription factor pathway [88]. Recently, this method was used to identify a preselection of highly relevant gene candidates for pancreatic cancer [89]. Unfixed, alive transfected cells served to investigate in-time cellular processes such as protein or organelle dynamics, kinetics, translocations, and redistribution or changes of phenotype over time [90,91]. Besides gene activation and overexpression, the study of gene downregulation and silencing can be performed with LCMAs as well. RNA interference (RNAi), established as a transient or as a stable process, has been used as a post-transcriptional loss-of-function tool since 2003. For a deeper look inside this method, we would like to refer to other reviews and articles [92,93,94]. In brief, synthetic small interfering RNAs (siRNAs) sequence-selectively suppress the genes of interest by base-pairing of the anti-sense nucleic acid to the target mRNA, which initiates its degradation [95]. Similar to cDNA transfected cell microarrays, siRNA was dissolved in aqueous gelatin mixed with the lipid transfection reagent and printed onto glass slides. Likewise, sucrose and fibronectin were added for enhanced transfection efficiency [96,97,98]. Nevertheless, many cell lines, especially primary cells with only a few exceptions, are obstacles to standard transfection methods. Thus, a stable integration and gene suppression tool was needed for RNAi screenings. Here, a viral transduction of short hairpin RNA (shRNA) vectors was implemented [78]. The vectors are taken up by the cell and integrated into the genome. Thereafter, shRNAs are produced in the cell nucleus as two complementary RNA sequences linked by a short loop. A RNAse III enzyme cleaves the shRNA into siRNAs, which then initiate mRNA degradation [93,99]. According to the cDNA cell microarrays described above, gene expression on RNAi cell microarrays was sequence-specific regulated, whereby the expression level was time- and dose-dependent and limited to the discrete spots [97]. The resulting cell microarrays are also known as solid-phase optimized transfection RNAi. An overview of the different high-throughput RNAi screenings in cultured cells can be found in reviews [78,100]. Additionally to reports where RNAi was used as a proof-of-concept to efficiently suppress the expression of reporter genes [97,101] or their fusion proteins [102], the research group of Erfle used transfected cell microarrays for the investigation of a variety of cellular regulation processes including specific endocytosis or secretory pathways [79,82,103]. Co-transfection using unspecific Cy3-labeled DNA oligonucleotides can be used to locate the exact position of transfected cells with only minor effects on uptake and efficacy [79]. In a recent study, researchers used stable fluorescent HeLa cells for the automatic and time-resolved investigation of cellular division, proliferation, survival, and migration via changes in phenotype after incorporation of RNAi [91]. Rantala et al. [81] and Fengler and coworkers [104] developed a cell spot microarray method, which supports the preparation of transfection cell microarrays and the analysis of RNAi screenings. Similar to Erfle and coworkers [96], they added fibronectin or a protein mixture for an enhanced, faster cell attachment at the oligonucleotide spots. Rantala et al. [81] additionally applied the transfection solution to a plate with hydrophobic polystyrene. After cell seeding and an adherence time of 5–20 min, a washing step was introduced into the microarray preparation procedure to remove all non-adhered cells between the spots. In this way, spots of transfected cells are separated from each other, which supports the automated analysis better compared to microarrays of a cell carpet [81,104]. Where data analysis are facilitated in this way, the differently treated cell groups on one microarray are not separated from each other, which may cause stimulation across the cell spots induced by secreted factors. This first step in the prevention of cross-contamination of different transfected and wildtype cells, however, is not usable for the investigation of different cell types on one slide. Further developments involving the increase of the variety of applied cell types are described in the following chapters. 2.3.3. Affinity-Based Immobilization The construction of a LCMA containing different cell types demands for stable immobilization of the cells. Therefore, ligands targeting cells with high affinity and specificity are helpful. The affinity-based immobilization of cells is an effective way to specifically capture cells that differ in type, stage of differentiation, or development during malignant transformation. Capturing specific cells provides important information for analyzing cellular processes. Immobilizing different capture molecules to microarrays allows the identification of cell types based on the expression of different cell surface molecules [43]. Another application of cell microarrays is the capture of specific cell types out of a complex cell mixture [105]. Cell-specific capture reagents include antibodies, proteins, aptamers, peptides, and small molecules [105]. Mammalian cells possess a complex array of glycans on the cell surface. The entity of glycans expressed in a cell is called the glycome. It comprises glycolipids, glycoproteins, and proteoglycans. It has been assumed that the expressed set of glycans differs between different cell types and different stages of cell development and differentiation [106]. Furthermore, malignant transformation alters glycosylation [106]. Lectins and glycans can be spotted onto microarrays for cell carbohydrate or lectin profiling [43]. A method for glycan profiling of living mammalian cells using lectin microarrays was established by Tateno et al. in 2007 [106]. Most of the interactions between carbohydrates and lectins on the cell surface are multivalent. This multivalency, as well as spotting lectins or carbohydrates in high concentrations, promotes cell binding even further. This simple method allows a rapid profiling of the cell surface glycome and is important for the understanding of glycome changes in cellular processes including cell-to-cell communication and immune response modulation [43,106]. Furthermore, the identification, analysis, characterization, and capture of different cell types with lectin microarrays is a useful tool in various fields, such as medicine and biology [105]. It is estimated that mammalian cells possess about 500 unique glycan structures and that the cell surface displays about 100 different lectins [43]. The microarray described by Tateno et al. [106] consisted of 43 lectins with distinct binding specificities, which were spotted separately and covalently bound to an epoxy array. The slide was incubated with Chinese Hamster Ovary (CHO) cells and their glycosylation-defective mutants (Lec1, Lec2, and Lec8), K562 cells after and before differentiation, and the splenocytes of wildtype and β1-3-N-acetylglucosaminyltransferase II knockout mice. The cells were labeled with Cell Tracker Orange and detected with an evanescent-field fluorescence scanner in the liquid phase. The experiments showed that lectin microarrays require a relatively low amount of cells, the glycome can be characterized in an intact state, experiments can be carried out in a high-throughput and rapid manner, and cells remain viable, which is useful for functional analysis [106]. Another method allowing the rapid and efficient analysis of cells is an immunoaffinity-based microarray. Milgram et al. [107] developed an antibody microarray for label-free cell-based applications in 2011. The study was focused on the microarray fabrication for SPR analysis of cells or cellular activity. Therefore, immunoglobulin G (IgG) antibodies, anti-CD90 (Cluster of Differentiation 90) against T-lymphocytes (LS102.9), and anti-I-A against B lymphocytes (3A9) were separately conjugated to N-Hydroxysuccinimide (NHS) pyrrole. The solutions of coupled antibodies were diluted in spotting buffer containing the free pyrrole. The antibody immobilization on the gold-covered glass slide occurred through electropolymerization of the pyrrol coupled to the antibodies with free pyrrole. Then, the slide was incubated with a mixture of B- and T-lymphocytes. With the help of a fluorescently labeled antibody (phycoerythrin anti-CD19), which specifically binds B-lymphocytes, the specific capturing of cells could be confirmed through optical and fluorescence microscopy. The polypyrrole-based chemistry proved to be an efficient and robust method immobilizing antibodies on biochips for label-free cellular analysis [107]. Apart from antibodies, alternative binders can be used to capture cells on LCMAs. Aptamers are single-stranded oligonucleotides, which bind their target molecules with high affinity and specificity. In comparison to antibodies, aptamers display higher long-term stability [11]. Another advantage is the low toxicity of aptamers [108]. Aptamers can be easily synthesized and modified for direct immobilization on microarrays [11]. For example, Song et al. [109] showed that circulating tumor cells (CTC) expressing the Epithelial Cell Adhesion Molecule (EpCAM) can be detected and captured by the DNA aptamer SYL3C, which they generated for this purpose. Anti-EpCAM antibody-functionalized surfaces already exist in microfluidic devices, but the use of the antibodies is restricted, because of their size and instability. Therefore, the 3′ terminal biotinylated aptamer SYL3C was immobilized through biotin-streptavidin interactions. To avoid interactions between the immobilized ligand and the surface, a polyethylene glycol (PEG) linker was incorporated to the aptamer between the nucleotides and the biotin moiety. The EpCAM-positive cell line Kato III and the EpCAM-negative cell line Ramos were used for the experiments. Aptamer SYL3C was shown to possess the potential for CTC enrichment and could be used on microarrays [109]. In another study performed by Chen et al. [110], an aptamer targeting Ramos cells was used to isolate and analyze single tumor cells with microwell arrays on a microfluidic device. The microwell array was coated with avidin to bind the biotinylated aptamers. The captured cells were observed under the microscope. Enzyme kinetics were analyzed using cell-permeable dye to monitor the intramolecular fluorescence of the cells with a fluorescence microscope. The aptamers-coated microwell array on the microfluidic device was shown to be capable of separation between specific tumor cells and could be of use for clinical samples, e.g., blood samples. In the future, personalized medicine could be established through specific tumor cell (isolated from the certain patient) enzyme kinetics in high-throughput assays. This novel method offers new opportunities in clinical applications [110]. In another aptamer-based study, anti-prostate-specific membrane antigen (PSMA) aptamers were immobilized on a microchip and fabricated into a high-throughput micro-sampling unit by Dharmasiri et al. [111] to capture rare circulating prostate tumor cells. Therefore, the poly(methyl methacrylate) surface of the microchip was activated using UV light. The generated carboxylic acid was then converted into a succinimidyl ester derivate via 1-Ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC) coupling, which then reacted with amino-modified aptamers. Anti-PSMA aptamers target LNCaP cells (prostate cancer cell line). LNCaP cells were counted, fluorescently labeled, and put into whole blood with CTC concentrations of 10 cells per mL. The cells could be successfully released after selecting and capturing via enzymatic digestion of the extracellular membrane (ECM) domain of PSMA using trypsin. The anti-PSMA aptamer was found to bind LNCaP specifically and is of use for capturing cells out of complex mixtures with high-throughput micro-sampling devices [111]. Small-molecule and peptide microarrays were used for screening living mammalian cells by Lee et al. [112]. Ligands were conjugated to the surface via polyethylene glycol—N-Hydroxysuccinimide (PEG-NHS). The PEG linker presents the ligand to its target with a distance to the microarray surface. All spotted ligands contained a primary amino group. The peptide cRGDyk is known to bind the integrin αvβ3 receptor. After binding the peptide cRGDyk to the surface, receptor positive and negative M21 cells were labeled with fluorescent dyes differently and incubated on the microarray to prove the specific binding of the peptide. Other specific ligands used by Lee et al. [112] to optimize the assay for high-throughput applications are known to be specific for the PSMA receptor, e.g., β-AG and Glycophosphatidylinositol. Another ligand specifically binding melanocortin-1 receptors is the α-melanocyte stimulating hormone, which was also used by Lee et al. [112]. Falsey et al. [113] prepared chemical microscope slides for the covalent immobilization of small molecules and peptides through site-specific oxime bond or thiazolidine ring ligation reaction. Glass slides were derivatized with (3-Aminopropyl)triethoxysilane (APTES). Two different ways to convert the amino slides to glyoxylyl derivates were used. One of them was to couple fluorenylmethoxycarbonyl serine followed by deprotection and oxidation. The second was to couple protected glyoxylic acid to the slide and deprotect it afterwards with hydrochloric acid. Biotin and the peptides were modified before spotting. The carboxyl terminus was coupled to a succinimic derivate as linker and an amino-oxy group or 1, 2-amino-thiol group. Stearic acids were added separately during the coupling step with glyoxylic acid to avoid unspecific binding. The peptide wGeyidvk was found to specifically bind to the surface idiotype of WEHI-231 cells (murine lymphoma). For the cell adhesion assay, Falsey et al. have immobilized the peptide wGeyidvk on the microarray. The labeled target cells were incubated on the microarray and were shown to specifically bind to the peptide. PEG5000—NHNH2 was essential to block the slide and avoid non-specific binding of cells and improve the background for cell adhesion assays [113]. Folic acid, another small molecule, was recently used to modify Montmorillonite clay (FA‑Mont) [114]. FA-Mont was tested as a cell culture material to provide cell adhesion of folate-positive cells on the clay surface. Cultivating folate-receptor-rich HeLa and folate-receptor-poor A-549 cells in 96-well plates coated with FA-Mont proved FA‑Mont’s ability to discriminate between the two cell lines. The cell adhesion was detected with the help of fluorescence and scanning electron microscopy. Potentially, this method could be applied for cell-on-chip studies by modifying the surface using different ligands, since clays are easy to combine with different functional organic groups. Furthermore, they possess low toxicity and are chemically stable [114]. The aim of affinity immobilization is not only enhanced cell attachment, but also the ability of bioactive molecules or proper combinations of them to activate cellular processes, such as phosphorylation, cell differentiation, the production of proteins, and apoptosis [42,63,115]. Therefore, gentle immobilization of the molecules is crucial, as molecules have to display bioactivity afterwards. Denaturation and conformational changes, as well as misorientation, lead to inefficient binding of cellular receptors and to decreased bioactivity. Thus, covalent binding and self-assembly of molecules to microarray surfaces are often used for long-term studies of cells [55]. The influence of lectin and carbohydrates on cell–cell interactions, communication, and immune response was shown by the specific attachment of T-cells to their corresponding sugar residues on a microarray with a variety of carbohydrates [116]. Immobilized peptides on microarrays likewise cause a response when cells bind to and incorporate these molecules. Here, tyrosine phosphorylation was induced in certain cell types [113]. Simultaneous immobilization of bioactive molecules and antibodies yielded the opportunity to induce protein secretion and detection of secreted proteins by the immobilized antibodies in a single experiment. As the volume of medium contained on a microarray is very small, the detection limit of antibody-protein binding is easily reached under standard conditions. A more detailed summary of the detection of cell activation on microarrays has been given elsewhere [43,51]. The examples summarized in this section demonstrate the diversity of ligands for specific cell capture on biochip surfaces, as well as the broad applicability of these ligands in the construction of cell microarrays. While most studies are rather on the proof-of-concept stage, other studies have already allowed the specific identification, attachment, and isolation of cells from complex biological samples. This specific enrichment of desired cell types on the microarray surface represents a major advantage of affinity‑based cell capture: clinical samples can be directly used for the LCMA fabrication with no need for the prior isolation of the cells of interest. Thus, it can be assumed that affinity-based cell capture will continue to be further developed and will contribute to the construction of future LCMAs. 2.3.4. Adsorption and Entrapment Instead of the printing of cell suspensions, cells can be applied to the complete microarray surface, which was modified previously. This is the common way to prepare living cell microarrays. These microarrays usually possess a glass or silicon surface that is modified via nanoprinter with several spots of cell-attachment-promoting biological molecules. Besides the aforementioned affinity-based immobilization of cells utilizing antibodies, aptamers, as well as specific proteins or peptides, further biomaterials such as polymers were printed to microarray surfaces to control cell attachment [42,113,116,117,118,119]. The binding of cells to these biomolecules is enhanced, whereas the remaining surface is structured or passivated, making it unattractive for cell attachment. In this way, it is ensured that cell groups are immobilized and separated from each other [61,62,63]. In the following Section, these methods are described in more detail. Structured Surfaces Cells are generally able to adsorb nonspecifically on many different kinds of solid substrates, while their viability and behavior may be affected either positively or negatively. This is exploited to mimic in vivo microenvironments or to prevent cross-contaminations and to separate cell spots from each other in an appropriate way. For these purposes, several methods were developed. The simplest method was already mentioned in Section 2.3.2, a micropatterned surface of cell-repellent and cell-adhesive regions. Here, a cell-adhesion-permitting transfection mixture was spotted onto a cell-repulsive hydrophobic polystyrene surface [81]. In this style, cell-repulsive coatings for surface passivation like PEG can be used for the passivation of surfaces between spots of cell attachment-enhancing proteins [63]. Moreover, elastomeric materials were used to form microwells utilizing micromolding-soft-lithography. Surface patterning via photolithography, soft-lithography via microcontact printing, and stencil-assisted techniques have been summarized elsewhere [120,121]. A silicone stamp is dipped into a solution of PEG-diacrylate on a treated glass slide. This solution was polymerized utilizing UV-radiation, resulting in the formation of a microwell-microarray after removal of the stamp. Cells were seeded and cultivated in these wells [122,123]. Additional hydrogels, such as hyaluronic acid or agarose, are also used for the formation of microwell-microarrays via soft-lithography [124]. An overview of strategies for the preparation of multiplexed LCMAs with separated cell groups of different origin was given in a recently published review [73]. A combination of surface chemistry, soft-lithography, and centrifugation was used to pattern different types of cells on a surface in a fast and controlled way [125]. 1-Hexadecanethiol was printed onto gold surfaces by microcontact printing to create hydrophobic areas. The remaining surface was passivated by ethylene glycol before fibronectin was adsorbed to the hydrophobic areas. After preparation, the treated surfaces were placed in centrifugation tubes. A minimal volume of cell suspension was added, and the tube was centrifuged at 2000 rpm for 1 min to force the cells to adhere. To immobilize co-cultures, a mask positioned at defined regions prevented the attachment of the first cell type, so that the second cell type could be patterned after removing the mask [125]. The working group of Kataoka [126,127,128] established a cell microarray chip for the detection and analysis of mixtures of several cell types via a confocal microarray laser scanner. Their chip consists of 20,944 microchambers (105 μm width, 50 μm depth and 300 μm distance) made of polystyrene. They were able to analyze antigen-specific B-cells, to detect malaria-infected erythrocytes within healthy erythrocytes, and to detect circulating carcinoma cells within a probe of leukocytes in a very specific and sensitive manner, while the tests themselves took less than 1 h [126,127,128]. Reymann et al. [129] applied cell arrays with 9216 microwells for reverse transfection studies and separated the cells physically in cavities of a titanium-coated glass slide. Thus, they prevented cross-contaminations and exploited the high-throughput advantages of transfected cell microarrays in a single system [129]. When the heterogeneity of cell populations or the individual cellular responses to stimuli are studied, single cells have to be isolated from each other and analyzed individually. Since several sizes of the stamp utilized for microcontact printing are often available, a small size in the range of a single cell can be used for the generation of single-cell microarrays. In another method, a highly defined hydrogel grid was applied, which was produced by dipping a nylon tissue into alginate solution, followed by a subsequent air- and freeze-drying on the microarray slide [130]. Cell behavior and cell fate of several cell types, such as stem cells or progenitor cells, are guided in vivo by the biomolecules of the cellular microenvironment, as well as by topographical cues [131]. The influence of a variety of geometries and sizes of surface structures were studied under the same conditions on one single topography array by Moe et al. [132]. The authors described an increased influence of these features on the neural and glial differentiation of primary murine neural progenitor cells, while they distinguished between anisotropic and isotropic topographies [132]. Biomaterials for Untargeted Cell Attachment and Stimulation Besides the targeted attachment of a specific cell type described in Section 2.3.3, the general enhancement of cell attachment itself is also focus of intense research. Several biomaterials, such as polymers, are under investigation for this purpose. An overview of common surface modification strategies is given in [133]. Different cell types (cell lines and embryonic stem cells) were cultivated on microarrays with spots of several polymers to investigate the cell adhesion properties of these polymers, resulting in new coatings for cell culture applications [59,134]. In this way, new synthetic biomaterials for tissue engineering applications were identified [117,134]. In addition to polymers, an agarose coating can be used to prevent unspecific cell attachment. As aforementioned, cell-repulsive characteristics were also found for a surface that is densely coated with PEG [122,135]. On top of this coating, polyurethane or polymers were printed to enhance cell attachment at these defined positions [59]. Anderson and coworkers demonstrated that their polymers support stem cell growth and proliferation in different levels and that only certain polymers were able to promote cell spreading and differentiation [134]. 2.3.5. Simulation of In Vivo Microenvironment Several approaches for microarray platforms are currently under development in order to create a physiologically relevant microenvironment. If cell reaction upon stimuli should be obtained, cells have to be provided with passive (stiffness, geometrical clues) and active (biological and chemical signals, cell–cell contacts) microenvironment conditions. For example, the in vivo microenvironment of cancer cells is composed of different types of neighboring cells, various chemical gradients, low oxygen concentration, high metabolite concentration, and a mechanical surrounding, resulting in a much more complex system than the one available in a simple monolayer of spotted cancer cells. Moreover, primary cells are sensitive to outer stimuli and change their physiology in simple monolayer cultures in comparison to a true in vivo situation. There are two major types of cell-based microarrays, which aim to mimic the in vivo microenvironment: (1) cell niche microarrays; (2) 3D cell culture microarrays. Stem Cell Niche The cell niche is a specific in vivo microenvironment to which cells are exposed depending on their location in the body [136,137]. In general, the cell niche is a combination of different physical, chemical, and biological factors. Physical factors include mechanical stiffness of substrate and pressure, chemical factors comprise a distribution of gases, pH values, and nutrients, while biological factors include the interplay of signaling molecules and cell–cell contacts. There are two extensively studied cell niches: the stem cell (SC) niche and the cancer stem cell (CSC) niche. For both, SCs and CSCs, oxygen tension plays an important role since in developing tumors, as well as at injury sites, oxygen tensions are much lower than ambient oxygen concentrations [136,137]. Stem cell niche plays an important role in the understanding of cell biology for regenerative medicine and provides an optimal in vitro microenvironment that reflects the in vivo situation. In order to simulate the stem cell niche, embryonic stem cells (ESCs) are often cultivated in vitro on various feeder cells: mouse embryonic fibroblasts, human fetal muscle and skin cells, adult skin and marrow cells, and foreskin fibroblasts [138]. Induced pluripotent stem cells (iPSCs) are also often cultured with the help of feeder cells. Although co-cultivation with feeder cells can provide cells with complex cytokines and signal molecules, this technique has serious limitations: presence of potential pathogens, risks of cross-contamination, as well as difficulties to distinguish between stem cells and feeder cells in response to external stimuli. The biochemical and physiochemical microenvironment can be simulated in vitro by the usage of ECM proteins and materials with variable mechanical properties. Therefore, several ECM arrays, biomaterial arrays, stiffness arrays, and topography arrays, as well as combinations of these were developed to simulate the complex stem cell niche in vitro [131,139,140]. Cellular transmembrane integrin receptors bind to ECM proteins, e.g., collagen and fibronectin. Thus, these molecules as well as polylysine or integrin-binding peptides like RGD (Arg–Gly–Asp, the cell adhesion promoting sequence of fibronectin) are commonly used substrates to control cell attachment. Since the ECM is important for cell adhesion and communication, the effects on cell behavior are well-investigated. Rasi Ghaemi et al. [63] investigated the cell adhesion qualities of several ECM proteins using mesenchymal stem cells (MSCs). The authors printed these proteins onto epoxy-silane-coated slides and passivated the residual surface with covalently bound bisamin-PEG. Collagen I was found to be most qualified for cell adhesion and growth, and was also used as surface modification for stem cells undergoing osteogenic differentiation [63]. Likewise, the adhesion profiles of several cell lines were investigated on ECM arrays [141]. Removable microfluidic channels, which are orthogonally aligned to fibronectin-coated gold strips, can be utilized for the generation of replicates of small isolated groups of different cell types on one single microarray [142]. To understand and mimic the microenvironment of cells in vivo, defined mixtures of ECM proteins and growth factors are printed onto the microarrays [115]. Microarrays modified with several combinations of ECM proteins and further factors (e.g., matrix stiffness) were used to screen for adhesion profiles of metastatic cell lines and primary tumor-derived cells. Thereby, the interactions between specific integrins of metastatic cells, and the ECM were shown [143]. Specific combinations of ECM proteins affected the differentiation of ESCs and the liver-specific function of hepatocytes [115]. Gobaa and colleagues developed an artificial niche microarray, where stem cell fate could be studied simultaneously under various conditions [140]. In this study, different types of cells (adherent and non-adherent) were spotted on a hydrogel-based microarray platform, coated with different types of proteins and exposed to different stiffnesses (shear moduli in the range of 1–50 kPa) in order to mimic diverse in vivo niches/conditions. The authors could reveal the influence of the mechanical environment on stem cell differentiation in terms of specific protein and mRNA expression. Appropriate simulation of the CSC niche can help to study the mechanisms of tumor development and find suitable treatment strategies. Indeed, most anti-cancer drugs, which were shown to be effective in in vitro experiments, demonstrated poor performance in subsequent animal studies and clinical trials. Although many studies have already been performed on the CSC niche and the influence of various parameters on cancer cell biology has been investigated, there are no reports on the establishment of CSC niche microarrays yet. Three-Dimensional Cell Constructs An efficient separation of cells on the same microarray is possible after encapsulation of the cells. Apart from separation, this method possesses another advantage, since cell growth conditions play a fundamental role in the cellular response to external stimuli [144]. Growth, metabolism, morphology, and organization of adherent cells in in vitro cultures differ greatly compared to the processes in vivo. In vivo cells do not grow in 2D monolayers, but form 3D structures, which can be imitated in vitro. Several 3D in vitro cell culture platforms have been developed over the last several years in order to improve the physiological relevance of the experimental results and fill the gap between 2D monolayer cell cultures and animal models [75,145]. Organoid cell cultures were first introduced in the early 1970s by radiobiologists in cancer research [146,147]. Being cultivated in cell spheroids, various tumor cell lines demonstrated enhanced resistance to chemotherapy and radiotherapy in terms of decreased apoptosis and increased clonogenecity [148,149]. Numerous studies on various cell lines, including primary cells, revealed significant changes in gene expression profiles for proliferation, angiogenesis, migration, invasion, or chemosensitivity of 3D relative to 2D cell culture conditions [144,150,151,152,153]. A comparison of epithelial cells growing in monolayers or in 3D constructs shows strong distinctions in gene and receptor expression, proliferation, and cell communication, as well as in their properties of differentiation [154,155,156]. Cells, which were grown in 2D culture, recover many of their lost characteristics after implanting them into an in vivo-like environment [157]. Due to this, microarrays promoting cell growth in 3D networks became more important in the last few years. There are three major 3D cell culture techniques that are routinely used today: (1) cells grown on a porous (anorganic or protein-based) matrix/scaffold; (2) cell spheroids (organoids); and (3) hydrogel-based cultures [158,159,160]. The choice of the 3D cell culture model is dependent on the intended application—if the aim is to study migration/adhesion in a 3D microenvironment, hydrogel constructs are the best choice. In contrast, for a simulation of a tumor-like environment, organoids are usually used. The main advantage of hydrogels is their transparency, enabling the online monitoring of cell behavior through the entire 3D construct. Several working groups extend traditional 2D living cell microarrays to the physiologically more relevant and more complex 3D systems [161]. Solid Scaffold Microarrays. Solid scaffolds can be created from a variety of materials. It is still discussed if such porous scaffolds provide true or pseudo 3D growth conditions since cells grow attached on the surface of the scaffold and, in general, it is a geometrically more complex structure than a cell culture flask bottom. Ock & Li presented 3D tissue model microarrays fabricated from biodegradable polylactic acid (PLA) [162]. PLA scaffolds were prepared with the help of the laser foaming technique and brain glioblastom T98G cells were seeded and cultivated over different periods of times (up to 120 h). Cells grown in 3D scaffolds demonstrated higher viability and aggregation, and they showed clusters of multiple cells. Moreover, cell–cell connections as well as the formation of microvilli and fibers were observed on 3D cultivated cells. Cell Spheroid Microarrays. Spontaneous cell–cell aggregation and creation of in vitro organoids represents a simple 3D cell culture model. Previously described PEG-microwells, which are formed by soft-lithography, can be used for the formation of 3D cell constructs. Cell repulsion is highest when PEG coating is performed on the complete microarray surface, and the microwells are formed by the hollows in this coating. Seeded cells cannot attach, resulting in cell aggregates, e.g., embryonic bodies of stem cells [123,163]. Cell spheroids of a certain size can be immobilized on the substrate in order to create 3D organoid microarrays. Wang and colleagues established such a microarray system, where they plotted MSC-spheroids on a microdomain patterned template on glass substrates [164]. Cells cultivated in this system were successfully differentiated towards adipocytes. Moreover, the differentiation efficiency in 3D structures was higher than in 2D monolayer controls. Consequently, 3D cultivation conditions are preferable for MSCs and microarray readouts will provide more relevant information about cell response to the stimuli. An alternative 3D spheroid-based approach (gel-free 3D microfluidic cell culture system) was developed by Ong et al. [165]. To create cell aggregates under dynamic cultivation conditions, the authors used the inter-cellular polymeric linker polyethyleneimine–hydrazide. Cell lines (A549 and C3A cells) and MSCs aggregates were immobilized in microchannels. Immobilized cells retained their proliferation and differentiation capacity and could be cultivated under dynamic conditions. Gel-based 3D Microarrays. In comparison to other 3D cell culture platforms, hydrogel-based culture is a relatively new and rapidly developing technique, which allows combination of organic, anorganic, and biological molecules in order to provide cells with an optimal microenvironment. The group of Dordick reported the establishment of 3D microarray platform based on MCF7 and Hep3B cells encapsulated in 60 nL alginate gel spots on modified glass slides for high-throughput toxicity screening [45]. In this study, the authors printed solutions of a cell-gel mixture to microarrays utilizing a non-contact nanoprinter. To polymerize the alginate solution, a mixture of barium chloride (BaCl2) and poly-l-lysine (PLL) was printed onto the microarray. The droplets of the cell-gel solution were then spotted over the same positions of previously printed BaCl2/PLL. PLL supports the attachment of alginate to the microarray surface, and BaCl2 causes the polymerization of the alginate. In a later study, the same working group encapsulated mouse embryonic stem cells (ESCs) in alginate on a microarray [166]. Using this microarray, ESCs could be expanded and differentiated in small format, and the influence of model signaling molecules (tretionin and FGF-4) on these processes was studied in dual slide configuration. The same working group established a 3D microarray for neural stem cell differentiation and toxicology [167] as well as for drug testing [168]. Neural stem cells cultivated in this microarray system could be expanded, differentiated, and used for cytotoxicity testing (dose-response curves) of neurotoxicants (retinoic acid, dexamethasone, and cadmium chloride). Moreover, on-chip in-cell immunofluorescence assay was performed using this microarray, which was comparable to conventional 2D assays. Pathel et al. [169] presented bioadhesive maleimide functionalized PEG hydrogel-based microarrays, where adherent and non-adherent cells can be encapsulated. The used microgels supported the spreading and cell growth of HeLa and TF-1a cell lines and could be used for high-throughput drug screening. In addition, the variation in microgel composition provided control of the stiffness for an investigation of the influence of the microenvironment on cell survival and function. Another working group used functionalized PEG hydrogel-based cell microarrays for the cultivation and study of epithelial ovarian carcinoma cell aggregates [47]. Here, the combination of two 3D culturing approaches (termed spheroids in hydrogels) was developed to study cell aggregation and to screen anti-cancer therapeutics. PEG-hydrogels were also used by Ranga and colleagues to create a 3D microarray system for the screening of different environments for mouse ESC cultivation [170]. The authors used an automatic liquid handling robot with nanopipetter head and studied the combination of five 3D microenvironmental factors: mechanical properties, proteolytic degradability, ECM proteins, cell–cell interaction, and the presence of soluble factors. Axelrod and colleagues entrapped single fibroblast cells surrounded by bacterial aggregates in hydrogels on a microarray slide [171]. This microarray system represents a co-culture approach with 3D microenvironment. Fibroblasts, hepatocytes, and macrophages encapsulated in PEG hydrogels, modified with RGD peptides, and immobilized on glass slides were reported by Koh et al. [172]. In this work, microgels with all three types of cells were immobilized on one slide in order to create a multiphenotype microarray. Suspensions consisting of cells and PEG were added into a network of microchannels on top of the slide, leading to a separation of different cell types without using a nanoprinter. Ozawa et al. [173] used electrodeposition for the fabrication of an alginate gel microwell array, where they cultured ESCs and HepG2 cells to construct spheroids. Moreover, they could create a co-culture system where mouse fibroblast cells were entrapped in alginate layers, and spheroids of another cell type were cultured above. This cultivation approach can also be used as an ESCs niche without the risk of cross-contamination. Although 3D cell microarrays possess numerous advantages in terms of their biological and physiological complexity, they do bring about challenges in terms of appropriate analytics (like all 3D cell culture systems). The estimation of cell viability, as well as specific staining of the complex structure of live 3D constructs, is still problematic. Moreover, issues of image acquisition, image analysis, and quantification represent additional challenges [174]. Concerning imaging, a sequence of images at a different focus depth (z-stack) is required to reveal the full complexity of the 3D organoids. Another approach for 3D microarray imaging is Raman micro-spectroscopy, but this needs further development. In the case of indirect cell viability assays, longer incubation times and often full cell lysis are required. Furthermore, cell assays, which rely on fluorescent and colorimetric assays (e.g., calcium flux), cannot be measured properly with existing methods [174]. 2.4. Specialized Microarray Systems Although the application of 3D tissues helps to approach physiological conditions, 3D cell cultivation is still far away from reflecting processes present in the human body. Due to static cultivation conditions, active transport processes are neglected and interaction of different cell types cannot be simulated adequately. To overcome these limitations, more specialized cell microarray systems have been developed. They allow the study of cell biology, including effects based on cell–cell interactions and fluid dynamics by utilizing microfluidic chip systems. 2.4.1. Microfluidic Systems While conventional multiwell plates are easy to use and are currently the gold standard in cell cultivation and investigation, the true composition of the medium is changing from medium exchange to medium exchange. Due to this, a controlled culture environment and a regulation of nutrient and metabolite concentration is not truly feasible [134,175]. Moreover, static multiwell plate systems can neither reflect the active transport of substances (nutrients, drugs, metabolites, etc.) present in the human body, e.g., by the blood circulation, nor imitate the interactions of different cells, tissues, and organs. To enable controlled cultivation conditions, perfusion reactors can be utilized, but they require large volumes of medium and supplements, resulting in expensive and impractical investigations. As a consequence, microfluidic cell culture systems using medium perfusion were developed to provide a controlled culture and differentiation environment [134,175]. Microfluidics allow the precise handling of μL volumes using μm channels in which the fluid flow can be controlled by miniaturized pumps and valves. Originally, microfluidics was introduced for lab-on-a-chip and micro-total analysis systems [176]. Manifold biological assays have already been successfully transferred to microfluidic devices including PCR [177] and protein separation and analysis [178]. For cell culture applications, most microfluidic devices are constructed of polydimethylsiloxane, thereby exploiting the air permeability, plasticity, and biocompatibility of this material [178,179]. In most cases, microfluidic cell culture chips contain at least two cell types at separate positions. The different cell types can, e.g., be cultivated in different chambers of the chip. To allow cell–cell interaction, microgaps can be introduced. These can support communication and migration of cells. For example, Businaro et al. [180] have utilized such a system to investigate the cross talk between cancer and immune cells. B16 melanoma cells and immune cells were seeded in different chambers of a microfluidic chip. A microgap interconnected both culture chambers and allowed the observation of immune cell migration towards the cancer cells. Another possibility to separate distinct cultivation chambers is the integration of porous membranes with different cell types seeded on both sides of the membrane [181]. While the membrane represents a physical barrier for the cells, thereby preventing mixing of different cell types, biomolecules like proteins and metabolites can pass through the pores, facilitating the indirect communication of the cells. Based on the experimental needs, microfluidic systems can be operated either in static or in dynamic mode. The example described above for the co-cultivation of cancer and immune cells was operated in a static mode in order to allow undisturbed migration of immune cells through the gap towards the cancer cells and to avoid effects of flow-induced sheer stress. Static mode is also often used during cell loading and cell adhesion in separate culture chambers. In contrast, other applications necessitate dynamic mode enabled by micropumps. This allows precise control of the cellular environment by ensuring a continuous supply of fresh media as well as a removal of metabolites. Moreover, the applied pressure and resulting sheer stress may also result in the formation of cell morphology closer to in vivo conditions. Within the context of the cellular systems summarized in this review article, dynamic mode microfluidics offers several key advantages: It allows for the development of highly miniaturized and parallelized dynamic cultivation platforms including reservoirs and channels for medium and supplement supply, as well as bypasses and outlets for the integration of analytics. Thereby, the consumption of samples, media, and supplements is minimized, and blood flow can be mimicked. Moreover, the dimensions of microfluidic systems are close to the scales of biological systems. This allows the simulation of the cellular microenvironment including, e.g., diffusion barriers and/or adsorption of substances and resulting concentration profiles. Besides, the major advantage of the microfluidic cell culture system is the better reflection of the interaction of different cell types or organs. In this scenario, one has to consider that, e.g., the administration of a drug does not result in an even distribution of the drug in all cell types and organs. In particular, the concentration will be highest at the point of administration while, in distinct regions of the body, the drug will occur in a diluted concentration. These differences in drug distribution will be further increased by adsorption of the drug within certain tissues resulting in lower bioavailability. Moreover, tissues do not only adsorb drugs in a passive way, cells also metabolize the drug. Therefore, peripheral cells might hardly see any active drug but rather be challenged by metabolites, which might seriously contribute to toxic effects. In summary, microfluidics allow for the design of co-culture chips helpful for the simulation of cell–cell interactions [178]. As an extension of these systems, OOC systems have been developed for a better understanding and simulation of complex networks present in the human body. 2.4.2. Organ-on-a-Chip Systems The US Food and Drug Administration (FDA) reported in 2004 that 92% of new potential drugs fail their approval in first clinical trials, although they pass successfully all in vitro and in vivo preclinical experiments [182]. At this point of development, the drug is applied to healthy people. Here, serious drawbacks of current drug screening systems become obvious: While conventional in vitro studies using human cells fail to imitate complex cellular networks, results of in vivo animal studies are not readily transferable to humans based on differences in physiology and metabolism. To improve the drug safety system and increase the efficiency of the drug screening process, as well as to reduce animal use in in vivo studies, novel well-designed human test systems are needed. In this context, OOC systems arose as a platform combining the advantages of TMA (namely the involvement of different cell types representing a functional unit of the tissue) and CMA (especially the application of living cells comprising metabolic activity), in combination with the benefits of microfluidic systems elaborated above. In the simplest case, the OOC aims to simulate one specific organ [183]. For example, microfluidic systems with integrated membranes have been used to simulate the human gut. In this case, the membrane was used as a matrix for human intestine epithelial cell adhesion. By applying fluid flow, sheer stress was introduced and cyclic strain was used to mimic peristaltic motion. Under these conditions, columnar epithelium developed and folded into structures recapitulating the structure of intestinal villi. This system was used to investigate the symbiotic relationship between epithelial cells and Lactobacillus rhamnosus as a commensal intestinal microbe [184]. Co-cultivation of epithelial cells with bacteria improved the barrier function as already observed for the human gut in vivo. Thus, the gut-on-a-chip was supposed to be a valuable tool for drug screening tests. By cultivation of renal cells, kidney-on-a-chip systems have been developed. For kidney cells, fluid shear stress has shown to be an important factor: Under physiological conditions, the cells are subjected to luminal fluid shear stress, which is a key modulator for cellular signal transduction [185], the organization of the cytoskeleton, and the formation of cellular junctions. By simulation of this fluid flow, kidney models with more realistic physiology can be constructed in a microfluidic device. Another organ already transferred to a microfluidic chip is the lung. By recapitulation of the complex microstructure of the tissue, and simulation of the mechanical stress induced by breathing, different lung-on-a-chip devices have been built and used for drug screening, as reviewed by Doryab et al. [186]. While these single-OOC devices have already been successfully applied to drug screening procedures and have been helpful towards a better understanding of diseases, the ultimate vision is a complete human-on-a-chip device. As a step forward to this direction, multi-organ-on-a-chip (multi-OOC) systems have attracted attention in recent years [187]. One major application of multi-OOC systems is pharmacokinetic drug toxicity screening. The action of a drug in the human body is governed by the complex interaction of several processes including the absorption of the drug by cells, its distribution in different cells, tissues, and organs, its metabolization within the different cells and organs, and its elimination (ADME). Moreover, in vivo microenvironments are often highly dynamic and heterogeneous with respect of blood flow, accounting for convective drug transport and diffusion barriers resulting in the formation of concentration profiles. These processes cannot be mimicked by static CMAs and single-OOC systems but can be modeled in multi-OOC systems. By the combination of different types of living tissues, and their interconnection by microfluidics, multi-OOCs are able to mimic in vivo drug response [188]. For example, Maschmeyer et al. [189] have developed a four-OOC system for the co-culture of human intestines, livers, skin, and kidneys. Therefore, human skin biopsies and 3D-villi-like structured intestine models have been integrated into the microfluidic systems. Liver was simulated by 3D spheroids and kidney by a monolayer of human proximal tubule epithelial cells on a membrane. Fluid excreted by the kidney cells was removed via the membrane. Using a peristaltic micropump, pulsatile media flow was applied to the channels interconnecting the four tissues. By the choice of integrated tissues, the authors were able to simulate several key parameters for in vivo drug fate: Through the kidney cells, drug excretion can be simulated, while the liver cells contribute a model for first path metabolism. The intestine compartment may act as a simulation unit for oral administration of the drug. The chip maintained homeostasis for at least 28 days and was therefore suggested as a device for ADME profiling and repeated-dose systemic toxicity testing of drug candidates [189]. This example demonstrates the power and potential of state-of-the-art multi-OOC devices. In the future, these efforts might result in a complete human-on-a-chip system allowing a mimicking of whole body responses. Such systems might not only be useful in comprehensive drug screening and toxicity assays, but may also help to gain better insights into the mechanisms of various diseases [190]. 3. Current Limitations The advantage of microarray technology is the very high specificity of each recognition event. Contrary to molecular identification, the analysis of cells is less specific, a fact mainly caused by cellular complexity. This disadvantage is balanced out by the main strength of using living systems: online and/or real-time investigation of biological effects. Despite this promising advantage, most investigations are unfortunately only proof-of-principle studies without further applications. Often, the complexity and scale of throughput are limited by the analytical technique, data acquisition, and data evaluation. These are own separate investigation parts within the process of microarray establishment. Moreover, high-throughput read out, e.g., with automated microscopes, automated segmentation, and analytic algorithms have to be adapted for the analysis of multiple signals. In particular, long-term studies of activity, viability, or stimulation of cells on LCMAs require these complex, well-matched techniques. One big challenge is the usage of more than one cell type on the same microarray. Often each cell type needs its individual cultivation condition. Cross-contaminations of different cell types through cell migration have to be avoided using physical or chemical borders for an adequate identification of the cell positions. Berthuy et al. [73] mainly discussed, in their recently published review, strategies for the preparation of multiplexed LCMAs. Additionally, the cross-talk between the cells is an important limitation. Cells influence each other via the secretion of cytokines or several other factors, which are taken up by other cells. If different cell types are not separated from each other, their biochemical signals may change cell responses and affect study results. This applies not only to cells of different tissue origin, but also to cells that are transfected with different genes. Assays analyzing the supernatant are not applicable in this case. Although this is a desired effect in co-cultivation studies, it has to be considered in all investigations, where different cells are cultivated next to each other. Commercially available DNA-microarrays are ready-to-use and storable; TMAs and CMAs of fixed cells are stable for several months. Even microarrays printed with plasmids or other nucleotides can be stored and used for transfected cell microarrays in another laboratory. In contrast, microarrays containing living cells have to be freshly prepared each time before an investigation can take place. Under consideration of this aspect, this type of microarray is more time-consuming and complex. The laboratories, which work with LCMAs, have to have a cell culture lab and adequate analytics matching their application, as well as access to microarray preparation techniques, such as lithography or nanoprinter devices. Contrary to this, laboratories applying samples onto TMAs and fixed CMAs do not need any kind of microarray preparation technique, such as tissue/cell isolation, hollow needles, and donor/acceptor blocks or a microtome. Instead, TMAs and CMAs can be commercially obtained. If LCMAs could be prepared and stored in one place and later used for investigations at a different laboratory, this would be extremely helpful and time-saving. Laboratories would also be independent from microarray preparation techniques and could thus be more focused on developing applications. 4. Conclusions and Future Directions New fields of application of microarray technology have been developed for use in medicine, biology, and toxicology. The fundamentals of microarray technology have been used to design new special techniques. Microarrays are no longer only based on oligonucleotides, antibodies, proteins, or aptamers. Even cells (fixed or alive) can be immobilized on slides creating special cell microarrays. Current cell arrays can be categorized into tissue, cell lines, primary cells, and organ biopsies. Furthermore, cellular microarrays can be classified depending on diverse criteria such as fixed tissue or systemic approaches, static or dynamic cultivation conditions, and 2D or 3D cell culture. Predominantly, living cell microarrays are currently used to analyze cellular behavior and for pharmaceutical drug testing. In this article, we focused on self-spotted glass slide living cell arrays, which are a popular format. The glass slides are coated with functional groups enabling the attachment of cells to the glass surface. After attachment, cells proliferate on the glass surface comparable to traditional plastic cultivation flasks. These microarrays can be stored under defined conditions. In the case of human tissues, which are rare and hard to obtain, strategies to enrich well characterized cells are important. LCMAs can be used to overcome these limitations. This approach is enabling minimal invasive analysis of patients and will be useful for biomarker detection, single-cell analysis, personalized medicine, and point-of-care diagnostic. Important applications of LCMAs include the discrimination between different tissue phenotypes and the discovery of previously unknown tissue subtypes, e.g., in cancer research. Usually, studies are performed with hundreds of thousands or even millions of cells. In some cases, it is not possible to collect large numbers of cells, which makes their analysis (e.g., transcriptome) almost impossible. LCMAs provide an alternative opportunity to work with only a few cells. To improve the detection and treatment of diseases and to analyze fundamental biological principles, new approaches to single-cell analysis must be developed. To create 3D cell cultures, drops of hydrogels, synthetic scaffolds, or cell organoids can be attached to the microarray platform. Being more physiologically relevant, these 3D systems possess challenges in analytics and image acquisition. However, 3D cell microarrays have great perspectives in personalized drug screening in the improvement of preclinical testing (e.g., for anti-cancer drugs and substances, which help to restore cell specific cell functions). With the help of this new method, a high-throughput screening for toxicology, protein expression and production, micro RNA expression, drug testing, or the stimulus-response relationships of living cells is possible. To acquire not only results of toxicology, but also a better understanding of stimulus-response assays, a suitable monitoring of living cells is essential. For example, a very common method to monitor a stimulus-response relationship is the use of impedance measurements (ECIS). Here, a change of the cell shape caused by a drug or chemical or any other kind of stimulus can be measured easily. This technique can show much more sensitive results of a dose-response than any kind of colorimetric method. Simple cell cultures or cell lines can be used in early screens and for more complex questions; e.g., organoid cultures are an adequate screening system. Finally, there is a clear value of living cell microarrays to enable the analysis of different cell types and treatment conditions on a miniaturized high-throughput platform, thereby minimizing costs, reagents, drugs, and experimental variance due to very small sample volumes. Acknowledgments This work was partially funded by the Niedersächsische Krebsgesellschaft e.V. and was carried out as an integral part of the BIOFABRICATION FOR NIFE Initiative, which is financially supported by the Lower Saxony ministry of Science and Culture and the Volkswagen Stiftung. NIFE is the Lower Saxony Center for Biomedical Engineering, Implant Research and Development in Hannover, a joint translational research center of the Hannover Medical School, the Leibniz University Hannover, the University of Veterinary Medicine Hannover and the Laser Center Hannover. We further thank Pepelanova for her critical proofreading of the manuscript. Conflicts of Interest The authors declare no conflict of interest. The funding sponsors had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results. Abbreviations 2D Two-dimensional 3D Three-dimensional ADME Absorption, distribution, metabolism, and excretion APTES (3-Aminopropyl)triethoxysilane CD Cluster of Differentiation CHO Chinese Hamster Ovary CMA Cell microarray CSC Cancer stem cell CTC Circulating tumor cells DIC Differential interference contrast DNA Deoxyribonucleic acid ECIS Electric cell-substrate impedance sensor ECM Extracellular matrix EDC 1-Ethyl-3-(3-dimethylaminopropyl)carbodiimide EpCAM Epithelial Cell Adhesion Molecule ESC Embryonic stem cell FA-Mont Folic acid modified montmorillonite clay FDA Food and Drug Administration GFP Green fluorescent protein IgG Immunoglobulin G iPSC Induced pluripotent stem cell LCMA Living cell microarray MEA Multi-electrode array MEMS Micro-electromechanical systems MRI Magnetic resonance imaging mRNA Messenger RNA MSC Mesenchymal stem cell NHS N‑Hydroxysuccinimide OOC Organ-on-a-chip PCR Polymerase chain reaction PEG Polyethylene glycol PLA Polylactic acid PLL Poly-l-lysine PSMA Prostate-specific membrane antigen RNA Ribonucleic acid RNAi RNA interference SC Stem cell shRNA Short hairpin RNA siRNA Small interfering RNA SPR Surface plasmon resonance TMA Tissue microarray UV Ultraviolet Figure and Scheme microarrays-05-00011-sch001_Scheme 1Scheme 1 Overview of living cell microarrays presented in this review. Figure 1 Figurative overview of living cell microarrays (LCMAs) described in this review. (A) 2D platforms; (i) undirected adsorption of cells on passivated LCMAs with partially unpassivated surface areas; (ii) LCMAs coated with biomaterials for enhanced cell attachment and stimulation, inclusively transfected cell microarrays; (iii) affinity-based immobilization of cells; (B) LCMAs for simulations of in vivo microenvironments and cell stimulation; (i) 3D cell constructs; (ii) microarrays with microstructured surfaces; (C) microfluidic devices with organoids. ==== Refs References 1. Schena M. Shalon D. Davis R.W. Brown P.O. Quantitative monitoring of gene expression patterns with a complementary DNA microarray Science 1995 270 467 470 10.1126/science.270.5235.467 7569999 2. Ehrenreich A. DNA microarray technology for the microbiologist: An overview Appl. Microbiol. Biotechnol. 2006 73 255 273 10.1007/s00253-006-0584-2 17043830 3. Singh A. Kumar N. A review on DNA microarray technology IJCRR 2013 5 01 05 4. Luo J. Zha S. Gage W.R. Dunn T.A. Hicks J.L. Bennett C.J. Ewing C.M. Platz E.A. Ferdinandusse S. Wanders R.J. Α-methylacyl-coa racemase: A new molecular marker for prostate cancer Cancer Res. 2002 62 2220 2226 11956072 5. Gerhold D. Rushmore T. Caskey C.T. DNA chips: Promising toys have become powerful tools Trends Biochem. Sci. 1999 24 168 173 10.1016/S0968-0004(99)01382-1 10322428 6. Pirrung M.C. How to make a DNA chip Angew. Chem. 2002 41 1276 1289 10.1002/1521-3773(20020415)41:8<1276::AID-ANIE1276>3.0.CO;2-2 19750752 7. Schwanhausser B. Busse D. Li N. Dittmar G. Schuchhardt J. Wolf J. Chen W. Selbach M. Global quantification of mammalian gene expression control Nature 2011 473 337 342 10.1038/nature10098 21593866 8. Schwanhausser B. Busse D. Li N. Dittmar G. Schuchhardt J. Wolf J. Chen W. Selbach M. Corrigendum: Global quantification of mammalian gene expression control Nature 2013 495 126 127 10.1038/nature11848 23407496 9. Walter J.G. Kokpinar O. Friehs K. Stahl F. Scheper T. Systematic investigation of optimal aptamer immobilization for protein-microarray applications Anal. Chem. 2008 80 7372 7378 10.1021/ac801081v 18729475 10. Lübbecke M. Walter J.G. Stahl F. Scheper T. Aptamers as detection molecules on reverse phase protein microarrays for the analysis of cell lysates Eng. Life Sci. 2012 12 144 151 10.1002/elsc.201100100 11. Witt M. Walter J.-G. Stahl F. Aptamer microarrays—Current status and future prospects Microarrays 2015 4 115 10.3390/microarrays4020115 12. Okamoto T. Suzuki T. Yamamoto N. Microarray fabrication with covalent attachment of DNA using bubble jet technology Nature Biotechnol. 2000 18 438 441 10748527 13. Rose S. Application of a novel microarraying system in genomics research and drug discovery J. Assoc. Lab. Autom. 1998 3 53 56 14. Battifora H. The multitumor (sausage) tissue block: Novel method for immunohistochemical antibody testing Lab. Investig. J. Tech. Methods Pathol. 1986 55 244 248 15. Kononen J. Bubendorf L. Kallioniemi A. Barlund M. Schraml P. Leighton S. Torhorst J. Mihatsch M.J. Sauter G. Kallioniemi O.P. Tissue microarrays for high-throughput molecular profiling of tumor specimens Nat. Med. 1998 4 844 847 10.1038/nm0798-844 9662379 16. Rimm D.L. Camp R.L. Charette L.A. Olsen D.A. Provost E. Amplification of tissue by construction of tissue microarrays Exp. Mol. Pathol. 2001 70 255 264 10.1006/exmp.2001.2363 11418004 17. Barlund M. Forozan F. Kononen J. Bubendorf L. Chen Y. Bittner M.L. Torhorst J. Haas P. Bucher C. Sauter G. Detecting activation of ribosomal protein s6 kinase by complementary DNA and tissue microarray analysis J. Nat. Cancer Inst. 2000 92 1252 1259 10.1093/jnci/92.15.1252 10922410 18. Sauter G. Simon R. Hillan K. Tissue microarrays in drug discovery Nat. Rev. Drug Dis. 2003 2 962 972 10.1038/nrd1254 14654795 19. Vogel U. Overview on techniques to construct tissue arrays with special emphasis on tissue microarrays Microarrays 2014 3 103 10.3390/microarrays3020103 20. Rubin M.A. Dunn R. Strawderman M. Pienta K.J. Tissue microarray sampling strategy for prostate cancer biomarker analysis Am. J. Surg. Pathol. 2002 26 312 319 10.1097/00000478-200203000-00004 11859202 21. Simon R. Mirlacher M. Sauter G. Tissue microarrays in cancer diagnosis Expert Rev. Mol. Diagn. 2003 3 421 430 10.1586/14737159.3.4.421 12877382 22. Kaplan J. Hukku B. Cell line characterization and authentication Methods Cell Biol. 1998 57 203 216 9648106 23. Pipas J.M. Sv40: Cell transformation and tumorigenesis Virology 2009 384 294 303 10.1016/j.virol.2008.11.024 19070883 24. Poulos S.P. Dodson M.V. Hausman G.J. Cell line models for differentiation: Preadipocytes and adipocytes Exp. Biol. Med. 2010 235 1185 1193 10.1258/ebm.2010.010063 20864461 25. Braunschweig T. Chung J.Y. Hewitt S.M. Tissue microarrays: Bridging the gap between research and the clinic Expert Rev. Proteom. 2005 2 325 336 10.1586/14789450.2.3.325 16000080 26. Waterworth A. Hanby A. Speirs V. A novel cell array technique for high-throughput, cell-based analysis In Vitro Cell. Dev. Biol. Anim. 2005 41 185 187 10.1290/0505032.1 16223332 27. Ferrer B. Bermudo R. Thomson T. Nayach I. Soler M. Sanchez M. Castillo M. Calvo J. Campo E. Fernandez P.L. Paraffin-embedded cell line microarray (peclima): Development and validation of a high-throughput method for antigen profiling of cell lines Pathobiol. J. Immunopathol. Mol. Cell. Biol. 2005 72 225 232 10.1159/000089416 16374066 28. Andersson A.C. Stromberg S. Backvall H. Kampf C. Uhlen M. Wester K. Ponten F. Analysis of protein expression in cell microarrays: A tool for antibody-based proteomics J. Histochem. Cytochem. 2006 54 1413 1423 10.1369/jhc.6A7001.2006 16957166 29. Kampf C. Andersson A.-C. Wester K. Björling E. Uhlen M. Ponten F. Antibody-based tissue profiling as a tool for clinical proteomics Clin. Proteom. 2004 1 285 299 10.1385/CP:1:3-4:285 30. La Spada A. Rainoldi B. De Blasio A. Biunno I. Application of tissue microarray technology to stem cell research Microarrays 2014 3 159 167 10.3390/microarrays3030159 31. Hart T. Zhao A. Garg A. Bolusani S. Marcotte E.M. Human cell chips: Adapting DNA microarray spotting technology to cell-based imaging assays PLoS ONE 2009 4 11 10.1371/journal.pone.0007088 19862318 32. Schwenk J.M. Stoll D. Templin M.F. Joos T.O. Cell microarrays: An emerging technology for the characterization of antibodies BioTechniques 2002 33 S54 S61 33. Masuda N. Ohnishi T. Kawamoto S. Monden M. Okubo K. Analysis of chemical modification of rna from formalin-fixed samples and optimization of molecular biology applications for such samples Nucleic Acids Res. 1999 27 4436 4443 10.1093/nar/27.22.4436 10536153 34. Werner M. Chott A. Fabiano A. Battifora H. Effect of formalin tissue fixation and processing on immunohistochemistry Am. J. Surg. Pathol. 2000 24 1016 1019 10.1097/00000478-200007000-00014 10895825 35. Schoenberg Fejzo M. Slamon D.J. Frozen tumor tissue microarray technology for analysis of tumor rna, DNA, and proteins Am. J. Pathol. 2001 159 1645 1650 10.1016/S0002-9440(10)63011-8 11696425 36. Stephan J.P. Schanz S. Wong A. Schow P. Wong W.L. Development of a frozen cell array as a high-throughput approach for cell-based analysis Am. J. Pathol. 2002 161 787 797 10.1016/S0002-9440(10)64238-1 12213706 37. Miyaji T. Hewitt S.M. Liotta L.A. Star R.A. Frozen protein arrays: A new method for arraying and detecting recombinant and native tissue proteins Proteomics 2002 2 1489 1493 10.1002/1615-9861(200211)2:11<1489::AID-PROT1489>3.0.CO;2-8 12442248 38. Ziauddin J. Sabatini D.M. Microarrays of cells expressing defined cDNAs Nature 2001 411 107 110 10.1038/35075114 11333987 39. Angres B. Cell microarrays Expert Rev. Mol. Diagn. 2005 5 769 779 10.1586/14737159.5.5.769 16149879 40. Elad T. Lee J.H. Belkin S. Gu M.B. Microbial whole-cell arrays Microbial. Biotechnol. 2008 1 137 148 10.1111/j.1751-7915.2007.00021.x 21261831 41. van der Meer J.R. Belkin S. Where microbiology meets microengineering: Design and applications of reporter bacteria Nat. Rev. Micro. 2010 8 511 522 10.1038/nrmicro2392 20514043 42. Yarmush M.L. King K.R. Living-cell microarrays Ann. Rev. Biomed. Eng. 2009 11 235 257 10.1146/annurev.bioeng.10.061807.160502 19413510 43. Papp K. Szittner Z. Prechl J. Life on a microarray: Assessing live cell functions in a microarray format Cell. Mol. Life Sci. 2012 69 2717 2725 10.1007/s00018-012-0947-z 22391673 44. Anglin E. Davey R. Herrid M. Hope S. Kurkuri M. Pasic P. Hor M. Fenech M. Thissen H. Voelcker N.H. Cell microarrays for the screening of factors that allow the enrichment of bovine testicular cells Cytom. A J. Int. Soc. Anal. Cytol. 2010 77 881 889 10.1002/cyto.a.20913 20803736 45. Lee M.Y. Kumar R.A. Sukumaran S.M. Hogg M.G. Clark D.S. Dordick J.S. Three-dimensional cellular microarray for high-throughput toxicology assays Proc. Natl. Acad. Sci. USA 2008 105 59 63 10.1073/pnas.0708756105 18160535 46. Seidel D. Krinke D. Jahnke H.-G. Hirche A. Kloß D. Mack T.G. Striggow F. Robitzki A. Induced tauopathy in a novel 3D-culture model mediates neurodegenerative processes: A real-time study on biochips PLoS ONE 2012 7 11 10.1371/journal.pone.0049150 23145103 47. Loessner D. Kobel S. Clements J. Lutolf M. Hutmacher D. Hydrogel microwell arrays allow the assessment of protease-associated enhancement of cancer cell aggregation and survival Microarrays 2013 2 208 10.3390/microarrays2030208 48. Mutiu A.I. Brandl C.J. RNA isolation from yeast using silica matrices J. Biomol. Tech. 2005 16 316 317 16522851 49. Achilles J. Stahl F. Harms H. Muller S. Isolation of intact rna from cytometrically sorted saccharomyces cerevisiae for the analysis of intrapopulation diversity of gene expression Nat. Protocols 2007 2 2203 2211 10.1038/nprot.2007.322 17853877 50. Xu T. Olson J. Zhao W. Atala A. Zhu J.-M. Yoo J.J. Characterization of cell constructs generated with inkjet printing technology using in vivo magnetic resonance imaging J. Manuf. Sci. Eng. 2008 130 021013 021013 10.1115/1.2902857 51. Cagnin S. Cimetta E. Guiducci C. Martini P. Lanfranchi G. Overview of micro- and nano-technology tools for stem cell applications: Micropatterned and microelectronic devices Sensors 2012 12 15947 10.3390/s121115947 23202240 52. Spira M.E. Hai A. Multi-electrode array technologies for neuroscience and cardiology Nat. Nano. 2013 8 83 94 10.1038/nnano.2012.265 23380931 53. Ross J.D. Connor S.M.O. Blum R.A. Brown E.A. DeWeerth S.P. Multielectrode impedance tuning: Reducing noise and improving stimulation efficacy Proceedings of the 26th Annual International Conference of the IEEE Engineering in Medicine and Biology Society San Francisco, CA, USA 1–5 September 2004 4115 4117 54. Asakura K. Hayashi S. Ojima A. Taniguchi T. Miyamoto N. Nakamori C. Nagasawa C. Kitamura T. Osada T. Honda Y. Improvement of acquisition and analysis methods in multi-electrode array experiments with ips cell-derived cardiomyocytes J. Pharmacol. Toxicol. Methods 2015 75 17 26 10.1016/j.vascn.2015.04.002 25910965 55. Michelini E. Roda A. Staying alive: New perspectives on cell immobilization for biosensing purposes Anal. Bioanal. Chem. 2012 402 1785 1797 10.1007/s00216-011-5364-x 21922308 56. Belkin S. Gu M. Whole Cell Sensing Systems I: Reporter Cells and Devices Springer Berlin/Heidelberg, Germany 2010 Volume 117 1 208 57. Belkin S. Gu M.B. Whole Cell Sensing Systems II: Applications Springer Berlin/Heidelberg, Germany 2010 Volume 118 1 222 58. Date A. Pasini P. Daunert S. Fluorescent and bioluminescent cell-based sensors: Strategies for their preservation Adv. Biochem. Eng. Biotechnol. 2010 117 57 75 20091290 59. Tourniaire G. Collins J. Campbell S. Mizomoto H. Ogawa S. Thaburet J.F. Bradley M. Polymer microarrays for cellular adhesion Chem. Commun. 2006 2118 2120 10.1039/b602009g 16703126 60. Jonczyk R. Timur S. Scheper T. Stahl F. Development of living cell microarrays using non-contact micropipette printing J. Biotechnol. 2016 217 109 111 10.1016/j.jbiotec.2015.11.013 26603124 61. Anglin E.J. Salisbury C. Bailey S. Hor M. Macardle P. Fenech M. Thissen H. Voelcker N.H. Sorted cell microarrays as platforms for high-content informational bioassays Lab Chip 2010 10 3413 3421 10.1039/c0lc00185f 20941408 62. Ghaedi M. Tuleuova N. Zern M.A. Wu J. Revzin A. Bottom-up signaling from hgf-containing surfaces promotes hepatic differentiation of mesenchymal stem cells Biochem. Biophys. Res. Commun. 2011 407 295 300 10.1016/j.bbrc.2011.03.005 21382341 63. Rasi Ghaemi S. Harding F. Delalat B. Vasani R. Voelcker N.H. Surface engineering for long-term culturing of mesenchymal stem cell microarrays Biomacromolecules 2013 14 2675 2683 10.1021/bm400531n 23767759 64. Suri S. Singh A. Nguyen A.H. Bratt-Leal A.M. McDevitt T.C. Lu H. Microfluidic-based patterning of embryonic stem cells for in vitro development studies Lab Chip 2013 13 4617 4624 10.1039/c3lc50663k 24113509 65. Yahya W. Kadri N. Ibrahim F. Cell patterning for liver tissue engineering via dielectrophoretic mechanisms Sensors 2014 14 11714 10.3390/s140711714 24991941 66. Huang Y. Joo S. Duhon M. Heller M. Wallace B. Xu X. Dielectrophoretic cell separation and gene expression profiling on microelectronic chip arrays Anal. Chem. 2002 74 3362 3371 10.1021/ac011273v 12139041 67. Xu T. Jin J. Gregory C. Hickman J.J. Boland T. Inkjet printing of viable mammalian cells Biomaterials 2005 26 93 99 10.1016/j.biomaterials.2004.04.011 15193884 68. Saunders R.E. Gough J.E. Derby B. Delivery of human fibroblast cells by piezoelectric drop-on-demand inkjet printing Biomaterials 2008 29 193 203 10.1016/j.biomaterials.2007.09.032 17936351 69. Yusof A. Keegan H. Spillane C.D. Sheils O.M. Martin C.M. O’Leary J.J. Zengerle R. Koltay P. Inkjet-like printing of single-cells Lab Chip 2011 11 2447 2454 10.1039/c1lc20176j 21655638 70. Wu X. Zahari M.S. Renuse S. Jacob H.K. Sakamuri S. Singal M. Gabrielson E. Sukumar S. Pandey A. A breast cancer cell microarray (CMA) as a rapid method to characterize candidate biomarkers Cancer Biol. Ther. 2014 15 1593 1599 10.4161/15384047.2014.961886 25535895 71. Cui X. Boland T. Human microvasculature fabrication using thermal inkjet printing technology Biomaterials 2009 30 6221 6227 10.1016/j.biomaterials.2009.07.056 19695697 72. Ferris C.J. Gilmore K.J. Beirne S. McCallum D. Wallace G.G. in het Panhuis M. Bio-ink for on-demand printing of living cells Biomaterials Sci. 2013 1 224 230 10.1039/C2BM00114D 73. Berthuy O.I. Blum L.J. Marquette C.A. Cells on chip for multiplex screening Biosens. Bioelectron. 2016 76 29 37 10.1016/j.bios.2015.04.024 25892543 74. Berthuy O.I. Mandon C.A. Corgier B.P. Octobre G.G. Ceccone G. Spampinato V. Blum L.J. Marquette C.A. Material surface engineering for multiplex cell culture in microwell J. Mater. Sci. 2014 49 4481 4489 10.1007/s10853-014-8145-z 75. Ruedinger F. Lavrentieva A. Blume C. Pepelanova I. Scheper T. Hydrogels for 3d mammalian cell culture: A starting guide for laboratory practice Appl. Microbiol. Biotechnol. 2015 99 623 636 10.1007/s00253-014-6253-y 25432676 76. Dababneh A.B. Ozbolat I.T. Bioprinting technology: A current state-of-the-art review J. Manuf. Sci. Eng. 2014 136 061016 10.1115/1.4028512 77. Murphy S.V. Atala A. 3D bioprinting of tissues and organs Nat. Biotechnol. 2014 32 773 785 10.1038/nbt.2958 25093879 78. Echeverri C.J. Perrimon N. High-throughput rnai screening in cultured cells: A user’s guide Nat. Rev. Genet. 2006 7 373 384 10.1038/nrg1836 16607398 79. Erfle H. Simpson J.C. Bastiaens P.I. Pepperkok R. Sirna cell arrays for high-content screening microscopy BioTechniques 2004 37 454 458 15470900 80. Yoshikawa T. Uchimura E. Kishi M. Funeriu D.P. Miyake M. Miyake J. Transfection microarray of human mesenchymal stem cells and on-chip sirna gene knockdown J. Control. Release 2004 96 227 232 10.1016/j.jconrel.2004.01.024 15081214 81. Rantala J.K. Mäkelä R. Aaltola A.-R. Laasola P. Mpindi J.-P. Nees M. Saviranta P. Kallioniemi O. A cell spot microarray method for production of high density sirna transfection microarrays BMC Genom. 2011 12 1 15 10.1186/1471-2164-12-162 21443765 82. Erfle H. Eskova A. Reymann J. Starkuviene V. Cell arrays and high-content screening Protein Microarrays Korf U. Humana Press New York, NY, USA 2011 Volume 785 277 287 83. Delehanty J.B. Shaffer K.M. Lin B. A comparison of microscope slide substrates for use in transfected cell microarrays Biosens. Bioelectron. 2004 20 773 779 10.1016/j.bios.2004.04.016 15522592 84. McConnell K.I. Schweller R.M. Diehl M.R. Suh J. Live-cell microarray surface coatings supporting reverse transduction by adeno-associated viruses BioTechniques 2011 51 255 258 21988691 85. Rajan S. Djambazian H. Dang H.C. Sladek R. Hudson T.J. The living microarray: A high-throughput platform for measuring transcription dynamics in single cells BMC Genom. 2011 12 115 10.1186/1471-2164-12-115 21324195 86. Palmer E.L. Miller A.D. Freeman T.C. Identification and characterisation of human apoptosis inducing proteins using cell-based transfection microarrays and expression analysis BMC Genom. 2006 7 1 17 87. Mannherz O. Mertens D. Hahn M. Lichter P. Functional screening for proapoptotic genes by reverse transfection cell array technology Genomics 2006 87 665 672 10.1016/j.ygeno.2005.12.009 16503394 88. Konrad A. Wies E. Thurau M. Marquardt G. Naschberger E. Hentschel S. Jochmann R. Schulz T.F. Erfle H. Brors B. A systems biology approach to identify the combination effects of human herpesvirus 8 genes on nf-κb activation J. Virol. 2009 83 2563 2574 10.1128/JVI.01512-08 19129458 89. Buchholz M. Honstein T. Kirchhoff S. Kreider R. Schmidt H. Sipos B. Gress T.M. A multistep high-content screening approach to identify novel functionally relevant target genes in pancreatic cancer PLoS ONE 2015 10 11 10.1371/journal.pone.0122946 25849100 90. Erfle H. Lisauskas T. Claas C. Reymann J. Starkuviene V. Cell arrays for the measurement of organelle dynamics in living cells Cell-Based Microarrays Palmer E. Humana Press New York, NY, USA 2011 Volume 706 73 81 91. Neumann B. Walter T. Hériché J.K. Bulkescher J. Erfle H. Conrad C. Rogers P. Poser I. Held M. Liebel U. Phenotypic profiling of the human genome by time-lapse microscopy reveals cell division genes Nature 2010 464 721 727 10.1038/nature08869 20360735 92. Mittal V. Improving the efficiency of rna interference in mammals Nat. Rev. Genet. 2004 5 355 365 10.1038/nrg1323 15143318 93. O’Keefe E.P. siRNAs and shRNAs: Tools for protein knockdown by gene silencing Word Lab. 2013 10.13070/mm.cn.3.197 94. Fellmann C. Lowe S.W. Stable rna interference rules for silencing Nat. Cell. Biol. 2014 16 10 18 10.1038/ncb2895 24366030 95. Siomi H. Siomi M.C. On the road to reading the rna-interference code Nature 2009 457 396 404 10.1038/nature07754 19158785 96. Erfle H. Pepperkok R. Production of sirna and cdna-transfected cell arrays on noncoated chambered coverglass for high-content screening microscopy in living cells Target Discovery and Validation Reviews and Protocols Sioud M. Humana Press New York, NY, USA 2007 Volume 360 155 161 97. Mousses S. Caplen N.J. Cornelison R. Weaver D. Basik M. Hautaniemi S. Elkahloun A.G. Lotufo R.A. Choudary A. Dougherty E.R. Rnai microarray analysis in cultured mammalian cells Genome Res. 2003 13 2341 2347 10.1101/gr.1478703 14525932 98. Erfle H. Neumann B. Liebel U. Rogers P. Held M. Walter T. Ellenberg J. Pepperkok R. Reverse transfection on cell arrays for high content screening microscopy Nat. Protocols 2007 2 392 399 10.1038/nprot.2006.483 17406600 99. Moore C.B. Guthrie E.H. Huang M.T.-H. Taxman D.J. Short hairpin rna (shrna): Design, delivery, and assessment of gene knockdown Methods Mol. Biol. 2010 629 141 158 20387148 100. Starkuviene V. Pepperkok R. Erfle H. Transfected cell microarrays: An efficient tool for high-throughput functional analysis Expert Rev. Proteom. 2007 4 479 489 10.1586/14789450.4.4.479 17705706 101. Fjeldbo C.S. Misund K. Günther C.C. Langaas M. Steigedal T.S. Thommesen L. Laegreid A. Bruland T. Functional studies on transfected cell microarray analysed by linear regression modelling Nucleic Acids Res. 2008 36 e97 10.1093/nar/gkn428 18628295 102. Kumar R. Conklin D.S. Mittal V. High-throughput selection of effective rnai probes for gene silencing Genome Res. 2003 13 2333 2340 10.1101/gr.1575003 14525931 103. Simpson J.C. Joggerst B. Laketa V. Verissimo F. Cetin C. Erfle H. Bexiga M.G. Singan V.R. Heriche J.-K. Neumann B. Genome-wide rnai screening identifies human proteins with a regulatory function in the early secretory pathway Nat. Cell Biol. 2012 14 764 774 10.1038/ncb2510 22660414 104. Fengler S. H Bastiaens P.I. Grecco H.E. Roda-Navarro P. Optimizing cell arrays for accurate functional genomics BMC Res. Notes 2012 5 1 8 10.1186/1756-0500-5-358 22214347 105. Chen D.S. Davis M.M. Molecular and functional analysis using live cell microarrays Curr. Opin. Chem. Biol. 2006 10 28 34 10.1016/j.cbpa.2006.01.001 16413817 106. Tateno H. Uchiyama N. Kuno A. Togayachi A. Sato T. Narimatsu H. Hirabayashi J. A novel strategy for mammalian cell surface glycome profiling using lectin microarray Glycobiology 2007 17 1138 1146 10.1093/glycob/cwm084 17693441 107. Milgram S. Bombera R. Livache T. Roupioz Y. Antibody microarrays for label-free cell-based applications Methods 2012 56 326 333 10.1016/j.ymeth.2011.10.016 22200606 108. Shigdar S. Lin J. Yu Y. Pastuovic M. Wei M. Duan W. RNA aptamer against a cancer stem cell marker epithelial cell adhesion molecule Cancer Sci. 2011 102 991 998 10.1111/j.1349-7006.2011.01897.x 21281402 109. Song Y. Zhu Z. An Y. Zhang W. Zhang H. Liu D. Yu C. Duan W. Yang C.J. Selection of DNA aptamers against epithelial cell adhesion molecule for cancer cell imaging and circulating tumor cell capture Anal. Chem 2013 85 4141 4149 10.1021/ac400366b 23480100 110. Chen Q. Wu J. Zhang Y. Lin Z. Lin J.-M. Targeted isolation and analysis of single tumor cells with aptamer-encoded microwell array on microfluidic device Lab Chip 2012 12 5180 5185 10.1039/c2lc40858a 23108418 111. Dharmasiri U. Balamurugan S. Adams A.A. Okagbare P.I. Obubuafo A. Soper S.A. Highly efficient capture and enumeration of low abundance prostate cancer cells using prostate-specific membrane antigen aptamers immobilized to a polymeric microfluidic device Electrophoresis 2009 30 3289 3300 10.1002/elps.200900141 19722212 112. Lee J.H. Bao K. Frangioni J.V. Choi H.S. Screening of small molecule microarrays for ligands targeted to the extracellular epitopes of living cells Microarrays 2015 4 53 63 10.3390/microarrays4010053 26435848 113. Falsey J.R. Renil M. Park S. Li S. Lam K.S. Peptide and small molecule microarray for high throughput cell adhesion and functional assays Bioconjugate Chem. 2001 12 346 353 10.1021/bc000141q 114. Bongartz R. Ag D. Seleci M. Walter J.-G. Yalcinkaya E.E. Demirkol D.O. Stahl F. Timur S. Scheper T. Folic acid-modified clay: Targeted surface design for cell culture applications J. Mater. Chem. B 2013 1 522 528 10.1039/C2TB00328G 115. Flaim C.J. Chien S. Bhatia S.N. An extracellular matrix microarray for probing cellular differentiation Nat. Methods 2005 2 119 125 10.1038/nmeth736 15782209 116. Nimrichter L. Gargir A. Gortler M. Altstock R.T. Shtevi A. Weisshaus O. Fire E. Dotan N. Schnaar R.L. Intact cell adhesion to glycan microarrays Glycobiology 2004 14 197 203 10.1093/glycob/cwh022 14638630 117. Hook A.L. Anderson D.G. Langer R. Williams P. Davies M.C. Alexander M.R. High throughput methods applied in biomaterial development and discovery Biomaterials 2010 31 187 198 10.1016/j.biomaterials.2009.09.037 19815273 118. Shadpour H. Sims C.E. Allbritton N.L. Enrichment and expansion of cells using antibody-coated micropallet arrays Cytom. A J. Int. Soc. Anal. Cytol. 2009 75 609 618 10.1002/cyto.a.20741 19504569 119. He J. Liu Y. Xie X. Zhu T. Soules M. DiMeco F. Vescovi A.L. Fan X. Lubman D.M. Identification of cell surface glycoprotein markers for glioblastoma-derived stem-like cells using a lectin microarray and lc-ms/ms approach J. Proteome Res. 2010 9 2565 2572 10.1021/pr100012p 20235609 120. Colpo P. Ruiz A. Ceriotti L. Rossi F. Surface functionalization for protein and cell patterning Whole Cell Sensing Systems I: Reporter Cells and Devices Belkin S. Gu B.M. Springer Berlin/Heidelberg, Germany 2010 109 130 121. Falconnet D. Csucs G. Grandin H.M. Textor M. Surface engineering approaches to micropattern surfaces for cell-based assays Biomaterials 2006 27 3044 3063 10.1016/j.biomaterials.2005.12.024 16458351 122. Oh E.H. Lee S.H. Lee S.H. Ko H.J. Park T.H. Cell-based high-throughput odorant screening system through visualization on a microwell array Biosens. Bioelectron. 2014 53 18 25 10.1016/j.bios.2013.09.039 24103575 123. Moeller H.C. Mian M.K. Shrivastava S. Chung B.G. Khademhosseini A. A microwell array system for stem cell culture Biomaterials 2008 29 752 763 10.1016/j.biomaterials.2007.10.030 18001830 124. Kim H. Cohen R.E. Hammond P.T. Irvine D.J. Live lymphocyte arrays for biosensing Adv. Funct Mater. 2006 16 1313 1323 10.1002/adfm.200500888 125. Barrett D.G. Yousaf M.N. Rapid patterning of cells and cell co-cultures on surfaces with spatial and temporal control through centrifugation Angew. Chem. 2007 46 7437 7439 10.1002/anie.200701841 17705322 126. Yamamura S. Kishi H. Tokimitsu Y. Kondo S. Honda R. Rao S.R. Omori M. Tamiya E. Muraguchi A. Single-cell microarray for analyzing cellular response Anal. Chem. 2005 77 8050 8056 10.1021/ac0515632 16351155 127. Yatsushiro S. Yamamura S. Yamaguchi Y. Shinohara Y. Tamiya E. Horii T. Baba Y. Kataoka M. Rapid and highly sensitive detection of malaria-infected erythrocytes using a cell microarray chip PLoS ONE 2010 5 11 10.1371/journal.pone.0013179 20967248 128. Yamamura S. Yatsushiro S. Yamaguchi Y. Abe K. Shinohara Y. Tamiya E. Baba Y. Kataoka M. Accurate detection of carcinoma cells by use of a cell microarray chip PLoS ONE 2012 7 11 10.1371/journal.pone.0032370 22396762 129. Reymann J. Beil N. Beneke J. Kaletta P.P. Burkert K. Erfle H. Next-generation 9216-microwell cell arrays for high-content screening microscopy BioTechniques 2009 47 877 878 19852772 130. Zawko S.A. Schmidt C.E. Simple benchtop patterning of hydrogel grids for living cell microarrays Lab Chip 2010 10 379 383 10.1039/B917493A 20091011 131. Ankam S. Teo B.K.K. Kukumberg M. Yim E.K.F. High throughput screening to investigate the interaction of stem cells with their extracellular microenvironment Organogenesis 2013 9 128 142 10.4161/org.25425 23899508 132. Moe A.A. Suryana M. Marcy G. Lim S.K. Ankam S. Goh J.Z. Jin J. Teo B.K. Law J.B. Low H.Y. Microarray with micro- and nano-topographies enables identification of the optimal topography for directing the differentiation of primary murine neural progenitor cells Small 2012 8 3050 3061 10.1002/smll.201200490 22807278 133. Hook A.L. Thissen H. Voelcker N.H. Surface manipulation of biomolecules for cell microarray applications Trends Biotechnol. 2006 24 471 477 10.1016/j.tibtech.2006.08.001 16919345 134. Anderson D.G. Levenberg S. Langer R. Nanoliter-scale synthesis of arrayed biomaterials and application to human embryonic stem cells Nat. Biotechnol. 2004 22 863 866 10.1038/nbt981 15195101 135. Hook A.L. Thissen H. Voelcker N.H. Advanced substrate fabrication for cell microarrays Biomacromolecules 2009 10 573 579 10.1021/bm801217n 19159278 136. Lavrentieva A. Majore I. Kasper C. Hass R. Effects of hypoxic culture conditions on umbilical cord-derived human mesenchymal stem cells Cell Commun. Signal. 2010 8 10.1186/1478-811X-8-18 20637101 137. Collet G. El Hafny-Rahbi B. Nadim M. Tejchman A. Klimkiewicz K. Kieda C. Hypoxia-shaped vascular niche for cancer stem cells Contemp. Oncol. 2015 19 A39 A43 10.5114/wo.2014.47130 25691820 138. Eiselleova L. Peterkova I. Neradil J. Slaninova I. Hampl A. Dvorak P. Comparative study of mouse and human feeder cells for human embryonic stem cells Int. J. Dev. Biol. 2008 52 353 363 10.1387/ijdb.082590le 18415935 139. Fu J. Wang Y.K. Yang M.T. Desai R.A. Yu X. Liu Z. Chen C.S. Mechanical regulation of cell function with geometrically modulated elastomeric substrates Nat. Methods 2010 7 733 736 10.1038/nmeth.1487 20676108 140. Gobaa S. Hoehnel S. Roccio M. Negro A. Kobel S. Lutolf M.P. Artificial niche microarrays for probing single stem cell fate in high throughput Nat. Methods 2011 8 949 955 10.1038/nmeth.1732 21983923 141. Kuschel C. Steuer H. Maurer A.N. Kanzok B. Stoop R. Angres B. Cell adhesion profiling using extracellular matrix protein microarrays BioTechniques 2006 40 523 531 10.2144/000112134 16629399 142. Zhou H. Zhao L. Zhang X. In-channel printing-device opening assay for micropatterning multiple cells and gene analysis Anal. Chem. 2015 87 2048 2053 10.1021/ac504823s 25630902 143. Reticker-Flynn N.E. Braga Malta D.F. Winslow M.M. Lamar J.M. Xu M.J. Underhill G.H. Hynes R.O. Jacks T.E. Bhatia S.N. A combinatorial extracellular matrix platform identifies cell-extracellular matrix interactions that correlate with metastasis Nat. Commun 2012 3 1122 10.1038/ncomms2128 23047680 144. Zschenker O. Streichert T. Hehlgans S. Cordes N. Genome-wide gene expression analysis in cancer cells reveals 3d growth to affect ecm and processes associated with cell adhesion but not DNA repair PLoS ONE 2012 7 11 10.1371/journal.pone.0034279 22509286 145. Page H. Flood P. Reynaud E.G. Three-dimensional tissue cultures: Current trends and beyond Cell. Tissue Res. 2013 352 123 131 10.1007/s00441-012-1441-5 22729488 146. Sutherland R.M. Inch W.R. McCredie J.A. Kruuv J. A multi-component radiation survival curve using an in vitro tumour model Int J. Radiat Biol Relat Stud. Phys. Chem Med. 1970 18 491 495 10.1080/09553007014551401 5316564 147. Sutherland R.M. Inch W.R. McCredie J.A. Phytohemagglutinin (pha)-induced transformation of lymphocytes from patients with cancer Cancer 1971 27 574 578 10.1002/1097-0142(197103)27:3<574::AID-CNCR2820270310>3.0.CO;2-7 5279329 148. Lee J.M. Mhawech-Fauceglia P. Lee N. Parsanian L.C. Lin Y.G. Gayther S.A. Lawrenson K. A three-dimensional microenvironment alters protein expression and chemosensitivity of epithelial ovarian cancer cells in vitro Lab. Investig. 2013 93 528 542 23459371 149. Friedrich J. Ebner R. Kunz-Schughart L.A. Experimental anti-tumor therapy in 3-D: Spheroids—old hat or new challenge? Int J. Radiat. Biol. 2007 83 849 871 10.1080/09553000701727531 18058370 150. Chang T.T. Hughes-Fulford M. Monolayer and spheroid culture of human liver hepatocellular carcinoma cell line cells demonstrate distinct global gene expression patterns and functional phenotypes Tissue Eng. A 2009 15 559 567 10.1089/ten.tea.2007.0434 18724832 151. Kim Y. Lasher C.D. Milford L.M. Murali T.M. Rajagopalan P. A comparative study of genome-wide transcriptional profiles of primary hepatocytes in collagen sandwich and monolayer cultures Tissue Eng. Part. C Methods 2010 16 1449 1460 10.1089/ten.tec.2010.0012 20412007 152. Yeh H.Y. Liu B.H. Sieber M. Hsu S.H. Substrate-dependent gene regulation of self-assembled human msc spheroids on chitosan membranes BMC Genom. 2014 15 10 10.1186/1471-2164-15-10 24387160 153. Luca A.C. Mersch S. Deenen R. Schmidt S. Messner I. Schafer K.L. Baldus S.E. Huckenbeck W. Piekorz R.P. Knoefel W.T. Impact of the 3D microenvironment on phenotype, gene expression, and egfr inhibition of colorectal cancer cell lines PLoS ONE 2013 8 11 10.1371/journal.pone.0059689 23555746 154. Schmeichel K.L. Bissell M.J. Modeling tissue-specific signaling and organ function in three dimensions J. Cell Sci. 2003 116 2377 2388 10.1242/jcs.00503 12766184 155. Streuli C. Extracellular matrix remodelling and cellular differentiation Curr. Opin. Cell Biol. 1999 11 634 640 10.1016/S0955-0674(99)00026-5 10508658 156. Weigelt B. Lo A.T. Park C.C. Gray J.W. Bissell M.J. HER2 signaling pathway activation and response of breast cancer cells to HER2-targeting agents is dependent strongly on the 3D microenvironment Breast Cancer Res. Treat. 2010 122 35 43 10.1007/s10549-009-0502-2 19701706 157. Edmondson R. Broglie J.J. Adcock A.F. Yang L. Three-dimensional cell culture systems and their applications in drug discovery and cell-based biosensors Assay Drug Dev. Technol. 2014 12 207 218 10.1089/adt.2014.573 24831787 158. Andersen T. Auk-Emblem P. Dornish M. 3D cell culture in alginate hydrogels Microarrays 2015 4 133 10.3390/microarrays4020133 159. Lin R.Z. Chang H.Y. Recent advances in three-dimensional multicellular spheroid culture for biomedical research Biotechnol. J. 2008 3 1172 1184 10.1002/biot.200700228 18566957 160. Carletti E. Motta A. Migliaresi C. Scaffolds for tissue engineering and 3D cell culture Methods Mol. Biol. 2011 695 17 39 21042963 161. Gidrol X. Fouque B. Ghenim L. Haguet V. Picollet-D’hahan N. Schaack B. 2D and 3D cell microarrays in pharmacology Curr. Opin. Pharmacol. 2009 9 664 668 10.1016/j.coph.2009.05.002 19520607 162. Ock J. Li W. Fabrication of a three-dimensional tissue model microarray using laser foaming of a gas-impregnated biodegradable polymer Biofabrication 2014 6 1 12 10.1088/1758-5082/6/2/024110 163. Karp J.M. Yeh J. Eng G. Fukuda J. Blumling J. Suh K.Y. Cheng J. Mahdavi A. Borenstein J. Langer R. Controlling size, shape and homogeneity of embryoid bodies using poly(ethylene glycol) microwells Lab Chip 2007 7 786 794 10.1039/b705085m 17538722 164. Wang W.J. Itaka K. Ohba S. Nishiyama N. Chung U.I. Yamasaki Y. Kataoka K. 3D spheroid culture system on micropatterned substrates for improved differentiation efficiency of multipotent mesenchymal stem cells Biomaterials 2009 30 2705 2715 10.1016/j.biomaterials.2009.01.030 19215979 165. Ong S.M. Zhang C. Toh Y.C. Kim S.H. Foo H.L. Tan C.H. van Noort D. Park S. Yu H. A gel-free 3D microfluidic cell culture system Biomaterials 2008 29 3237 3244 10.1016/j.biomaterials.2008.04.022 18455231 166. Fernandes T.G. Kwon S.J. Bale S.S. Lee M.Y. Diogo M.M. Clark D.S. Cabral J.M. Dordick J.S. Three-dimensional cell culture microarray for high-throughput studies of stem cell fate Biotechnol. Bioeng. 2010 106 106 118 10.1002/bit.22661 20069558 167. Meli L. Barbosa H.S. Hickey A.M. Gasimli L. Nierode G. Diogo M.M. Linhardt R.J. Cabral J.M. Dordick J.S. Three dimensional cellular microarray platform for human neural stem cell differentiation and toxicology Stem Cell Res. 2014 13 36 47 10.1016/j.scr.2014.04.004 24816401 168. Meli L. Jordan E.T. Clark D.S. Linhardt R.J. Dordick J.S. Influence of a three-dimensional, microarray environment on human cell culture in drug screening systems Biomaterials 2012 33 9087 9096 10.1016/j.biomaterials.2012.08.065 22998815 169. Patel R.G. Purwada A. Cerchietti L. Inghirami G. Melnick A. Gaharwar A.K. Singh A. Microscale bioadhesive hydrogel arrays for cell engineering applications Cell. Mol. Bioeng. 2014 7 394 408 10.1007/s12195-014-0353-8 25328548 170. Ranga A. Gobaa S. Okawa Y. Mosiewicz K. Negro A. Lutolf M.P. 3D niche microarrays for systems-level analyses of cell fate Nat. Commun. 2014 5 4324 10.1038/ncomms5324 25027775 171. Akselrod G.M. Timp W. Mirsaidov U. Zhao Q. Li C. Timp R. Timp K. Matsudaira P. Timp G. Laser-guided assembly of heterotypic three-dimensional living cell microarrays Biophys J. 2006 91 3465 3473 10.1529/biophysj.106.084079 16891375 172. Koh W.G. Itle L.J. Pishko M.V. Molding of hydrogel microstructures to create multiphenotype cell microarrays Anal. Chem. 2003 75 5783 5789 10.1021/ac034773s 14588018 173. Ozawa F. Ino K. Arai T. Ramon-Azcon J. Takahashi Y. Shiku H. Matsue T. Alginate gel microwell arrays using electrodeposition for three-dimensional cell culture Lab Chip 2013 13 3128 3135 10.1039/c3lc50455g 23764965 174. Charwat V. Schütze K. Holnthoner W. Lavrentieva A. Gangnus R. Hofbauer P. Hoffmann C. Angres B. Kasper C. Potential and limitations of microscopy and raman spectroscopy for live-cell analysis of 3D cell cultures J. Biotechnol. 2015 205 70 81 10.1016/j.jbiotec.2015.02.007 25687101 175. Cimetta E. Figallo E. Cannizzaro C. Elvassore N. Vunjak-Novakovic G. Micro-bioreactor arrays for controlling cellular environments: Design principles for human embryonic stem cell applications Methods 2009 47 81 89 10.1016/j.ymeth.2008.10.015 18952171 176. Choi J.R. Song H. Sung J.H. Kim D. Kim K. Microfluidic assay-based optical measurement techniques for cell analysis: A review of recent progress Biosens. Bioelectron. 2016 77 227 236 10.1016/j.bios.2015.07.068 26409023 177. Ma S. Loufakis D.N. Cao Z. Chang Y.W. Achenie L.E.K. Lu C. Diffusion-based microfluidic PCR for “one-pot” analysis of cells Lab Chip 2014 14 2905 2909 10.1039/C4LC00498A 24921711 178. Li R. Lv X. Zhang X. Saeed O. Deng Y. Microfluidics for cell-cell interactions: A review Front. Chem. Sci. Eng. 2016 10 90 98 10.1007/s11705-015-1550-2 179. Sia S.K. Whitesides G.M. Microfluidic devices fabricated in poly(dimethylsiloxane) for biological studies Electrophoresis 2003 24 3563 3576 10.1002/elps.200305584 14613181 180. Businaro L. De Ninno A. Schiavoni G. Lucarini V. Ciasca G. Gerardino A. Belardelli F. Gabriele L. Mattei F. Cross talk between cancer and immune cells: Exploring complex dynamics in a microfluidic environment Lab Chip 2013 13 229 239 10.1039/C2LC40887B 23108434 181. Ramadan Q. Jafarpoorchekab H. Huang C.B. Silacci P. Carrara S. Koklu G. Ghaye J. Ramsden J. Ruffert C. Vergeres G. Nutrichip: Nutrition analysis meets microfluidics Lab Chip 2013 13 196 203 10.1039/C2LC40845G 23184124 182. U.S. F.D.A. Innovation or stagnation: Challenge and Opportunity on the Critical Path to New Medical Products Challenges and Opportunities Report—March 2004 FDA Silver Spring, MD, USA 2004 183. Bhise N.S. Ribas J. Manoharan V. Zhang Y.S. Polini A. Massa S. Dokmeci M.R. Khademhosseini A. Organ-on-a-chip platforms for studying drug delivery systems J. Control. Release 2014 190 82 93 10.1016/j.jconrel.2014.05.004 24818770 184. Kim H.J. Huh D. Hamilton G. Ingber D.E. Human gut-on-a-chip inhabited by microbial flora that experiences intestinal peristalsis-like motions and flow Lab Chip 2012 12 2165 2174 10.1039/c2lc40074j 22434367 185. Wilmer M.J. Ng C.P. Lanz H.L. Vulto P. Suter-Dick L. Masereeuw R. Kidney-on-a-chip technology for drug-incuced nephrotoxiciy screening Trends Biotechnol. 2016 34 156 170 10.1016/j.tibtech.2015.11.001 26708346 186. Doryab A. Amoabediny G. Salehi-Najafabadi A. Advances in pulmonary therapy and drug development: Lung tissue engineering to lung-on-a-chip Biotechnol. Adv. 2016 in press 10.1016/j.biotechadv.2016.02.006 26875777 187. Wang Z.J. Samanipour R. Koo K.I. Kim K. Organ-on-a-chip platforms for drug delivery and cell characterization: A review Sens. Mater. 2015 27 487 506 10.18494/SAM.2015.1086 188. Ghaemmaghami A.M. Hancock M.J. Harrington H. Kaji H. Khademhosseini A. Biomimetic tissues on a chip for drug discovery Drug Discov. Today 2012 17 173 181 10.1016/j.drudis.2011.10.029 22094245 189. Maschmeyer I. Lorenz A.K. Schimek K. Hasenberg T. Ramme A.P. Hubner J. Lindner M. Drewell C. Bauer S. Thomas A. A four-organ-chip for interconnected long-term co-culture of human intestine, liver, skin and kidney equivalents Lab Chip 2015 15 2688 2699 10.1039/C5LC00392J 25996126 190. Lee J.B. Sung J.H. Organ-on-a-chip technology and microfluidic whole-body models for pharmacokinetic drug toxicity screening Biotechnol. J. 2013 8 1258 1266 10.1002/biot.201300086 24038956

PMC005xxxxxx/PMC5003488.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020012microarrays-05-00012ReviewAdvantages of Array-Based Technologies for Pre-Emptive Pharmacogenomics Testing Shahandeh Al 1Johnstone Daniel M. 2Atkins Joshua R. 1Sontag Jean-Marie 1Heidari Moones 1Daneshi Nilofar 1Freeman-Acquah Elvis 3Milward Elizabeth A. 1*Alekseyev Yuriy Academic EditorLiu Gang Academic Editor1 School of Biomedical Sciences and Pharmacy, The University of Newcastle, Callaghan 2308, Australia; Ali.Shahandeh@uon.edu.au (A.S.); Joshua.Atkins@uon.edu.au (J.R.A.); Jean-Marie.Sontag@newcastle.edu.au (J.-M.S.); Moones.Heidari@uon.edu.au (M.H.); Nilofar.Daneshi@uon.edu.au (N.D.)2 The Bosch Institute and Discipline of Physiology, University of Sydney, Sydney 2006, Australia; Daniel.Johnstone@sydney.edu.au3 Institute of Genetics and Cytology, School of Life Sciences, Northeast Normal University, Changchun 130024, China; davfreeman82@yahoo.com* Correspondence: Liz.Milward@newcastle.edu.au; Tel.: +61-249-215-16728 5 2016 6 2016 5 2 1229 2 2016 17 5 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).As recognised by the National Institutes of Health (NIH) Precision Medicine Initiative (PMI), microarray technology currently provides a rapid, inexpensive means of identifying large numbers of known genomic variants or gene transcripts in experimental and clinical settings. However new generation sequencing techniques are now being introduced in many clinical genetic contexts, particularly where novel mutations are involved. While these methods can be valuable for screening a restricted set of genes for known or novel mutations, implementation of whole genome sequencing in clinical practice continues to present challenges. Even very accurate high-throughput methods with small error rates can generate large numbers of false negative or false positive errors due to the high numbers of simultaneous readings. Additional validation is likely to be required for safe use of any such methods in clinical settings. Custom-designed arrays can offer advantages for screening for common, known mutations and, in this context, may currently be better suited for accredited, quality-controlled clinical genetic screening services, as illustrated by their successful application in several large-scale pre-emptive pharmacogenomics programs now underway. Excessive, inappropriate use of next-generation sequencing may waste scarce research funds and other resources. Microarrays presently remain the technology of choice in applications that require fast, cost-effective genome-wide screening of variants of known importance, particularly for large sample sizes. This commentary considers some of the applications where microarrays continue to offer advantages over next-generation sequencing technologies. microarraynext-generation sequencingpharmacogenomicspersonalized healthcare ==== Body 1. Introduction Many reviews cover the advantages of emerging genome-scale sequencing technologies in diverse contexts and these will not be revisited here. Yet, in the rush to adopt these promising new technologies, researchers and funding bodies sometimes fail to recognize that, in some contexts, the use of these platforms is unjustifiable and that high-density genotyping arrays continue to be a far more appropriate choice. Fortunately, since these sequencing technologies are frequently still very costly, there is increasing awareness that the newer approaches are not always better. For example, the practical advantages of high-density genotyping arrays over genome-scale sequencing in studies involving large numbers of samples have been acknowledged in the lead-up to implementation of President Obama’s PMI. This is summarized in the September 2015 report of the PMI Working Group to the Advisory Committee to the Director of the US NIH [1]. The Working Group noted that, in most circumstances, issues with the cost, imperfect results and expectation of technology obsolescence made genome-scale sequencing approaches presently inappropriate for large numbers of individuals. The Working Group recommends ongoing monitoring to assess when the balance of the scientific value over the costs and capabilities of such methods reaches a “tipping point”. Meanwhile, as recognized by the Committee, reasonable utility can be achieved at an affordable cost by high-density genome-wide arrays testing common and rare gene variants. As addressed elsewhere in the Special Issue on “Microarrays in the Era of Next Generation Sequencing”, microarrays continue to be used in diverse applications. Examples include detection of chromosomal abnormalities in cytogenetics using array-based comparative genome hybridization, in providing rapid turnaround for prenatal investigations using limited amounts of DNA and in replacing other technologies such as fluorescence in situ hybridization (FISH) for some oncology applications. In the present article we will focus primarily on the area of pharmacogenomics, where arrays are now being widely used both in basic research and in research into the effective translation of pharmacogenomics into clinical practice. While microarrays do not enable new gene variants to be discovered and are, therefore, generally not well-suited to clinical genetics applications seeking to identify novel disease-associated variants, high throughput genome-wide array technology can still provide the capacity to simultaneously assess essentially all single nucleotide polymorphisms (SNPs) of known functional importance in the human genome [2,3,4,5,6]. This level of genomic coverage is sufficient for many current applications in medicine and research, including most pharmacogenomics applications, as will be discussed in more detail below. The rapid output, affordability, and availability of microarray technology, along with its high accuracy and established and validated pipelines for data analysis and variant calling, make it the logical choice for such applications [7]. This is particularly true for large sample sizes, for example in large genome wide association studies (GWAS), where microarrays have been, and in most instances continue to be, the only economically viable option. Using microarrays, scientists from around the globe can contribute data of various kinds (including genomic, epigenetic, and transcriptomic data) to massive consortium project, even when only able to afford to study a small number of samples. Although sample size restricts the capacity to detect association in small studies, analyses of collective pooled sample sets can be extremely powerful. Hence, while next-generation sequencing (NGS) technologies are essential for discovery-driven research focused on the identification of novel sequences, it may often be unnecessary and even potentially a fiscally irresponsible misuse of research funding [7] to use such methods for profiling common variants (e.g., SNPs) across a large number of patients or in a variety of other applications where detecting novel sequences is not the primary goal. Although less relevant in the context of this article, array technology is not only still useful for genomic studies but also continues to offer many advantages for various other kinds of high-throughput studies, including transcriptomics, where it remains the platform of choice for many studies. For example, in 2014, RNA-seq data was uploaded into the Gene Expression Omnibus (GEO) database for around 9000 samples whereas microarray data was uploaded for over 54,000 samples [8]. Microarray-based clinical tests provide a powerful tool for simultaneous measurement of the relative expression levels of a large number of well-established clinically relevant genes in the context of disease or drug responses. There is a wide range of applications for gene expression microarrays in providing RNA profiles associated with different disease states for various purposes, including monitoring pharmacological responses in clinical trial participants and identifying suitable drug treatments for individual patients, as reviewed elsewhere [9,10,11]. In view of such considerations, the relatively high costs of sequencing often appear hard to justify in a climate where increasing numbers of researchers are losing funding. Even ignoring the often higher cost of consumables and equipment for NGS as compared to microarray, the greatest cost often lies in the labour. The cost of next-generation whole-genome and transcriptome sequencing is dropping rapidly, and may one day match the cost of microarray-based methods. However the frequent claims of the $1000 genome or even of costs comparable to those of arrays usually do not adequately take into account the cost of time and human resources in sample preparation, sequence alignment, and filtering through huge volumes of data to catalogue SNPs or other information of interest, let alone the infrastructure required for sequencing, data processing, and storage [7]. Microarrays, therefore, continue to provide a highly cost-effective choice in contexts involving samples from relatively large groups of individuals, such as pharmacogenomics. This review will primarily consider the enduring value of microarrays in pharmacogenomics. We will, first, very briefly review the current status of pharmacogenomics in clinical practice before going on to consider criteria that a genotyping platform will need to meet to be relevant to clinical pharmacogenomics in the future. We will consider how microarrays measure up to these criteria and briefly discuss some examples of successful applications of microarray in research into the effective translation of pharmacogenomics into clinical practice. 2. Pharmacogenomics in Practice As defined by the Food and Drug Administration (FDA), pharmacogenomics studies variations of DNA (genomic) and RNA (transcriptomic) characteristics as related to drug response, providing information which can be used to inform appropriate drug selection or dosage regimens for individual patients [12]. This relies on the identification of SNPs and other variants in genes known to be important in pharmacokinetics or pharmacodynamics. Considerable ongoing research focuses on identifying and profiling these variants; however, currently only a few gene variants are considered to have a firm evidence base for clinical actionability. For most drugs, information on clinically-actionable gene variants (for which either a change of medication or a change of dose are recommended) can be obtained by screening only a small portion of the genome. The guidelines of the Clinical Pharmacogenetics Implementation Consortium (CPIC), supported by the US NIH and available through the Pharmacogenomics Knowledge Base (PharmGKB) [13], list only 17 genes with “high” evidence (Level 1A or 1B) of a drug-modifying effect (see Table 1), with “moderate” evidence (Level 2A or 2B) for an additional 40 genes. While further research is likely to reveal a number of other variants that modify the pharmacokinetic or pharmacodynamics profiles of new or existing drugs, the degree of screening required is, therefore, unlikely to extend beyond the capabilities of microarray technology for some time into the future. Several microarray-based tests that simultaneously examine variations in multiple genes are approved by the FDA and have entered practice. These include AmpliChip CYP450 from Roche and MammaPrint from Agendia. Although whole genome sequencing and whole exome sequencing of potential pharmacogenomic gene variants have been reported previously [33,34], as far as we are aware the first and, to date, the only FDA-cleared NGS platform for in vitro diagnostic testing is a single gene test only, specifically a cystic fibrosis mutation detection test utilizing the Illumina MiSeqDx System [35]. For these and other reasons described elsewhere in this article, microarrays are likely to continue to be relevant and beneficial for clinical practice for some time into the future. 3. Minimum Criteria for a Clinically Useful Pharmacogenomics Platform In practical terms, irrespective of the technology used, the ideal pharmacogenomics platform should meet the following minimum criteria [36,37]. 3.1. Analytical Validity Ideally, a pharmacogenomics test should have high analytical specificity and sensitivity, with appropriate laboratory quality assurance and assay robustness. The data generated should be highly accurate with minimal errors in calling of gene variants. However, accuracy issues continue to restrict the usefulness of NGS. Even the most advanced sequencing platforms still have a base call error rate that, although usually proportionately small compared to many other technologies, is amplified by the large number of reads performed in an NGS experiment. This can make it difficult to distinguish polymorphisms from sequencing errors [38,39,40]. Various kinds of bias affect the analytic validity of NGS data [38,41]. Systematic bias involves non-random errors arising because of inaccuracies inherent in the platform and associated protocols, including errors deriving from the methods used to generate the original sequencing library [38]. Systematic errors can also reflect coverage bias, which may occur in regions where the genome sequence, chemistry, or conformation affects data output. This form of bias can in part be platform-dependent but can also occur across platforms and the error involved can be substantial—for example, a 2012 study by Quail and colleagues [42] of three platforms, Ion Torrent Personal Genome Machine, Pacific Biosciences PacBio RS, and Illumina MiSeq, found that output from sequencing extremely AT-rich genomes contained high levels of bias and errors with no coverage of almost 30% of the genomes investigated. Another important component of systematic bias—pertinent to laboratory quality assurance and assay robustness—isbatch effects relating to external factors such as reagent variability [38,41]. Sequencing accuracy for leading longer established technologies such as Illumina is often over 99% [43,44]. For single nucleotide variants differing from the reference genotype, the error rates for whole-genome and whole-exome sequencing of Illumina HiSeq or Complete Genomics have been estimated to be up to 0.1% or 0.6%, respectively, using replicate high-coverage sequencing of human blood and saliva DNA samples [39] and advances such as the HiSeq X Ten model and the Complete Genomics Long Fragment Reads technology [45] are achieving considerably better rates. These are relatively well-established technologies which have been in use and evolving for some time, facilitating development of expertise and optimisation of protocols. However, in general, these technologies tend to be relatively costly compared to some of the other platforms, although these are sometimes less accurate with higher error rates [38,43,46]. Sequencing accuracy of the PacBio platform has been reported to be in the range of 80%–90% [43,47], with a study comparing three important platforms—Ion Torrent Personal Genome Machine, Pacific Biosciences PacBio RS and Illumina MiSeq (reviewed in more detail in the study in question)—on a set of four microbial genomes observing error rates of below 0.4% for the Illumina platform, 1.78% for Ion Torrent and 13% for PacBio sequencing [42]. The number of error-free reads, without a single mismatch or insertion and deletion (indel), was 76.45%, 15.92%, and 0% for MiSeq, Ion Torrent, and PacBio, respectively. The PacBio errors were evenly distributed, whereas MiSeq produced more errors after long (>20-base) homopolymer tracts or for GC-rich motifs. The affordable and widely used Ion Torrent platform produced erroneous base numbers for homopolymers >8 bases long and failed to generate reads entirely for long (>14-base) homopolymer tracts, along with strand-specific errors that were not associated with any obvious motif. The long reads and low error rates of early NGS platforms such as Roche 454 sequencers made error correction relatively unimportant [38]. Most error-correction programs have primarily addressed substitution errors, since these have been an issue for the widely used Illumina machines; however, short read platforms, such as Ion Torrent, are more prone to other sorts of errors, such as indels [48,49,50]. As approaches to error correction are refined for each emerging technology, the accuracy of the output is likely to improve considerably; one of the potentially most exciting new developments, the MinION, has a raw sequencing error rate of about 12% which can be improved to 0%–3% with hybrid or de novo error correction [51]. Such errors are often relatively unimportant for discovery-based applications in research settings or clinical investigations to identify disease-related mutations in families, where candidate variants can be validated using a range of other approaches. However, such errors are more of a problem in clinical pharmacogenomic contexts requiring fast and reliable decisions about medications, where microarrays and, in particular, validated custom-designed arrays for pharmacogenomics and other applications can offer more reliable options [10,34,52]. 3.2. Clinical Validity and Utility While, as discussed above, a test must be able to evaluate the measure of interest accurately (analytical validity), a test only has clinical validity if what is being measured correlates closely with some clinical outcome of interest. In the context of pharmacogenomics, clinical validity translates to the ability of a genomic test to detect or predict the response to a drug correctly and with high specificity and sensitivity. Even though a test may have analytical and clinical validity, it still need not necessarily be clinically useful if, for example, the information provided by the test does not serve any useful purpose for the physician, the patient or other relevant stakeholders. The clinical utility of a genomic test, in the broadest sense, can be considered to refer to the usefulness of the information it provides in enabling clinicians, patients or other stakeholders to make appropriate health-related decisions; a comprehensive list of factors influencing clinical utility is provided by the Centres for Disease Control and Prevention (CDC) [53]. Relevant considerations include whether appropriate equipment, expertise and validated educational materials are available to allow effective use of test results in healthcare decision making. In the present context the concept of clinical utility is used primarily with regard to the requirement that a pharmacogenomics test should provide information of value to decision making by health professionals or patients. For example, the test may help decide on applicable interventional approaches. The availability and accessibility of a test also affect the extent of its usefulness. Ideally, pharmacogenomic test results would be accessible by professionals working at point-of-care (e.g., doctors or pharmacists), using a simple database query that would link the prescribed drug with a patient’s genotype to identify any recommended modifications to the treatment or dose. In our experience, and as evidenced by large-scale pre-emptive clinical pharmacogenomics testing programs described in more detail below [54], pharmacogenomics test results can be relatively easily and rapidly extracted from DNA array data using potentially automatable procedures, and can be interpreted by personnel with relatively little training and experience. In contrast, one of the main practical barriers to implementation of NGS in clinical practice is the relative scarcity of personnel capable of handling and interpreting NGS data. Expertise in recognizing errors from true calls (which will also reflect analytical validity) is critical in ensuring the correct prescription is received by patients, and avoiding possible claims of negligence. The interpretation of NGS data presently involves extensive and time consuming analyses that require expert human judgement. While similar considerations also apply for microarray data, at present there are relatively well-established automated algorithms and pipelines for array data processing [55], whereas NGS data analysis still more commonly requires considerable human input and judgement [46,56]. 4. Additional Technical Issues Speed is another concern that is often raised. The entire turnaround time, from DNA collection to reporting, should be no longer and, if possible, less than that of standard pathology tests. However, while the speed of the new sequencing technologies continues to increase, the requirements of error correction and other analytical factors have not kept pace and continue to cause bottlenecks in this regard. For example, Yang and colleagues (2013) note the need to improve the run-time of error correction algorithms and validation procedures involving hybrid datasets generated by multiple platforms [38]. However, while fast turnaround will be important for future applications of epigenomics, transcriptomics, or proteomics in diagnostic contexts, where profiles are dynamic and there is a need for rapid assessment of a person’s current status, the genome effectively remains unchanged over time and can therefore be determined in advance (“pre-emptively”) before health problems arise, making turnaround time essentially irrelevant. So, in the context of pharmacogenomics, speed may be a relatively unimportant criterion for pre-emptive testing, although it remains an issue if rapid clinical decision making is required for a patient who has not been previously genotyped. Infrastructure also requires consideration. Ideally, it should ultimately be possible to perform genotyping and data interrogation in close proximity to the point-of-care e.g., within a hospital laboratory or in community settings such as a pharmacy or GP office. Equipment for running the assay should, therefore, be user-friendly and self-contained, while data analysis and reporting should be compatible with the computing power provided by a standard desktop or laptop computer. Technologies such as Oxford MinION, while still evolving, hold considerable promise in this context [57]. Data storage and linkage are among the most important limiting factors in many clinical contexts. For example, the relevant data generated by a platform should be sufficiently compact to link with electronic health records. Given that current microarray platforms can screen approximately 500,000–1 million or more SNPs or 1 million probes for detecting copy number variation for a few hundred dollars, and that the resulting SNP data can be compressed to a few megabytes, this technology would appear the most appropriate fit based on the above criteria [58]. The need for infrastructure that can handle “big data” is another potential limitation of NGS that makes it currently less feasible for clinical applications. In terms of data storage, the raw compressed fastq files from a single whole genome sequencing run at 30× sequencing depth amount to ~100 gigabytes; on top of this, the aligned, processed files require storage of around 1–1.5 terabytes per patient [59]. While advances in cloud storage and data transmission technologies will help overcome problems associated with storing these massive volumes of data, at least with regard to pharmacogenomics, it is debatable whether this is an economically intelligent use of resources given the small number of known variants that modify drug responses. 5. Ethical, Legal, and Social Issues (ELSI) Several other issues also need to be taken into account in addition to the above considerations, most notably ELSI. In this regard there are a range of potential concerns, including privacy and unanticipated results. Irrespective of the technology, some of these concerns can be addressed by appropriate consent procedures such as “traffic light” systems similar to those being used in related contexts, such as consent for the release of an individual’s genomic information for research purposes [60]. Such systems can be used to enable people to give different levels of consent for different categories of genomic information. For example, with respect to privacy, a person might elect not to provide certain kinds of genomic information for commercial purposes but may be happy to provide this information for other research purposes. With respect to unanticipated results, a person might elect not to be informed about a mutation predisposing to an untreatable condition. However, it is worth noting that, irrespective of the technology used (microarray or NGS), these issues are far less likely to present concerns for pharmacogenomic tests than for genomic diagnostic testing. The presence of a pharmacogenomically actionable mutation is generally innocuous unless a person takes a medicine affected by the mutation, in which case the information becomes potentially advantageous, reducing the likelihood of an adverse reaction or therapeutic failure. Another important ethical concern surrounding pharmacogenomics relates to justice and equity. As pharmacogenomics and precision medicine begin to assume pivotal roles in healthcare, there will be increasing need to ensure fair, even distribution of benefits to prevent further widening of the gaps that exist between individuals of different socioeconomic status and, in particular, people from developing or resource-limited countries, who may already be disadvantaged with respect to healthcare. Such countries are also least able to sustain any inefficiencies in healthcare systems arising as a result of drugs that were originally developed for use in other populations not working appropriately because of genomic differences. Countries from Europe, Africa, Asia, and the Pacific have now initiated the implementation of genomic approaches in health care. In view of the foregoing discussion and taking into consideration the issues of inadequate funds, lack of access to technology, and scarcity of well-trained health experts in developing or resource-limited countries, currently the only viable approach to ensure that these countries are given equitable opportunities to benefit from these new initiatives would appear to be the utilization of microarray techniques, rather than next-generation approaches. Initiatives implementing large-scale pharmacogenomics are now starting to appear worldwide. For example, the European Ubiquitous Pharmacogenomics network [61] project provides information on the prevalence and effects of pharmacogenomically relevant gene variants in Europe with particular focus on developing countries, in order to generate locally-relevant drug dose recommendations [62]. Pharmacogenomics networks are also being set up in Asia, for example the Asian Network for Pharmacogenomics [63]. The Human Heredity and Health in African (H3Africa) initiative, backed by the US NIH, the UK Wellcome Trust, and the African Society of Human Genetics, aims to develop the capacity of African scientists to apply genomic and epidemiological approaches in locally-relevant clinical contexts [62,64]. 6. Using Microarrays for Pre-Emptive Pharmacogenomics Testing One example of successful array-based pharmacogenomics in practice is the Pharmacogenomic Resource for Enhanced Decisions in Care and Treatment (PREDICT) program at Vanderbilt University Medical Centre [54]. The PREDICT program uses panel-based genotyping to identify specific SNPs that are known to have drug-response associations, allowing tailored clinical decision support to be provided for each participant. In an initial study of almost 10,000 participants, which focused on only five well-established drug-gene interactions (clopidogrel—CYP2C19; simvastatin—SLCOB1; warfarin—CYP2C9 and VKORC1; thiopurines—TPMT; tacrolimus—CYP3A5), one or more actionable variants were identified in 91% of genotyped participants [54]. The clinical utility of this approach is, therefore, likely to be considerably greater than single gene tests. In addition, the pre-emptive, panel-based genotyping approach used in this study enabled substantial reduction in the testing burden compared to single gene assays and facilitated the provision of results at the point of care. This study testifies that custom-designed arrays are appropriate for accurate identification of common SNPs in accredited, quality-controlled pharmacogenomic screening services. As was also noted in the introduction, a recent review has highlighted the success of array-based pre-emptive pharmacogenomics testing in several other US medical centers, namely St Jude’s Children’s Research Hospital, University of Florida and Shands Hospital, the Mayo Clinic, and Mount Sinai Medical Centre [65]. While the genotyping platform varied among the centres, as did the number of genes assayed (ranging from 34 to 230), each program identified a high prevalence of actionable variants. In fact, when considering only 12 pharmacogenes, it is estimated that over 97% of the population of the US have at least one actionable high-risk diplotype [65]. The experience of these trailblazing centres in instituting pre-emptive pharmacogenomics testing has already highlighted challenges and solutions to implementation, paving the way for smooth deployment in other locations. 7. Conclusions Advances in sequencing technologies have revolutionised genomic discovery in the lab, while concomitant reductions in cost will increase the feasibility of employing such technologies in routine clinical or pharmacy practice. Yet the availability of new technologies should not dictate that its predecessors be discarded for every application. In the context of widespread pharmacogenomics profiling of large numbers of individuals, existing microarray technology offers considerable advantages over sequencing with respect to cost of infrastructure, ease of analysis, interpretation, and logistics of data storage and interrogation. In contrast, NGS offers no obvious advantages over array-based methods for screening large numbers of common variants. While the urge to embrace an exciting new technology as a panacea can sometimes seem irresistible, we hope that common sense will prevail in judging the utility of genomics technologies in a context-dependent manner. Author Contributions All authors contributed to literature searching and to writing the review. Conflicts of Interest The authors declare no conflict of interest. microarrays-05-00012-t001_Table 1Table 1 Genes considered to have high levels of evidence for effects on drug responses according to PharmGKB and CPIC Gene and Drug Guidelines. Genes Drugs CPIC PharmGKB CPIC Publications HLA-B Abacavir; allopurinol; phenytoin; carbamazepine A 1A [14,15,16,17] CYP2C19 Amitriptyline; clopidogrel; imipramine *; trimipramine *; citalopram; escitalopram A 1A [18,19,20] CYP2D6 Amitriptyline; codeine; desipramine; doxepin; fluvoxamine; imipramine; nortriptyline; paroxetine; trimipramine A 1A [18,20,21] UGT1A1 Atazanavir A 1A [22] TPMT Azathioprine; mercaptopurine; thioguanine A 1A [23,24] DPYD Capecitabine;fluorouracil; tegafur A 1A [25] CFTR Ivacaftor A 1A [26] CYP2C9 Warfarin; phenytoin ** A 1A [16,27] G6PD Rasburicase A 1A [28] SLCO1B1 Simvastatin A 1A [29,30] CYP3A5 Tacrolimus A 1A [31] VKORC1 Warfarin A 1A [27] IFNL3 Peginterferon alfa-2a; peginterferon alfa-2b; ribavirin; telaprevir A 1A [32] CYP2B6 Efavirenz B 1B CYP4F2 Warfarin B 1B ANKK1 Bupropion D 1B GRIK4 Citalopram D 1B * PharmGKB level of evidence 2A; ** PharmGKB level of evidence 1B. ==== Refs References 1. Precision Medicine Initiative (PMI) Working Group Report to the Advisory Committee to the Director, NIH The Precision Medicine Initiative Cohort Program-Building a Research Foundation for 21st Century Medicine NIH Bethesda, MD, USA 2016 2. Wang D.G. Fan J.B. Siao C.J. Berno A. Young P. Sapolsky R. Ghandour G. Perkins N. Winchester E. Spencer J. Large-scale identification, mapping, and genotyping of single-nucleotide polymorphisms in the human genome Science 1998 280 1077 1082 10.1126/science.280.5366.1077 9582121 3. Cutler D.J. Zwick M.E. Carrasquillo M.M. Yohn C.T. Tobin K.P. Kashuk C. Mathews D.J. Shah N.A. Eichler E.E. Warrington J.A. High-throughput variation detection and genotyping using microarrays Genome Res. 2001 11 1913 1925 11691856 4. Sund K.L. Zimmerman S.L. Thomas C. Mitchell A.L. Prada C.E. Grote L. Bao L. Martin L.J. Smolarek T.A. Regions of homozygosity identified by SNP microarray analysis aid in the diagnosis of autosomal recessive disease and incidentally detect parental blood relationships Genet. Med. 2013 15 70 78 10.1038/gim.2012.94 22858719 5. Kumar P. Al-Shafai M. Al Muftah W.A. Chalhoub N. Elsaid M.F. Aleem A.A. Suhre K. Evaluation of SNP calling using single and multiple-sample calling algorithms by validation against array base genotyping and mendelian inheritance BMC Res. Notes 2014 7 12 10.1186/1756-0500-7-747 24398031 6. Perez-Enciso M. Rincon J.C. Legarra A. Sequence- vs. Chip-assisted genomic selection: Accurate biological information is advised Genet. Sel. Evol. 2015 47 43 10.1186/s12711-015-0117-5 25956961 7. Sboner A. Mu X.J. Greenbaum D. Auerbach R.K. Gerstein M.B. The real cost of sequencing: Higher than you think! Genome Biol. 2011 12 125 10.1186/gb-2011-12-8-125 21867570 8. Su Z. Fang H. Hong H. Shi L. Zhang W. Zhang W. Zhang Y. Dong Z. Lancashire L.J. Bessarabova M. An investigation of biomarkers derived from legacy microarray data for their utility in the RNA-seq era Genome Biol. 2014 15 523 10.1186/s13059-014-0523-y 25633159 9. Bates S. The role of gene expression profiling in drug discovery Curr. Opin. Pharmacol. 2011 11 549 556 10.1016/j.coph.2011.06.009 21752712 10. Milward E.A. Daneshi N. Johnstone D.M. Emerging real-time technologies in molecular medicine and the evolution of integrated “pharmacomics” approaches to personalized medicine and drug discovery Pharmacol. Ther. 2012 136 295 304 10.1016/j.pharmthera.2012.08.008 22951096 11. Anderson D.C. Kodukula K. Biomarkers in pharmacology and drug discovery Biochem. Pharmacol. 2014 87 172 188 10.1016/j.bcp.2013.08.026 24001556 12. Savers S. Guidance for industry E15 definitions for genomic biomarkers, pharmacogenomics, pharmacogenetics, genomic data, and sample coding categories Biotechnol. Law Rep. 2008 27 359 363 13. PharmGKB Pharmacogenomics Knowledge Implementation Available online: https://www.pharmgkb.org/ (accessed on 23 May 2016) 14. Martin M.A. Klein T.E. Dong B.J. Pirmohamed M. Haas D.W. Kroetz D.L. Clinical pharmacogenetics implementation consortium guidelines for HLA-B genotype and abacavir dosing Clin. Pharmacol. Ther. 2012 91 734 738 10.1038/clpt.2011.355 22378157 15. Hershfield M.S. Callaghan J.T. Tassaneeyakul W. Mushiroda T. Thorn C.F. Klein T.E. Lee M.T. Clinical pharmacogenetics implementation consortium guidelines for human leukocyte antigen-B genotype and allopurinol dosing Clin. Pharmacol. Ther. 2013 93 153 158 10.1038/clpt.2012.209 23232549 16. Caudle K.E. Rettie A.E. Whirl-Carrillo M. Smith L.H. Mintzer S. Lee M.T. Klein T.E. Callaghan J.T. Clinical pharmacogenetics implementation consortium guidelines for CYP2C9 and HLA-B genotypes and phenytoin dosing Clin. Pharmacol. Ther. 2014 96 542 548 10.1038/clpt.2014.159 25099164 17. Leckband S.G. Kelsoe J.R. Dunnenberger H.M. George A.L. Jr. Tran E. Berger R. Muller D.J. Whirl-Carrillo M. Caudle K.E. Pirmohamed M. Clinical pharmacogenetics implementation consortium guidelines for HLA-B genotype and carbamazepine dosing Clin. Pharmacol. Ther. 2013 94 324 328 10.1038/clpt.2013.103 23695185 18. Hicks J.K. Swen J.J. Thorn C.F. Sangkuhl K. Kharasch E.D. Ellingrod V.L. Skaar T.C. Muller D.J. Gaedigk A. Stingl J.C. Clinical pharmacogenetics implementation consortium guideline for CYP2D6 and CYP2C19 genotypes and dosing of tricyclic antidepressants Clin. Pharmacol. Ther. 2013 93 402 408 10.1038/clpt.2013.2 23486447 19. Scott S.A. Sangkuhl K. Gardner E.E. Stein C.M. Hulot J.S. Johnson J.A. Roden D.M. Klein T.E. Shuldiner A.R. Clinical pharmacogenetics implementation consortium guidelines for cytochrome P450-2C19 (CYP2C19) genotype and clopidogrel therapy Clin. Pharmacol. Ther. 2011 90 328 332 10.1038/clpt.2011.132 21716271 20. Hicks J.K. Bishop J.R. Sangkuhl K. Muller D.J. Ji Y. Leckband S.G. Leeder J.S. Graham R.L. Chiulli D.L. A L.L. Clinical pharmacogenetics implementation consortium (CPIC) guideline for CYP2D6 and CYP2C19 genotypes and dosing of selective serotonin reuptake inhibitors Clin. Pharmacol. Ther. 2015 98 127 134 10.1002/cpt.147 25974703 21. Crews K.R. Gaedigk A. Dunnenberger H.M. Klein T.E. Shen D.D. Callaghan J.T. Kharasch E.D. Skaar T.C. Clinical pharmacogenetics implementation consortium (CPIC) guidelines for codeine therapy in the context of cytochrome P450 2D6 (CYP2D6) genotype Clin. Pharmacol. Ther. 2012 91 321 326 10.1038/clpt.2011.287 22205192 22. Gammal R.S. Court M.H. Haidar C.E. Iwuchukwu O.F. Gaur A.H. Alvarellos M. Guillemette C. Lennox J.L. Whirl-Carrillo M. Brummel S. Clinical pharmacogenetics implementation consortium (CPIC) guideline for UGT1A1 and Atazanavir prescribing Clin. Pharmacol. Ther. 2016 99 363 369 10.1002/cpt.269 26417955 23. Relling M.V. Gardner E.E. Sandborn W.J. Schmiegelow K. Pui C.H. Yee S.W. Stein C.M. Carrillo M. Evans W.E. Klein T.E. Clinical pharmacogenetics implementation consortium guidelines for thiopurine methyltransferase genotype and thiopurine dosing Clin. Pharmacol. Ther. 2011 89 387 391 10.1038/clpt.2010.320 21270794 24. Relling M.V. Gardner E.E. Sandborn W.J. Schmiegelow K. Pui C.H. Yee S.W. Stein C.M. Carrillo M. Evans W.E. Hicks J.K. Clinical pharmacogenetics implementation consortium guidelines for thiopurine methyltransferase genotype and thiopurine dosing: 2013 update Clin. Pharmacol. Ther. 2013 93 324 325 10.1038/clpt.2013.4 23422873 25. Caudle K.E. Thorn C.F. Klein T.E. Swen J.J. McLeod H.L. Diasio R.B. Schwab M. Clinical pharmacogenetics implementation consortium guidelines for dihydropyrimidine dehydrogenase genotype and fluoropyrimidine dosing Clin. Pharmacol. Ther. 2013 94 640 645 10.1038/clpt.2013.172 23988873 26. Clancy J.P. Johnson S.G. Yee S.W. McDonagh E.M. Caudle K.E. Klein T.E. Cannavo M. Giacomini K.M. Clinical pharmacogenetics implementation consortium (CPIC) guidelines for ivacaftor therapy in the context of CFTR genotype Clin. Pharmacol. Ther. 2014 95 592 597 10.1038/clpt.2014.54 24598717 27. Johnson J.A. Gong L. Whirl-Carrillo M. Gage B.F. Scott S.A. Stein C.M. Anderson J.L. Kimmel S.E. Lee M.T. Pirmohamed M. Clinical pharmacogenetics implementation consortium guidelines for CYP2C9 and VKORC1 genotypes and warfarin dosing Clin. Pharmacol. Ther. 2011 90 625 629 10.1038/clpt.2011.185 21900891 28. Relling M.V. McDonagh E.M. Chang T. Caudle K.E. McLeod H.L. Haidar C.E. Klein T. Luzzatto L. Clinical pharmacogenetics implementation consortium (CPIC) guidelines for rasburicase therapy in the context of G6PD deficiency genotype Clin. Pharmacol. Ther. 2014 96 169 174 10.1038/clpt.2014.97 24787449 29. Wilke R.A. Ramsey L.B. Johnson S.G. Maxwell W.D. McLeod H.L. Voora D. Krauss R.M. Roden D.M. Feng Q. Cooper-Dehoff R.M. The clinical pharmacogenomics implementation consortium: CPIC guideline for SLCO1B1 and simvastatin-induced myopathy Clin. Pharmacol. Ther. 2012 92 112 117 10.1038/clpt.2012.57 22617227 30. Ramsey L.B. Johnson S.G. Caudle K.E. Haidar C.E. Voora D. Wilke R.A. Maxwell W.D. McLeod H.L. Krauss R.M. Roden D.M. The clinical pharmacogenetics implementation consortium guideline for SLCO1B1 and simvastatin-induced myopathy: 2014 Update Clin. Pharmacol. Ther. 2014 96 423 428 10.1038/clpt.2014.125 24918167 31. Birdwell K.A. Decker B. Barbarino J.M. Peterson J.F. Stein C.M. Sadee W. Wang D. Vinks A.A. He Y. Swen J.J. Clinical pharmacogenetics implementation consortium (CPIC) guidelines for CYP3A5 genotype and tacrolimus dosing Clin. Pharmacol. Ther. 2015 98 19 24 10.1002/cpt.113 25801146 32. Muir A.J. Gong L. Johnson S.G. Lee M.T. Williams M.S. Klein T.E. Caudle K.E. Nelson D.R. Clinical pharmacogenetics implementation consortium (CPIC) guidelines for IFNL3 (IL28B) genotype and PEG interferon-α-based regimens Clin. Pharmacol. Ther. 2014 95 141 146 10.1038/clpt.2013.203 24096968 33. Mizzi C. Peters B. Mitropoulou C. Mitropoulos K. Katsila T. Agarwal M.R. van Schaik R.H. Drmanac R. Borg J. Patrinos G.P. Personalized pharmacogenomics profiling using whole-genome sequencing Pharmacogenomics 2014 15 1223 1234 10.2217/pgs.14.102 25141897 34. Chua E.W. Cree S.L. Ton K.N. Lehnert K. Shepherd P. Helsby N. Kennedy M.A. Cross-comparison of exome analysis, next-generation sequencing of amplicons, and the iPLEX® ADME PGx panel for pharmacogenomic profiling Front. Pharmacol. 2016 7 1 10.3389/fphar.2016.00001 26858644 35. Sheridan C. Milestone approval lifts Illumina’s NGS from research into clinic Nat. Biotechnol. 2014 32 111 112 10.1038/nbt0214-111 24509734 36. Grosse S.D. Khoury M.J. What is the clinical utility of genetic testing? Genet. Med. 2006 8 448 450 10.1097/01.gim.0000227935.26763.c6 16845278 37. Haddow J.E. Palomaki G.E. ACCE: A model process for evaluating data on emerging genetic tests Hum. Genome Epidemiol. 2004 217 233 38. Yang X. Chockalingam S.P. Aluru S. A survey of error-correction methods for next-generation sequencing Brief. Bioinform. 2013 14 56 66 10.1093/bib/bbs015 22492192 39. Wall J.D. Tang L.F. Zerbe B. Kvale M.N. Kwok P.Y. Schaefer C. Risch N. Estimating genotype error rates from high-coverage next-generation sequence data Genome Res. 2014 24 1734 1739 10.1101/gr.168393.113 25304867 40. Daber R. Sukhadia S. Morrissette J.J. Understanding the limitations of next generation sequencing informatics, an approach to clinical pipeline validation using artificial data sets Cancer Genet. 2013 206 441 448 10.1016/j.cancergen.2013.11.005 24528889 41. Taub M.A. Corrada Bravo H. Irizarry R.A. Overcoming bias and systematic errors in next generation sequencing data Genome Med. 2010 2 87 10.1186/gm208 21144010 42. Quail M.A. Smith M. Coupland P. Otto T.D. Harris S.R. Connor T.R. Bertoni A. Swerdlow H.P. Gu Y. A tale of three next generation sequencing platforms: Comparison of Ion Torrent, Pacific Biosciences and Illumina MiSeq sequencers BMC Genomics 2012 13 12 10.1186/1471-2164-13-341 22233519 43. Hackl T. Hedrich R. Schultz J. Forster F. Proovread: Large-scale high-accuracy PacBio correction through iterative short read consensus Bioinformatics 2014 30 3004 3011 10.1093/bioinformatics/btu392 25015988 44. Dohm J.C. Lottaz C. Borodina T. Himmelbauer H. Substantial biases in ultra-short read data sets from high-throughput DNA sequencing Nucleic Acids Res. 2008 36 e105 10.1093/nar/gkn425 18660515 45. Peters B.A. Kermani B.G. Sparks A.B. Alferov O. Hong P. Alexeev A. Jiang Y. Dahl F. Tang Y.T. Haas J. Accurate whole-genome sequencing and haplotyping from 10 to 20 human cells Nature 2012 487 190 195 10.1038/nature11236 22785314 46. O’Rawe J. Jiang T. Sun G. Wu Y. Wang W. Hu J. Bodily P. Tian L. Hakonarson H. Johnson W.E. Low concordance of multiple variant-calling pipelines: Practical implications for exome and genome sequencing Genome Med. 2013 5 28 10.1186/gm432 23537139 47. Ono Y. Asai K. Hamada M. PBSIM: PacBio reads simulator—Toward accurate genome assembly Bioinformatics 2013 29 119 121 10.1093/bioinformatics/bts649 23129296 48. Alic A.S. Tomas A. Medina I. Blanquer I. Muffinec: Error correction for de novo assembly via greedy partitioning and sequence alignment Inf. Sci. 2016 329 206 219 10.1016/j.ins.2015.09.012 49. Nakamura K. Oshima T. Morimoto T. Ikeda S. Yoshikawa H. Shiwa Y. Ishikawa S. Linak M.C. Hirai A. Takahashi H. Sequence-specific error profile of Illumina sequencers Nucleic Acids Res. 2011 39 e90 10.1093/nar/gkr344 21576222 50. Hoffmann S. Otto C. Kurtz S. Sharma C.M. Khaitovich P. Vogel J. Stadler P.F. Hackermuller J. Fast mapping of short sequences with mismatches, insertions and deletions using index structures PLoS Comput. Biol. 2009 5 12 10.1371/journal.pcbi.1000502 19750212 51. Hargreaves A.D. Mulley J.F. Assessing the utility of the Oxford Nanopore MinION for snake venom gland cDNA sequencing PeerJ. 2015 3 e1441 10.7717/peerj.1441 26623194 52. Johnson J.A. Burkley B.M. Langaee T.Y. Clare-Salzler M.J. Klein T.E. Altman R.B. Implementing personalized medicine: Development of a cost-effective customized pharmacogenetics genotyping array Clin. Pharmacol. Ther. 2012 92 437 439 10.1038/clpt.2012.125 22910441 53. Centers for Disease Control and Prevention Genomic Testing Available online: http://www.cdc.gov/genomics/gtesting/acce/index.htm (accessed on 23 May 2016) 54. Van Driest S.L. Shi Y. Bowton E.A. Schildcrout J.S. Peterson J.F. Pulley J. Denny J.C. Roden D.M. Clinically actionable genotypes among 10,000 patients with preemptive pharmacogenomic testing Clin. Pharmacol. Ther. 2014 95 423 431 10.1038/clpt.2013.229 24253661 55. Johnstone D.M. Riveros C. Heidari M. Graham R.M. Trinder D. Berretta R. Olynyk J.K. Scott R.J. Moscato P. Milward E.A. Evaluation of different normalization and analysis procedures for Illumina gene expression microarray data involving small changes Microarrays 2013 2 131 152 10.3390/microarrays2020131 56. Baker M. De novo genome assembly: What every biologist should know Nat. Method 2012 9 333 10.1038/nmeth.1935 57. Oxford Nanopore Technologies Learn About Minion Available online: https://www.nanoporetech.com/products-services/minion-mki (accessed on 23 May 2016) 58. LaFramboise T. Single nucleotide polymorphism arrays: A decade of biological, computational and technological advances Nucleic Acids Res. 2009 37 4181 4193 10.1093/nar/gkp552 19570852 59. Eisenstein M. Big data: The power of petabytes Nature 2015 527 S2 S4 10.1038/527S2a 26536222 60. Ethics and Genetics Committee Ethics and Genetics Report 2013—A Shift in Privacy Law and the Attendant Risks Ethics and Genetics Glascow, Scotland 2013 61. The U-PGx Consortium Ubiquitous Pharmacogenomics Available online: upgx.eu (accessed 23 May 2016) 62. Mitropoulos K. Al Jaibeji H. Forero D.A. Laissue P. Wonkam A. Lopez-Correa C. Mohamed Z. Chantratita W. Lee M.T. Llerena A. Success stories in genomic medicine from resource-limited countries Hum. Genomics 2015 9 11 10.1186/s40246-015-0033-3 26081768 63. Asian Network For Pharmacogenomics Research Overview Available online: http://www.asianpr.org/ (accessed on 23 May 2016) 64. Rotimi C. Abayomi A. Abimiku A. Adabayeri V.M. Adebamowo C. Adebiyi E. Ademola A.D. Adeyemo A. Adu D. Affolabi D. Research capacity. Enabling the genomic revolution in Africa Science 2014 344 1346 1348 24948725 65. Dunnenberger H.M. Crews K.R. Hoffman J.M. Caudle K.E. Broeckel U. Howard S.C. Hunkler R.J. Klein T.E. Evans W.E. Relling M.V. Preemptive clinical pharmacogenetics implementation: Current programs in five US medical centers Annu. Rev. Pharmacol. Toxicol. 2015 55 89 106 10.1146/annurev-pharmtox-010814-124835 25292429

PMC005xxxxxx/PMC5003489.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020013microarrays-05-00013ArticleSmall-Molecule Inhibition of Rho/MKL/SRF Transcription in Prostate Cancer Cells: Modulation of Cell Cycle, ER Stress, and Metastasis Gene Networks Evelyn Chris R. 1†Lisabeth Erika M. 2Wade Susan M. 1Haak Andrew J. 1Johnson Craig N. 3Lawlor Elizabeth R. 4Neubig Richard R. 12*Negrini Massimo Academic Editor1 Department of Pharmacology, University of Michigan, Ann Arbor, MI 48109, USA; Chris.Evelyn@cchmc.org (C.R.E.); suemwade@umich.edu (S.M.W.); Haak.Andrew@mayo.edu (A.J.H.)2 Department of Pharmacology & Toxicology, Michigan State University, East Lansing, MI 48824, USA; matheser@msu.edu3 University of Michigan Microarray Core, University of Michigan Comprehensive Cancer Center, Ann Arbor, MI 48109, USA; johnscrn@umich.edu4 Departments of Pediatrics and Pathology, University of Michigan, Ann Arbor, MI 48109, USA; elawlor@umich.edu* Correspondence: rneubig@msu.edu; Tel.: +1-517-353-7145† Current address: Division of Experimental Hematology and Cancer Biology, Cincinnati Children’s Research Foundation, Cincinnati, OH 45229, USA. 28 5 2016 6 2016 5 2 1330 3 2016 16 5 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Metastasis is the major cause of cancer deaths and control of gene transcription has emerged as a critical contributing factor. RhoA- and RhoC-induced gene transcription via the actin-regulated transcriptional co-activator megakaryocytic leukemia (MKL) and serum response factor (SRF) drive metastasis in breast cancer and melanoma. We recently identified a compound, CCG-1423, which blocks Rho/MKL/SRF-mediated transcription and inhibits PC-3 prostate cancer cell invasion. Here, we undertook a genome-wide expression study in PC-3 cells to explore the mechanism and function of this compound. There was significant overlap in the genes modulated by CCG-1423 and Latrunculin B (Lat B), which blocks the Rho/MKL/SRF pathway by preventing actin polymerization. In contrast, the general transcription inhibitor 5,6-dichloro-1-β-d-ribofuranosyl-1H-benzimidazole (DRB) showed a markedly different pattern. Effects of CCG-1423 and Lat B on gene expression correlated with literature studies of MKL knock-down. Gene sets involved in DNA synthesis and repair, G1/S transition, and apoptosis were modulated by CCG-1423. It also upregulated genes involved in endoplasmic reticulum stress. Targets of the known Rho target transcription factor family E2F and genes related to melanoma progression and metastasis were strongly suppressed by CCG-1423. These results confirm the ability of our compound to inhibit expression of numerous Rho/MKL-dependent genes and show effects on stress pathways as well. This suggests a novel approach to targeting aggressive cancers and metastasis. metastasistranscriptioncell cycleRho ==== Body 1. Introduction Metastasis is a complex multi-step process that enables tumor cells to disseminate from their site of origin to colonize distant tissue and organ sites [1,2,3,4,5]. Recent evidence has identified alterations in gene expression related to metastasis with several classes of genes implicated [4]. Furthermore, numerous gene signatures have been identified that can predict the aggressiveness and metastatic potential of numerous cancers [6,7]. Metastasis initiation genes enable tumor cells to invade the surrounding tissue, disseminate, and intravasate [4]. This includes genes related to epithelial-mesenchymal transition (EMT, e.g., Twist1, Snai1, and Snai2, [8]) as well as hepatocyte growth factor (HGF) pathway-dependent genes [9]. Genes that enable tumor cells to extravasate and survive at the secondary tissue site contribute to metastasis progression including prostaglandin G/H synthase 2 (PTGS2), epiregulin, and angiopoietin-like 4 (ANGPTL4) [4,10,11]. Lastly, metastasis virulence genes, such as parathyroid hormone-related protein (PTHRP) and interleukin 11 (IL11) enable tumor cells to colonize and survive at the distant secondary tumor site [4,12,13,14]. Therefore, gene expression plays a vital role in the complex process of cancer metastasis. The small GTPases RhoA and RhoC are strongly implicated in metastasis [15,16]. While most attention has been focused on the cytoskeletal actions of RhoA and RhoC [16], they also produce transcriptional effects [17,18,19,20]. This is mediated through the actin-dependent translocation of the transcriptional co-activator MKL (MAL, MTRF, or BSAC) to the nucleus where MKL binds with serum response factor (SRF) at serum response elements (SRE). Transcriptional targets include c-Fos, matrix metalloproteinases, genes related to EMT [21], and genes responsive to E2F, c-Myc, and c-Jun transcription factors [18]. Recently, MKL and SRF were shown to play a key role in metastasis in melanoma and breast cancer xenograft systems [22] providing a strong link to the transcriptional effects of RhoA or RhoC in these processes. In addition, we identified CCG-1423, a nM-potency small-molecule inhibitor of RhoA/C transcriptional signaling [23]. This compound inhibits SRE promoter activation by RhoA and -C and MKL in both HEK-293T cells and in PC-3 prostate cancer cells. It also inhibits Matrigel® invasion by PC-3 cells but does not inhibit invasion by the Gαi/ras-dependent SKOV-3 ovarian cancer cells [23]. This supports the role for RhoC in invasion by PC-3 cells [24] and is consistent with the recent demonstration of a role for MKL [22] implicating transcriptional outputs from the Rho pathway in invasion by certain aggressive cancers. To gain further insight into the mechanism of action of CCG-1423, we undertook a genome-wide gene expression analysis of its effects on PC-3 prostate cancer cells utilizing conditions employed in our PC-3 cell Matrigel invasion assays. There were strong overlaps between genes regulated by CCG-1423 and by Lat B, which is known to disrupt Rho-mediated gene transcription. Furthermore, targets of the Rho-regulated transcription factor E2F related to G1/S transition were strongly suppressed by CCG-1423. Finally, there was a clear correlation between CCG-1423-regulated genes and those modulated by RNAi-mediated suppression of the transcriptional co-activator MKL—especially in melanoma where Rho/MKL function is strongly implicated in metastasis. Beyond effects on the Rho/MKL pathways, CCG-1423 also stimulated expression of stress-responsive genes such as activated transcription factors 3 and 4 (ATF3, ATF4) and DNA-damage-induced transcript 3 (DDIT3, also referred to as CHOP) suggesting that alternative mechanisms contribute to the actions of this compound. 2. Materials and Methods 2.1. Cell Lines and Reagents Dimethyl Sulfoxide (DMSO), the marine toxin Latrunculin B (Lat B), and the RNA Polymerase carboxyl-terminal domain kinases inhibitor DRB (5,6-Dichlorobenzimidazole 1-β-d-ribofuranoside) were all purchased from Sigma (St. Louis, MO, USA). The RhoA/C transcriptional pathway inhibitor, CCG-1423 (N-(2-(4-chloroanilino)-1-methyl-2-oxoethoxy)-3,5-bis(trifluoromethyl) benzamide), was from Cayman Chemical (Ann Arbor, MI, USA). The PC-3 cell line was a kind gift from Kenneth Pienta (University of Michigan, Ann Arbor, MI, USA). The RNAqueous® RNA isolation kit was purchased from Ambion (Austin, TX, USA). The TURBO DNAse-free kit was purchased from Ambion (Austin). The TaqMan® Reverse-Transcription Reagents Kit was purchased from Applied Biosystems (Foster City, CA, USA). The Brilliant® II SYBR® Green QPCR Master Mix with low Rox dye was purchased from Stratagene (La Jolla, CA, USA). 2.2. Cell Culture The PC-3 prostate cancer epithelial cell line was normally maintained in DMEM (Invitrogen, Carlsbad, CA, USA, cat.#: 11995) containing 10% fetal bovine serum (FBS), 100 units/mL penicillin, and 100 µg/mL streptomycin at 37 °C in 5% CO2. 2.3. Microarray Analysis PC-3 cells (9 × 105) were seeded into individual 60 mm dishes. Twenty-four hours later, they were serum-starved (0.5% FBS) for twenty-four hours. Cells were then treated for an additional two or twenty-four hours with DMSO (1.0%), 3 µM of CCG-1423, 0.5 µM of Lat B, or 50 µM of DRB under serum-starved conditions (0.5% FBS). Cells from three dishes for each condition were lysed and RNA was isolated using the RNAqueous® kit from Ambion following the manufacturer’s directions. cDNA and biotin-labeled cRNA synthesis, hybridization to the human U133 plus 2.0 Affymetrix gene chips, scanning of the gene chips, and analysis of the data images were all performed at the University of Michigan Comprehensive Cancer Center (UMCCC) Affymetrix and Microarray Core Facility in Ann Arbor, MI, USA. Expression values for each probeset on the chip were calculated using a robust multi-array average (RMA) [25] and were expressed as log2 transformed data. Fold-change values for each probeset were calculated by subtracting the average log2 (expression) values for the compound samples from the average log2 (expression) values of the DMSO samples. For simple comparisons of different treatment results and preliminary candidate selection, a record for each probeset showing at least a two-fold change (i.e., log2-fold change >1 for a decrease or <−1 for an increase in expression) was imported into a MySQL® database (Sun Microsystems, Santa Clara, CA, USA). Literature microarray datasets [20,22] were also imported for comparisons. Queries of those datasets were utilized to obtain counts for genes with altered gene expression, identify overlaps between datasets, identify highly regulated genes (log2-fold change >3 or <−3), filter by gene ontology (GO) categories to generate Table 1 and Table 2 and the Venn diagram in Figure 1. 2.4. Bioinformatic Analysis Gene set enrichment analysis (GSEA) was also done to identify both KEGG categories and literature gene expression datasets that were selectively modulated by our compound [26]. Gene sets were curated and organized in the Broad Institutes Molecular Signature Database version 3.0. GSEA was performed using the R statistical environment and the PGSEA Bioconductor package. For each gene set a Z-score was created testing whether the mean fold changes (genes of gene set/average of controls) were statistically different from the mean fold changes of the entire microarray data set. Linear models were then used on to find gene sets with statistical different Z-scores between the two experimental groups. Finally, to assess enrichment of specific GO categories using an alternate method, genes that were significantly induced or repressed by CCG-1423 with log2-fold changes of greater than ±1 (absolute fold-change >2) were uploaded for analysis using DAVID bioinformatics resources [27]. GO biologic process, molecular function and cell component category representation were determined for both gene lists and GO categories with Benjamini corrected p-values (false discovery rate, FDR) of <0.05 were considered to be significantly over-represented. 2.5. Quantitative Real-Time Polymerase Chain Reaction (qRT-PCR) To confirm mRNA changes from the microarray results for a set of candidate genes, PC-3 cells (9 × 105) were seeded into 60 mm dishes. Twenty-four hours after plating, PC-3 cells were serum-starved (0.5% FBS). Cells were then treated for an additional twenty-four hours with DMSO (0.03% or 1.0%) or compound under serum-starved conditions (0.5% FBS) for PC-3 cells. Cells were lysed and RNA was isolated using the RNAqueous® kit from Ambion following the manufacturer’s directions. RNA (10 µg) was treated with DNAse using the TURBO DNA-free kit from Ambion following the manufacturer’s directions. DNAse-treated RNA (1 µg) was used as a template for synthesizing cDNA utilizing the Taqman® Reverse-Transcription Reagents kit. The components used for the reverse-transcription reaction were 1× reverse transcription buffer, 5.6 mM MgCl2, 2 mM dNTPs, oligo dT, RNAse inhibitor, reverse transcriptase enzyme, sterile water, and 1 µg of RNA. The sequential order of reaction conditions was: (i) 25 °C for 10 min; (ii) 48 °C for 30 min; and (iii) 95 °C for 5 min and carried out in a Biometra TGradient thermocycler (Analytikijena, Gottingen, Germany). The qRT-PCR reaction was performed using 5 µL of cDNA sample per well in a 25 µL final reaction volume in a 96-well qPCR plate, and the Brilliant® II SYBR® Green QPCR Master Mix with low Rox dye from Stratagene in a Stratagene MX3000P® QPCR System. The reaction mix consisted of 1× of the Brilliant® II SYBR® Green QPCR Master Mix with low Rox dye, 200 nM each of both forward and reverse primers, and sterile water. The sequential order of reaction conditions was: (i) 95 °C for 10 min; (ii) 95 °C for 30 s; (iii) 55 °C for 1 min; and (iv) 72 °C for 30 s. Steps 2 through 4 were carried out for 50 cycles. The primer sequences used were GAPDH: Fwd-5′-GGAAGGACTCATGACCACAG-3′, Rev-5′-ACAGTCTTCTGGGTGGCAGTGATG-3′ (base-pairs were corrected to match human sequence) [20]; RGS4: Fwd-5′-TTCCCACAACAAGAAGGACAAAG-3′, Rev-5′-TGATTCAGCCCATTTCTTGAC-3′ (base-pairs were corrected to match human sequence) [28]; RGS7: Fwd-5′-CCTTCTAACCCATGGCTGTC-3′, Rev-5′-TTTTTCAGGTCCTCCACTGC-3′ [29]; CTGF: Fwd-5′-CAGAGTGGAGCGCCTGTTC-3′, Rev-5′-CTGCAGGAGGCGTTGTCAT-3′ [30]; SOX9: Fwd-5′-CAACCAGAATTCCCTTTGGA-3′, Rev-5′-TGCTCCATTTAGCCAAGGTT-3′ [31]; ATF3: Fwd-5′- TCAAGGAAGAGCTGAGGTTTGCCA-3′, Rev-5′-CTTCTTGTTTCGGCACTTTGAGC-3′; ATF4: Fwd-5′-CTGCTGAATGCCGTGAGAA-3′, Rev-5′-GCGTATTAGGGGCAGCAGT-3′; DDIT3 (CHOP): Fwd-5′-CCATCTCTGCAGTTGGATCA-3′, Rev-5′-CCAAAATCAGAGCTGGAACC-3′. GAPDH gene expression was utilized as an internal control. The relative mRNA gene expression was calculated using the following formula: 2−(∆Ct) where the ∆Ct value = Ct value of Sample—Ct value of GAPDH. Fold changes were calculated by dividing the 2−(∆Ct) of the compound sample by the 2−(∆Ct) of the DMSO control sample. 3. Results 3.1. Microarray Analysis For our microarray analysis, we chose to use unstimulated PC-3 cells (grown in 0.5% serum) to mimic the conditions in which functional inhibition of PC-3 cell matrix invasion was demonstrated [23]. We also utilized the actin cytoskeleton inhibitor, Lat B, as a positive control compound that inhibits the RhoA/C/MKL/SRF transcriptional pathway via disruption of actin polymerization and the transcription elongation inhibitor, DRB, as a “non-specific” control to inhibit general RNA Polymerase II-mediated transcription. After two hours of compound treatment, we observed only a few transcripts with >2-fold changes (Table 1). Interestingly, both CCG-1423 and Lat B had primarily stimulatory effects on gene expression. CCG-1423 increased levels of 48 transcripts out of 49 total transcripts changed, while Lat B increased expression of 15 genes out of 18 total changed gene transcripts (Table 1 and Table S1). In contrast, the transcription inhibition control compound, DRB, had a marked effect in the opposite direction at 2 h, primarily reducing mRNA levels (714 gene transcripts showed decreased levels out of 719 total gene transcripts changed—Table 1). Of the 15 genes up-regulated by Lat B at 2 h, five of them were also up-regulated by CCG-1423 (JUN, ERRFI1, FBXO32, GDF15, and GEM). This is consistent with the depletion of MKL in invasive breast cancer and melanoma cell lines, where the authors also observed an increase in Jun and FBXO32 RNA levels [22]. In addition, CCG-1423 induced the stress-responsive genes ATF3 and DDIT3 (also referred to as CHOP) while Lat B did not. However, at 24 h, a large number of genes had their expression altered by the three compounds by ≥2 fold (Table 1 and Figure 1). Not surprisingly, DRB modulated expression of the greatest number of genes (3020), while CCG-1423 also affected a substantial number (2142). Lat B had the most selective effect, modulating expression of only 608 genes. Unlike the 2 h time point, at 24 h, all three compounds downregulated 1.5–2 times more transcripts than they up-regulated (Table 1). There is a clear relation between the genes modulated by CCG-1423 and by Lat B (Figure 1): 73% of genes modulated by Lat B were also altered by CCG-1423. Furthermore, 427 of the 444 genes regulated by both Lat B and CCG-1423 changed in the same direction (Table S1). While DRB modulated almost 50% more genes than did CCG-1423, DRB only shared 247 modulated genes in common with Lat B vs. 444 for CCG-1423. The similarity of the effects of CCG-1423 and Lat B is confirmed in a principal components analysis (Figure 1B). The data points for Lat B lie essentially on the line connecting the DMSO control and the CCG-1423 samples, consistent with a high proportion of shared genes but the more limited effect of Lat B. In contrast, the DRB samples are clearly in a different sector of the principal components graph. It is not surprising that CCG-1423 and Lat B have strongly overlapping transcriptional effects since CCG-1423 was initially discovered in a screen for inhibition of a RhoA/C-responsive, MKL/SRF dependent luciferase reporter (SRE.L) [23]. Indeed, Lat B was used as a positive control for the screen due to its ability to block RhoA/C-SRE-mediated gene transcription by preventing actin polymerization. Although CCG-1423 does not block the RhoA/C transcription pathway at the level of actin polymerization [23], it clearly does modulate the same set of genes as does Lat B (although it also targets a broader array). To further assess whether CCG-1423 could modulate the same target genes as reported for MKL-1, we compared our gene expression data with results from two studies in the literature documenting MKL-dependent gene expression. Prywes and colleagues tested the effect of overexpressing a dominant-negative MKL1 protein on serum-induced gene expression in NIH-3T3 cells [20]. Treisman and colleagues tested the effect of MKL1/2 (also referred to as MRTF A/B) shRNA knockdown on global gene expression in invasive breast cancer (MDA-MB-231) and melanoma (B16F2) cell lines [22]. CCG-1423 regulated 18%–26% of MKL-dependent genes (Table 2). Surprisingly, Lat B only regulated a very small percentage of the MKL-dependent genes from those three studies (0%–17%) while DRB regulated a similar percentage of genes as did CCG-1423 (18%–25%, Table 2). Since the overlap between MKL-1 dependent genes and the genes regulated by CCG-1423 is relatively low, this suggests that CCG-1423 may affect a broader transcriptional mechanism. Although CCG-1423 did not affect any more MKL-regulated genes than did DRB, the direction of effects by CCG-1423 on the two cancer cell lines correlated with that of MKL knockdown (Figure 2A,D). In contrast, the effects of DRB showed no directionality (Figure 2C,F). Furthermore, even the few MKL-dependent genes affected by Lat B showed the same directionality and a significant correlation in magnitude with the effects of MKL knock-down (Figure 2B,E). Both CCG-1423 and Lat B showed more consistent effects on the melanoma cell line (Table 2 and Figure 2) than they did on the breast cancer cells, which is in line with our prior data on the strong biological effect of CCG-1423 on invasive melanoma [23]. The CCG-1423 modulated genes were also enriched in those predicting aggressive behavior of clinical melanomas (see below). Although CCG-1423 clearly modulates a broader set of genes than does the genetic suppression of MKL function, our data do support an effect on MKL-dependent genes—most significantly those identified in the B16F2 melanoma system. 3.2. Candidate Gene Analysis To help focus our analysis of the large number of modulated genes, we used several criteria to identify potential genes of interest with respect to the ability of CCG-1423 to block PC-3 cell matrix invasion. Utilizing metastasis-related gene ontology (GO) categories as a filter, we identified 203 gene transcripts (Table S2) out of the 2142 total gene transcripts that are regulated by CCG-1423 at 24 h. To further narrow the candidate genes to follow-up upon, we chose those where CCG-1423 caused at least an eight-fold change in gene expression leaving us with 16 candidate gene transcripts. Of those, we chose four that the literature provided evidence related to invasion or metastasis. These included regulator of G-protein signaling 4 (RGS4) and 7 (RGS7), which were both upregulated by CCG-1423, and CTGF (connective tissue growth factor), and SOX9 (sex determining region Y-box 9, also referred to as campomelic dysplasia, autosomal sex reversal) which were down-regulated. Several studies have shown a relationship between RGS protein expression and cancer invasion. RGS4 was previously found to be down-regulated in invasive cancers compared to either non-invasive cancer or normal epithelial cells for both breast and ovarian cancer systems [32,33]. In NIH3T3 mouse fibroblast cells, constitutively activated Gαo-subunit stimulated colony formation in a Stat3-dependent manner [34]. RGS7-mediated suppression of Gαo signaling could reduce cancer progression dependent upon Gαo signaling pathways. Interestingly, RGS4 has several predicted SRF binding sites located within its promoter [35], suggesting that modulation of RGS4 mRNA levels by CCG-1423 may be through a SRF-dependent mechanism. CTGF and SOX9, both of which were found to be downregulated in this study, have also been implicated in cancer invasion. Expression of CTGF is known to be induced by RhoA signaling and it is over-expressed in many types of cancers, including head and neck squamous cell carcinoma [36], pancreatic cancer [37], gastric cancer [38], and in pre-B acute lymphoblastic leukaemia [39]. In gastric cancer, CTGF overexpression correlated with poor patient survival [38] and in pancreatic cancer CTGF stimulated tumor growth in vitro and in vivo [37]. SOX9 is overexpressed in both colorectal cancer [40] and hormone-refractory prostate cancer [41] and has been correlated with poor patient survival in colorectal cancer [40] and enhanced in vivo tumor growth, angiogenesis, and invasion in a LNCaP prostate cancer xenograft model [42]. Therefore, inhibition of CTGF and/or SOX9 expression by CCG-1423 could contribute to reduced PC-3 cell invasion. In order to confirm the microarray results, we tested by qRT-PCR the effect of CCG-1423 on levels of RGS4, RGS7, CTGF, and SOX9 mRNA in PC-3 prostate cancer cells. As expected, CCG-1423 increased mRNA levels of RGS4 and RGS7 and decreased CTGF and SOX9 gene expression (Table 3). Overall, the magnitudes of CCG-1423-induced changes were similar by qRT-PCR and microarray quantification for all four candidate genes. 3.3. Time-Course of CCG-1423 Regulated Gene Expression In order to better understand the timing of CCG-1423-regulated gene expression, we studied the effect of CCG-1423 on RGS4, RGS7, CTGF, and SOX9 gene expression in PC-3 prostate cancer cells from 1 to 24 h (Figure 3). RGS4 gene expression increased in a time-dependent manner peaking at 12–24 h while RGS7 gene expression reached its maximum by 6 h (Figure 3B). Interestingly, CCG-1423 stimulated CTGF gene expression at six hours but it subsequently decreased at 12 and 24 h (Figure 3D) to produce the strong reduction observed in the microarray data. Finally, CCG-1423 inhibited SOX9 gene expression reaching peak suppression by 12–24 h (Figure 3C). Consequently it takes between 6 h (RGS7) and 24 h (RGS4 and CTGF) of CCG-1423 treatment to obtain maximal effects on expression of these candidate mRNAs. 3.4. Global Analysis In addition to the candidate gene approach, we undertook three global analyses of gene expression to assess the possibility of actions at the level of gene networks. This would not be surprising, in light of complex gene programs induced by RhoA family members in NIH-3T3 cells [18]. Comparisons of the effects of CCG-1423 and the control compounds (Lat B and DRB) were done for functionally defined gene sets (KEGG and GO pathways) and literature microarray datasets (Broad MSigDB version 3.0). Genes modulated by CCG-1423 in PC-3 cells showed a significant correlation with 30 KEGG pathway sets (adjusted p value < 0.05). Similarly, Lat B and DRB showed significant correlations with 21 and 34 categories, respectively. A heat-map for the top 25 KEGG gene categories regulated by CCG-1423 is shown in Figure 4. All three compounds (CCG-1423, Lat B, and DRB) produced highly significant effects on genes involved in purine and pyrimidine metabolism and DNA repair. The heat map for Lat B is very similar to that for CCG-1423, and both had marked effects on genes involved in cell cycle and DNA replication. (Figure S1A). Indeed, all 10 of the top categories for Lat B were within the top 15 for CCG-1423. In contrast, the top 25 gene categories regulated by DRB were quite different (Figure S1B) and only five of the top 10 gene categories modulated by DRB are within even the top 25 for CCG-1423. The concordance between effects of CCG-1423 and Lat B on KEGG pathway genes, and specifically on cell cycle and DNA replication, is consistent with the principal components analysis in Figure 1C. In addition, treatment of LPA stimulated PC3 cells with CCG-1423 reduced DNA replication, which further validates our microarray data [23]. We also assessed the most strongly modulated genes according to GO category (biological process, molecular function, and cellular compartment) using David [43]. This analysis separately assesses genes that are up-regulated and those that are down-regulated. The biological processes that showed the greatest effects (Table S3, false discovery rate, FDR < 10−8, red font) were all down-regulated and were related to cell cycle and DNA replication and repair. This further validates the results from the GSEA/KEGG analysis. With cell cycle, DNA replication, and DNA repair as the biological process, the molecular functions with lowest FDR, as expected, were related to nucleotide/nucleoside binding and the cellular compartments were chromosome, nucleus, and mitotic spindle. The most highly significant up-regulated genes (FDR < 10−5) were related to apoptosis followed closely by endoplasmic reticulum stress and unfolded protein response (FDR < 10−4). Markers of the unfolded protein response (ATF3, ATF4, and DDIT3—also referred to as CHOP) were retested by qRT-PCR and their mRNA levels were markedly increased at 24 h (10–80-fold, Figure 5). The literature datasets in MSiGDB (GSEA, Broad Institute, version 3.0) also revealed important similarities with the effects of CCG-1423 on gene expression. Of the 50 related gene datasets (Figure S2), 36 were not significantly affected by DRB (Table S4) so we consider them more related to the CCG-1423 mechanism. Of those, 25 were shared with Lat B further confirming the close similarity of the effects of CCG-1423 and Lat B. In this analysis, we will focus on those gene sets (36/50) that were not also associated with the effects of the transcription elongation inhibitor DRB. As would be predicted from the KEGG and GO analysis, the greatest number of related sets had to do with proliferation, DNA synthesis, and cell cycle. Several other commonalities were also apparent in those gene sets specific for CCG-1423 but not DRB. For example, genes that are targets of the transcription factor family E2F were strongly downregulated (marked blue in Table S4). Specifically DHFR, TK1, and CCNE2 (cyclin E2) were all downregulated either ~2-fold (DHFR) or ~8-fold (TK1 and CCNE2) by treatment with CCG-1423 and ~2-fold (DHFR) or ~4-fold (TK1 and CCNE2) by Lat B. Interestingly, E2F-regulated transcription has been shown to be one of three key gene programs (E2F, Myc, and Jun) induced by RhoA or RhoC in NIH-3T3 cells [18], [23]. The ability of CCG-1423 to shut down Rho-regulated gene transcription, including suppression of the E2F transcription pathway, is unique and may represent an important new tool for disrupting Rho-mediated transformation and invasion. The effect of our compound also had significant overlap with gene sets found in metastatic cancers. Two gene sets of particular interest relate to melanoma since CCG-1423 was found to be most potent at inhibiting the highly metastatic variant A375M2 line [23]. WINNEPENNINCKX_MELANOMA_METASTASIS_UP and KAUFFMANN_MELANOMA_ RELAPSE_UP gene sets (marked red in Table S4) were both strongly downregulated by CCG-1423 in PC-3 cells. Strong downregulation of genes associated with metastasis and relapse could provide an important approach to preventing those processes and/or killing cells engaging in those processes. Another very highly associated gene set (marked yellow in Table S4) includes those that are upregulated in human gastric cancer cells after selection for resistance to doxorubicin (KANG_DOXORUBICIN_RESISTANCE_UP). Genes upregulated by doxorubicin resistance but downregulated by CCG-1423 include TOP2A and RRM1. This suggests the potential for synergy between doxorubicin and our compound. Finally, two gene sets (also marked yellow in Table S4) that were strongly up-regulated by CCG-1423 relate to effects of agents that represent novel potential cancer therapies. 4. Discussion RhoA and RhoC have been strongly implicated in the aggressiveness and metastasis of a number of cancer types, including breast, prostate, and melanoma [15,16,44], but the precise mechanisms underlying those effects are not clear. Furthermore, there are no effective targeted therapies to combat Rho-mediated functions. Recent evidence [22] points to a critical role for gene transcriptional mechanisms downstream of Rho (i.e., MKL and SRF) and we have identified a series of small molecule compounds, including CCG-1423, that block Rho/MKL-stimulated gene transcription [23,45]. In the present study, we used mRNA microarray analysis in the RhoC-dependent aggressive PC-3 human prostate cancer cell line [24,46] to show that CCG-1423 does effectively perturb Rho-regulated transcription pathways. While no single gene was identified whose modulation explains the effects of our compounds, we demonstrate a strong connection to E2F family transcription factor-regulated genes. This is consistent with data showing that RhoA and RhoC-mediated transformation of NIH 3T3 cells is dependent on activation of E2F transcriptional mechanisms [18]. This study identifies other biological mechanisms (cell cycle, DNA replication and repair, G1/S checkpoint, apoptosis, and ER stress) influenced by our novel Rho transcription pathway inhibitor, CCG-1423, in PC-3 prostate cancer cells. It is clear that this compound does not produce a broad or general inhibition of transcription since its effect is quite distinct from that of the transcription elongation inhibitor DRB. However, the number of genes modulated by CCG-1423 is significantly greater than those affected by actin cytoskeletal block by Lat B (this report) as well as those modulated by the ROCK inhibitor Y-27632 in NIH 3T3 cells [18]. This is consistent with the concept that Rho activation can engage multiple transcription programs but only a subset are blocked by existing targeted Rho pathway inhibitors [18]. Alternatively, the molecular mechanism of CCG-1423 may involve a more general cellular target than just Rho or its downstream transcription coactivator MKL. We have identified several transcriptional programs that are modulated in PC-3 human prostate cancer cells by our Rho/MKL/SRF transcription inhibitor CCG-1423. In particular the strong suppression of genes responsive to the E2F family of transcription factors is of great interest. E2F plays a key role in the G1/S checkpoint resulting in increased DNA synthesis and cancer cell proliferation and transformation [47]. Our observation that CCG-1423 suppresses E2F-modulated genes fits well with the recently defined ability of RhoA or C to induce E2F-regulated genes in NIH-3T3 cells [18]. Furthermore, the E2F-related gene clusters are also at least partially suppressed by Lat B which blocks Rho-induced gene expression by a different mechanism. Finally, the role of the E2F pathway in Rho-mediated transformation in the 3T3 system suggests that this will have important functional implications [18]. As noted in the Results Section, substantial information suggests relevance of our compound’s effects to melanoma. Microarray data sets from primary melanomas identified genes related to the propensity of a tumor to metastasize and/or relapse [6,48]. CCG-1423 reduced expression (in PC-3 cells) of those same gene sets. Whether those regulated genes are causal in the metastasis/relapse or are just markers, it will be of interest to determine whether compounds of this sort could have a beneficial effect in preclinical models. Even if those genes are markers, they are likely to be driven by a signaling mechanism or gene program that, if abrogated, could have a beneficial effect. Two key publications further support the importance of Rho/MKL/SRF signaling in melanoma metastasis. First, Hynes and colleagues found markedly increased RhoC expression in the A375M2 cell line, which had been selected by twice isolating pulmonary metastases of the parental A375 melanoma line. We previously demonstrated that A375M2 cells were especially sensitive to the growth inhibition and apoptosis-inducing effects of CCG-1423 [23]. More recently, Treisman and colleagues showed that MKL and SRF were critical for metastasis of BF16F2 melanoma cells [22]. This latter result is critical to the concept that Rho-stimulated gene transcription is a major element in cancer aggressiveness and metastasis. That was precisely the rationale for pursuing compounds with Rho transcription inhibition activity such as those in the CCG-1423 series. Similar types of information implicate a role for Rho-regulated gene transcription in breast cancer. RhoC is highly over-expressed in inflammatory breast cancer, arguably the most aggressive form of breast cancer [49]. Also, RhoC−/− mice show a much lower frequency of lung metastases from breast tumors caused by polyoma middle T-antigen expression despite similarly sized primary tumors [50]. Most relevant to Rho-stimulated gene transcription is the suppressing effect of MKL knock-down on metastasis of MDA-MB-231 xenografts [22]. We see a correlation of gene expression effects by CCG-1423 in the prostate cancer cells with the MKL-dependent genes from Treisman’s study of melanoma and breast cancer (Figure 2). Specifically MYH9 and MYL9 were among the few shared MKL-dependent genes in those two tumor types and our compound reduces expression of both (ca. 2–3-fold) in PC-3 cells. This suggests that CCG-1423 represses processes that are active in numerous cancer cell types. There remain a number of questions to be answered regarding CCG-1423. What is the direct molecular target(s) of CCG-1423 and more recently identified analogs [45]. What is the significance of the “additional” genes modulated by the compound compared to the relatively smaller number of genes that appear to be MKL-dependent in the studies from Prywes [19] and Treisman [22]. E2F-regulated genes may play a key role in the action of CCG-1423 given their strong induction by RhoA and RhoC in NIH-3T3 cells [18]. It is not known, however, if those are actually MKL-dependent. The ability of CCG-1423 to suppress expression of genes activated by the RhoA/C pathway confirms our previous conclusion that it is a Rho/MKL/SRF pathway inhibitor. Furthermore, the strong suppression of E2F-regulated genes provides additional rationale for its testing in aggressive cancer models. Future work remains to definitively identify the direct molecular target of this compound and to understand the mechanism behind the fairly broad scope of genes whose expression it modulates. Additional studies in melanoma and breast cancer, which are strongly influenced by RhoA/C pathways, as well as identification of the properties of tumor subsets that might make them particularly responsive to this family of compounds will be important future directions. 5. Conclusions In this study, we have uncovered several key cellular gene expression networks that are perturbed by our compound CCG-1423, including cell cycle mediators, Rho-mediated gene expression, and ER stress mechanisms. In addition, modulation of MKL/SRF dependent metastasis networks in both breast cancer and melanoma cell lines with our compound highlights the importance of pharmacologic inhibitors of this pathway as potential future therapeutics in cancer biology. Acknowledgments Supported by NIH GM39561 (RRN) and NCI SARC Sarcoma SPORE 1U01-CA114757 (ERL). Additional funds were provided by Michigan State University and the microarray core of the University of Michigan Comprehensive Cancer Center. No funds were provided for open access publication. Supplementary Materials The following are available online at http://www.mdpi.com/2076-3905/5/2/13/s1. Click here for additional data file. Author Contributions Chris R. Evelyn and Richard R. Neubig conceived the study, planned experiments, and did primary manuscript preparation; Erika M. Lisabeth reviewed data and made substantial manuscript revisions; Chris R. Evelyn performed the microarray studies, RT-PCR, and candidate gene analyses; Susan M. Wade performed functional studies of candidate genes; Andrew J. Haak performed stress response pathway studies; Craig N. Johnson analyzed initial microarray data and undertook GSEA analyses; and Elizabeth R. Lawlor undertook the David analyses, identified key pathways, and contributed design ideas. All authors reviewed the manuscript and made significant contributions to the analysis. Microarray data are available on GEO Accession number: GSE30188. Conflicts of Interest The authors declare no conflict of interest. Abbreviations The following abbreviations are used in this manuscript: MKL Megakaryocytic Leukemia SRF Serum Response Factor Lat B Latrunculin B DRB 5,6-dichloro-1-β-d-ribofuranosyl-1H-benzimidazole ER Endoplasmic reticulum Figure 1 Microarray analysis of regulated genes at 24 h. PC-3 cells were treated with 3 µM of CCG-1423, 0.5 µM of Lat B, and 50 µM of DRB under serum-starved conditions (0.5% FBS) for 24 h and gene expression assessed with an Affymetrix gene chip. (A) The Venn diagram shows the number of genes regulated (stimulated or inhibited) by CCG-1423, Lat B, and DRB with ≥2-fold change. The percentages of genes coordinately regulated by CCG-1423 and Lat B, or Lat B, and DRB are indicated; (B) A principal components analysis of the differentially regulated genes shows distinct patterns for Lat B and CCG-1423 vs. DRB. Figure 2 Magnitude of expression changes observed with compounds compared to effects of MKL/MTRF suppression. Gene sets and the fold-change (log2) were compiled from our microarray data sets and from the data of Medjkane et al. [22]. For the overlapping genes, the magnitude of the change by CCG-1423 is plotted relative to the magnitude of the change induced by combined shRNA suppression of MRTF-A and MRTF-B (also referred to as MKL-1 and MKL-2). Linear regression analysis was performed to provide an indication of the extent to which the trends were similar for the two different interventions. Numbers in lower left quadrant are mean ± SD of slope and that in the lower right quadrant is the p value when significant. (A–C) Comparison of compound addition on gene expression to the fold change observed in MDA-MB-231 breast carcinoma cells with shRNA against MRTF-A and MRTF-B; (D–F) Same analysis but compared to B16F2 melanoma cells with shRNA against MRTF-A and MRTF-B. Figure 3 Time-course of CCG-1423 effect on PC-3 cell gene expression. PC-3 cells were treated with 3 µM of CCG-1423 under serum-starved (0.5% FBS) conditions for 1, 2, 6, 12, and 24 h. Four genes: RGS4 (A); RGS7 (B); SOX9 (C); and CTGF (D), identified as metastasis candidates, were tested using qRT-PCR as described in the Materials and Methods. (A–D) Data represent mean ± SEM fold-change values (over DMSO Control) for three separate experiments. Figure 4 Concept map of gene expression changes in PC-3 cells induced by three treatments. KEGG pathways, defined by MSigDB (Broad Institute version 3.0), were examined using Gene set enrichment analysis (GSEA) and the entire microarray data set. The top 25 KEGG pathways significantly associated with changes induced by CCG-1423 (3 µM) are shown along with the related changes induced by Lat B (0.5 µM) and DRB (30 µM). All three compounds produced significant effects on genes involved in purine and pyrimidine metabolism while CCG-1423 and Lat B had more selective effects on cell cycle and DNA replication. Colors represent Z-scores with blue indicating that genes of the set were downregulated and red indicating that genes of the set were up-regulated. The darkness represents the level of significance. White means that the adjusted p value is >0.05. One of the 24-h Lat B samples was lost during processing for the microarray so there are only two replicates. Figure 5 qRT-PCR confirmation of ER Stress-related gene expression. PC-3 cells were treated with 3 µM CCG-1423 for the indicated times. RNA was isolated and qRT-PCR performed as described. Expression of (A) ATF3, (B) ATF4, and (C) CHOP (also referred to as DDIT3) is expressed as the fold change compared to the DMSO control samples (24-h with DMSO). All expression levels were normalized relative to the expression of GAPDH. microarrays-05-00013-t001_Table 1Table 1 Number of genes regulated by compounds. Treatment 2 h 24 h Stimulated Inhibited Total Stimulated Inhibited Total CCG-1423 48 1 49 837 1307 2142 Lat B 15 3 18 225 383 608 DRB 5 714 719 1058 2026 3020 * This table shows the number of genes whose expression changed 2-fold after treatment of PC-3 prostate cancer cells under serum-starved conditions (0.5% FBS) with 3 µM CCG-1423, 0.5 µM Lat B, and 50 µM DRB. * The total number of genes changes does not equal the sum of stimulated and inhibited gene totals because some genes have probesets that are stimulated and others that are inhibited. microarrays-05-00013-t002_Table 2Table 2 Comparison of genes regulated by CCG-1423 with MKL1-dependent genes from the literature. Gene Categories CCG-1423 Lat B DRB References Total Genes Regulated 2142 608 3020 FBS-MKL1-Dependent Genes (#) (NIH3T3) 5 out of 28 0 out of 28 7 out of 28 [20] FBS-MKL1-Dependent Genes (%) (NIH3T3) 17.8% 0.0% 25.0% [20] MKL1-Dependent Genes (#) (MDA-MB-231) 273 out of 1070 83 out of 1070 271 out of 1070 [22] MKL1-Dependent Genes (%) (MDA-MB-231) 25.5% 7.8% 25.3% [22] MKL1-Dependent Genes (#) (B16F2) 73 out of 323 54 out of 323 58 out of 323 [22] MKL1-Dependent Genes (%) (B16F2) 22.6% 16.7% 18.0% [22] Shown are the number and percentage of genes regulated by CCG-1423, Lat B, and DRB that share identity with genes regulated by serum (10% FBS)-induced in an MKL1-dependent manner in NIH3T3 mouse fibroblast cells [20] or those suppressed by knock-down of MTRF-A/B (i.e., MKL1/2-dependent) in aggressive MDA-MB-231 breast cancer cells or B16F2 melanoma cells [22]. microarrays-05-00013-t003_Table 3Table 3 Comparison of microarray versus qRT-PCR results. Gene Microarray qRT-PCR (Fold-change) (Fold-change) RGS4 21.7, 21.6, 19.4 * 12.4 RGS7 9.6 9.9 CTGF 0.05 0.12 SOX9 0.12, 0.13 * 0.19 PC-3 cells were treated with 3 µM of CCG-1423 under serum-starved conditions (0.5% FBS) for 24 h and gene expression was assessed with the Affymetrix array or in a quantitative real-time-PCR (qRT-PCR) reaction assay. The genes RGS4, RGS7, CTGF, and SOX9, which were identified as candidates, were confirmed using a qRT-PCR assay. The mean fold-change values (over DMSO Control) for expression of these genes (n = 3) are displayed in the table. For the Microarray data, the log2 transformed expression values were retransformed by exponentiating using 2x, where x equals the log2 expression value. Then, the fold-change values were calculated by dividing the expression values for the CCG-1423-treated samples by the expression values for the DMSO samples. * indicates mean fold-change values of different probesets for RGS4 or SOX9. ==== Refs References 1. Christofori G. New signals from the invasive front Nature 2006 441 444 450 10.1038/nature04872 16724056 2. Fidler I.J. The pathogenesis of cancer metastasis: The “seed and soil” hypothesis revisited Nat. Rev. Cancer 2003 3 453 458 10.1038/nrc1098 12778135 3. Gupta G.P. Massague J. Cancer metastasis: Building a framework Cell 2006 127 679 695 10.1016/j.cell.2006.11.001 17110329 4. Nguyen D.X. Bos P.D. Massague J. Metastasis: From dissemination to organ-specific colonization Nat. Rev. Cancer 2009 9 274 284 10.1038/nrc2622 19308067 5. Sawyer T.K. Cancer metastasis therapeutic targets and drug discovery: Emerging small-molecule protein kinase inhibitors Expert Opin. Investig. Drugs 2004 13 1 19 10.1517/13543784.13.1.1 14680449 6. Winnepenninckx V. Lazar V. Michiels S. Dessen P. Stas M. Alonso S.R. Avril M.F. Ortiz Romero P.L. Robert T. Balacescu O. Gene expression profiling of primary cutaneous melanoma and clinical outcome J. Natl. Cancer Inst. 2006 98 472 482 10.1093/jnci/djj103 16595783 7. Clark E.A. Golub T.R. Lander E.S. Hynes R.O. Genomic analysis of metastasis reveals an essential role for RhoC Nature 2000 406 532 535 10.1038/35020106 10952316 8. Yang J. Weinberg R.A. Epithelial-mesenchymal transition: At the crossroads of development and tumor metastasis Dev. Cell 2008 14 818 829 10.1016/j.devcel.2008.05.009 18539112 9. Hu G. Chong R.A. Yang Q. Wei Y. Blanco M.A. Li F. Reiss M. Au J.L. Haffty B.G. Kang Y. MTDH activation by 8q22 genomic gain promotes chemoresistance and metastasis of poor-prognosis breast cancer Cancer Cell 2009 15 9 20 10.1016/j.ccr.2008.11.013 19111877 10. Gupta G.P. Nguyen D.X. Chiang A.C. Bos P.D. Kim J.Y. Nadal C. Gomis R.R. Manova-Todorova K. Massague J. Mediators of vascular remodelling co-opted for sequential steps in lung metastasis Nature 2007 446 765 770 10.1038/nature05760 17429393 11. Padua D. Zhang X.H. Wang Q. Nadal C. Gerald W.L. Gomis R.R. Massague J. TGFβ primes breast tumors for lung metastasis seeding through angiopoietin-like 4 Cell 2008 133 66 77 10.1016/j.cell.2008.01.046 18394990 12. Kang Y. Siegel P.M. Shu W. Drobnjak M. Kakonen S.M. Cordon-Cardo C. Guise T.A. Massague J. A multigenic program mediating breast cancer metastasis to bone Cancer Cell 2003 3 537 549 10.1016/S1535-6108(03)00132-6 12842083 13. Mundy G.R. Metastasis to bone: Causes, consequences and therapeutic opportunities Nat. Rev. Cancer 2002 2 584 593 10.1038/nrc867 12154351 14. Yin J.J. Selander K. Chirgwin J.M. Dallas M. Grubbs B.G. Wieser R. Massague J. Mundy G.R. Guise T.A. TGF-β signaling blockade inhibits pthrp secretion by breast cancer cells and bone metastases development J. Clin. Investig. 1999 103 197 206 10.1172/JCI3523 9916131 15. Sahai E. Marshall C.J. Rho-GTPases and cancer Nat. Rev. Cancer 2002 2 133 142 10.1038/nrc725 12635176 16. Narumiya S. Tanji M. Ishizaki T. Rho signaling, ROCK and mDia1, in transformation, metastasis and invasion Cancer Metastasis Rev. 2009 28 65 76 10.1007/s10555-008-9170-7 19160018 17. Miralles F. Posern G. Zaromytidou A.I. Treisman R. Actin dynamics control SRF activity by regulation of its coactivator MAL Cell 2003 113 329 342 10.1016/S0092-8674(03)00278-2 12732141 18. Berenjeno I.M. Nunez F. Bustelo X.R. Transcriptomal profiling of the cellular transformation induced by Rho subfamily GTPases Oncogene 2007 26 4295 4305 10.1038/sj.onc.1210194 17213802 19. Cen B. Selvaraj A. Burgess R.C. Hitzler J.K. Ma Z. Morris S.W. Prywes R. Megakaryoblastic leukemia 1, a potent transcriptional coactivator for serum response factor (SRF), is required for serum induction of srf target genes Mol. Cell. Biol. 2003 23 6597 6608 10.1128/MCB.23.18.6597-6608.2003 12944485 20. Selvaraj A. Prywes R. Expression profiling of serum inducible genes identifies a subset of SRF target genes that are MKL dependent BMC Mol. Biol. 2004 5 10.1186/1471-2199-5-13 15329155 21. Morita T. Mayanagi T. Sobue K. Dual roles of myocardin-related transcription factors in epithelial mesenchymal transition via slug induction and actin remodeling J. Cell Biol. 2007 179 1027 1042 10.1083/jcb.200708174 18056415 22. Medjkane S. Perez-Sanchez C. Gaggioli C. Sahai E. Treisman R. Myocardin-related transcription factors and SRF are required for cytoskeletal dynamics and experimental metastasis Nat. Cell Biol. 2009 11 257 268 10.1038/ncb1833 19198601 23. Evelyn C.R. Wade S.M. Wang Q. Wu M. Iniguez-Lluhi J.A. Merajver S.D. Neubig R.R. CCG-1423: A small-molecule inhibitor of rhoa transcriptional signaling Mol. Cancer Ther. 2007 6 2249 2260 10.1158/1535-7163.MCT-06-0782 17699722 24. Yao H. Dashner E.J. van Golen C.M. van Golen K.L. RhoC GTPase is required for PC-3 prostate cancer cell invasion but not motility Oncogene 2006 25 2285 2296 10.1038/sj.onc.1209260 16314838 25. Irizarry R.A. Hobbs B. Collin F. Beazer-Barclay Y.D. Antonellis K.J. Scherf U. Speed T.P. Exploration, normalization, and summaries of high density oligonucleotide array probe level data Biostatistics 2003 4 249 264 10.1093/biostatistics/4.2.249 12925520 26. Kim S.Y. Volsky D.J. Page: Parametric analysis of gene set enrichment BMC Bioinform. 2005 6 10.1186/1471-2105-6-144 27. Huang da W. Sherman B.T. Lempicki R.A. Systematic and integrative analysis of large gene lists using david bioinformatics resources Nat. Protoc. 2009 4 44 57 10.1038/nprot.2008.211 19131956 28. Wang X. Adams L.D. Pabon L.M. Mahoney W.M. Jr. Beaudry D. Gunaje J. Geary R.L. Deblois D. Schwartz S.M. RGS5, RGS4, and RGS2 expression and aortic contractibility are dynamically co-regulated during aortic banding-induced hypertrophy J. Mol. Cell. Cardiol. 2008 44 539 550 10.1016/j.yjmcc.2007.11.019 18207159 29. Chertkow Y. Weinreb O. Youdim M.B. Silver H. Gene expression changes in peripheral mononuclear cells from schizophrenic patients treated with a combination of antipsychotic with fluvoxamine Prog. Neuropsychopharmacol. Biol. Psychiatry 2007 31 1356 1362 10.1016/j.pnpbp.2007.04.016 17662512 30. Li M.H. Sanchez T. Pappalardo A. Lynch K.R. Hla T. Ferrer F. Induction of antiproliferative connective tissue growth factor expression in wilms’ tumor cells by sphingosine-1-phosphate receptor 2 Mol. Cancer Res. 2008 6 1649 1656 10.1158/1541-7786.MCR-07-2048 18922980 31. Vincourt J.B. Vignaud J.M. Lionneton F. Sirveaux F. Kawaki H. Marchal S. Lomazzi S. Plenat F. Guillemin F. Netter P. Increased expression of matrilin-3 not only in osteoarthritic articular cartilage but also in cartilage-forming tumors, and down-regulation of SOX9 via epidermal growth factor domain 1-dependent signaling Arthritis Rheumatol. 2008 58 2798 2808 10.1002/art.23761 18759284 32. Hurst J.H. Mendpara N. Hooks S.B. Regulator of g-protein signalling expression and function in ovarian cancer cell lines Cell. Mol. Biol. Lett. 2009 14 153 174 10.2478/s11658-008-0040-7 18979070 33. Xie Y. Wolff D.W. Wei T. Wang B. Deng C. Kirui J.K. Jiang H. Qin J. Abel P.W. Tu Y. Breast cancer migration and invasion depend on proteasome degradation of regulator of G-protein signaling 4 Cancer Res. 2009 69 5743 5751 10.1158/0008-5472.CAN-08-3564 19549919 34. Ram P.T. Horvath C.M. Iyengar R. STAT3-mediated transformation of NIH-3T3 cells by the constitutively active q205l GALPHAO protein Science 2000 287 142 144 10.1126/science.287.5450.142 10615050 35. Champion Chip Transcription Factor Search Portal Available online: http://www.sabiosciences.com/chipqpcrsearch.php?app=TFBS (accessed on 27 March 2016) 36. Mullis T.C. Tang X. Chong K.T. Expression of connective tissue growth factor (CTGF/CCN2) in head and neck squamous cell carcinoma J. Clin. Pathol. 2008 61 606 610 10.1136/jcp.2007.052795 18441156 37. Bennewith K.L. Huang X. Ham C.M. Graves E.E. Erler J.T. Kambham N. Feazell J. Yang G.P. Koong A. Giaccia A.J. The role of tumor cell-derived connective tissue growth factor (CTGF/CCN2) in pancreatic tumor growth Cancer Res. 2009 69 775 784 10.1158/0008-5472.CAN-08-0987 19179545 38. Liu L. Li Z. Feng G. You W. Li J. Expression of connective tissue growth factor is in agreement with the expression of VEGF, VEGF-C, -D and associated with shorter survival in gastric cancer Pathol. Int. 2007 57 712 718 10.1111/j.1440-1827.2007.02162.x 17922682 39. Boag J.M. Beesley A.H. Firth M.J. Freitas J.R. Ford J. Brigstock D.R. de Klerk N.H. Kees U.R. High expression of connective tissue growth factor in PRE-B acute lymphoblastic leukaemia Br. J. Haematol. 2007 138 740 748 10.1111/j.1365-2141.2007.06739.x 17760805 40. Lu B. Fang Y. Xu J. Wang L. Xu F. Xu E. Huang Q. Lai M. Analysis of SOX9 expression in colorectal cancer Am. J. Clin. Pathol. 2008 130 897 904 10.1309/AJCPW1W8GJBQGCNI 19019766 41. Wang H. McKnight N.C. Zhang T. Lu M.L. Balk S.P. Yuan X. SOX9 is expressed in normal prostate basal cells and regulates androgen receptor expression in prostate cancer cells Cancer Res. 2007 67 528 536 10.1158/0008-5472.CAN-06-1672 17234760 42. Wang H. Leav I. Ibaragi S. Wegner M. Hu G.F. Lu M.L. Balk S.P. Yuan X. SOX9 is expressed in human fetal prostate epithelium and enhances prostate cancer invasion Cancer Res. 2008 68 1625 1630 10.1158/0008-5472.CAN-07-5915 18339840 43. Dennis G. Jr. Sherman B.T. Hosack D.A. Yang J. Gao W. Lane H.C. Lempicki R.A. David: Database for annotation, visualization, and integrated discovery Genome Biol. 2003 4 P3 10.1186/gb-2003-4-5-p3 12734009 44. Zohn I.M. Campbell S.L. Khosravi-Far R. Rossman K.L. Der C.J. Rho family proteins and RAS transformation: The rhoad less traveled gets congested Oncogene 1998 17 1415 1438 10.1038/sj.onc.1202181 9779988 45. Evelyn C.R. Bell J.L. Ryu J.G. Wade S.M. Kocab A. Harzdorf N.L. Hollis Showalter H.D. Neubig R.R. Larsen S.D. Design, synthesis and prostate cancer cell-based studies of analogs of the Rho/MKL1 transcriptional pathway inhibitor, CCG-1423 Bioorg. Med. Chem. Lett. 2010 20 665 672 10.1016/j.bmcl.2009.11.056 19963382 46. Sequeira L. Dubyk C.W. Riesenberger T.A. Cooper C.R. van Golen K.L. Rho gtpases in PC-3 prostate cancer cell morphology, invasion and tumor cell diapedesis Clin. Exp. Metastasis 2008 25 569 579 10.1007/s10585-008-9173-3 18461284 47. Chen H.Z. Tsai S.Y. Leone G. Emerging roles of E2FS in cancer: An exit from cell cycle control Nat. Rev. Cancer 2009 9 785 797 10.1038/nrc2696 19851314 48. Kauffmann A. Rosselli F. Lazar V. Winnepenninckx V. Mansuet-Lupo A. Dessen P. van den Oord J.J. Spatz A. Sarasin A. High expression of DNA repair pathways is associated with metastasis in melanoma patients Oncogene 2008 27 565 573 10.1038/sj.onc.1210700 17891185 49. Van Golen K.L. Wu Z.F. Qiao X.T. Bao L.W. Merajver S.D. RhoC GTPase, a novel transforming oncogene for human mammary epithelial cells that partially recapitulates the inflammatory breast cancer phenotype Cancer Res. 2000 60 5832 5838 11059780 50. Hakem A. Sanchez-Sweatman O. You-Ten A. Duncan G. Wakeham A. Khokha R. Mak T.W. RhoC is dispensable for embryogenesis and tumor initiation but essential for metastasis Genes Dev. 2005 19 1974 1979 10.1101/gad.1310805 16107613

PMC005xxxxxx/PMC5003490.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020014microarrays-05-00014ArticleRetrospective Proteomic Analysis of Cellular Immune Responses and Protective Correlates of p24 Vaccination in an HIV Elite Controller Using Antibody Arrays Perera Suneth S. 1Wang Bin 1Damian Arturo 2Dyer Wayne 3Zhou Li 1Conceicao Viviane 1Saksena Nitin K. 1*Borrebaeck Carl A. K. Academic EditorWingren Christer Academic EditorAndreasson Ulrika Academic Editor1 Department of Medicine, University of Sydney, Sydney 2000, Australia; perera.suneth@gmail.com (S.S.P.); bin.wang1@unsw.edu.au (B.W.); li.zhou@health.nsw.gov.au (L.Z.); vickanc@hotmail.com (V.C.)2 Department of Cytogenetics, Children’s Hospital at Westmead, Sydney 2000, Australia; Arturo.damian@health.nsw.gov.au3 Australian Red Cross Blood Service, 17 O’Riordan Street, Alexandria NSW 2015 and School of Medical Sciences, (Faculty of Medicine) University of Sydney, Sydney 2000, Australia; Wayne.dyer@sydney.edu.au* Correspondence: nitin.saksena@sydney.edu.au or nitin.saksena@bigpond.com; Tel.: +61-2-88243571 or +61-43196015802 6 2016 6 2016 5 2 1423 10 2015 25 1 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Background: HIV p24 is an extracellular HIV antigen involved in viral replication. Falling p24 antibody responses are associated with clinical disease progression and their preservation with non-progressive disease. Stimulation of p24 antibody production by immunization to delay progression was the basis of discontinued p24 vaccine. We studied a therapy-naive HIV+ man from Sydney, Australia, infected in 1988. He received the HIV-p24-virus like particle (VLP) vaccine in 1993, and continues to show vigorous p24 antigen responses (>4% p24-specific CD4+ T cells), coupled with undetectable plasma viremia. We defined immune-protective correlates of p24 vaccination at the proteomic level through parallel retrospective analysis of cellular immune responses to p24 antigen in CD4+ and CD8+ T cells and CD14+ monocytes at viremic and aviremic phases using antibody-array. We found statistically significant coordinated up-regulation by all three cell-types with high fold-changes in fractalkine, ITAC, IGFBP-2, and MIP-1α in the aviremic phase. TECK and TRAIL-R4 were down-regulated in the viremic phase and up-regulated in the aviremic phase. The up-regulation of fractalkine in all three cell-types coincided with protective effect, whereas the dysfunction in anti-apoptotic chemokines with the loss of immune function. This study highlights the fact that induction of HIV-1-specific helper cells together with coordinated cellular immune response (p < 0.001) might be important in immunotherapeutic interventions and HIV vaccine development. Antibody microarrayHIVimmune responseschemokinescytokinesAcquired immune deficiency syndrome (AIDS)p24 vaccinemonocyteCD4+ TCD8+ T cells ==== Body 1. Introduction HIV-1 p24 response is known to occur in HIV-infected individuals, but it wanes with time in concomitance with viremia [1,2]. In contrast, this response is, more intact in aviremic therapy-naïve HIV+ long-term non-progressors [1]. Taking this into account, it is believed that therapeutic immunization HIV-1 p24 antigen in HIV infection will induce HIV-specific immune responses, which will be durable and capable of controlling HIV disease progression. Moreover, it is known that cellular immune responses against HIV Gag capsid protein p24 are durable and robust and therefore important at various stages of HIV disease. Gag-specific T-cell responses associate with favorable clinical course [3,4,5,6]. However, the correlates of this protective immunity remain to be elucidated. As waning p24 antibody responses are associated with clinical disease progression [1], stimulating p24 antibody production with the use of therapeutic immunization may delay clinical progression. This was the basis of a p24 vaccination trial for HIV-infected individuals initiated in the early 1990s, but the vaccine trial was inconclusive [7]. Therapeutic vaccination has been proposed as a mechanism for stimulating the immune response in patients with chronic HIV infection [8]. Studies suggest that immunization with HIV antigens can delay CD4+ T cell decline and lower the rate of viral DNA increase in asymptomatic HIV-1 infected people [9]. However, other studies have failed to show any significant improvement in surrogate markers with the use of a therapeutic vaccination [10,11]. We recently found an HIV-infected, therapy-naïve non-progressor, who was enrolled in the HIV p24 VLP vaccine trial in 1993 in Australia. Patient still maintains CD4+ T cell counts (700/µL blood), CD8+ T cell counts (1400/µL blood), <20 copies of HIV RNA/mL plasma (January 2013) and harbors strong HIV gag-p24 specific proliferative responses, with >4% of antigen-specific cells [12,13] after 20 years of HIV-p24 vaccination, but it is unclear what his p24 responses were prior to vaccination in 1993. Here we describe cellular responses to p24 at the protein level in CD4+, CD8+ T cells and monocytes of this p24-vaccinated individual at the level of 274 cytokines, chemokines, their receptors, ligands, and associating proteins in order to define protective correlates of p24 vaccine. Throughout the course of HIV infection, the study individual has been therapy naïve with below detectable levels of HIV RNA in plasma until 2011. In a rare clinical event, in mid-2011, the study individual experienced a viremia blip showing 563 copies of HIV RNA/mL plasma, which naturally subsided fairly quickly to below detectable levels. We have analyzed p24 antigen responses at both viremic and aviremic phases of this individual collected in 2011. To our knowledge this study shows, for the first time, the protective immune correlates of HIV-p24 vaccination at the protein level in a rare therapy naive HIV+ non-progressor, who may have responded to the p24 vaccine. Contrary to these previous studies, which focused on a single arm of the immune response, this study simultaneously analyzed CD4+, CD8+ T cells, and CD14+ monocytes. Our data are unique in showing a detailed snapshot of the immune correlates functioning during viremic and aviremic phases of this p24-vaccinated individual with a clear distinction between the cell types governing each of these phases of HIV disease and the vital role of coordinated immune response of all three cell types in HIV disease and its control. 2. Methods and Materials 2.1. Clinical History of the Study Subject The blood samples were obtained from a 64-year-old male, an elite controller who was presumed infected with HIV in December 1988 through homosexual transmission from his partner, who died of AIDS on 22 January 1993. He was diagnosed seropositive by both ELISA and Western blot in 1993. The study subject has since remained asymptomatic, and plasma viral loads have remained undetectable (<50 copies/mL) since the first sample (1993) was tested. In May 2011 the subject had a transient viral blip that lasted for 4–5 months, where his viral load in plasma was recorded at 500–600 copies/mL. We termed this phase the viremic phase and 70 mL of patient blood was collected in 10 mL EDTA tubes on 18 May 2011 and on 7 June 2011. After this, the subject fully recovered and the viral load was below detectable levels (BDL) along with normal CD4+ and CD8+ T cell counts. We termed this the aviremic phase and collected 70 mL of blood on 7 November 2011. This work was approved by the Human Ethics Committee of Westmead Hospital. Figure 1 shows the summary of the study subject’s CD4+ and CD8+ T cell counts over time in relation to plasma viremia in the absence of HAART treatment. Previously this patient has undergone several therapeutic trials. He was enrolled in a trial of Imiquimod (an IFN-α inducing agent) during 1992 [14]. He also received two sessions of whole-body 434-MHz microwave therapy in April and August 1992, in addition to the p24 virus-like particle (p24VLP) vaccine trial between January and June 1993–1994 [15]. The p24 vaccine trial was inconclusive as the patient developed adverse effects to Azidothymidine (AZT) mono-therapy and only received two of the six scheduled vaccine doses. The patient underwent further two courses of microwave therapy in 1994. During his visit to the clinic, a viral blip was discovered on 18 May 2011, which coincided with gallbladder (GB) inflammation; this sample was not analyzed in the study. Only upon monitoring inflammation and termination of treatment with steroids was another sample collected on 7 June 2011, at which time GB inflammation had subsided completely; the viral blip persisted in this sample. This was deemed appropriate for the protein analyses. Following this collection, the viral blip subsided completely and the patient has maintained below detectable levels of viremia and healthy lymphocyte counts subsequent and prior to the event in 2011 [12]. 2.2. PBMCs Separation and Sample Preparation Whole peripheral blood mononuclear cells (PBMCs) were extracted using Ficoll separation method. PBMCs were washed once with PBS, pelleted and were divided into two fractions. This step was repeated for both the viremic and the aviremic phases individually in order to assay p24 antigen-stimulated and un-stimulated fractions. Equal amount of cells were re-suspended for each fraction in RPMI medium 1640 containing 10% FBS in a non-adherent “Nunc® dishes, low cell binding” to yield the maximum amount of cells. One fraction of PBMCs was stimulated with HIV capsid protein p24 at 2 ng/mL. This is the standard concentration used for assaying IFNγ production using Enzyme-linked immuno-sorbent spot (ELISPOT) assay. Both the un-stimulated and stimulated fractions were incubated overnight at 37 °C with 5% CO2. 2.3. Macs Column Separation of CD4+, CD8+ T Cells and Monocytes Following overnight stimulation (12 h) of patient PBMCs from p24-stimulated and un-stimulated fractions were extracted by gently pipetting up and down to detach the cells from the non-adherent plate into 15 mL BD falcon tube. PBMCs were centrifuged at 1000 g to remove cell debris. Cell pellet was further washed twice with PBS to remove excess antigen. After aspirating all the PBS, the cell pellet was further washed with 10 mL of Magnetic cell-sorting (MACS) buffer (50 mL PBS + 250 µL of Fetal bovine Serum (FBS) + 200 µL of 2 mM EDTA) by centrifuging at 300× g for 10 min at 4 °C. Cells were washed with PBS in order remove any inherent cytokine/chemokine expression. A baseline was obtained using the un-stimulated fraction as both stimulated and un-stimulated cells were conditioned using the same method. MACS® Column Technology was used for separating PBMCs into different cell subsets. Using MACS®MicroBeads CD14+ monocytes, CD4+ helper T cells, and CD8+ cytotoxic T cells were extracted according to the manufacturer’s specifications using the positive selection Magnet Activated Cell Sorting method (MACS), Miltenyi Biotech (Marburg, Germany). CD14+ monocytes were extracted first to remove the by CD4+CD14+ T cell sub-population, followed by CD4+CD14− helper T cells and CD8+ cytotoxic T cells. As per instructions, PBMC pellet was washed with MACS buffer and re-suspended in 80 µL of MACS buffer and 20 µL CD14+ beads per 107 total cells. Cells were incubated for 15 min at 4 °C–8 °C re-suspended in 2 mL MACS buffer and spun at 300× g for 10 min to wash of the unbound/excess beads. Supernatant was pipetted completely and cells were re-suspended in 500 µL of MACS buffer. Magnetic separation was carried out with MS columns. Extracted cells were centrifuged for 10 min at 300× g to obtain a purified cell pellet and further washed with PBS. This process was repeated in the order of CD14+, CD4+CD14− and CD8+ to extract cells. Flow cytometric verification of cell purity of CD4+, CD8+ T cells, and monocytes was evaluated using flow cytometry, as previously described by us [16]. 2.4. Protein Extraction Whole cellular proteins were extracted from both p24-stimulated and un-stimulated fractions of purified CD14+ monocytes, CD4+ helper T cells and CD8+ cytotoxic T-cells from each phase as per RayBiotech© (Norcross, GA, USA) instructions. The 2×RayBio®Cell Lysis Buffer was diluted with H2O. 1:1 dilution of 150µL of 2× Raybio cell lyses buffer was used to lyse equal input of cells (5 × 106) throughout our assays for protein extraction. One percent protease inhibitor (Sigma Aldrich, St Louis, MO, USA) was also added to prevent protein from degrading. Each of the samples was homogenized by sonication for one minute and the cell debris was removed by centrifuging at 10,000× g for 5 min. Lysates were stored at −80°C and were used within a week. 2.5. Quantification Using the BioRad DC-Protein Assay A Bio-Rad DC-protein assay kit and the reagent pack (Bio-Rad, Gladesville, NSW, Australia) were used to quantify proteins as per manufacturer’s instructions. Five microliters of sample volume was used and 0.25, 0.5, 0.75, 1, and 1.5 mg/mL standards were used to get the standard curve. Bio RAD SmartSpec Plus spectrometer was used in measuring the absorbance at 750 nm. Final total protein concentrations in lysates of each cell type ranged from 1000 to1500 µg/mL, which were within the required analytical range for the protein array assay. 2.6. RayBio® Combination of Human Cytokine Antibody Array G Series Proteomic analysis of 274 cytokines, chemokines, their receptors, ligands and associating proteins from the CD4+, CD8+ T cells, and monocytes was carried out as per instructions from RayBio® Combination of Human Cytokine Antibody Array G Series kit (Ref: AAH-CYT-G4000-8), which was purchased from RayBiotech®. Glass slides were blocked with blocking buffer for 30 min. 50 µL of sample mixture (27 µg protein) from each of unstimulated and p24-stimulated fractions from each of the viremic and non-viremic phases were added to the sample wells along with 1 μL of Internal control and incubated overnight at 4 °C. After washing with buffers I and II, slides were incubated with biotin conjugated antibodies for 2 h at room temperature. Slides were washed and incubated for a further 1 h at room temperature in the dark with 70 µL of fluorescent dye conjugated Strepavidin. Slides were washed and dried by centrifugation at 1000 rpm for 3 min. Slides were scanned using an Axon Genepix scanner using the cy3 channel or at excitation frequency of 532 nm. 2.7. Data Normalization and Analysis After normalizing the expression values as per instructions from RayBiotech© using the positive controls, the fold-changes (FC) were obtained by comparing the un-stimulated cell fraction against the p24-stimulated cell fraction. This step was repeated for both phases and across all three cell-types to obtain the fold-change comparisons between the two phases. As per instructions from RayBiotech, data were processed in Microsoft Excel 2007, where background was subtracted for the expression value to obtain the true expression. These protein expression values were then normalized according to the RayBiotech guidelines using the positive control from the reference array as the standard. The following equation was used to normalize the data between the two phases: X(Ny) = X(y) × P1/P(y), (1) where P1 = the average signal expression value of the positive control spots on the reference array; P(y) = the average signal expression value of the positive control spots on Array y; X(y) = the signal expression value for a particular spot on Array for sample “y”; and X(Ny) = the normalized value for that particular spot “X” on Array for sample “y”. The fold-changes were obtained by dividing the expression values of the un-stimulated fraction by the p24-stimulated fraction for each corresponding protein. These fold changes were converted to log scale and a negative sign was given to depict a down-regulation. 2.8. Statistical Analysis Using K-Way and Log Linear Models to Discover K–Gene Interactions We used a combination of association rule mining and log-linear modeling to discover k–gene interactions. Using this technique we could discover interactions among k-genes that cannot be explained by the combined effects of any of the subsets of those genes. Our results reveal some previously unknown associations between cell-types and proteins that point to solid biological explanations. Further, the Log-linear analysis technique was used to examine the relationship between more than two categorical variables—in this case three-cell types CD4+, CD8+ T cells, and CD14+ monocytes and their encoded proteins. The technique was used for both hypothesis testing and possible model building. A Pearson’s chi-square test was used instead of log-linear analysis, but this only allowed for two of the variables to be compared at a time [17]. Log-linear analysis uses a likelihood ratio statistic X2: [18] that has an approximate chi-square distribution when the sample size is large: (2) x2=2∑OijlnOijEij where ln = natural algorithm; Oij = observed frequency in cellij (i = row and j = column); Eij = expected frequency in cellij; and X2 = the deviance for the model [19]. 2.9. Follow-up Tests Once the model of best fit was determined, the highest-order interaction was examined by conducting chi-square analyses at different levels of one of the variables. For the chi-square analyses, it needed to break the model down into a 2 × 2 or 2 × 1 contingency table. For example, if one is examining the relationship among four variables, and the model of best fit contained one of the three-way interactions, one would examine its simple two-way interactions at different levels of the third variable. The results of log linear analyses are shown in Table 1 below. 3. Results Our results are unique in showing the first snapshot of the length and breadth of cellular p24-antigen responses at the cellular and proteomic level for 274 cytokines, chemokines, their receptors, ligands, and associating proteins simultaneously in CD4+, CD8+ T cells, and monocytes of an HIV+ Long-term non-progressors (LTNP) at the viremic and aviremic phases after 20 years of receiving the HIVp24 vaccination. Proliferative responses to p24 antigen are still robust in this individual, as per our previous report [1,12]. In the current study, the data show evidence not only for the differential regulation of cytokines/chemokines and associating proteins in diverse leukocytes at viremic and aviremic phases in response to p24 antigen, but also their differential expression (DE) within the same cell type at these two disease phases in this individual. As per manufacturer’s instructions, the fold-changes between >1.5 and <−1.54 were deemed non-significant. At the individual cell level, using this cutoff, the CD4+T cells yielded 59, monocytes 53, and the CD8+ T cells 52 DE proteins, respectively, from a total of 274 proteins analyzed on the protein array plate. 3.1. Overlapping Proteins that Functionally Bind CD4+, CD8+ T Cells,and CD14+Monocytes: Evidence for Crosstalk between Cell Types Proteins that overlapped between all three cell-types were visualized first because of their coordinated response to p24 antigen, which has vital functional significance at different stages of HIV infection and secondly because a coordinated cellular response has not been previously visualized at the proteomic level. A combined analysis of all three cell-types showed 68 significant DE proteins in the aviremic phase and 62 in the viremic phase (Figure 2A,B). Of 62 DE proteins at the viremic phase, six proteins—Dtk, MIP-1α, MMP-13, PIGF, TECK, and TRAIL R4—were overlapping between the three cell-types. Of these six proteins, four—Dtk, MIP-1α, MMP-13, and PIGF—expressed the same trend in two of the cell types but a reverse trend was observed in the other. However, two proteins, TECK and TRAIL-R4, showed similar expression trend in all three cell-types in this phase (Figure 2A,B). These critical differences in the expression of these proteins clearly segregated the viremic and aviremic phases, notably, the high-fold change (FC) in Dtk (17 FC), MMP13 (10 FC), and PIGF (−17 FC) in monocytes during the viremic phase. Moreover, the Dtk (−15-fold) and MIP-1α (nine-fold) also showed high expression in CD8+ T cells at viremic phase. This differential regulation of the same proteins in diverse cell types at the viremic phase is further suggestive of possible cross-talk between cell types in modulating the expression of these proteins at a given disease phase. Full summary of the 62 DE proteins in the viremic phase is given in Supplementary Table S1. 3.2. HIV Aviremic Phase At the aviremic phase, based on all three cell-types, 14 proteins—Eotaxin-2, Fractalkine, FSH, Cathepsin-S, IGFBP-2 and 4, EDA-A2, IGFBP-4, I-TAC, CCL14a, M-CSF, MIP-1α, TGF-β-3, and VEGF—were differentially expressed. Of these 14 proteins, five in particular (Fractalkine, Cathepsin-S, CC14a, and IGFBP-2 and 4) showed systematic discriminatory expression trends between viremic and aviremic phases. Fractalkine, in particular showed consistently high fold-changes across all three cell-types, which has not been reported previously; these differences between viremic and aviremic phases further stress their functional significance in HIV disease. In contrast, the other nine proteins, although segregated between the two disease phases, appear to be regulated differentially between cell types. Chemokine MIP1-α was the only protein overlapping for all three cell-types and also for both phases of disease; it was up-regulated in all three cell-types with significant fold-change, except in monocytes during the viremic phase. Its highest up-regulation (9 FC) was seen in the CD8+ T cells during viremia. MIP-1α and MIP-1β have been shown to be increased in the LTNPs and are ligands for the CCR5 receptor [20]. I-TAC was up-regulated in all three cell-types but was unique in its high expression in CD4+ T (7 FC) and CD8+ T cells (13 FC) during aviremic phase, but its levels of expression were comparable in monocytes between the two phases. MCSF was significant for aviremia for all three cell-types, but was highly up-regulated (11 FC) uniquely in CD8+ T cells during aviremia. The notable feature of these analyses was that during the aviremic phase the majority of responses emanated from the CD8+ T cells, followed by CD4+ T cells, which was evident from high fold-changes in the expression of seven of 11 overlapping proteins in this phase. This feature was consistent with the previously described role of CD8+ T cells in both non-progressive HIV disease and also in containing viremia. What is more significant about these proteins specific for viremic and aviremic phases is that such proteins have not been described in combination as shown here, and they may hold greater significance in guiding these disease phases (see the Supplementary Table S2 for the summary of 68 DE proteins at the aviremic phase). 3.3. Cell-Specific Differentially Expressed Proteins during Viremic and Aviremic Phases 3.3.1. Proteins Segregating Viremic and Aviremic Phases Based on CD4+ T Cells Apart from the coordinated response all three cell-types mounted (as discussed above), we also observed cell-specific responses discriminating between the viremic and aviremic phases of HIV infection. For instance, in CD4+ T cells 59DE proteins showed significant fold-changes. Forty DE proteins were seen in the viremic phase, whereas 43 DE proteins were specific to the aviremic phase (Figure 2A). Of the 40 DE proteins for the viremic phase, DTK (−9.966 FC), Endoglin (−5.557), ICAM 3 (12.185), PIGF (4.513), IL21-R (−15.041), LIF (9.118), Osteoprotegerin (−15.166), TECK (−8.937), TGF-β3 (31.911), and VEGF-D (−29.987) showed high fold-change expression values. Whereas, of the 43 DE proteins for the aviremic phase, DTK (−4.916), Eotaxin-2 (−5.706), Fractalkine (14.720), FSH (6.718), IGFBP-2 (8.358), IL-4 (−7.651), I-TAC (7.138), and XEDAR (8.613) showed high fold-changes in expression (Figure 2A). Another notable feature was the mirror regulation of certain cytokines, such as the up-regulation of EDA-A2, FGF-7, IL-4, IL-1β, IGFBP-4, and ALCAM proteins during viremic phase in the CD4+ T cells, and their down-regulation in the aviremic phase, suggesting their strong functional significance. Similarly, GRO, IGFBP-3, Osteoprotegerin, TECK, TIMP-1, CCL-28, BTC, VEGF-D, VEGF, and TRAIL R4 were up-regulated in the aviremic CD4+ T cells and were down-regulated in the viremic phase for the same cell type (Figure 2A,B). Apart from this, some proteins were also significantly expressed in both phases in CD4+ T cells, displaying similar expression trends, but the differences in fold-changes of these proteins were actually robust in discriminating between viremic and aviremic phases, as evident from high expression of Dtk, Endoglin, and PIGF during the viremic phase and FSH, MIP-1α, MIP-1β, and TGF-β3 during the aviremic phase (Figure 2A,B). In the CD4+ T cells, the 59DE proteins showed significantly expressed fold-changes in both phases. Of 59 DE proteins, 40 were expressed in the viremic phase and 43 in the non-viremic phase (Figure 3A). Among the 40 DE proteins in the viremic phase ICAM 3 (12.185), IL21-R (−15.041), LIF (9.118), TECK (−8.937) and TGF-α3 (31.911) had very high fold-change expression values. Similarly, among the 43 fold-change expressions in the aviremic phase Eotaxin-2 (−5.706), Fractalkine (14.720), IGFBP-2 (8.358), MIP1-δ (41.3), I-TAC (7.138), and XEDAR (8.613) had very high fold- change expression values. Thus, there was a clear distinction between DE proteins expressed in aviremic vs. the viremic phase in CD4+ T cells—an observation not previously reported. Again the notable feature was the mirror regulation of certain cytokines, such as the up-regulation of EDA-A2, FGF-7, IL-4, IL-1β, IGFBP-4, and ALCAM proteins during viremic phase in the CD4+ T cells, and their down-regulation in the aviremic phase (Figure 3A). Similarly, GRO, IGFBP-3, Osteoprotegerin, TECK, TIMP-1, CCL-28, BTC, VEGF-D, VEGF, and TRAIL R4 were up-regulated in aviremic CD4+ T cells and down-regulated in the viremic phase (Figure 3A). This reverse trend of expression was also observed in both CD14+ monocytes and CD8+ T cells, supporting their functional significance of mirror regulation of these cytokines/chemokines in modulating viremic and aviremic phases of HIV disease. Also noteworthy is that some fold-changes in proteins were significantly expressed in both phases with a similar expression trend; however, this expression was very high only in one phase and lower in the other phase. This was evident with the expressions of Dtk, Endoglin, and PIGF with high fold-changes during the viremic phase and FSH, MIP-1α, MIP-1β, and TGF-β3 during the aviremic phase (Supplementary Table S3). 3.3.2. Proteins Segregating Viremic and Aviremic Phases Based on CD14+ Monocytes A total of 53 DE proteins were expressed in monocytes, with 40 in the viremic phase and 36 in the aviremic phase (Figure 3B). The most prominent proteins expressed in the viremic phase were DAN (−3.444), Dtk (37.758), EGF-R (−3.409), IGFBP-3(4.399), MMP-13 (10.3250), PIGF (−17.390), SCF R (8.777), TECK (−8.183), TGF-α (8.389), TIMP-1 (−6.598), TRAIL R3 (−3.902), and VEGF-D (−55.239). In contrast, the highly expressed proteins in the aviremic phase were fractalkine (11.945), FSH (7.178), GCP-2 (6.344), GM-CSF (4.715), CathepsinS (−7.38), and CCL14a (−5.04). The CD14+ monocytes showed up-regulation of proteins MMP-7, EPCAM, and IGFBP-2 in viremic phase and up-regulation of MIP-1δ, PDGF-BB, MCP-4, and MIP-1α in the aviremic phase. In addition to this, certain proteins were significantly expressed in both phases with a similar trend, although the degree of expression was higher only in one phase. This was observed with the expression of TIMP-1, ALCAM, ENA-78, Endoglin, GITR, GRO, IL-13 R α1, IP-10, I-TAC, and uPAR in the viremic phase and NAP-2, TACE, and IL-17c in the aviremic phase (Supplementary Table S4). 3.3.3. Proteins Segregating Viremic and Aviremic Phases Based on CD8+ T Cells Unlike the two other cell types, the expression fold-change in CD8+ T cells was more dominant in the aviremic phase. Of the 53 differentially-expressed proteins, 25 were expressed in the viremic phase and 42 in the non-viremic phase (Figure 3C and Supplementary Table S5). For instance, the Dtk (−15.493), EGF-R (7.874), FSH (4.195), and MIP-1 δ (−5.584) were notable in the viremic phase with their high-fold change expressions, while ACE-2 (5.191), BMP-4 (−7.524), CCL14a (18.690), EDA-A2 (−4.202), EGF (−33.681), FGF-7 (20.007), FLRG (−9.492), Fractalkine (4.651), GCP-2 (−4.230), GM-CSF (5.085), IGFBP-4 (−7.524), I-TAC (13.259), MCP-4 (6.9474), M-CSF (11.791), NT-4 (−9.349), PARC (−19.778), PDGF-BB (−10.598), SCF (36.281), TACE (−86.264), TGF-β 3 (−3.977), and VEGF (−5.110) were specific to the aviremic phase. The reverse trend was observed in CD8+ cytotoxic T-cells with MIP-1β, HGF, Leptin R, XEDAR, and IL-13Rα, being up-regulated in the viremic phase and down-regulated in the aviremic phase. Moreover, the mirror regulation was also noted for the proteins MIP-3g, Fcr RIIB/C, MMP-10 and ErbB2, with up-regulation during the aviremic phase (Figure 3C). 3.4. CD4+ T Cells and Monocytes Are Key Players during Viremia, Whereas CD4+, CD8+ T Cells and Monocytes Are Equal Partners during the Aviremic Phase In this analysis, we teased out where the most immune responses during viremic and aviremic phases are emanating from. For this, we compared side-by-side all three cell types for the differentially-expressed proteins and calculated the percentages of proteins representing each cell type at a given phase of each cell type. These analyses showed that CD4+ T cells and monocytes mounted the greatest response during viremia, with 40/62 DE proteins (68%) each for monocytes and CD4+ T cells involved in this response, as opposed to 25/62 (40%) of DE proteins from CD8+ T cells. This suggests that although all three cell-types have a significant role during viremia, the dominant partners were the CD4+ T cells and monocytes. In contrast, during aviremia the immune responses mounted by CD8+ T cells were much higher than observed in viremic phase, consistent with their cytotoxic role and in non-progressive HIV disease, with 42/68 DE proteins (62%) involved in this response. But when CD8+ T cells were compared against CD4+ T cells (63%) and monocytes (53%), it was apparent that all three cell-types were dominant partners during aviremia, which has never been shown before at the protein level and signifies functional cross-talk between cell-types in controlling viremia (Supplementary Table S6). 3.5. Q-RTPCR Validation of Proteomic Expression: High Concordance between Protein and Genomic Expression Trends In order to validate the functional value of these proteins, we attempted a reverse validation (protein to gene). Using q-RTPCR, we performed genomic validation of six DE proteins—MIP-1β, Eotaxin-2, IP-10, I-TAC, MIG, and MCP-2—by designing appropriate gene-specific primers. We observed a consistent trend of expression between our proteins and the genomic data (Table 2) with the exception of protein MIP-3β, which showed a reverse expression trend consistent with previous observations that protein expression does not always correlate with gene expression, as some genes are only transcriptionally active or vice versa. This in no way changes the interpretation of this study. 3.6. K-Way and Log-Linear Model Analysis Reveal Joint Expression of Genes from All Three Cell Types Significant in Two- and Three-Way Interactions The k-way and log linear model analysis solidifies one critical observation: that proteins tend to express in pairs or all three of CD8, CD4, or monocyte rather than with only a single one of these. Overall we observed far fewer proteins expressing just only of CD8, CD4, or monocytes than would be expected if these act independently (p < 0.001). Instead we observed many more proteins than expected expressing none of CD8, CD4 or monocyte, or two or three of these simultaneously (chi-square p < 0.001). In the log linear model of the joint expression of genes with CD8, CD4, and monocytes, all two-way and three-way interactions were significant (p < 0.001 for each interaction, see Table 1). Thus, our observations on coordinated cellular response by all three cell-types in a given phase of HIV disease are unambiguously supported by these statistical analyses. 4. Discussion In this study we examined HIV-1-specific immune responses at the level of 274 cytokines, chemokines, their receptors, ligands, and associated proteins post-immunization with an HIV-1 immunogen p24 in a rare elite controller. Since immune responses against HIV-1 Gag are known to be associated with improved control of HIV infection [21], it is believed that Gag-like peptides may be good candidates for an effective therapeutic vaccine. Even though peptide-specific cellular immune responses were induced in the study subject, the protective correlates of effective immunity to p24 have never been defined in such detail as attempted in this study using antibody arrays. This could partly be attributed to the p24 vaccination trial being inconclusive; no follow-up was done due to the failure of the vaccine. The continued presence of robust proliferative responses and strong cellular responses to p24 antigen post-vaccination in the study subject prompted this study to holistically define cellular responses by visualizing them at the protein level. This rare HIV+ non-progressor who received the p24 vaccine in 1993 continues to live disease- and viremia-free in the absence of HAART therapy with high CD4+ T cell counts (700/µL blood taken in January, 2013) [1,12]. Moreover, the continued presence of p24 responses after 20 years of p24 vaccination raises the possibility that p24 vaccination in this individual may have worked, although we do not have pre-vaccination data on p24 responses dating back to 1992. The main stimulus to carry out such a detailed study of p24 antigen-induced cellular responses at the protein level emanates from a notable occurrence of a rare plasma viremia blip (563 HIV RNA copies/mL plasma) in 2011, which naturally subsided within a few weeks without therapeutic intervention. Although his viral blip coincided with gallbladder inflammation, we discarded the first sample collected close to the gallbladder inflammation (18 May 2011) and used the second sample collected (7 June 2011) when the inflammation subsided upon treatment. This allowed us a rare window of opportunity to study cellular immune responses uniquely related to the p24 antigen in both the viremic and aviremic phases from blood samples collected in 2011. Our comprehensive and simultaneous analyses of cellular responses to p24 antigen in the patient’s CD4+, CD8+ T cells, and monocytes provided the first snapshot of the protective immune correlates of p24 vaccination. The notable finding is that the proteins associated with coordinated cellular responses and cell-specific immune responses clearly discriminated between proteins associated with the viremic and aviremic phases of HIV disease. Through these analyses, we were not only able to show cohesive and intimate interaction between CD4+, CD8+ T cells, and CD14+ monocytes, but were also successful in deciphering the proteomic basis of the broad crosstalk between these cell-types that segregated these two phases. The proteins—Dtk, MIP-1α, MMP-13, PIGF, TECK, and TRAIL-R4—were specific for the viremic phase, whereas the proteins Eotaxin-2, Fractalkine, FSH, IGFBP-2 and 4, Cathepsin-S, EDA-A2, I-TAC, CCL14a, M-CSF, MIP-1α, TGFβ3, and VEGF were specific to the aviremic phase. Many of these phase-specific proteins have not been previously described in the context of HIV disease staging; our data sheds a new light on their role in guiding viremic and aviremic phases through coordinated regulation across cell types. Further, apart from the coordinated immune responses from all three cell-types, some of the proteins were highly expressed in two of three or one of three cell-types analyzed, suggesting that variable expression trends between cell types had less bearing on the outcome in controlling viremia or aviremia. It was the coordinated immune response and cross-talk between cell-types that was vital in providing functional synergy and guiding the quality of the immune response required in a given phase of HIV disease. Both for the viremic and aviremic phases, many of the cytokines, chemokines, their receptors, ligands, and associating proteins shown here have not been shown before, because most previous studies have analyzed a single cytokine or a few cytokines at best. Analyzing proteins in an antibody array format has provided a new paradigm in understanding and visualizing novel proteins involved in HIV-specific cellular responses that may prove to be important in future vaccine strategies for HIV. One significant example is Fractalkine (CX3CL1), which is a low-molecular-weight protein and is the unique ligand for the chemokine receptor CX3CR1 expressed on monocytes, natural killer cells, and T cells [22] and is involved in chemotaxis, adhesion of leukocytes, and proliferation [22,23]. Previously, increased levels of Fractalkine in the cerebrospinal fluid have also been implicated in the progression of dementia in HIV-positive patients [24]. However, recent publications have shown that Fractalkine may play a role in promotion of cell survival during homeostatic and inflammatory conditions [25,26]. In our study we observed Fractalkine levels were up-regulated in all three cell types throughout the aviremic phase, suggesting its protective effect, which is consistent with the studies by Landsman et al. [25] and Karlmark et al. [26]. We believe that the protective effect of Fractalkine could also be mediated by its interaction with partner proteins (Cathepsin-S, CCL14a, and IGFBP-2 and 4) during the aviremic phase, but it remains to be proven how these proteins work in concert. Further, we also found that during the aviremic phase antiviral, pro-inflammatory, and anti-apoptotic cytokines were up-regulated, suggesting that the host is able to mount an effective T-cell-derived immunity without leading to T cell exhaustion. Thus, it is plausible to hypothesize that the up-regulation of proteins inducing a protective effect during aviremic phase are possibly vital for maintaining healthy T-cell populations and any imbalance or deterioration in the immunological health of any of the cell types may tilt this balance in favor of viremia—an aspect reminiscent of HIV disease. Our study showed increased levels of ICAM 3 (intercellular adhesion molecule 3) expression in CD4+ helper T cells during the viremic stage, which is in accordance with previous findings [10,27] where increased ICAM3 is associated with plasma viremia. As a protein involved in co-stimulation, increased levels of ICAM-3 have not only been shown to increase HIV-1 transcription and viral production, but also to augment the infection of resting CD4+T cells [10]. We observed a 12-fold increase of ICAM3 expression in CD4+ T cells at the viremic stage, consistent with its involvement of co-stimulatory mechanisms in viral entry. The natural ligands for CCR5 are macrophage inflammatory protein 1α and 1β (MIP-1α and MIP-1β), human CC chemokine ligand 3 like 1 gene (CCL3L1), and RANTES and the only ligand for CXCR4 is SDF-1 [28]. These natural chemokine ligands are known to act as competitive inhibitors of HIV-1 during viral entry and to down-regulate the expression of these co-receptors on the cell surface, thereby becoming associated with the control of HIV1 disease progression [28,29]. CD8+ T cells and macrophages secrete most of the aforementioned chemokines, while RANTES is known to be secreted by platelets [28,30]. In this study we observed down-regulation of both MIP-1α and MIP-1β expression in the viremic phase CD4+ T cells and CD14+ monocytes when compared to the aviremic phase. However, in CD8+ T cells MIP-1α and MIP-1β expressions were highly up-regulated in the viremic phase compared to the aviremic phase, suggesting that CD8+T cells play a crucial role in viral suppression and highlighting their significance in viremia control during HIV infection. In genomic validation of these findings, we noted a similar expression trend for MIP-1α and MIP-1β, but it was also notable that even though MIP-1β was up-regulated in both phases in CD4+ T cells, its higher up-regulation at the aviremic phase carried much more functional weight and is consistent with their antiviral properties. Similarly, patterns were also observed for others in the macrophage inflammatory protein (MIP) family. Even though the cumulative relevance of cytokines and their associating proteins has never been evaluated in the viremic and aviremic phases, we believe that these proteins work in concert and play an important part in the homeostasis of these cell-types. MIP-1δ/CCL15 is the ligand for chemokine receptors like CCR1 and CCR3 and is a known chemo-attractant for neutrophils, monocytes, and lymphocytes. Downregulation of MIP-1δ was observed in the aviremic CD4+ T cells and viremic CD14+ monocytes and CD8+ T cells. Up-regulation was observed in the aviremic CD14+ monocytes. Other inflammatory chemokines and recruiters of lymphocytes [31] of the MIP family such as CCL20/MIP-3α and CCL19/MIP-3β were also significantly expressed in both our studies. CCL20/MIP-3α was up-regulated in the aviremic CD4+ T cells and viremic CD8+ T cells, and also detected in aviremic CD8+ T cells. Similarly, CCL19/MIP-3β up-regulation in both phases in all three cell-types was also consistent with the genomic dataset (Table 2). This trend was also observed in the aviremic CD4+ T cells and in viremic monocytes. MIP-3β is known to regulate CD4+ T-cell immune responses in the secondary lymphoid organs by promoting activation-induced cell death [32]. The effect of MIP-3β is two-fold as, in the presence of MIP-3β, antigen-stimulated CD4+ T cells are highly activated and MIP-3β also promotes activated T cell differentiation into memory T cells [32]. Further to this theory, Kim et al. [33] have shown that CCR7, a receptor associated with MIP-3β, protects CD8+ T cells from apoptosis. The multiple roles played by CD4+ T cells during HIV infection as well as the fact of them being the prime target of HIV make them one of the most effective immune cells in dealing with the invading virus [34]. The cytotoxic potential of antiviral CD4+ T cells is critical for maintaining the homeostasis of the CD8+ T cells and antibody-producing B cells. It is known that many cytokines such as IFN-γ (interferon γ), IL-2, and Tumor Necrosis Factor α (TNFα) are produced by CD4+ T cells [34]. They are the first indicators of T-cell exhaustion, which leads to impaired function of CD8+ T cells and progressive loss of IL-2 production, followed by the loss of TNFα production [34]. Therefore, the loss of function of CD4+ T cells due to infections, such as seen in HIV/AIDS, is largely responsible for the impairment of antiviral immunity in the HIV-infected host, which possibly also causes an imbalance in the coordinated immune response with its partner cells. The increased IL4 to INF-γ ratio has been correlated with high levels of plasma viremia and rapid loss of CD4+ T cells [8]. In CD4+ T cells, we observed an up-regulation of IL4 during the viremic phase and a 7.5-fold down-regulation during the aviremic stage. However, we did not observe a significant expression of INFγ, which was quite stable in both phases in this patient. CXC chemokine ligand 9 (CXCL9)/MIG, chemokines CXCL10/IFN-inducible protein 10 (IP-10), and CXCL11/IFN-inducible T cell α chemoattractant (I-TAC) are chemokines belonging to the IFN-2 inducible family. These chemokines are known to attract and activate CXCR3-bearing cells such as peripheral memory T, B and, natural killer cells, and in vitro T helper type 1 (Th1) cells, but not naive T cells [35,36,37]. These three chemokines have an antagonistic activity on CCR5 and may counteract the action of inflammatory chemokines such as MIP-1α that act via CCR5 [38]. Similarly, it is known that increased expression of these chemokines promotes recruitment of susceptible T cells, which in turn might enhance the sequestration of T cells in infected lymphoid organs and the spread of infection between cells, contributing to the immunopathology of AIDS [39]. Interestingly, IP-10 was up-regulated in all three cell-types in both phases, but this up-regulation was more prominent in viremic CD4+ T cells, CD14+ monocytes and in aviremic phase CD8+ T cells. Not much is known about MIG, I-TAC, and IP-10 in the context of HIV. However, a recent paper by Lajoie et al. [40] has shown that HIV-1-exposed seronegative (HESN) individuals have significantly lower expression of MIG and IP-10 in their genital mucosa compared with HIV-infected commercial sex workers. This is quite significant: since both LTNPs and HIV exposed but seronegative (HESN) individuals are able to control the infection naturally, they may express a similar cytokine profile; given the potential of these two chemokines, the down-regulation of these chemokines during the aviremic phase may offer some protective effect. Along with MIG, IP-10, and I-TAC, five other chemokine antagonists including MIP-1β and Eotaxin-2 are known in the literature [39]. Other than MIP-1β, the functional relevance of the other proteins is not known in the context of HIV, but it would be interesting to see how these chemoattractant proteins in concert play a part in the activation and recruitment of lymphocytes during HIV infection. Another suggestion would be that the up-regulation of chemokine antagonists may be involved in the TH1 to TH2 shift, thereby favoring pro-viral activity or better regulation of antiviral cytokines to prevent lymphocyte recruitment and activity during viral entry. Both Eotaxin 1 and 3, which are known chemoattractants [41,42,43], were stable in both phases; however, Eotaxin-2 was down-regulated in the aviremic phase in all three cell-types and in viremic CD14+ and CD8+ T cells. This pattern was mirrored in the genomic data as well, where it was down-regulated in all three cell types in both phases, suggesting its vital functional role. However, its down-regulation was much more prominent in the viremic phase. MCP-4 (Monocyte chemotactic protein-4) was up-regulated in both CD14+ and CD8+ during the aviremic phase, while it was down-regulated in CD14+ in the viremic phase. MCP-4 shares 56%–61% sequence identity with the three known mono cyto-chemotactic proteins including MCP-3 (a known chemokine antagonist for CCR5 receptor) and is 60% identical with Eotaxin. MCP-4 is functionally similar to both Eotaxin and MCP-3. MCP-4, like MCP-3, is a chemoattractant of high efficacy for monocytes and T lymphocytes [44]. It needs to be iterated that HIV disease progression is due, in part, to accelerated rates of apoptotic cell death of infected, as well as non-infected bystander cells [9,45]. One of the main proteins involved in this apoptosis process is TRAIL (TNF-related apoptosis inducing ligand) as well as its cognate receptors [45]. TRAIL is a member of the TNF superfamily with ligands that include Fas ligand [46] (Huang et al. [47] 2006). TRAIL has five known receptors, surface expressed TRAIL-R1 through TRAIL-R4 and the soluble Osteoprotegerin. Two of the receptors, TRAIL-R1 and TRAIL-R2, are known inducers of apoptotic signal, while TRAIL-R3, TRAIL-R4, and osteoprotegerin are known to act as decoy receptors [46].TRAIL-R3 and TRAIL-R4 lack the intracellular regions necessary for signal transduction, while osteoprotegerin, a receptor heavily involved in bone remodeling, acts as a soluble inhibitor of RANK ligand, thereby preventing the apoptotic signal [46,48,49]. We saw a reverse trend between TRAIL-R3, TRAIL-R4, and Osteoprotegerin in CD4+ T cells, where their expression was down-regulated in the viremic phase but up-regulated in the aviremic phase. In fact, TRAIL-R4 levels were down-regulated throughout the viremic stage in all three cell types. TRAIL-R3 was also down-regulated in CD14+ T cells during the viremic phase, while Osteoprotegerin was down-regulated throughout the viremic phase in CD8+ T cells. HIV-associated T cell depletion is mediated, at least in part, by disordered apoptosis [9]. This patient seems to efficiently regulate TRAIL-mediated apoptosis and maintain the CD4+ T cell levels by utilizing TRAIL-R3, TRAIL-R4, and, to some degree, Osteoprotegerin. This protective effect may be due to the increased expression of TRAIL-R3, TRAIL-R4, and Osteoprotegerin during the aviremic phase. However, during the viremic phase the down-regulation of these receptors seems to coincide with depletion of CD4+ T cells, which has not been shown before. This also highlights that simultaneous modulation of these proteins at any given phase is more important functionally than the modulation of a single protein in vivo during HIV infection. The proposal that protective cytokines play a major role in maintaining this patient’s T cell levels during the course of the infection is further reinforced by the fact that lesser-known chemokines such as TECK/CCL25 were down-regulated in the viremic phase in all three cell types, while being up-regulated during the aviremic phase in monocytes. This observation coincides with results from macaques infected with Simian immunodeficiency virus (SIV), where a decrease in TECK expression was associated with increased apoptosis in lymphoid tissues [50], suggesting that dysfunction in anti-apoptotic chemokines might be a mechanism that contributes to loss of immune function following pathogenic HIV infection. TECK (thymus-expressed chemokine) is known to play a role in the development of T-cells and mucosal immunity in mice [51,52]. Overall, this study is unique in providing the first snapshot of immune correlates of HIV-p24 antigen responses and their simultaneous analysis in CD4+, CD8+ T cells, and monocytes in a rare therapy-naïve HIV+ LTNP, who underwent p24 vaccination for HIV. Our analyses show for the first time how these immune correlates discriminate between viremic and aviremic phases at the level of broad cellular and cell-specific responses. These analyses have also shown that, although the CD8+ T cells play a significant role in aviremia, the equal participation of monocytes and CD4+ T cells is also needed. In contrast, the CD4+ T cells and CD14+ monocytes played an important role in the viremic phase in terms of the pro-inflammatory cytokine expression. These analyses are valuable in shedding light on mechanisms that regulate the immune homeostasis, especially in the aviremic phase of CD4+ and CD8+ T cells and CD14+ monocytes, and showing how the protective effect of anti-apoptotic proteins, evident during the infection with an up-regulation of these factors in the aviremic phase and a down-regulation during the viremic phase, is significant in modulating HIV disease stages. Overall, these analyses stress that a coordinated effort of three cell types and the crosstalk between them may be vital in functionally guiding a potent immune response at a given HIV disease phase. 5. Conclusions In conclusion, our data provide the functional significance of the multifaceted simultaneous innate and adaptive immune responses to p24 antigen in CD4+, CD8+ T cells, and monocytes during viremic and aviremic phases of a rare HIV+ therapy naïve non-progressor, who received p24 vaccination. Along with providing the proteomic basis of cellular immune response that may have come into play post-p24 vaccination, understanding its impact on cellular defense system is sorely needed for future HIV vaccines. High-throughput technologies will allow for in-depth analysis and holistic understanding of immune responses that play a vital role during HIV disease. These analyses will pave the way for the development of novel approaches, which need integration in future vaccine strategies. It will be particularly critical to harness key innate immune mediators that contribute to control of viremia in HIV-infected individuals. Although we have analyzed only three cell types for evaluating cellular immune response to p24 antigen, the immune responses prompted by other cell types, such as NK cells and B cells, which also play a vital role in HIV infection, cannot be understated. More work is needed to createa global snapshot of HIV-specific cellular immune responses. Acknowledgments The abstract for this study was published in the Lancet (13 November 2013) for the HIV Cure Meeting, San Francisco organized by the journals Lancet and Cell, as “Retrospective proteomic analysis of cellular immune-responses and protective correlates of p24 vaccination in an HIV elite controller”(by Nitin K. Saksena and Suneth S. Perera). Nitin K. Saksena acknowledges the NHMRC for the Development Grant Funding for this project. Nitin K. Saksena and all authors acknowledge the written consensual participation of patients in the study. Suneth S. Perera is thankful to the Arin Apcaring Scholarship for his Master of Science Project. Supplementary Materials The supplementary materials are available online at http://www.mdpi.com/2076-3905/5/2/14/s1. Click here for additional data file. Author Contributions The work presented in this manuscript is a part of Suneth S. Perera’s Master of Medicine thesis (University of Sydney). Suneth S. Perera carried out all the experimental work, wrote the manuscript and analyzed the dataset. Bin Wang provided intellectual input, archival sample collection and clinical follow-up. Arturo Damian helped with the protein analysis of differentially expressed (DE) proteins and its interpretation. Wayne Dyer provided the intellectual input and sample collection. Li Zhou and Viviane Conceicao helped with protein analysis and cell culture experiments. Suneth S. Perera conducted his Master of Medicine (M. Med) thesis under the supervision of Nitin K. Saksena, who conceptualized the study, guided it, actively participated in writing, analysis and its interpretation. Conflicts of Interest The authors declare no conflict of interest. Figure 1 Summary of the study subject’s CD4+ and CD8+ T cell counts and plasma viral load between 1996 and 2013. Figure 2 Venn diagram. Interaction and relationship between CD4+T, CD8+T cells, and CD14+ monocytes during the viremic (A) and aviremic phases (B). Insets within the circles show the DE genes and their breakdown within an individual cell type. Overlapping proteins between cell types are shown within the overlapping circles in figures (A) and (B). Figure 3 Log scale CD4+ (A), CD14+ (B), CD8+ (C) T cell foldchange during viremic and aviremic phases in log scale, where a negative log value has been shown to depict down-regulation in relation to up-regulation. microarrays-05-00014-t001_Table 1Table 1 K-way and log-linear model statistical analyses. K-Way and Higher-Order Effects K df Likelihood Ratio Chi-Square Sig. Pearson Chi-Square K-way and Higher Order Effects a 1 7 616.1 0 1056.511 2 4 104.972 0 147.811 3 1 22.957 0 25.08 K-way Effects b 1 3 511.128 0 908.7 2 3 82.015 0 122.731 3 1 22.957 0 25.08 a Tests that K-way and higher order effects are zero. b Tests that K-way effects are zero. Monocyte than would be expected if these act independently (p < 0.001); Instead we observed many more proteins than expected expressing with none of CD8, CD4, or monocyte, or with two or all three of these simultaneously (chi-sq, p < 0.001); In the log linear model of the joint expression of genes with CD8, CD4, and monocyte, all two-way and three-way interactions were significant (p < 0.001 for each interaction, see below). microarrays-05-00014-t002_Table 2Table 2 Expression trends of co-expressed immune modulators between the qRTPCR and protein array for all three cell-types at viremic and aviremic phases. Cytokines PCR Array Expression Protein Array Expression Viremic CD4+ Aviremic CD4+ Viremic CD14+ Aviremic CD14+ Viremic CD8+ Aviremic CD8+ Viremic CD4+ Aviremic CD4+ Viremic CD14+ Aviremic CD14+ Viremic CD8+ Aviremic CD8+ CCL18/PARC 156.28 – – – – – 1.72 – – – – −19.78 CCL19/MIP-3B 5.42 14.05 14.89 47.90 12.06 10.93 – 1.77 1.76 – – – CCL24/eotaxin-2 −6.46 −6.81 −20.45 −4.19 −44.26 −7.52 – −5.71 −2.96 −2.37 −1.58 −1.91 CCL4/MIP-1- β 5.97 6.16 – – – – 1.63 3.23 – – 14.75 −2.67 CCL5/RANTES – – −107.00 16.94 – – – −1.76 – 2.73 – – CCL8/MCP-2 199.19 39.18 68.40 25.49 18.40 50.56 – – 2.49 12.47 1.71 – CD80 – – 4.38 4.88 – – – – 1.53 3.79 – – CXCL10/IP-10 83.17 16.13 31.04 18.40 67.74 15.89 – – 4.17 3.46 – – CXCL11/I TAC 180.77 66.81 78.58 103.39 311.27 41.93 – 7.14 9.88 8.26 – 13.26 CXCL13/BLC 6.05 26.03 −5.29 – 4.41 8.69 2.32 5.90 1.54 – – 1.55 CXCL2/GRO −10.72 – −29.94 −7.20 −13.34 – −1.58 1.59 −6.47 −1.82 – – CXCL5/ENA-78 −205.36 −17.36 −1041.18 −37.22 −14.91 −18.64 – 1.98 −14.91 −2.82 – – CXCL9/MIG 79.23 28.29 11.52 43.77 94.48 34.06 – – 3.44 1.61 – – IL8 – – −11.66 – – – – 1.55 −2.24 – 1.57 – VCAM1 32.85 15.89 109.29 60.38 – – – – 1.55 – – – NB: FC expressions left blank represent no significantly expressed genes/proteins. For the PCR, array genes with ±4 fold changes were deemed significant as per the manufacturer’s instructions. For the protein array, protein fold changes ≤−1.53 and ≤1.5 were deemed significant as per the manufacturer’s instructions. Bold fold change represent trends that are akin in both arrays, whereas normal fold changes represent immune modulators that are significantly expressed only in one array or when the trends are not matching. ==== Refs References 1. Dyer W.B. Zaunders J.J. Yuan F.F. Wang B. Learmont J.C. Geczy A.F. Saksena N.K. McPhee D.A. Gorry P.R. Sullivan J.S. Mechanisms of HIV non-progression; robust and sustained CD4+ T-cell proliferative responses to p24 antigen correlate with control of viremia and lack of disease progression after long-term transfusion-acquired HIV-1 infection Retrovirology 2008 5 10.1186/1742-4690-5-112 19077215 2. Pontesilli O. Carotenuto P. Kerkhof-Garde S.R. Roos M.T. Keet I.P. Coutinho R.A. Goudsmit J. Miedema F. Lymphoproliferative response to HIV type 1 p24 in long-term survivors of HIV type 1 infection is predictive of persistent AIDS-free infection AIDS Res. Hum. Retroviruses 1999 15 973 981 10.1089/088922299310485 10445809 3. Kiepiela P. Ngumbela K. Thobakgale C. Ramduth D. Honeyborne I. Moodley E. Reddy S. de Pierres C. Mncube Z. Mkhwanazi N. CD8+ T-cell responses to different HIV proteins have discordant associations with viral load Nat. Med. 2007 13 46 53 10.1038/nm1520 17173051 4. Pettersen F.O. Tasken K. Kvale D. Combined Env- and Gag-specific T cell responses in relation to PD-1 and CD4+ T cell loss rates in HIV-1 infection Clin. Exp. Immunol. 2010 161 315 323 20491784 5. Rolland M. Heckerman D. Deng W. Rousseau C.M. Coovadia H. Bishop K. Goulder P.J. Walker B.D. Brander C. Mullins J.I. Broad and Gag-biased HIV-1 epitope repertoires are associated with lower viral loads PLoS ONE 2008 3 14 10.1371/journal.pone.0001424 18183304 6. Zuniga R. Lucchetti A. Galvan P. Sanchez S. Sanchez C. Hernandez A. Sanchez H. Frahm N. Linde C.H. Hewitt H.S. Relative dominance of Gag p24-specific cytotoxic T lymphocytes is associated with human immunodeficiency virus control J. Virol. 2006 80 3122 3125 10.1128/JVI.80.6.3122-3125.2006 16501126 7. Smith D. Gow I. Colebunders R. Weller I. Tchamouroff S. Weber J. Boag F. Hales G. Adams S. Patou G. Therapeutic vaccination (p24-VLP) of patients with advanced HIV-1 infection in the pre-HAART era does not alter CD4 cell decline HIV Med. 2001 2 272 275 10.1046/j.1468-1293.2001.00080.x 11737409 8. Altfeld M. Addo M.M. Kreuzer K.A. Rockstroh J.K. Dumoulin F.L. Schliefer K. Leifeld L. Sauerbruch T. Spengler U. TH1 to TH2 shift of cytokines in peripheral blood of HIV-infected patients is detectable by reverse transcriptase polymerase chain reaction but not by enzyme-linked immunosorbent assay under nonstimulated conditions J. Acquir. Immune Defic. Syndr. 2000 23 287 294 10.1097/00126334-200004010-00001 10836750 9. Babu C.K. Suwansrinon K. Bren G.D. Badley A.D. Rizza S.A. HIV Induces TRAIL Sensitivity in Hepatocytes PLoS ONE 2009 4 10.1371/journal.pone.0004623 19247452 10. Barat C. Gervais P. Tremblay M.J. Engagement of ICAM-3 provides a costimulatory signal for human immunodeficiency virus type 1 replication in both activated and quiescent CD4+ T lymphocytes: Implications for virus pathogenesis J. Virol. 2004 78 6692 6697 10.1128/JVI.78.12.6692-6697.2004 15163761 11. Casartelli N. Di Matteo G. Potesta M. Rossi P. Doria M. CD4 and major histocompatibility complex class I down-regulation by the human immunodeficiency virus type 1 Nef protein in pediatric AIDS progression J. Virol. 2003 77 11536 11545 10.1128/JVI.77.21.11536-11545.2003 14557639 12. Wang B. Dyer W.B. Zaunders J.J. Mikhail M. Sullivan J.S. Williams L. Haddad D.N. Harris G. Holt J.A. Cooper D.A. Comprehensive analyses of a unique HIV-1-infected nonprogressor reveal a complex association of immunobiological mechanisms in the context of replication-incompetent infection Virology 2002 304 246 264 10.1006/viro.2002.1706 12504566 13. Zaunders J.J. Dyer W.B. Wang B. Munier M.L. Miranda-Saksena M. Newton R. Moore J. Mackay C.R. Cooper D.A. Saksena N.K. Identification of circulating antigen-specific CD4+ T lymphocytes with a CCR5+ , cytotoxic phenotype in an HIV-1 long-term nonprogressor and in CMV infection Blood 2004 103 2238 2247 10.1182/blood-2003-08-2765 14645006 14. Goldstein D. Hertzog P. Tomkinson E. Couldwell D. McCarville S. Parrish S. Cunningham P. Newell M. Owens M. Cooper D. Aadministration of imiquimod, an interferon inducer, in asymptomatic human immunodeficiency virus-infected persons to determine safety and biologic response modification J. Infect. Dis. 1998 178 858 861 10.1086/515343 9728559 15. Kelleher A.D. Roggensack M. Jaramillo A.B. Smith D.E. Walker A. Gow I. McMurchie M. Harris J. Patou G. Cooper D.A. Safety and immunogenicity of a candidate therapeutic vaccine, p24 virus-like particle, combined with zidovudine, in asymptomatic subjects. Community HIV Research Network Investigators AIDS 1998 12 175 182 10.1097/00002030-199802000-00007 9468366 16. Potter S.J. Lemey P. Dyer W.B. Sullivan J.S. Chew C.B. Vandamme A.M. Dwyer D.E. Saksena N.K. Genetic analyses reveal structured HIV-1 populations in serially sampled T lymphocytes of patients receiving HAART Virology 2006 348 35 46 10.1016/j.virol.2005.12.031 16455126 17. Howell D.C. Statistical Methods for Psychology 7th ed. Cengage Learning Belmot, CA, USA 2009 630 655 18. Field A. Discovering Statistics Using SPSS 2nd ed. Sage Publications Thousand Oaks, CA, USA 2005 695 718 19. Agresti A. An Introduction to Categorical Data Analysis 2nd ed. Wiley Inter-Science Hoboken, NJ, USA 2007 212 20. Poropatich K. Sullivan D.J. Jr. Human immunodeficiency virus type 1 long-term nonprogressors: The viral, genetic and immunological basis for disease non-progression J. General. Virol. 2011 92 247 268 10.1099/vir.0.027102-0 21106806 21. Imai T. Hieshima K. Haskell C. Baba M. Nagira M. Nishimura M. Kakizaki M. Takagi S. Nomiyama H. Schall T.J. Identification and molecular characterization of fractalkine receptor CX3CR1, which mediates both leukocyte migration and adhesion Cell 1997 91 521 530 10.1016/S0092-8674(00)80438-9 9390561 22. Imami N. Pires A. Hardy G. Wilson J. Gazzard B. Gotch F. A balanced type 1/type 2 response is associated with long-term nonprogressive human immunodeficiency virus type 1 infection J. Virol. 2002 76 9011 9023 10.1128/JVI.76.18.9011-9023.2002 12186885 23. White G.E. Greaves D.R. Fractalkine: A survivor’s guide: Chemokinesas antiapoptotic mediators Arterioscler. Thromb. Vasc. Biol. 2012 32 589 594 10.1161/ATVBAHA.111.237412 22247260 24. Cotter R. Williams C. Ryan L. Erichsen D. Lopez A. Peng H. Zheng J. Fractalkine (CX3CL1) and brain inflammation: Implications for HIV-1-associated dementia J. Neurovirol. 2002 8 585 598 10.1080/13550280290100950 12476352 25. Landsman L. Bar-On L. Zernecke A. Kim K.W. Krauthgamer R. Shagdarsuren E. Lira S.A. Weissman I.L. Weber C. Jung S. CX3CR1 is required for monocyte homeostasis and atherogenesis by promoting cell survival Blood 2009 113 963 972 10.1182/blood-2008-07-170787 18971423 26. Karlmark K.R. Zimmermann H.W. Roderburg C. Gassler N. Wasmuth H.E. Luedde T. Trautwein C. Tacke F. The fractalkine receptor CXCR1 protects against liver fibrosis by controlling differentiation and survival of infiltrating hepatic monocytes Hepatology 2010 52 1769 1782 10.1002/hep.23894 21038415 27. Geijtenbeek T.B. van Kooyk Y. DC-SIGN: A novel HIV receptor on DCs that mediates HIV-1 transmission Curr. Top. Microbiol. Immunol. 2003 276 31 54 12797442 28. Gallo R.C. Garzino-Demo A. DeVico A.L. HIV infection and pathogenesis: What about chemokines? J. Clin. Immunol. 1999 19 293 299 10.1023/A:1020539524373 10535605 29. Mikhail M. Bin W. Saksena N.K. Mechanisms involved in non-progressive HIV disease AIDS Rev. 2003 3 230 230 15012002 30. Kornbluth R.S. Kee K. Richman D.D. CD40 Ligand (CD154) stimulation of macrophages to produce HIV-1-suppressive β-chemokines Proc. Natl. Acad. Sci. USA 1998 95 5205 5210 10.1073/pnas.95.9.5205 9560254 31. Hieshima K. Imai T. Opdenakker G. van Damme J. Kusuda J. Tei H. Sakaki Y. Takatsuki K. Miura R. Yoshie O. Molecular cloning of a novel human CC chemokine liver and activation-regulated chemokine (LARC) expressed in liver. Chemotactic activity for lymphocytes and gene localization on chromosome 2 J. Biol. Chem. 1997 272 5846 5853 10.1074/jbc.272.9.5846 9038201 32. Yasuda T. Kuwabara T. Nakano H. Aritomi K. Onodera T. Lipp M. Takahama Y. Kakiuchi T. Chemokines CCL19 and CCL21 promote activation-induced cell death of antigen-responding T cells Blood 2007 109 449 456 10.1182/blood-2006-04-018101 16973962 33. Kim J.W. Ferris R.L. Whiteside T.L. Chemokine C receptor 7 expression and protection of circulating CD8+ T lymphocytes from apoptosis Clin. Cancer Res. 2005 11 7901 7910 10.1158/1078-0432.CCR-05-1346 16278415 34. Kannanganat S. Ibegbu C. Cheenreddi L. Robinson H.L. Amara R.R. Multiple-cytokine-producing antiviral CD4 T cells are functionally superior to single-cytokine-producing cells J. Virol. 2007 81 8468 8476 10.1128/JVI.00228-07 17553885 35. Muller M. Carter S. Hofer M.J. Campbell I.L. Review: The chemokine receptor CXCR3 and its ligands CXCL9, CXCL10 and CXCL11 in neuroimmunity-a tale of conflict and conundrum Neuropathol. Appl. Neurobiol. 2010 36 368 387 10.1111/j.1365-2990.2010.01089.x 20487305 36. Qin S. Rottman J.B. Myers P. Kassam N. Weinblatt M. Loetscher M. Koch A.E. Moser B. Mackay C.R. The chemokine receptors CXCR3 and CCR5 mark subsets of T cells associated with certain inflammatory reactions J. Clin. Investig. 1998 101 746 754 10.1172/JCI1422 9466968 37. Sallusto F. Lenig D. Mackay C.R. Lanzavecchia A. Flexible programs of chemokine receptor expression on human polarized T helper 1 and 2 lymphocytes J. Exp. Med. 1998 187 875 883 10.1084/jem.187.6.875 9500790 38. Foley J.F. Yu C.R. Solow R. Yacobucci M. Peden K.W. Farber J.M. Roles for CXC chemokine ligands 10 and 11 in recruiting CD4+ T cells to HIV-1-infected monocyte-derived macrophages, dendritic cells, and lymph nodes J. Immunol. 2005 174 4892 4900 10.4049/jimmunol.174.8.4892 15814716 39. Petkovic V. Moghini C. Paoletti S. Uguccioni M. Gerber B. ITAC/CXCL11 is a natural antagonist for CCR5 J. Leukoc. Biol. 2004 76 701 708 10.1189/jlb.1103570 15178708 40. Lajoie J. Juno J. Burgener A. Rahman S. Mogk K. Wachihi C. Mwanjewe J. Plummer F.A. Kimani J. Ball T.B. A distinct cytokine and chemokine profile at the genital mucosa is associated with HIV-1 protection among HIV-exposed sero-negative commercial sex workers Mucosal Immunol. 2012 5 277 287 10.1038/mi.2012.7 22318497 41. Osakwe C.E. Bleotu C. Chifiriuc M.C. Crancea C. Otelea D. Paraschiv S. Petrea S. Dinu M. Baicus C. Streinu-Cercel A. TH1/TH2 cytokine levels as an indicator for disease progression in human immunodeficiency virus type 1 infection and response to antiretroviral therapy Roum. Arch. Microbiol. Immunol. 2010 69 24 34 21053781 42. Ogilvie P. Bardi G. Clark-Lewis I. Baggiolini M. Uguccioni M. Eotaxin is a natural antagonist for CCR2 and an agonist for CCR5 Blood 2001 97 1920 1924 10.1182/blood.V97.7.1920 11264152 43. Ogilvie P. Paoletti S. Clark-Lewis I. Uguccioni M. Eotaxin-3 is a natural antagonist for CCR2 and exerts a repulsive effect on human monocytes Blood 2003 102 789 794 10.1182/blood-2002-09-2773 12689946 44. Uguccioni M. Loetscher P. Forssmann U. Dewald B. Li H. Lima S.H. Li Y. Kreider B. Garotta G. Thelen M. Monocyte chemotactic protein 4 (MCP-4), a novel structural and functional analogue of MCP-3 and eotaxin J. Exp. Med. 1996 183 2379 2384 10.1084/jem.183.5.2379 8642349 45. Schnepple D.J. Shepard B. Bren G.D. Cummins N.W. Natesampillai S. Trushin S. Algeciras-Schimnich A. Meng X.W. Sainski A.M. Rizza S.A. Isolation of a TRAIL antagonist from the serum of HIV-infected patients J. Biol. Chem. 2011 286 35742 35754 10.1074/jbc.M111.274639 21859711 46. Huang Y. Erdmann N. Peng H. Herek S. Davis J.S. Luo X. Ikezu T. Zheng J. TRAIL-mediated apoptosis in HIV-1-Infected macrophages is dependent on the inhibition of Akt-1 phosphorylation J. Immunol. 2006 177 2304 2313 10.4049/jimmunol.177.4.2304 16887991 47. Hu W.H. Johnson H. Shu H.B. Tumor necrosis factor-related apoptosis-inducing ligand receptors signal NF-κB and JNK activation and apoptosis through distinct pathways J. Biol. Chem. 1999 274 30603 30610 10.1074/jbc.274.43.30603 10521444 48. Emery J.G. McDonnell P. Burke M.B. Deen K.C. Lyn S. Silverman C. Dul E. Appelbaum E.R. Eichman C. DiPrinzio R. Osteoprotegerin is a receptor for the cytotoxic ligand TRAIL J. Biol. Chem. 1998 273 14363 14367 10.1074/jbc.273.23.14363 9603945 49. Holen I. Croucher P.I. Hamdy F.C. Eaton C.L. Osteoprotegerin (OPG) is a survival factor for human prostate cancer cells Cancer Res. 2002 62 1619 1623 11912131 50. Qin S. Sui Y. Murphey-Corb M.A. Reinhart T.A. Association between decreased CXCL12 and CCL25 expression and increased apoptosis in lymphoid tissues of cynomolgus macaques during SIV infection J. Med. Primatol. 2008 37 46 54 10.1111/j.1600-0684.2008.00327.x 19187430 51. Vicari A.P. Figueroa D.J. Hedrick J.A. Foster J.S. Singh K.P. Menon S. Copeland N.G. Gilbert D.J. Jenkins N.A. Bacon K.B. TECK: A novel CC chemokine specifically expressed by thymic dendritic cells and potentially involved in T cell development Immunity 1997 7 291 301 10.1016/S1074-7613(00)80531-2 9285413 52. Wurbel M.A. McIntire M.G. Dwyer P. Fiebiger E. CCL25/CCR9 interactions regulate large intestinal inflammation in a murine model of acute colitis PLoS ONE 2011 6 14 10.1371/journal.pone.0016442 21283540

PMC005xxxxxx/PMC5003491.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020015microarrays-05-00015ArticleEnhancing Interpretability of Gene Signatures with Prior Biological Knowledge Squillario Margherita †Barbieri Matteo †Verri Alessandro *Barla Annalisa Ruskin Heather J. Academic EditorDIBRIS, University of Genoa, Via Dodecaneso 35, I-16146 Genova, Italy; margherita.squillario@unige.it (M.S.); matteo.barbieri@dibris.unige.it (M.B.); annalisa.barla@unige.it (A.B.)* Correspondence: alessandro.verri@unige.it; Tel.: +39-010-353-6601† These authors contributed equally to this work. 08 6 2016 6 2016 5 2 1505 10 2015 31 5 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Biological interpretability is a key requirement for the output of microarray data analysis pipelines. The most used pipeline first identifies a gene signature from the acquired measurements and then uses gene enrichment analysis as a tool for functionally characterizing the obtained results. Recently Knowledge Driven Variable Selection (KDVS), an alternative approach which performs both steps at the same time, has been proposed. In this paper, we assess the effectiveness of KDVS against standard approaches on a Parkinson’s Disease (PD) dataset. The presented quantitative analysis is made possible by the construction of a reference list of genes and gene groups associated to PD. Our work shows that KDVS is much more effective than the standard approach in enhancing the interpretability of the obtained results. gene expressionfunctional characterizationvariable selectionsparse regularizationestablished domain knowledgeKDVSParkinson’s diseasegene ontology ==== Body 1. Introduction Gene expression measures allow for the study of complex diseases such as neurodegenerative diseases and tumors that, unlike Mendelian disorders, depend on the concerted misregulation of several genes. The analysis of microarray data aims at finding a gene signature able to discriminate between groups of samples (e.g., cases and controls, responding or not responding to a specific treatment) and the associated gene functional modules for a pathology of interest. These modules, defined in terms of the established domain knowledge, allow for the assessment of the degree of involvement of the gene signature in relevant pathways, processes, or functions. The most common approach to tackle this endeavor, which we refer to as standard pipeline, finds the gene signature and the associated gene functional modules in two steps (see Figure 1). In the first data analysis step, a variable selection method of choice yields a gene signature. In the second step, the obtained signature is functionally characterized by means of an enrichment analysis [1], which aims at recovering biologically relevant genes possibly discarded in the variable selection process. When using Gene Ontology (GO) [2] as the established domain knowledge, the enriched gene modules are GO terms. The obtained results are interpreted by domain experts who evaluate the significance of the selected GO terms by means of the established biological knowledge on the pathology of interest. Recently, Knowledge Driven Variable Selection (KDVS) [3], an alternative pipeline that uses GO a priori as the established domain knowledge, has been proposed. The KDVS pipeline (see Figure 1) performs data analysis and functional characterization at the same time, providing, as a final result, a list of GO terms and associated gene signatures relevant for the pathology of interest. This enhances the biological interpretability of the obtained results in terms of functional gene modules. The aim of this work is to assess quantitatively the effectiveness of KDVS with respect to the standard pipeline in the analysis of a gene expression microarray dataset. We restricted our attention to Parkinson’s Disease (PD) as a case study. To this purpose, we built benchmark lists of GO terms and genes by using the Kyoto Encyclopeadia of Genes and Genomes (KEGG) [4], Gene Prospector [5], and the Gene Ontology Annotations (GOA). The obtained benchmark lists allowed us to measure the selection performance in terms of Precision, Recall and F-Measure for both pipelines. The remainder of this paper is organized as follows. We describe material and methods in Section 2, illustrate the results in Section 3, present our comments in Section 4, and state our final remarks in Section 5. The identified GO terms, genes, and benchmark lists can be found as tables in the Supplementary Material (see Tables S1–S5). 2. Experimental Section In this section, we describe materials and methods of our work. We start with the dataset and the normalization procedure we used, and then we describe the experimental framework, the standard and the KDVS pipeline, and the construction of the benchmark lists. Finally, we illustrate the metrics we used to assess performance. 2.1. Data and Preprocessing We devised a binary classification problem of PD cases and controls by using four public microarray datasets stored in the Gene Expression Omnibus (GEO) repository [7]: GSE7621 [8], GSE20292, GSE20291 and GSE20168 [9,10]. All datasets measure the expression on post-mortem brain tissue from patients affected by PD and controls. Specifically, GSE7621 is composed by microarray measures of 16 cases and nine controls deriving from the substantia nigra tissue measured on the HG-U133 Plus 2 platform, characterized by 54,713 probesets. The other three datasets belong to the Superseries GSE20295 and use the HG-U133A platform characterized by 22,283 probesets. GSE20292 is composed by 11 cases and 18 controls from the same brain tissue, the GSE20291 is composed by 15 cases and 20 controls deriving from the putamen brain region, and GSE20168 is composed by 14 cases and 15 controls deriving from the prefrontal area nine brain region. Normalization of gene expression values was performed on each data matrix using the Robust Multichip Average method [11], with an R script included in the aroma package [12]. After normalization, we discarded the control probesets and merged the four preprocessed matrices into one single p×n matrix X, where p=22215 is the number of common probesets and n=118 is the total number of samples (56 cases and 62 controls). An n-dimensional vector Y of binary labels distinguishes between cases and controls. In the remainder of the paper, a dataset will be a pair of the type (X,Y). 2.2. Methods 2.2.1. Experimental Framework The statistical analysis of microarray data (like any small set of samples in high-dimensional space) can easily lead to biased results [13]. In order to perform an unbiased analysis, we adopted a two nested cross-validation procedure [14], which we briefly describe here for the sake of completeness. The full dataset (X,Y) is first split in B chunks (external split) obtaining B datasets (Xb,Yb) with b=1,…,B each consisting of B-1 chunks. An optimal model (i.e., a gene signature (actually, a probeset signature) and a classifier) is then obtained for each of the B datasets by means of a B-1-fold cross-validation (internal split). Each of the B models leads to a possibly different list of selected features; the final aggregate list is obtained by including only those variables appearing in at least a given number of those B lists. 2.2.2. The Standard Pipeline The standard pipeline reflects the classical approach to extract relevant biological features from normalized high-throughput data sets. It is composed of two steps: data analysis and functional analysis (Figure 1). Data Analysis In order to assess the reproducibility of the produced results with the standard pipeline we considered several methods. Fifteen lists of discriminant probesets were obtained by combining three feature selection methods with five classifiers within the unbiased framework described above through the software library PyXPlanner [15]. The three feature selection methods were FilterKBest [16], which selects the top-k features with the highest F-value from a one-way ANOVA test, LASSO [17] and Elastic Net (ENET) [18], which selects the features corresponding to the nonzero components of the vector β minimizing the functional ∥Xβ-Y∥22+τ∥β∥1 and ∥Xβ-Y∥22+ατ∥β∥1+(1-α)τ∥β∥22, respectively. The five classification algorithms were k-Nearest Neighbors (k-NN), Logistic Regression (LR), Linear Support Vector Machines (LSVM), Ordinary Least Squares (OLS), and Regularized Least Squares (RLS). A sixteenth list was obtained by means of the univariate method most commonly used in the analysis of this kind of data, the Bonferroni corrected t-test. The last method we used, ℓ1ℓ2FS, is an embedded regularization method based on ENET, studied in [19,20] and successfully applied in the analysis of high-throughput molecular data [21,22,23,24]. The algorithm, embedded in the unbiased framework of above, is implemented in L1L2Signature [25], a tool in Python based on the L1L2Py [26] and PPlus [27] libraries. Functional Analysis The functional characterization of the gene signature identified with the standard pipeline was performed through enrichment analysis using the online toolkit WebGestalt [28,29]. WebGestalt takes as input a list of relevant genes/probesets and performs an enrichment analysis based on a hypergeometric test, providing several methods to correct for multiple hypothesis and using several databases (e.g., KEGG or GO) for identifying the most relevant pathways and ontologies in each signature. In other words, given a GO term and a reference set (such as the entire human genome or the list of genes in a microarray platform), the enrichment is based on the comparison between the fraction of signature genes in the GO term and the fraction of GO term genes in the reference set. The signature is enriched in the GO term if the former is larger than the latter fraction. In our experiments, we enriched each signature using GO, selecting the HG-U133A platform as a reference set, 0.05 as the level of significance, the Bonferroni correction and three as the minimum number of genes in each GO term considered. 2.2.3. The KDVS Pipeline Let us present the KDVS pipeline of Figure 1. For a more detailed description see [3]. KDVS [30], implemented in Python, is based on the prototype presented in [31]. It uses the established domain knowledge (Gene Ontology release 20100110 [32]) before the actual feature selection step and provides users with a list of discriminant GO terms each coupled with a list of discriminant genes. KDVS consists of three stages: the local integration, knowledge retrieval and post–processing. The local integration stage accepts the gene expression dataset (X,Y), the microarray annotations (e.g., from GEO), and the representation of biological knowledge (GO). By using the microarray annotation, KDVS builds the mapping from the probeset list to the GO terms and vice versa to allow fast querying in both directions. Then, for each GO term t, it generates a ps×n submatrix of gene expression data, with ps≪p, where only the expression values related to genes annotated to t are retained [3]. By construction, the overlap of each pair of submatrices is the same of the corresponding GO terms. In the knowledge retrieval stage, ℓ1ℓ2FS is performed on each submatrix (GO term), obtaining the classification error as well as the list of selected variables (in our case probesets) that are the most discriminant between the two classes). For all nodes for which ps<6, no feature selection is performed. Finally, the post–processing stage selects the GO terms for which the classification error is below a fixed threshold. Since KDVS processes one GO domain at a time—Molecular Function (MF), Biological Process (BP) or Cellular Component (CC)—we performed three runs using the same PD dataset. The output, therefore, was obtained by pooling in a single list the three lists of discriminant GO terms as well as the lists of selected probesets. 2.2.4. Benchmark Lists The benchmark lists were obtained through the workflow depicted in Figure 2. First, we queried KEGG and Gene Prospector [5]. KEGG is a database of curated biological pathways of the human genome, in addition to other organisms. Gene Prospector, instead, is a tool that allows users to search for genes associated with human diseases, risk factors, and other phenotypes, and may include both experimentally verified and not yet verified biological knowledge. We retrieved genes (1) from the Parkinson’s disease—Homo sapiens pathway of the KEGG PATHWAY database (ID: hsa05012); (2) from the Parkinson’s Disease (PD) entry of the KEGG DISEASE database (ID: H00057); and (3) by querying Gene Prospector for Parkinson’s Disease. The final list contained 482 genes. Next, by means of Gene Ontology Annotations (GOA) compiled for Homo sapiens, we extracted the list of GO terms associated to each of the 482 genes. Evidence codes are provided to motivate each association [6]. Finally, we filtered both lists retaining only the associations based on the following tags: the Experimental Evidence Codes EXP, IDA, IPI, IMP and IGI, IEP, the Traceable Author Statement, and the Inferred by Curator category, which we deemed as the most reliable. In the case of multiple associations between the same gene and GO term we retained the most recent. We obtained benchmark lists of 2121 GO terms (of which 1447 are BP terms, 446 are MF terms and 228 are CC terms) and 444 genes, see Table S1. 2.2.5. Performance Metrics In customary notation, the true positives (TP) are the benchmark GO terms or genes retained by the pipeline, while the false negatives (FN) are those discarded despite being present in the benchmark. The false positive (FP) are the retained GO terms or genes not in the corresponding benchmark list and the true negatives (TN) those discarded while not in the list. We evaluated the prediction performance through the mean test error and the Matthews Correlation Coefficient (MCC), which is defined as follows: MCC=TP×TN-FP×FN(TP+FP)(TP+FN)(TN+FP)(TN+FN). The MCC, unaffected by the presence of unbalanced classes, ranges between −1 and +1. The greater the MCC, the better the prediction with negative score marking below chance performance. The performance of GO terms and genes selection was measured in terms of Precision, Recall and F-measure with: Precision=TPTP+FP, Recall=TPTP+FN, F-measure=2×Precision×RecallPrecision+Recall. By definition, Precision, Recall, and F-measure range between 0 and 1, with greater values associated with better performance. High Precision is achieved when the large majority of retained GO terms (or genes) are in the benchmark list, while high Recall is achieved when most of the GO terms (or genes) in the benchmark list are retained. Clearly, by retaining all the terms, it is always possible to obtain perfect Recall at the expense of extremely low Precision values. Therefore, the F-measure, which is high if both Precision and Recall are high, is the score of choice to find the optimal trade-off between Precision and Recall. For KDVS, we computed Precision and Recall for the cumulative list of GO terms and genes and for each domain separately. 3. Results First, we describe the results obtained with the standard pipeline. We divided the dataset in B=9 chunks and performed 8-fold cross-validation. In Table 1, we report the four best test errors obtained from the sixteen methods along with the corresponding MCCs. The aggregate list of genes for each experiment was obtained by retaining only those that have been selected at least five out of nine times and then enriched according to the procedure described in Section 2.2.2. We then ran KDVS on the same dataset, using, for each GO term, the same experimental setting: B=9,K=8 and cutoff on gene frequency at 50% (5 out of 9). Based on the test error and standard deviation in Table 1, we decided to retain GO terms associated with a test error less than 31.7%, that is, the ℓ1ℓ2FS mean test error (23.1%) plus its standard deviation (8.6%). Table S2 and S3 report the list of discriminant GO terms and aggregate list of selected genes for the KDVS pipeline, while Table S4 reports the GO terms and gene lists for the best performing methods of the standard pipeline. The comparison of the results against the benchmark, in terms of Precision, Recall and F-measure, of KDVS and of the four top performing methods for the standard pipeline is reported in Table 2. We also added the result of the enrichment analysis performed on the list of genes provided by the t-test. While in Table 2, the results for KDVS are relative to the three GO domains together, Figure 3 shows three Receiver Operating Characteristic (ROC) curves, one for each domain, where we observe how sensitivity and specificity vary for different values of the error threshold. 4. Discussion 4.1. Statistical Analysis Let us first discuss the results illustrated in Table 1. By inspection, we note that the test errors are comparable and well below the chance error (47%). The large values of the standard deviation are likely to be related to the relatively small sample size. Given the complexity of the disease, it is not surprising that the prediction performance of all methods is below 80%. All of the MCC scores indicate a significant correlation between gene expression levels and classes. As for the results displayed in Table 2, we note that the KDVS pipeline F-measure, for comparable Precision values, is between 40 and 200 times greater than the F-measures obtained with the standard pipeline. Interestingly, the performance of the standard pipeline does not change much with the variable selection method (including the widely used t-test). Since the best performance of the standard pipeline is obtained by means of the ℓ1ℓ2FS feature selection method, we conclude that the actual gain of KDVS, which uses ℓ1ℓ2FS as the variable selection engine, is about 40-fold. Let us comment on the results in terms of absolute figures instead of percentages. In Table S5, we listed the TPs for all the methods in Table 2. While the TPs for KDVS are 270, the number of TPs for each of the five methods of the standard pipeline range from one to five. All in all, of the seven different GO terms collectively identified by the five methods, four are also in KDVS list, and two are direct ancestors of two KDVS GO terms. Clearly, in order to be profitably explored by domain experts, the KDVS list needs to be refined. On the other hand, the variability of the GO terms returned by the standard pipeline questions the reliability of the produced results. It is also interesting to consider the results in Table 2 from the gene point of view. In the standard pipeline, the gene enrichment produces a GO term list starting from a gene list. Not surprisingly, for all methods in the standard pipeline, the GO term F-measure is significantly smaller than the corresponding genes F-measure, while the opposite holds for KDVS, consistently with the underlying concept. Finally, the ROC curves in Figure 3 show that the considerable edge of KDVS vs. the standard pipeline remains true in each of the three GO domains considered separately. 4.2. Biological Significance Here, we comment on the results of the KDVS pipeline from a biological viewpoint. From the ROC curves shown in Figure 3, we note that the CC domain terms yield a better performance than MF and BP terms with respect to both specificity and sensitivity. By construction, the benchmark list may contain GO terms with broad meaning. The thorough review for each GO domain presented in the remainder of this Section shows that the the biological features of the selected GO terms common to the benchmark (see Table S1) are often relevant for a neurodegenerative disease such as PD. For the CC domain, the overlap consist of 69 terms, mainly related to: (i) mitochondrion (e.g., matrix, crista, outer and inner cellular membranes, mitochondrial respiratory chain, mitochondrial proton–transporting ATP synthase complex); (ii) neurons (e.g., synapse, synaptic vescicle, axon, dendrite and dendritic shaft); (iii) various cell regions like cell-cell junctions, proteinaceous extracellular matrix, cell cortex, filopodium, actin and microtubule cytoskeleton; and (iv) cytoplasmatic vescicles and several organelles such as the nucleus, endoplasmatic reticulum, Golgi, centrosomes and lysosomes. For the MF domain, the overlap consists of 71 terms, mainly related to: (i) binding of motor proteins; (ii) ions and groups (i.e., zinc, calcium, magnesium, iron manganese, copper, sodium, potassium, ATP, GTP); (iii) nucleotidic acids (i.e., chromatin, single- and double-stranded DNA, mRNA); (iv) integrins, signaling proteins, low-density lipoproteins, tyrosine kinase; (v) specific proteins or proteins categories like polyubiquitin, apoliprotein E, dopamine, heat shock proteins, NF-kappaB, protein N and C-terminus, SH3 domains, piridoxal phosphate, phosphatidylinositol; and (vi) unfolded proteins. The molecular functions related to the selected GO terms involve enzymes (e.g., hydrolase, peptidase, especially serine and cysteine-type peptidase), calcium channel, small conjugating protein ligase ubiquitin, cytochrome-c oxidase, NADH dehydrogenase and ubiquinol-cytochrome-c reductase. For the BP domain, the overlap consists of 130 terms, mainly related to: (i) various kind of metabolic processes concerning lipids, carbohydrates (e.g., glycogen), ATP or dopamine; (ii) development of the central nervous system, the forebrain, the heart and the skeletal tissue; and (iii) defense response, in particular from unfolded proteins and from viruses that prompt the differentiation of B cells, and from inflammation (i.e., acute-phase), oxidative stress, hypoxia, DNA damage, heat and tumor necrosis factors. The BP terms control cell adhesion, differentiation (i.e., B and myeloid), migration, signaling, cell cycle arrest, respiration, growth, differentiation and proliferation. The involved pathways concern Notch receptors, which regulate cell–cell communication in several ways (acting, in particular, in the central nervous system and in the heart) and the nerve growth factors, fundamental for the growth, maintenance, and survival of neurons. The involvement of the mitochondrion is essential as confirmed by the GO terms: mitochondrial electron transport, NADH to ubiquinone and regulation of mitochondrial membrane potential. Among the regulation processes related to PD, it is important to underline neurone differentiation, the positive regulation of anti-apoptosis, and the negative regulation of axonogenesis and of locomotion. 5. Conclusions The main aim of this work was to assess the effectiveness of the KDVS pipeline with respect to the standard pipeline for the analysis of microarray data. While the standard pipeline first selects the relevant variables and then uses the established biological domain knowledge to reconstruct relevant functional modules, KDVS obtains relevant functional modules by embedding the domain knowledge in the variable selection process. We considered PD as a case study and constructed lists of GO terms and genes, obtained by means of the available PD knowledge, which we use as benchmark. Our analysis shows that, for comparable values of precision, the recall and F-measure of KDVS are significantly higher (about two orders of magnitude) than the standard pipeline. Furthermore, KDVS, providing GO terms as output, enhances the biological interpretability suggesting an explanation of the phenomenon under study in terms of functional gene modules rather than single molecular variables. On the basis of the obtained results, we believe that the proposed approach can be regarded as a first step toward the construction of a data and knowledge driven process for the discovery of novel associations. Acknowledgments The authors would like to thank Grzegorz Zycinski for the implementation of KDVS, Salvatore Masecchia for L1L2Signature, L1L2Py and PPlus and Barbara Di Camillo for insightful discussions. Supplementary Materials The following are available online at http://www.mdpi.com/2076-3905/5/2/15/s1. Table S1: Benchmark GO terms and genes, Table S2: GO terms selected by KDVS, Table S3: Genes selected by KDVS, Table S4: GO terms and genes selected by the standard pipeline, Table S5: True positives. Click here for additional data file. Author Contributions Margherita Squillario and Matteo Barbieri designed the study and performed the experiments. Margherita Squillario provided the biological insight. Matteo Barbieri designed and developed the PyXPlanner. Annalisa Barla developed the statistical methodology and supervised KDVS development. Alessandro Verri developed the statistical methodology and supervised the study. Margherita Squillario and Matteo Barbieri contributed equally to the manuscript. Conflicts of Interest The authors declare that they have no competing interests. Abbreviations PDParkinson’s Disease GEOGene Expression Omnibus KDVSKnowledge Driven Variable Selection ℓ1ℓ2FSℓ1ℓ2 feature selection framework GOGene Ontology FiltKBestFilter K-Best selection based on ANOVA LASSOLasso ENETElastic Net KNNK Nearest Neighbor LRLogistic Regression LSVMLinear Support Vector Machine OLSOrdinary Least Square RLSregularized least square DAGDirected Acyclic Graph MFMolecular Function BPBiological Process CCCellular Component KEGGKyoto Encyclopaedia of Genes and Genomes GOAGene Ontology Annotations TASTraceable Author Statement ICInferred by Curator TPTrue Positive TNTrue Negative FPFalse Positive FNFalse Negative Figure 1 Knowledge Driven Variable Selection (KDVS) and standard pipelines. KDVS embeds the Gene Ontology (GO) domain knowledge into the variable selection step, providing as output a list of discriminant GO terms and genes. The standard pipeline, instead, first selects a gene signature and then performs an enrichment analysis in GO obtaining a discriminant GO term list. Figure 2 This scheme shows the workflow used to obtain the benchmark gene and GO terms lists. The benchmark gene list is composed of 444 genes and the benchmark GO term list is composed of 2121 terms: 1447 from Biological Process (BP), 446 from Molecular Function (MF) and 228 from Cellular Component (CC). Figure 3 ROC curves for the three GO domains. The plots show the ROC curves (Sensitivity vs. 1-Specificity, defined as FP/(TN+FP)) for the KDVS GO terms, for varying values of the threshold error. The highlighted point on the curve is associated with the highest F-measure, reported in the green box. microarrays-05-00015-t001_Table 1Table 1 Top performing methods for the standard pipeline. For each method, the average test error, standard deviation (SD), and MCC are reported. Experiment Test Error ± SD (%) MCC ℓ1ℓ2FS 23.1 ± 8.6 0.54 FiltKBest & LR 22.0 ± 9.7 0.56 LASSO & LR 22.0 ± 8.2 0.56 ENET & LR 24.6 ± 7.1 0.51 microarrays-05-00015-t002_Table 2Table 2 Selection performance of Knowledge Driven Variable Selection (KDVS) and five different instances of the standard pipeline vs. the benchmark. Precision, Recall and F-measure are reported for KDVS, the best four methods of Table 1 and the t-test for GO terms and genes. GO Terms Genes Experiments Precision (%) Recall (%) F-measure (×10-3) Precision (%) Recall (%) F-measure (×10-3) KDVS all domains 44.0 12.7 197.4 7.5 25.5 115.5 ℓ1ℓ2FS 71.4 0.2 4.8 10.4 1.1 20.4 FiltKBest & LR 50.0 0.1 1.0 3.5 0.5 8.0 LASSO & LR 50.0 0.1 2.8 18.8 0.7 13.1 ENET & LR 62.5 0.2 4.8 16.7 0.9 17.1 t-test 50.0 0.1 1.0 2.5 0.2 4.2 ==== Refs References 1. Huang D. Sherman B. Lempicki R. Bioinformatics enrichment tools: Paths toward the comprehensive functional analysis of large gene lists Nucleic Acids Res. 2009 37 1 13 10.1093/nar/gkn923 19033363 2. Ashburner M. Ball C. Blake J. Botstein D. Butler H. Cherry J. Davis A. Dolinski K. Dwight S. Eppig J. Gene ontology: Tool for the unification of biology. The Gene Ontology Consortium Nat. Genet. 2000 25 25 29 10.1038/75556 10802651 3. Zycinski G. Barla A. Squillario M. Sanavia T. Di Camillo B. Verri A. Knowledge Driven Variable Selection (KDVS)—A new approach to enrichment analysis of gene signatures obtained from high-throughput data Source Code Biol. Med. 2013 8 2 10.1186/1751-0473-8-2 23302187 4. Kanehisa M. Goto S. KEGG: Kyoto encyclopedia of genes and genomes Nucleic Acids Res. 2000 28 27 30 10.1093/nar/28.1.27 10592173 5. Yu W. Wulf A. Liu T. Khoury M. Gwinn M. Gene Prospector: An evidence gateway for evaluating potential susceptibility genes and interacting risk factors for human diseases BMC Bioinform. 2008 9 15 10.1186/1471-2105-9-528 19063745 6. Gene Ontology Annotations Documentation Available online: ftp://ftp.geneontology.org/go/www/GO.gettingStarted.shtml (accessed on 3 June 2016) 7. Edgar R. Domrachev M. Lash A. Gene Expression Omnibus: NCBI gene expression and hybridization array data repository Nucleic Acids Res. 2002 30 207 210 10.1093/nar/30.1.207 11752295 8. Lesnick T.G. Papapetropoulos S. Mash D.C. Ffrench-Mullen J. Shehadeh L. de Andrade M. Henley J.R. Rocca W.A. Ahlskog J.E. Maraganore D.M. A Genomic Pathway Approach to a Complex Disease: Axon Guidance and Parkinson Disease PLoS Genet. 2007 3 15 10.1371/journal.pgen.0030098 17571925 9. Zhang Y. James M. Middleton F.A. Davis R.L. Transcriptional analysis of multiple brain regions in Parkinson’s disease supports the involvement of specific protein processing, energy metabolism, and signaling pathways, and suggests novel disease mechanisms Am. J. Med. Genet. Part B Neuropsychiatr. Genet. 2005 137B 5 16 10.1002/ajmg.b.30195 15965975 10. Zheng B. Liao Z. Locascio J.J. Lesniak K.A. Roderick S.S. Watt M.L. Eklund A.C. Zhang-James Y. Kim P.D. Hauser M.A. PGC-1α , A Potential Therapeutic Target for Early Intervention in Parkinson’s Disease Sci. Transl. Med. 2010 2 52ra73 52ra73 10.1126/scitranslmed.3001059 20926834 11. Irizarry R. Bolstad B. Collin F. Cope L. Hobbs B. Speed T. Summaries of Affymetrix GeneChip probe level data Nucleic Acids Res. 2003 31 e15 10.1093/nar/gng015 12582260 12. The Aroma Project Available online: http://www.aroma-project.org (accessed on 3 June 2016) 13. Ambroise C. McLachlan G.J. Selection bias in gene extraction on the basis of microarray gene-expression data Proc. Natl. Acad. Sci. USA 2002 99 6562 6566 10.1073/pnas.102102699 11983868 14. Barla A. Mosci S. Rosasco L. Verri A. A method for robust variable selection with significance assessment Proceedings of the ESANN 2008 Bruges, Belgium 23–25 April 2008 15. PyXPlanner Documentation Available online: http://slipguru.disi.unige.it/Software/PyXPlanner (accessed on 3 June 2016) 16. Everitt B. The Cambridge Dictionary of Statistics Cambridge University Press Cambridge, UK 2006 432 17. Tibshirani R. Regression Shrinkage and Selection via the Lasso J. R. Stat. Soc. Ser. B 1996 58 267 288 18. Zou H. Hastie T. Regularization and variable selection via the elastic net J. R. Stat. Soc. Ser. B 2005 67 301 320 19. De Mol C. Mosci S. Traskine M. Verri A. A Regularized Method for Selecting Nested Groups of Relevant Genes from Microarray Data J. Comput. Biol. 2009 16 677 690 10.1089/cmb.2008.0171 19432538 20. De Mol C. De Vito E. Rosasco L. Elastic Net Regularization in Learning Theory J. Complex. 2009 25 201 230 10.1016/j.jco.2009.01.002 21. Fardin P. Barla A. Mosci S. Rosasco L. Verri A. Varesio L. The l1-l2 regularization framework unmasks the hypoxia signature hidden in the transcriptome of a set of heterogeneous neuroblastoma cell lines BMC Genom. 2009 10 15 10.1186/1471-2164-10-474 19832978 22. Fardin P. Barla A. Mosci S. Rosasco L. Verri A. Versteeg R. Caron H. Molenaar J. Ora I. Eva A. A biology-driven approach identifies the hypoxia gene signature as a predictor of the outcome of neuroblastoma patients Mol. Cancer 2010 9 185 10.1186/1476-4598-9-185 20624283 23. Squillario M. Barla A. A computational procedure for functional characterization of potential marker genes from molecular data: Alzheimer’s as a case study BMC Med. Genom. 2011 4 15 10.1186/1755-8794-4-55 21726470 24. Mascelli S. Barla A. Raso A. Mosci S. Nozza P. Biassoni R. Morana G. Huber M. Mircean C. Fasulo D. Molecular fingerprinting reflects different histotypes and brain region in low grade gliomas BMC Cancer 2013 13 10.1186/1471-2407-13-387 23947815 25. L1L2Signature Documentation Available online: http://slipguru.disi.unige.it/Software/L1L2Signature (accessed on 3 June 2016) 26. L1L2Py Documentation Available online: http://slipguru.disi.unige.it/Software/L1L2Py (accessed on 3 June 2016) 27. PPlus Documentation Available online: http://slipguru.disi.unige.it/Software/PPlus (accessed on 3 June 2016) 28. WebGESTALT Homepage Available online: http://bioinfo.vanderbilt.edu/webgestalt/ (accessed on 3 June 2016) 29. Zhang B. Kirov S. Snoddy J. WebGestalt: An integrated system for exploring gene sets in various biological contexts Nucleic Acids Res. 2005 33 W741 W748 10.1093/nar/gki475 15980575 30. KDVS code repository Available online: https://bitbucket.org/slipguru/kdvs (accessed on 3 June 2016) 31. Zycinski G. Barla A. Verri A. SVS: Data and knowledge integration in computational biology Proceedings of the 2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society Boston, MA, USA 30 August–3 September 2011 6474 6478 32. Gene Ontology Consortium Available online: http://geneontology.org/page/download-ontology (accessed on 3 June 2016)

PMC005xxxxxx/PMC5003492.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020016microarrays-05-00016ArticleEvaluation of Solid Supports for Slide- and Well-Based Recombinant Antibody Microarrays Gerdtsson Anna S. 12Dexlin-Mellby Linda 12Delfani Payam 12Berglund Erica 1Borrebaeck Carl A. K. 12Wingren Christer 12*Negrini Massimo Academic Editor1 Department of Immunotechnology, Lund University, Medicon Village, Lund SE-22381, Sweden; anna.sandstrom_gerdtsson@immun.lth.se (A.S.G.); linda.dmellby@gmail.com (L.D.-M.); payam.delfani@immun.lth.se (P.D.); erica.berglund@immun.lth.se (E.B.); carl.borrebaeck@immun.lth.se (C.A.K.B.)2 Clinical Cancer Research using Emerging Advanced Technologies for Health (CREATE), Lund University, Medicon Village, Lund SE-22381, Sweden* Correspondence: christer.wingren@immun.lth.se; Tel.: +46-46-222-4323; Fax: +46-46-222-420008 6 2016 6 2016 5 2 1614 10 2015 09 12 2015 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).Antibody microarrays have emerged as an important tool within proteomics, enabling multiplexed protein expression profiling in both health and disease. The design and performance of antibody microarrays and how they are processed are dependent on several factors, of which the interplay between the antibodies and the solid surfaces plays a central role. In this study, we have taken on the first comprehensive view and evaluated the overall impact of solid surfaces on the recombinant antibody microarray design. The results clearly demonstrated the importance of the surface-antibody interaction and showed the effect of the solid supports on the printing process, the array format of planar arrays (slide- and well-based), the assay performance (spot features, reproducibility, specificity and sensitivity) and assay processing (degree of automation). In the end, two high-end recombinant antibody microarray technology platforms were designed, based on slide-based (black polymer) and well-based (clear polymer) arrays, paving the way for future large-scale protein expression profiling efforts. antibody microarraysaffinity proteomicssolid supportarray-in-wellsource platemicroarray scanning ==== Body 1. Introduction Affinity proteomics, mainly represented by antibody microarrays, have emerged as an important tool within proteomics, providing unique opportunities for multiplexed protein expression profiling in both health and disease [1,2,3]. In this context, we have developed a recombinant antibody microarray technology platform for protein profiling of crude, directly-labelled proteomes [4,5,6]. The design and performance of antibody (protein) microarrays and how they are processed are dependent on several factors, of which the interplay between the antibodies (proteins) and the solid surfaces plays a pivotal role [5,7,8,9,10,11,12,13]. In more detail, the solid surfaces, including both the source plate to which the antibodies are loaded prior to printing and the surface onto which the antibodies are printed, will have a direct impact on the printing performances, array format (slide and well based), assay performances, assay processing (degree of automation) and sensing (slide- and plate-based scanners) [8,14,15]. Early development work by us and others has, however, focused almost entirely on the design and evaluation of slide-based planar solid supports for antibody microarrays [5,7,8,9,10,11,12]. Albeit successful, resulting in high-performing antibody microarray set-ups, e.g., [5,16], the potential of (novel) solid surfaces has not been (re-)evaluated in recent years. Additionally and most importantly, a comprehensive view, addressing all of the above surface-related issues in a combined manner, remains to be presented, leaving room for additional array technology platform development and further fostering of our understanding of the antibody-surface interplay. Ideally, the source plate should be inert and consistently display low protein-binding properties, but in the context of antibody microarrays, published data demonstrating this key feature is absent. In fact, we have recently generated data questioning the printing performances due to significant and inconsistent antibody binding properties displayed by a common and frequently-used source plate. In contrast, the solid microarray surfaces should display high antibody binding capacity and biocompatibility, while any non-specific (background) binding should be minimized [5,7,8,9,10,11]. In our efforts, a hydrophilic polymer slide, to which the antibodies are randomly adsorbed, have so far served as the best-performing solid support [5]. In fact, hydrophilic surfaces have often been reported as favorable for protein adsorption, as they yield homogeneous spots and stable arrays that can be stored for several months [8,17]. Recently, the possibility to use also enzyme-linked immunosorbent assay (ELISA) plates with flat bottoms as solid microarray supports was demonstrated [18,19], representing an attractive assay format more compatible with high-throughput clinical efforts. Notably, the sensing (scanner availability and performance) and compatibility of such well-based array layouts with, in particular, recombinant antibody microarray set-ups remain to be demonstrated. In this study, we have taken on the first comprehensive view and investigated the combined impact of the solid surfaces on the printing performances, assay format, assay performances, assay processing and sensing of recombinant antibody microarrays. A schematic layout of the array set-up and the technical features that have been addressed are shown in Figure S1. To this end, we have evaluated the use of eight source plates and fourteen solid microarray supports, including both slide- and well-based designs. In parallel, we have also compared manual and semi-automatic array handling, as well as three slide- and/or plate-based scanners for sensing. The results showed that the impact of the solid surfaces is significant and that the overall microarray technology platform performances could be further improved by considering these fundamental phenomena. The observed interplay between the antibodies and the solid surfaces of both the source plate and microarray support, and its impact on array performances in general, is discussed. 2. Materials and Methods 2.1. Surfaces Eight 384-well plates were evaluated as source plates, including clear polypropylene (PP) ABgene natural V-wells (ABgene, Epsom, UK), Corning clear polystyrene (PS) non-binding surface (NBS) treated (Corning, NY, USA), Corning white PS NBS treated (Corning), Genetix clear PS (Molecular Devices, Sunnyvale, CA, USA), Genetic clear PP (Molecular Devices), NUNC clear PS (NUNC, Roskilde, Denmark), NUNC black PP (NUNC) and PerkinElmer PS black ProxiPlate (PerkinElmer Life & Analytical Sciences, Wellesley, MA, USA) (Table 1). Eleven planar slides and three 96-well plates displaying a range of surface chemistries, surface geometries and binding chemistries were evaluated as antibody microarray solid support (Table 2). The slides were supplied by Corning, Schott (Jena, Germany), PolyAn (Berlin, Germany), Whatman (Florham Park, NY, USA), Arrayit (Sunnyvale, CA, USA), NUNC and Sigma (St Louise, MO, USA). The 96-well plates were supplied by Scienion (Berlin, Germany), Costar and NUNC. Clear MaxiSorp slides were used as a reference when evaluating plate-based supports. The surfaces were selected to represent a set of potentially high-performing surfaces for antibody microarrays, based on different surface properties and surface chemistries and provided by different vendors. 2.2. Antibodies In total, 102 recombinant single-chain fragment variable (scFv) antibodies (Table S1), stringently selected from an in-house-designed phage display library, were produced in 100 mL Escherichia coli cultures. In brief, the antibodies were purified from the cell supernatant using affinity chromatography on Ni2+-NTA agarose (Qiagen, Hilden, Germany) and eluted in 250 mM imidazole. The buffer was changed to PBS by extensive dialysis, and the antibodies were stored at 4 °C until used for microarray production. The protein concentration was determined by measuring the absorbance at 280 nm, and the degree of purity and integrity of the scFv antibodies was verified with 10% SDS-PAGE (Invitrogen, Carlsbad, CA, USA). 2.3. Samples Four well-characterized, de-identified human serum samples were used as model samples, including NS80 (a large pool of healthy controls), C1qD (C1q and properdin deficient), C3D (C3 deficient) and C4D (C4 deficient). While the former (healthy) sample was used for a majority of the experiments, the latter three were only used in experiments evaluating antibody specificities. All samples were collected at Skåne University Hospital (Lund, Sweden). Crude serum samples were diluted 1:45 in PBS and labelled with 0.6 mM biotin (EZ-link Sulfo-NHS-Biotin, Pierce, Rockford, IL, USA) for 2 h on ice, as previously described [4,5]. Unconjugated biotin was removed by extensive dialysis against PBS, whereafter the samples were aliquoted and stored at −20 °C. When used for microarray analysis, the labelled samples were diluted 2.5–160 times (10 times in the standard assay) in 1% (w/v) milk (Semper, Sundbyberg, Sweden) and 1% (v/v) Tween-20 (Merck, Whitehouse Station, NJ, USA) in PBS (PBS-MT). Pure BSA (SaveenWerner, Malmö, Sweden) and C3 (Quidel, San Diego, CA, USA) were biotinylated at a molar ratio of biotin:protein of 3:1 as described above. Non-specific surface binding was evaluated by applying 1 μg/mL Alexa-647-labelled streptavidin (SA647) (Invitrogen) or 0.2 mg/mL biotinylated BSA. Biotinylated C3 (5 μg/mL) was used in well-based array experiments. 2.4. Microarray Production and Processing Microarrays were produced by printing 12–60 different scFvs antibodies (at a protein concentration of (40–400 μg/mL) per array, using a non-contact printer based on piezo technology (SciFLEXARRAYER S11, Scienion). Biotinylated BSA (b-BSA) was used as a position marker. Each individual scFv antibody was printed as single droplets (~300 pL/drop) in four to eight replicates at a spot-to-spot distance of 200 or 300 μm. The arrays were stored overnight at room temperature (RT) prior to use. In the long-term storage experiments, the printed antibody microarray slides were stored for up to 42 days prior to analysis. For the evaluation of source plates, b-BSA (5 μg/mL) from 6–12 different wells was printed in 20 spot replicates on 1–14 individual subarrays. The source plates were incubated 1.5 h prior to printing. The antibody microarray slides were processed either manually or semi-automatically in a Protein Array Workstation (PAW) (PerkinElmer). For manual processing, individual sub-arrays were created by using silicon incubation chambers (Schott) or a hydrophobic pen (Dako, Glostrup, Denmark) for slides that did not fit the incubation chambers (Silane-Prep (Sigma) and GAPSII (Corning)). Arrays were blocked for 1 h at RT. While well-based arrays were only blocked with 1% (w/w) milk, 1% (v/v) Tween-20 in PBS (PBS-MT), several different types of blocking agents (SaveenWerner, BDH Chemicals (Poole, Great Britain), Semper, Merck and Sigma) were evaluated for slide-based solid supports (Table S2). After blocking, the arrays were processed at RT (4 °C for Protein array Workstation (PAW) processing). The slides were first washed in 0.05% (v/v) Tween-20 in PBS (PBS-T) (0.5% (v/v) Tween-20 in PBS for PAW processing) before the labelled sample was added for 1 h. After washing in PBS-T, the arrays were incubated for 1 h with 1 μg/mL SA647. Finally, the slides were washed in PBS-T and immediately dried under a stream of nitrogen gas (slides) or in air (plates) and subsequently scanned. The slides were scanned using a confocal microarray scanner (ScanArrayExpress HT, PerkinElmer,) at three to five different settings of photomultiplier tube (PMT) gain and laser power (LP) and quantified using the ScanArray Express software 4.0 (PerkinElmer) with the fixed circle method. Plates were scanned in an light-emitting diode/charge-couple device (LED/CCD) plate scanner (Sensovation, Radolfzell, Germany), with adjustable LED power and exposure time settings, and automatically quantified in the built-in Genomica software (Genomica, Madrid, Spain). In one experiment, plates and slides were scanned in the confocal LS reloaded combined plate and slide laser scanner (Tecan, Männedorf, Switzerland), with different gain settings (fixed laser power). In this particular case, the plates had to be scanned upside down through the bottom polymer layer due to laser angle restrictions, while slides were scanned with arrays facing upwards. For each spot, the local background was subtracted. In the case of more than four printed replicate spots, the highest and lowest replicates were first automatically excluded in order to compensate for any local defects, so that each data point used represents the mean value of four replicates. The intensity is given as mean intensity per pixel. For the evaluation of storage stability, arrays were normalized between days of analysis, using a semi-global normalization approach, as previously described [20,21]. It should be noted that in no case was signal overlap from neighboring wells observed when measuring the fluorescence from different wells. 2.5. Antibody Footprint The Web Antibody Modelling procedure [22] was used to generate structural homology models for five of our scFv antibodies in order to estimate the size of their foot print. The mean foot print was estimated to be 50 × 40 Å. 3. Results 3.1. Evaluation of Source Plates Eight different 384-well plates were evaluated as a protein (antibody) source plate by printing the same stock solution of biotinylated BSA picked from multiple wells (n = 12) on each plate and subsequently determining the signal intensity of the deposited spots (Table 1). The results showed that the reproducibility, expressed as the coefficient of variation (CV), of the printing process decreased in the order of NUNC black PP < Genetix PS < Genetix PP < ABgene PP < Corning clear PS < NUNC clear PS < PerkinElmer < Corning white PS and ranged from 3%–16%. Furthermore, the maximum percentage difference in signal intensity between spots ranged from 11%–55%, again with the NUNC black PP, Genetix PS and Genetix PP plates displaying the smallest variations (Table 1). Hence, the data showed large well-to-well variations in protein binding for some of the source plates, indicating significant surface heterogeneity. Noteworthy, the data also showed that observed spot signal intensities differed (up to 100%) depending on which source plate the BSA was picked from, demonstrating large differences in unwanted protein binding (Figure 1). The highest signal intensities (i.e., the lowest protein binding) were observed when BSA was collected from NUNC PS/NUNC PP and the lowest intensities (i.e., the highest protein binding) from Corning white PS/Corning clear PS plates. The observed protein binding behavior could neither be explained by the nature of the surfaces (e.g., PP vs. PS) (Table 1 and Figure 1) nor by the performances of the printer and/or the solid support on which the protein was dispensed (data not shown). Taken together, the data showed that the NUNC black PP plate was the preferred choice as the source plate, while many of the other source plates displayed significant and inconsistent protein binding properties. 3.2. Slide-Based Solid Supports: Surface Fouling The ability to block the slide-based solid supports from non-specific background binding, i.e., surface fouling, was then examined. To this end, the eleven supports were treated with twenty different blocking solutions (Table S2), and the level of non-specific protein binding to the surface after exposure to a crude, labelled serum sample was determined. Representative results illustrated for five slides and sixteen blocking agents are shown in Figure 2A. The result showed that two (FAST and SuperProtein) of the supports could not be adequately blocked to be compatible with the scanner parameters (laser power and PMT gain) required for the sensitive detection of the recombinant scFv microarrays, and they were therefore excluded from further testing (Figure 2A and Figure S2). In contrast, four supports (black MaxiSorp, Nexterion H, Nexterion P and GAPSII) could be satisfactorily blocked using a few selected blocking agents, while five supports (epoxy polymer, epoxy glass, NHS glass, NHS polymer and Silane-Prep) were efficiently blocked irrespective of the blocking solution used (Figure 2A and Figure S2). Hence, out of eleven supports, nine could be effectively blocked from surface fouling and were selected for further testing. 3.3. Slide-Based Solid Supports: Surface Fouling and Spot Properties Next, the remaining nine solid supports were further compared with respect to surface fouling and spot properties (size, morphology and intensity). Focused antibody microarrays were printed, and the slides were blocked with the top four to six blocking agents identified for each support (Figure S2); a crude, directly-labelled serum sample was analyzed. Antibody spots with the highest specific signal intensities (range 1400–30,900 Relative Fluorescsnce Unit (RFU), background corrected) were obtained when the solid supports were blocked with 5% (w/v) BSA (Silane-Prep, Nexterion P and GAPSII), 1% (w/v) fat-free milk (Nexterion H, NHS polymer and black MaxiSorp) or 0.5% (v/v) Tween-20 (epoxy glass, epoxy polymer and NHS glass) (Table 3 and Figure 2B). The surface fouling was found to be low (range 150–1100 RFU) and increased in the order of Nexterion P < Nexterion H < epoxy polymer < NHS polymer < NHS glass < black MaxiSorp < Silane-Prep < epoxy glass < GAPSII. However, the surface fouling varied across some of the slides, indicating adverse surface irregularities, a key feature that increased in the order of black MaxiSorp < Nexterion P < epoxy polymer < Nexterion H < NHS polymer < NHS glass < epoxy glass < GAPSII < Silane-Prep (Table 3). The specific antibody binding pattern also varied considerably across the supports (Figure 2B and Figure S3), demonstrating the impact of the solid support on the antibody array assay. The results showed that the average antibody signal intensity decreased in the order of epoxy glass > epoxy polymer > NHS glass > Nexterion H > black MaxiSorp > GAPSII > Nexterion P > Silane-Prep > NHS polymer (Table 3). Further, the spot size was uniform (varied ≤1.1-times) on five of nine supports, but differed (≤2.8-times) on all four NHS-functionalized supports (Nexterion H, Nexterion P, NHS glass and NHS polymer) (Figure 2B and Table 3). As could be expected, the average spot size differed across the supports (90–165 μm), indicating different wetting properties. Moreover, circular and homogeneous spots were in general observed on all supports, with the exception of Nexterion P and Nexterion H, which predominantly displayed poor, non-uniform spot morphologies (Table 3). The latter two slides also showed ring-shaped spot drying effects, as did Silane-Prep, GAPSII and epoxy glass slides (Figure 2B). Taken together, the supports were ranked based on surface fouling (surface regularity), spot intensity, spot size and spot morphology, resulting in three of nine slides being selected for further testing, including epoxy glass, epoxy polymer and black MaxiSorp. 3.4. Slide-Based Solid Supports: Reproducibility, Sensitivity and Dynamic Range The assay reproducibility was determined by analyzing the same serum sample on four identical antibody subarrays printed on two slides of each solid support. The results showed that the spot-to-spot, array-to-array and slide-to-slide reproducibility all decreased in the order of black MaxiSorp > epoxy polymer > epoxy glass (Table 4). Hence, the data implied that the black MaxiSorp slides displayed a more consistent surface well adapted for microarray analysis. Next, the assay sensitivity was investigated by analyzing seven serial dilutions of a serum sample, with a total protein concentration ranging from 12.5–800 μg/mL, on a focused antibody microarray. The results showed that even low-abundant analytes (e.g., IFN-γ, IL-12 and IL-4) could be detected in the most diluted sample on black Maxisorp and epoxy polymer slides, while a four-times (IFN-γ and IL-12) or eight-times (IL-4) higher total protein concentration was required for detection on the epoxy glass slides (Figure S4). However, the observed signal intensities were stronger on the two epoxy supports than on the black MaxiSorp slides, but on the epoxy supports, the signals were also more saturated at a lower protein concentration, indicating a more limited working dynamic range. Overall, the data implied that the performance of the supports decreased in the order of black MaxiSorp > epoxy polymer > epoxy glass. 3.5. Slide-Based Solid Supports: Specificity and Stability To investigate the influence of the solid support on the assay specificity, four well-characterized serum samples were profiled, including NS80 (a pool of healthy controls), C1qD (C1q and properdin deficient), C3D (C3 deficient) and C4D (C4 deficient) (Figure 3). As expected, no signals were detected for either C1q/properdin, C3 or C4 when the corresponding deficient serum samples were profiled on black MaxiSorp and epoxy polymer slides. While similar results were observed for the C1qD sample on epoxy glass support, five of six C3-specific antibodies and three of four C4-specific antibodies generated significant signals when profiling the C3D and C4D samples, respectively. Of note, no signals were observed on either of the supports if the antibody microarrays were hybridized with only biotinylated BSA and Alexa-647-labelled streptavidin (SA647) or only SA647 (data not shown). Thus, the results implied that the antibody specificity was maintained on the black Maxisorp and epoxy polymer supports, while the antibody-surface interplay in part impaired the specificity on the epoxy glass support. The long-term stability of the printed arrays was then tested by printing all arrays on Day 0 and then storing them in a dried out state for 0, 3, 10 and 42 days prior to assay processing. The results showed that the activity of the arrayed antibodies increased with storage time up to ten days on all three supports, whereafter the activity levelled off (Figure S5). The observed increase in antibody activity (signal intensities) was larger on the epoxy glass support than on the other two supports, but this support also gave rise to more non-specific binding to the printed antibodies (cf. Figure 3 and Figure S5). 3.6. Slide-Based Solid Supports: Semi-Automatic Array Handling Next, we turned from manual to semi-automatic array handling by processing the slides in a Protein Array Workstation (PAW). First, titration of a labelled serum sample showed that a total protein concentration of 1.0 mg/mL (0.2 mg/mL in manual handling) was the preferred concentration to enable also low-abundant analytes to be targeted (data not shown). Second, the spot properties were found to differ considerably across the three supports (Figure 4). While small, homogeneous spots of adequate morphology were observed on both black MaxiSorp and epoxy polymer supports, larger and often ring-shaped spots were obtained on epoxy glass supports (not observed during manual handling, cf. Figure 2 and Figure 4). Hence, the black MaxiSorp and epoxy polymer supports, but not epoxy glass, were compatible with automated PAW analysis. Third, the binding patterns were again found to differ depending on the solid support (cf. Figure 4 and Figure S3), demonstrating the importance of the antibody-surface interplay on the antibody activity. While more similar binding patterns were observed on the black MaxiSorp and epoxy polymer slides, the support identified to generate the most non-specific spots, epoxy glass (Figure 3), was also found to generate more apparent detectable spot features. Overall, the performances of the supports were found to decrease in the order of black MaxiSorp > epoxy polymer > epoxy glass. 3.7. Well-Based Solid Supports: Surface and LED/CCD Plate Scanner Three 96-well plates, SciPLEXPLATE, Costar and clear MaxiSorp, were evaluated as well-based solid supports. To this end, 23-plex antibody microarrays were printed and applied for serum protein profiling, using an LED/CCD-based plate scanner for sensing. The results showed that adequate spot morphology, low surface fouling and similar binding patterns were obtained on all three supports (Figure 5A). Still, the clear MaxiSorp plates generated higher signal intensities for at least a subset of antibodies (e.g., C1q, C3 and properdin), and this plate was therefore selected as the well-based solid support for the remaining part of the study. Next, we compared the performance of well-based arrays on clear MaxiSorp plates (LED/CCD plate scanner) with that of slide-based arrays on black MaxiSorp slides (confocal slide scanner, denoted PerkinElmer (PE) scanner). The results showed that similar signal intensities were obtained targeting high- to medium-abundant serum analytes (e.g., complement proteins), indicating equal performance (data not shown). However, when instead, predominantly low-abundant serum analytes (e.g., cytokines) were targeted, the observed binding patterns differed, with less analytes being detected on well-based arrays (Figure 5B) than on planar arrays (Figure 5C). Hence, the data indicated that the assay sensitivity was higher for the slide-based array set-up than for the well-based array set-up. 3.8. Well-Based Solid Supports: Confocal LS Slide/Plate Scanner In an attempt to increase the sensitivity for the well-based arrays, we introduced a confocal LS slide/plate scanner (denoted LS scanner). Consequently, we repeated the above experiments and compared the performance of well-based antibody arrays (clear MaxiSorp plates, LS scanner) (Figure 6A) with that of slide-based antibody arrays (black MaxiSorp slides, PE scanner) for serum protein profiling focusing on twenty mainly low-abundant analytes. While all but three analytes (IL-4, MCP-4 and TNF-β) now were detected on both set-ups, the results showed that higher and more dynamic signal intensities were still obtained on slide-based arrays (Figure 6B). Finally, we investigated whether the observed differences in performance were due to the support (clear MaxiSorp vs. black MaxiSorp) and/or scanner (PE scanner vs. LS scanner). To this end, well-based arrays (clear MaxiSorp) (Figure 6C) and slide-based arrays (clear MaxiSorp and black MaxiSorp), based on serial dilutions of six C3-specific antibodies, were produced and probed with pure, labelled C3 and scanned in the LS scanner and/or PE scanner. First, the results showed that higher and more dynamic signal intensities were obtained when slide-based arrays were scanned using the PE scanner compared to the LS scanner (Figure 6D). Second, similar signal intensities were obtained for slide-based arrays on clear and black MaxiSorp slides, irrespective of the scanner used. In this context, it might be of interest to note that the surface fouling was somewhat higher on clear MaxiSorp than on black MaxiSorp slides, often resulting in lower signal-to-noise ratios (data not shown). Noteworthy, when the same scanner (LS scanner) was used, equal or higher signal intensities were obtained for well-based arrays than for slide-based arrays produced on the same surface (clear MaxiSorp) (Figure 6D). Taken together, the results suggested that the observed differences in performance mainly reflected the choice of scanner. Still, slide-based arrays based on the black MaxiSorp support scanned in the PE scanner appeared to represent the best performing antibody microarray set-up. 4. Discussion To design high-performing antibody microarrays, it will be absolutely necessary to develop an integrated technology platform that has been optimized in all processing steps, where the interplay between the surfaces and the antibody is anticipated to play a central role [9,14]. Despite major efforts, several limitations in the development process have been pin-pointed, leaving room for additional technical improvements and further fostering of our understanding of the antibody-surface interplay. First, very few studies have been reported taking on such a comprehensive view; for a review, see [6,14,23]. Second, in previous studies, a major focus has been placed on the choice of solid array support, while the impact of the source plate has not been reported [14]. Third, the potential of (novel) array surfaces and array/assay formats has not been evaluated or re-evaluated in recent years (to explore potential cooperative effects due to other parallel developments of the array technology). Fourth, the sensing (scanner performance) and compatibility of well-based array layouts with in particular recombinant antibody microarrays are features that remain to be explored and exploited. In this study, we have therefore re-visited and further investigated the antibody-surface interplay, including both source plates and array surfaces. Uniquely, we took on a comprehensive approach and demonstrated the impact of the antibody-surface interplay on each step of the microarray process, including the printing process, array/assay performance (e.g., functionality, sensitivity and stability), degree of automation, array/assay format and sensing. It was demonstrated that the source plate had a profound influence on the printing performance, as reflected by significant and inconsistent protein binding, indicating unexpectedly high protein binding properties and adverse well-to-well surface irregularities. Hence, the actual printing concentration of an antibody differed not only from plate-to-plate, but also from well-to-well of the same plate, in the end impairing the overall assay reproducibility and sensitivity. The hydrophobicity of the surfaces is an important feature when discussing non-specific protein loss due to adsorption [24]. Although the protein loss could not simply be explained by the nature of the surfaces (PP vs. PS, PS being more hydrophobic in general), the highest protein loss was indeed observed on two PS surfaces. This lack of correlation might to some extent be explained by the fact that the surfaces were subjected to various surface treatments during their production process, which might influence their protein loading capacity in different ways. Hence, further experiments will be required to unravel their protein binding behavior. To the best of our knowledge, published data addressing this fundamental choice of source plate for antibody (protein) microarrays is still very limited [14]. Here, we thus present the first validated (inert, low and consistent protein binding properties) source plate, based on black polypropylene, suitable for antibody microarray production. Turning to the solid array support, we used the ability to block the surfaces for non-specific background binding, or surface fouling, as a key cut-off parameter [5,7,8,9,10,11]. Hence, increasing the signal-to-noise ratio is a vital platform parameter. The blocking step might be problematic in that: (i) large blocking molecules could sterically hinder binding to smaller probes (the molecular weight of our scFv antibodies is 28 kDa); (ii) the blocking molecules might not be completely non-fouling; and (iii) the blocking molecules might interact non-specifically with the sample. In the end, we found protein- (BSA or milk) and/or Tween-20-, but not polymer-based, blocking buffers to be the preferred blocking agents (for the surfaces tested here). While nine of 11 slides could be adequately passivated, we failed completely to block two supports, FAST slides (nitrocellulose surface with high protein binding capacity [25,26,27] and SuperProtein slides (hydrophobic surface based on a 150 μm-thick polymer membrane). We have, however, repeatedly observed lower sensitivity and higher background binding on nitrocellulose- and polymer matrix-based substrates [5], implying that these surfaces might not be compatible with recombinant scFv arrays targeting crude serum proteomes, but rather seem more suitable for reversed antibody arrays [28,29]. Noteworthy, hydrophobic surfaces have in general also been found to yield a higher degree of denatured immobilized proteins and non-specific binding [30], while we found a hydrophilic polymer support (black MaxiSorp) to be the best performing support in both this and a previous study [5]. Six of the slide-based solid supports displayed uneven background signals (e.g., GAPSII and Silane-Prep) and/or non-homogeneous spots (e.g., GAPSII, Silane-Prep, Nexterion P and Nexterion H) of different sizes (e.g., Nexterion H, Nexterion P, NHS glass and NHS polymer) and intensities (Silane-Prep and NHS polymer), indicating unfavorable surface irregularities impairing the immobilization and/or functionality of the arrayed antibodies. Silanized surfaces (GAPSII and Silane-Prep) are inherently hydrophobic, which might, as discussed above, cause protein to denature [30], while supplementing the printing buffer with glycerol or other additives to protect the antibodies might be beneficial [8,13]. We did not, however, optimize the printing buffer for each individual surface within this study, due to logistical limitations and technical issues (e.g., spotter compatibility with glycerol), but rather aimed to find a support compatible with the standard buffer (PBS) predominantly used in our overall work schedule. Consequently, three of eleven slide-based supports, including black MaxiSorp, epoxy polymer and epoxy glass, were thus subjected to a more detailed evaluation. While the spot-to-spot reproducibility was excellent [14] on all three supports indicating adequate local surface regularity, the array-to-array and slide-to-slide reproducibility was only adequate on black MaxiSorp, indicating more pronounced surface irregularities across/between slides of the two epoxy supports and, in particular, the epoxy glass slides. In accordance, the ability of black MaxiSorp to act as planar support for highly reproducible antibody array assays have previously been demonstrated [5]. In addition, the two best-performing substrates, black MaxiSorp and epoxy polymer, were both found to enable selected low-abundant serum analytes to be detected in crude, directly biotinylated serum diluted 7200 times, corresponding to a total protein concentration of about 12.5 μg/mL. This assay sensitivity is at the very high end of what have previously been reported for other array set-ups, e.g., [5,16]. The superior behavior of the black MaxiSorp and epoxy polymer supports was further manifested by the fact that these two surfaces, but not epoxy glass, were compatible with the PAW instrument (more stringent washes, etc.), enabling semi-automatic array handling. The protein loading capacity of array surfaces is a key feature, which in direct experimental terms is very difficult to assess. The footprint of our scFv antibodies was estimated to be 50 × 40 Å, suggesting a theoretical loading of 5 × 1010 protein molecules per mm2, assuming a monolayer. However, as we used surfaces with both a 2D and 3D surface geometry, the relevance of this number could be argued. Although important, we therefore rather used the on-chip functional activity, which can be experimentally determined in terms of intensity per spot, as an array surface ranking tool. In accordance with previous results [11,12,14,31], the on-chip functionality of the arrayed antibodies appeared to increase with storage time on all three supports and levelled off after ten days. The cause(s) for this is still not clear, but could at least partly be explained by reasoning that those probes that denatured upon deposition onto the solid support have refolded and thereby regained their functional activity. Most importantly, on Day 0, the activity of the arrayed antibodies and the observed binding patterns differed across these three top supports, indicating striking differences in the antibody-surface interplay. Although strong(er) signals in many cases were observed on the epoxy glass support, many of these spots also appeared to be false-positive, indicating unfavorable surface-antibody interactions. Additional experiments will be required in order to elucidate the underlying reasons (e.g., electrostatic and/or hydrophobicity effects) for these key observations. In addition, even though this platform is based on recombinant scFv antibodies constructed around the same identical scaffold, microarray adapted by molecular design [5,14,23], thus exhibiting similar molecular properties (differing only in their complementarity determining regions), we observed antibody clone-dependent differences. In accordance with previously-published work, this highlighted the importance of carefully evaluating the impact of the solid support on the functional on-chip activity/stability of the arrayed antibodies [5,7,8,9,10,11,12] and of using microarray-adapted antibody probes that are as similar as possible to minimize any bias introduced by the probes [23]. Well-based antibody microarrays are an attractive approach, especially from a clinical assay implementation point of view, and the first generation of mainly low-to-medium dense set-ups have been presented [18,19,32,33,34,35]. Here, we designed the very first well-based recombinant scFv antibody microarray set-up. Compared to slide-based solid supports, the availability of plates with specific surface (3D) coatings represents a limitation. While classical ELISA supports, such as hydrophobic polystyrene in some cases, have been found to cause partial protein denaturation [36], we tested three polystyrene-based plates that all resulted in highly functional antibody microarrays, with clear MaxiSorp plates representing the top candidate. Unfortunately, black MaxiSorp plates could not be evaluated due to plate scanner incompatibility. The overall performance (e.g., sensitivity) of the well-based arrays were found to be adequate, in particular when targeting high- to medium-abundant serum analytes, clearly indicating the potential of the set-up. Compared to the slide-based array set-up, the results first showed lower and less dynamic signal intensities, indicating inherently lower performance and, thus, potential limitations towards targeting low-abundant serum analytes. This scenario was, however, refined, and additional experiments showed that equal or higher signal intensities were obtained for well-based arrays than for slide-based arrays when the key technical differences (choice of surface and scanner) were neutralized, suggesting that the initially observed differences in performance mainly reflected the choice of scanner (and its performance). Of note, the plates had to be scanned upside down in the LS scanner due to laser angle restrictions associated with standard format 96-well plates, while the sensitivity might be even further improved if scanner-compatible plates could be utilized (not pursued due to logistical limitations). Still, the results clearly outlined the potential of well-based recombinant antibody microarrays for protein expression profiling. 5. Conclusions In conclusion, we have taken on the first comprehensive view and investigated the impact of the solid surfaces on the printing performance, assay format, assay performance, assay execution and sensing for recombinant antibody microarrays. The results clearly demonstrated the importance of considering the surface-antibody interplay when designing high-performing antibody microarrays. The practical output is two high-end array platforms, based on slide-based arrays (black polystyrene) and well-based arrays (clear polystyrene), paving the way for future protein expression profiling efforts. Acknowledgments This study was supported by grants from the Swedish Research Council (VR-NT), Sweden´s Innovation Agency (VINNOVA) and the Foundation of Strategic Research (Strategic Centre for Translational for Translational Cancer Research—CREATE Health (www.createhealth.lth.se). The authors gratefully acknowledge Genomica, Sensovation, and Tecan for making scanners available. Supplementary Materials The following are available online at http://www.mdpi.com/2076-3905/5/2/16/s1. Figure S1: Schematic illustration of the recombinant antibody microarray technology platform, also highlighting the main technical issues addressed in the study, Figure S2: Evaluation of blocking buffers for slide-based solid supports. The slides were blocked with 16 different blocking solutions (Table S2) and then exposed to crude, directly-biotinylated serum sample, whereafter any non-specific background binding (surface fouling) was detected, Figure S3: Comparison of antibody binding pattern on three different slide-based solid supports. A serum sample was profiled on a 14-plex antibody microarray printed on slide-based supports, including black MaxiSorp, epoxy polymer and epoxy glass. Quantified signals were collected from the highest possible scanner setting without any spot saturation (60% PMT gain/90% LP for black MaxiSorp slides and 50% PMT gain/90% LP for the epoxy-coated slides), Figure S4: Comparison of assay sensitivity and the dynamic range of slide-based antibody microarrays. Serial dilutions of a serum sample, ranging from a total protein concentration of 12.5–800 µg/mL, was profiled on black MaxiSorp, epoxy polymer and epoxy glass, Figure S5: Evaluation of long-term storage on the activity of slide-based antibody microarrays. Arrays were printed on black MaxiSorp, epoxy polymer and epoxy glass slides. All antibody microarrays were printed on Day 0 and then stored dried out for 0, 3, 10 or 42 days prior to array handling (blocking, washing, sample incubation, etc.). The antibody activity for four representative antibodies, targeting IL-4, TNF-α, IFN-γ and C1q, was assessed, Table S1: Antibodies used, Table S2: Blocking agents evaluated for blocking of slide-based solid supports. Click here for additional data file. Author Contributions Anna S. Gerdtsson performed the experiments and supervised; Erica Berglund participated in the planning of the project and the writing of the paper; Linda Dexlin-Mellby performed parts of the experiments, participated in the writing of the paper; Payam Delfani Performed parts of the experiments, participated in the writing of the paper; Erica Berglund performed parts of the experiments; Carl A. K. Borrebaeck participated in the planning of the project and the writing of the paper; Christer Wingren planned and supervised the project, wrote the paper. Conflicts of Interest The authors declare no conflict of interest. Abbreviations b-BSA: biotin-BSA; EA: ethanolamine; LP: laser power; NaB: sodium borate; NBS: non-binding surface; NC: nitrocellulose; NTA: nitrilotriacetic acid; PAW: Protein Array Workstation; PP: polypropylene; PS: polystyrene; RFU: relative fluorescence units; SA647: streptavidin-Alexa647; scFv: single-chain fragment variable. Figure 1 Evaluation of 384-well plates as protein (antibody) source plates for the production of antibody microarrays. The same stock solution of biotinylated BSA was loaded into 12 wells on each source plate and printed on black Maxisorp slides (six subarrays/slide). The spot signal intensities were determined, and the mean value over all six subarrays per well uptake was plotted for the source plates for which the two highest and two lowest signals were obtained. Figure 2 Evaluation of slide-based solid supports for antibody microarrays. The supports were compared with respect to blocking buffer, surface fouling, spot morphology and signal strength. (A) The slides (no antibodies printed) were first blocked and then incubated with crude, labelled serum and scanned after washing. Any observed signal intensity represents non-specific background binding. Representative scans of FAST, black MaxiSorp, Nexterion H, epoxy polymer and NHS glass slides are shown. The blocking agents tested are shown to the right; (B) Scanned microarray images of 14 × 8 antibody microarrays on epoxy glass, epoxy polymer, NHS glass, Nexterion H, black MaxiSorp, GAPSII, Nexterion P, Silane-Prep and NHS polymer slides. The printed antibody arrays were blocked and then incubated with crude, labelled serum before scanning. Figure 3 Influence of the slide-based solid support on the specificity of the arrayed antibodies. A 12-plex antibody microarray (6 × C3, 4 × C4, 1 × C1q and 1 × properdin) was used to profile four well-characterized serum samples, including NS80 (large pool from healthy donors), C1qD (C1q and properdin deficient), C3D (C3 deficient) and C4D (C4 deficient). Figure 4 Effect of semi-automatic array handling on slide-based array performance. A serum sample was profiled on a 34 × 8 antibody microarray printed on epoxy glass, epoxy polymer and black MaxiSorp slides that were semi-automatically processed in a Protein Array Workstation. All printed antibodies (n = 34) were unique, but several of the clones targeted the same protein (but most likely different epitopes). Figure 5 Evaluation of 96-well plates as support for 23-plex antibody microarrays interfaced with an LED/CCD plate scanner. (A) Signal intensities from a well-based microarray targeting high- and low-abundant serum proteins on Clear MaxiSorp, Costar and SciPLEXPLATE supports. The data were automatically quantified and processed to a 0–1 scale in an integrated scanner software; (B) Microarray image of serum profiling on a well-based antibody microarray targeting predominantly low-abundant analytes, produced in clear MaxiSorp plates and scanned in an LED/CCD scanner (90% LED power, 1200-ms exposure); (C) Microarray image of a slide-based antibody microarray (identical to that in (B)) produced on black MaxiSorp slides and scanned in a confocal microarray slide laser scanner (80% PMT gain/80% LP). Aliquots of the same biotinylated serum samples were used in both (B) and (C). The array layout used in (B) and (C) is shown below the scan images. Figure 6 Evaluation of the solid support and scanner on antibody microarray performance. Antibody microarrays, based on 20 antibodies, were printed on slides (clear MaxiSorp and black MaxiSorp) and plates (clear MaxiSorp) and scanned using confocal laser scanners (PE slide scanner and LS plate/slide scanner). (A) Microarray image of a 20-plex antibody well-based (clear MaxiSorp) microarray, hybridized with labeled serum and scanned at 120% gain; (B) Comparison of signal intensities and spot images (20 antibodies) for well-based arrays (clear MaxiSorp plates, LS scanner) and slide-based arrays (black MaxiSorp slides, PE scanner). Quantified signals were collected from the highest scanning intensity settings used still generating non-saturated spots (180% gain in the LS scanner and 90% PMT gain/90% LP in the PE scanner). The dashed line represents the lower limit of detection (local background plus three standard deviations) for the LS scanner; (C) Array of six different C3 antibodies in five dilutions printed in a clear MaxiSorp plate, hybridized with pure, labelled C3 (5 μg/mL) and scanned at 120% gain (LS scanner); (D) Comparison of signal intensities from the same C3 assay as in (C), run on well-based arrays (clear MaxiSorp plates, LS scanner) and slide-based arrays (black or clear MaxiSorp slides and LS or PE scanner). The quantified signals were taken from scans with the highest possible intensity setting without any spot saturation (60% PMT gain/90% LP for the PE scanner, 144% gain for the slides and 120% gain for the plate in the LS scanner). microarrays-05-00016-t001_Table 1Table 1 Evaluation of 384-well plates as protein (antibody) source plates for microarray production. The same stock solution of biotinylated bovine serum albumin (BSA) was loaded into 12 wells on each source plate and printed on black MaxiSorp slides (14 subarrays/slide). The spot signal intensities were determined and reported in terms of mean CV-value for all spots and the maximum signal intensity difference. Source Plate CV-Value Maximum Signal Difference (1) PP/black/NUNC (2) 3% 11% PP/clear/Genetix (3) 6% 20% PP/clear/ABgene (4) 7% 36% PS/clear/Genetix (5) 5% 19% PS/clear/NUNC (6) 8% 26% PS, NBS-treated/clear/Corning (7) 8% 55% PS/black/PerkinElmer (8) 10% 30% PS, NBS-treated/white/Corning (9) 16% 55% (1) ((highest signal intensity—lowest signal intensity)/lowest signal intensity) × 100%; (2) Stated to display lower binding capacity so that proteins and DNA will not bind, allowing for complete protein recovery; (3) no additional surface property information at hand; (4) stated to be a highly polished surface making it low protein binding and chemically inert; (5) no additional surface property information at hand; (6) stated to display lower binding capacity so that proteins and DNA will not bind, allowing for complete protein recovery; (7) stated to be a nonionic hydrophilic surface (polyethylene oxide-like) that minimizes molecular interactions; (8) no additional surface property information at hand; (9) Stated to be a nonionic hydrophilic surface (polyethylene oxide-like) that minimizes molecular interactions. microarrays-05-00016-t002_Table 2Table 2 Slide- and well-based surfaces evaluated as solid supports for antibody microarrays. Format Name Surface Chemistry Surface Geometry a Binding Chemistry Supplier Slides Nexterion H NHSb polymer 3D Covalent Schott Nexterion P Hydrophilic NHS polymer 3D Covalent Schott GAPS II Aminopropylsilane 2D Ionic Corning NHS glass NHS glass 3D Covalent PolyAn NHS polymer NHS polymer 3D Covalent PolyAn Epoxy glass Epoxy glass 3D Covalent PolyAn Epoxy polymer Epoxy polymer 3D Covalent PolyAn FAST Nitrocellulose 3D Adsorption Whatman SuperProtein Hydrophobic polymer 2D Adsorption Arrayit Silane-Prep Aminoalkylsilane glass 2D Ionic Sigma Black MaxiSorp Hydrophilic polymer 2D Adsorption NUNC Clear MaxiSorp* Hydrophilic polymer 2D Adsorption NUNC 96-well plates Clear MaxiSorp Hydrophilic polymer 2D Adsorption NUNC SciPLEXPLATE Polystyrene 2D Adsorption Scienion Costar Polystyrene 2D Adsorption Corning a 3D—three-dimensional, 2D—two-dimensional; b N-Hydroxysuccinimide; * This slide was only used as a reference when evaluating plate-based surfaces. microarrays-05-00016-t003_Table 3Table 3 Comparison of the best blocking solution for nine solid supports. Slide Blocking Solution Background Intensity (1) Background Homogeneity (2) Spot Intensity (3) Spot Size (μm) Spot Morphology (4) Black MaxiSorp 5% (w/v) milk PBS 500 + + + 11,600 125–140 + + Epoxy glass 0.5% (v/v) Tween-20 PBS 800 − − 30,900 130–140 + Epoxy polymer 0.5% (v/v) Tween-20 PBS 250 + 14,700 130–150 + + + GAPSII 5% (w/v) BSA PBS 1100 − − − 6300 150–160 + + Nexterion H 1% (w/v) milk TBS 200 + 8900 60–120 − − Nexterion P 5% (w/v) BSA PBS 150 + + 5300 60–170 − − − NHS glass 0.5% (v/v) Tween-20 PBS 350 − 13,900 70–160 + NHS polymer 1% (w/v) milk TBS 300 − 1400 60–160 + + Silane-Prep 1% (w/v) BSA PBS 700 − − − 3700 160–170 − (1) The mean intensity (RFU) of the non-specific background binding intensity measured at four randomly-selected positions (70% PMT gain and 70% laser power); (2) the background homogeneity is graded from best (+ + +) to worst (− − −); (3) mean spot signal intensity (RFU) for anti-C1q, anti-C3, anti-C4, anti-CD40, anti-IL-8, anti-properdin and anti-VEGF (70% PMT gain and 70% laser power); (4) the spot morphology is graded from best (+ + +) to worst (− − −). microarrays-05-00016-t004_Table 4Table 4 Reproducibility of the antibody microarray set-up, expressed as coefficient of variation (CV), for non-normalized data. The same serum sample was analysed on four identical 14 × 8 antibody subarrays printed on two separate slides of each solid support. Array Features Epoxy Glass Epoxy Polymer Black MaxiSorp Spot-to-spot 4.7 ± 7.7 4.1 ± 5.2 3.3 ± 3.0 Array-to-array 26.4 ± 15.0 19.5 ± 6.5 3.6 ± 0.6 Slide-to-slide 34.3 ± 11.6 29.0 ± 15.7 12.2 ± 3.6 ==== Refs References 1. Ayoglu B. Haggmark A. Neiman M. Igel U. Uhlen M. Schwenk J.M. Nilsson P. Systematic antibody and antigen-based proteomic profiling with microarrays Expert Rev. Mol. Diagn. 2011 11 219 234 10.1586/erm.10.110 21405972 2. Sanchez-Carbayo M. Antibody microarrays as tools for biomarker discovery Methods Mol. Biol. 2011 785 159 182 21901599 3. Borrebaeck C.A. Wingren C. High-throughput proteomics using antibody microarrays: An update Expert Rev. Mol. Diagn. 2007 7 673 686 10.1586/14737159.7.5.673 17892372 4. Ingvarsson J. Larsson A. Sjoholm A.G. Truedsson L. Jansson B. Borrebaeck C.A. Wingren C. Design of recombinant antibody microarrays for serum protein profiling: Targeting of complement proteins J. Proteome Res. 2007 6 3527 3536 10.1021/pr070204f 17696517 5. Wingren C. Ingvarsson J. Dexlin L. Szul D. Borrebaeck C.A. Design of recombinant antibody microarrays for complex proteome analysis: Choice of sample labeling-tag and solid support Proteomics 2007 7 3055 3065 10.1002/pmic.200700025 17787036 6. Borrebaeck C.A. Wingren C. Antibody array generation and use Methods Mol. Biol. 2014 1131 563 571 24515491 7. Steinhauer C. Ressine A. Marko-Varga G. Laurell T. Borrebaeck C.A. Wingren C. Biocompatibility of surfaces for antibody microarrays: Design of macroporous silicon substrates Anal. Biochem. 2005 341 204 213 10.1016/j.ab.2004.10.036 15907865 8. Kusnezow W. Hoheisel J.D. Solid supports for microarray immunoassays J. Mol. Recognit. 2003 16 165 176 10.1002/jmr.625 12898667 9. Seurynck-Servoss S.L. White A.M. Baird C.L. Rodland K.D. Zangar R.C. Evaluation of surface chemistries for antibody microarrays Anal. Biochem. 2007 371 105 115 10.1016/j.ab.2007.07.010 17718996 10. Angenendt P. Glokler J. Sobek J. Lehrach H. Cahill D.J. Next generation of protein microarray support materials: Evaluation for protein and antibody microarray applications J. Chromatogr. A 2003 1009 97 104 10.1016/S0021-9673(03)00769-6 13677649 11. Angenendt P. Glokler J. Murphy D. Lehrach H. Cahill D.J. Toward optimized antibody microarrays: A comparison of current microarray support materials Anal. Biochem. 2002 309 253 260 10.1016/S0003-2697(02)00257-9 12413459 12. Kusnezow W. Jacob A. Walijew A. Diehl F. Hoheisel J.D. Antibody microarrays: An evaluation of production parameters Proteomics 2003 3 254 264 10.1002/pmic.200390038 12627378 13. Bergeron S. Laforte V. Lo P.S. Li H. Juncker D. Evaluating mixtures of 14 hygroscopic additives to improve antibody microarray performance Anal. Bioanal. Chem. 2015 407 8451 8462 10.1007/s00216-015-8992-8 26345442 14. Borrebaeck C.A. Wingren C. Design of high-density antibody microarrays for disease proteomics: Key technological issues J. Proteom. 2009 72 928 935 10.1016/j.jprot.2009.01.027 19457338 15. Hu S. Xie Z. Qian J. Blackshaw S. Zhu H. Functional protein microarray technology Wiley Interdiscip. Rev. Syst Biol. Med. 2011 3 255 268 10.1002/wsbm.118 20872749 16. Kusnezow W. Banzon V. Schroder C. Schaal R. Hoheisel J.D. Ruffer S. Luft P. Duschl A. Syagailo Y.V. Antibody microarray-based profiling of complex specimens: Systematic evaluation of labeling strategies Proteomics 2007 7 1786 1799 10.1002/pmic.200600762 17474144 17. Piehler J. Brecht A. Geckeler K.E. Gauglitz G. Surface modification for direct immunoprobes Biosens. Bioelectron. 1996 11 579 590 10.1016/0956-5663(96)83293-3 8652111 18. Mendoza L.G. McQuary P. Mongan A. Gangadharan R. Brignac S. Eggers M. High-throughput microarray-based enzyme-linked immunosorbent assay (ELISA) Biotechniques 1999 27 778 788 10524321 19. Matson R.S. Milton R.C. Cress M.C. Chan T.S. Rampal J.B. Printing low density protein arrays in microplates Methods Mol. Biol. 2007 381 339 361 17984528 20. Ingvarsson J. Wingren C. Carlsson A. Ellmark P. Wahren B. Engstrom G. Harmenberg U. Krogh M. Peterson C. Borrebaeck C.A. Detection of pancreatic cancer using antibody microarray-based serum protein profiling Proteomics 2008 8 2211 2219 10.1002/pmic.200701167 18528842 21. Carlsson A. Wingren C. Ingvarsson J. Ellmark P. Baldertorp B. Ferno M. Olsson H. Borrebaeck C.A. Serum proteome profiling of metastatic breast cancer using recombinant antibody microarrays Eur J. Cancer 2008 44 472 480 10.1016/j.ejca.2007.11.025 18171612 22. Whitelegg N.R. Rees A.R. WAM: an improved algorithm for modelling antibodies on the WEB Protein Eng. 2000 13 819 824 10.1093/protein/13.12.819 11239080 23. Borrebaeck C.A. Wingren C. Recombinant antibodies for the generation of antibody arrays Methods Mol. Biol. 2011 785 247 262 21901605 24. Dixit C.K. Vashist S.K. MacCraith B.D. O’Kennedy R. Evaluation of apparent non-specific protein loss due to adsorption on sample tube surfaces and/or altered immunogenicity Analyst 2011 136 1406 1411 10.1039/c0an00689k 21267470 25. Stillman B.A. Tonkinson J.L. Fast slides: A novel surface for microarrays Biotechniques 2000 29 630 635 10997277 26. Kukar T. Eckenrode S. Gu Y. Lian W. Megginson M. She J.X. Wu D. Protein microarrays to detect protein-protein interactions using red and green fluorescent proteins Anal. Biochem. 2002 306 50 54 10.1006/abio.2002.5614 12069413 27. Joos T.O. Schrenk M. Hopfl P. Kroger K. Chowdhury U. Stoll D. Schorner D. Durr M. Herick K. Rupp S. A microarray enzyme-linked immunosorbent assay for autoimmune diagnostics Electrophoresis 2000 21 2641 2650 10.1002/1522-2683(20000701)21:13<2641::AID-ELPS2641>3.0.CO;2-5 10949141 28. Madoz-Gurpide J. Wang H. Misek D.E. Brichory F. Hanash S.M. Protein based microarrays: A tool for probing the proteome of cancer cells and tissues Proteomics 2001 1 1279 1287 10.1002/1615-9861(200110)1:10<1279::AID-PROT1279>3.0.CO;2-W 11721639 29. Paweletz C.P. Charboneau L. Bichsel V.E. Simone N.L. Chen T. Gillespie J.W. Emmert-Buck M.R. Roth M.J. Petricoin I.E. Liotta L.A. Reverse phase protein microarrays which capture disease progression show activation of pro-survival pathways at the cancer invasion front Oncogene 2001 20 1981 1989 10.1038/sj.onc.1204265 11360182 30. Piehler J. Brecht A. Valiokas R. Liedberg B. Gauglitz G. A high-density poly(ethylene glycol) polymer brush for immobilization on glass-type surfaces Biosens. Bioelectron. 2000 15 473 481 10.1016/S0956-5663(00)00104-4 11419642 31. Steinhauer C. Wingren C. Khan F. He M. Taussig M.J. Borrebaeck C.A. Improved affinity coupling for antibody microarrays: Engineering of double-(His)6-tagged single framework recombinant antibody fragments Proteomics 2006 6 4227 4234 10.1002/pmic.200600036 16826567 32. Moody M.D. Van Arsdell S.W. Murphy K.P. Orencole S.F. Burns C. Array-based ELISAS for high-throughput analysis of human cytokines Biotechniques 2001 31 186 194 11464511 33. Wiese R. Belosludtsev Y. Powdrill T. Thompson P. Hogan M. Simultaneous multianalyte ELISA performed on a microarray platform Clin. Chem. 2001 47 1451 1457 11468236 34. Liew M. Groll M.C. Thompson J.E. Call S.L. Moser J.E. Hoopes J.D. Voelkerding K. Wittwer C. Spendlove R.S. Validating a custom multiplex ELISA against individual commercial immunoassays using clinical samples Biotechniques 2007 42 327 328 10.2144/000112332 17390539 35. Backen A.C. Cummings J. Mitchell C. Jayson G. Ward T.H. Dive C. “Fit-for-purpose” validation of searchlight multiplex ELISA of angiogenesis for clinical trial use J. Immunol. Methods 2009 342 106 114 10.1016/j.jim.2009.01.003 19174166 36. Butler J.E. Solid supports in enzyme-linked immunosorbent assay and other solid-phase immunoassays Methods 2000 22 4 23 10.1006/meth.2000.1031 11020313

PMC005xxxxxx/PMC5003493.txt

==== Front Microarrays (Basel)Microarrays (Basel)microarraysMicroarrays2076-3905MDPI 10.3390/microarrays5020017microarrays-05-00017ArticleSNPConvert: SNP Array Standardization and Integration in Livestock Species Nicolazzi Ezequiel Luis 1*Marras Gabriele 1Stella Alessandra 12Louhelainen Jari Academic Editor1 Bioinformatics Core Facility, PTP Science Park, Via Einstein—Loc. Cascina Codazza 26900 Lodi, Italy; gabriele.marras@ptp.it (G.M.); stella@ibba.cnr.it (A.S.)2 Istituto di Biologia e Biotecnologia Agraria—Consiglio Nazionale della Ricerca, Via Einstein—Loc. Cascina Codazza 26900 Lodi, Italy* Correspondence: ezequiel.nicolazzi@ptp.it; Tel.: +39-0371-466-233309 6 2016 6 2016 5 2 1728 1 2016 02 6 2016 © 2016 by the authors; licensee MDPI, Basel, Switzerland.2016This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).One of the main advantages of single nucleotide polymorphism (SNP) array technology is providing genotype calls for a specific number of SNP markers at a relatively low cost. Since its first application in animal genetics, the number of available SNP arrays for each species has been constantly increasing. However, conversely to that observed in whole genome sequence data analysis, SNP array data does not have a common set of file formats or coding conventions for allele calling. Therefore, the standardization and integration of SNP array data from multiple sources have become an obstacle, especially for users with basic or no programming skills. Here, we describe the difficulties related to handling SNP array data, focusing on file formats, SNP allele coding, and mapping. We also present SNPConvert suite, a multi-platform, open-source, and user-friendly set of tools to overcome these issues. This tool, which can be integrated with open-source and open-access tools already available, is a first step towards an integrated system to standardize and integrate any type of raw SNP array data. The tool is available at: https://github. com/nicolazzie/SNPConvert.git. single nucleotide polymorphismarraysoftwarestandardizationintegration ==== Body 1. Introduction For over fifty years, “traditional” livestock breeding has combined phenotypes and statistical models to infer the genetic value of individuals. Since its first application, SNP array technology revolutionized this scenario for the animal genetics industry and research. In fact, for the first time, it was possible to obtain consistent and cost-effective genetic information directly from dense genome-wide markers. A review of the impact of the genomic revolution on domestic animals is presented in [1]. The interest raised towards such technology created new opportunities for companies providing these service (e.g., Illumina and Affymetrix). In a short time, a large series of products were produced to fit market requests. As a consequence, a wide variety of low-, medium- and high-density SNP arrays, and their updates carrying slight modifications in the number and content of SNPs, were marketed. Unfortunately, this rapid expansion of availability of genomic information was not matched by a collective effort to standardize data, file formats, or type of information provided to end-users. In cattle livestock alone, three companies have produced more than 15 commercial SNP arrays in less than 10 years, and this number would more than double if the “private” SNP arrays (e.g., those produced by specific companies/consortia but not openly accessible to the public) were considered. Each genotyping platform is associated to a specific genotyping software to produce a range of output formats. By default, Illumina provides two possible file formats: the row format (where all the information for each SNP for each individual is provided on each row) and the matrix format (where the genotype call of each SNP for all individuals is provided on each row). The “ExampleData” folder of this tool, provides examples for each type of data. Using specific plug-ins, other output formats can also be accommodated. Affymetrix’s genotyping software for Windows users returns a range of output formats, whereas only the Affymetrix format is available for Linux/Mac users (individuals by row and SNP dosage—0, 1, 2 corresponding to the number of “B” alleles—by column). Having full access to the software and raw data would make the standardization and integration of data relatively easy. However, a highly common scenario is that researchers and industry do not have access to raw data or the genotyping platform software, or simply exchange genomic data produced in multiple formats, which makes post-hoc standardization and integration of SNP data a difficult but essential need. Consistent SNP IDs, allele-coding formats (e.g., A/B, Top, Forward etc.), map positions in a specific reference genome assembly, and other required information were not readily accessible to the general public until very recently, when producers and researchers started collaborating and sharing information. This collaboration resulted in the “SNPchimp” tool [2], an open-access multi-species online database that stores all the above information from the source (e.g., producers) and ensures full user-friendly access to the information. SNPchimp is a strategic tool for data analysis, but custom programs are still needed to standardize genotype allele coding and map coordinates. Thus, although a step forward was made towards the standardization of SNP array data, effective standardization and integration of formats, allele coding and map information was still unaccounted for end users with limited or no programming skills. Here, we present the SNPConvert suite, a simple set of programs to convert any Illumina raw file format (both row and matrix formats) to a PLINK basic format (see [3,4] for specifics on this format) and to modify the allele coding and update the genomic coordinates of any PLINK file, irrespectively of the technology used to produce the genotypes. This set of programs, released both as multi-platform source code intended for command line access and as a simplified graphical user interface (GUI) for Windows and Mac users, is able to solve most common problems any researcher, irrespectively of its programming skills, has to face when integrating multiple SNP array datasets. Although we report on file format conversion for the Illumina technology, similar tools are available for Affymetrix-based genotypes [5,6,7]. However, note that Affymetrix and Illumina data cannot be integrated directly, as SNP IDs and the allele calling process are not consistent in the two genotyping technologies. 2. Materials and Methods The SNPConvert suite was designed and developed in Python 2.7. The suite contains a set of three Python utilities that convert any Illumina row and matrix raw file formats to PLINK format (“PEDDA_ROW” and “PEDDA_MATRIX”, respectively) and can automatically modify the allele coding format and update the genomic coordinates of any PLINK file (“iConvert”). All three programs are multi-platform (e.g., they run on any operative system). A deliberate design choice was to use only Python’s built-in libraries to avoid dependencies of the whole structure and thus enhance portability. The use is restricted, from the end user’s point of view, to the modification of a simplified parameter file and the execution of the program(s) from the command line. An example flowchart is shown in Figure 1. 2.1. Utility n.1: PEDDA_ROW PEDDA_ROW software converts files in Illumina “row” format to PLINK (ped and map) format. To obtain this result, the user is asked to compile a parameter file to include the following: (i) The paths to two input files: the FinalReport file (in row format) and the SNP map file, provided, together with a series of other files, by genotyping labs using the Illumina (H)iScan technology; (ii) The selected allele coding output (generally, Illumina row file formats contain multiple allele coding); (iii) The separator of the file (comma, tab or space); (iv) The position of the SNP ID in the FinalReport file; (v) The position of the Individual ID in the FinalReport file; (vi) (Optional) the population ID (referred in PLINK as FID—family ID), displayed as the first column of the PED file; (vii) (Optional) the output file name. Based on these parameters, the software reads and processes/writes the genotypes (in the chosen allele coding format) of a single individual at a time (to reduce the amount of memory required). 2.2. Utility n.2: PEDDA_MATRIX Similarly, the PEDDA_MATRIX software converts files into the Illumina “matrix” format to PLINK format. Since the matrix format is much more standardized and can only accommodate one type of allele coding, the number of parameters to be set by the end-user is reduced to 4: (i) the paths to the same two input files as in PEDDA_ROW, but in this case the FinalReport file should be in matrix format; (ii) the separator of the file (comma, tab or space); (iii) the optional population ID; and (iv) the optional output file name. In this case, considering all individual genotypes are present on each line, the software stores the genotypes of all individuals before writing the output, thus requiring a larger amount of memory. 2.3. Utility n.3: iConvert The iConvert software converts allele coding formats (e.g., Illumina TOP/Illumina or Affymetrix FORWARD/Illumina or Affymetrix AB) and, if required, updates genomic coordinates from any PLINK file, irrespectively of the technology used. The only constraint is the use of a SNPchimp output file, which makes this tool directly accessible by the animal genetics community only. However, retrieving the required information and creating a file with a similar format to SNPchimp output (using a spreadsheet such as Microsoft Excel) can by-pass this drawback for species/chips not included in SNPchimp. Again, the user is asked to compile a parameter file to specify: (i) the PLINK PED and MAP input files; (ii) the missing values for genotypes in the PLINK file; (iii) the SNPchimp file containing the input and output allele formats and the genomic coordinates to be used (if required); (iv) the input and output allele formats (should be present in (iii)); (v) the choice of updating or not the genomic coordinates. This program simply produces updated PLINK PED and MAP files. Similarly to PEDDA_ROW, each individual is processed singularly; thus, the requirements in terms of memory are low. In order to facilitate access to these tools for Windows and Mac users (64 bit processors only), updated versions of the three aforementioned tools were wrapped into a graphical user interface (GUI). PyQt v.4 [8], a Python binding to the QT cross-platform project [9], was used to design the GUI and its functionalities. PyInstaller [10] was then used to create the two executable released. All the source codes used to create the GUI are freely available. The GUI was named SNPConvert v1.0, and it is a simple graphical application of the three software previously described, with the only difference of a further reduced set of parameter. Each of the three programs provides a full runtime log, where the user can check that all the data is read and handled as expected. It is important to note that, for GUI users, Python 2.7 is not a dependency. 3. Results SNPConvert is a publicly available set of user-friendly tools. Both source codes and GUI are provided, including source codes used to obtain the GUI. Ease of installation and use is ensured by the fact that SNPConvert (GUI or source code) is multiplatform. It was coded using only built-in Python packages and has only one generic dependency (e.g., Python 2.7) when running programs from the command line. The open-source nature of the tools aims at enhancing community-driven development and enhancement of the capabilities of the software. The only limitation of SNPConvert is the amount of random access and virtual memory available. Program requirements are highly variable depending on the file format (e.g., the program used) and the number of markers and individuals. For example, the amount of memory requirement of PEDDA_MATRIX or its GUI counterpart is higher than in PEDDA_ROW or iConvert. The reason is that because the file format contains multiple individual genotypes on a single row, PEDDA_MATRIX reads and stores in memory all genotypes of all individuals, writing the output at the end of the whole process. On the contrary, PEDDA_ROW has a much lower use of resources. In fact, since Illumina row format files store genotypes of each individual consecutively, PEDDA_ROW retains all the SNPs of a single individual in memory prior to writing the output file. Similarly, iConvert reads and writes converted alleles over single individuals; thus, the use of resources is limited. Both source codes and GUI were tested on “average” datasets, obtaining satisfactory results for an average user. For example, a full plate (96 samples) in matrix form was converted in less than 5 s (wall time) on a MacOs with 8Gb RAM and a 2.8 GHz Intel Core i7 processor. Using the same computer, but converting 777,962 SNPS (BovineHD) genotyped over 40 individuals reported in a row format file (e.g., a 4.75 Gb file), wall time increased to 2 min and 56 s to produce the output. Finally, remapping SNPs, converting allele coding of 417 BovineHD genotyped individuals (e.g., ~424.4 million conversions) and producing new genotype files took 8 min and 14 s of wall time. No appreciable differences in wall time were observed when running the programs from source code or from the GUI. 4. Discussion The set of tools reported here are able to solve most of the standardization issues of formats, allele coding, and mapping of Illumina SNP arrays. In addition, if combined with another already available open-source tools [3,4,5] or the proprietary Windows-only Affymetrix GenotypingConsole [11], such full standardization extends also to the Affymetrix technology. However, it is important to note that, currently, such user-friendly tools to standardize data without the need of programming skills or advanced bioinformatics tools are technology-specific. Since Illumina and Affymetrix are based on different technology, with different protocols to call and identify alleles, advanced bioinformatics methodologies must be applied when cross-technology standardization is required. This tool specifically addresses users without (or with limited) programming skills, as advanced and skilled users usually code their own programs to overcome these difficulties. In fact, special effort was put into making this tool multi-platform and user-friendly. These tools were coded using only built-in libraries in order to minimize installation requirements. The trade-off of this choice is some loss in the efficiency of some of SNPConvert’ functionalities. For example, iConvert could perform much faster using a multi-processor approach, which would require the installation of specific Python libraries. A reasonable increase in wall time was not considered a high cost in spite of its ease of installation/use. In any case, since source codes are released (e.g., including source codes to create the GUI), advanced users can use these scripts as a starting point for more advanced (and performing) applications. Storing data in PLINK format might not seem the best choice, especially considering that PLINK ped and map files are not the most efficient formats in terms of disk memory usage. Again, a trade-off between memory usage and easy (or easier) access to genomic analysis was considered. The choice of PLINK format as standard can be beneficial because users can perform a number of analyses based on the same format, irrespectively of the technology used to obtain the gentoypes. In fact, PLINK format is the only common format in output for Illumina and Affymetrix software. Furthermore, PLINK ped and map files allow direct access to any PLINK functionality and any tool accepting PLINK file formats as input file [12]. This means that any user, experienced or not, is able to readily perform data quality checks and even advanced analysis (with some limitations). Finally, multi-file format converters for genomic data, such as PGDspider [13], can convert PLINK file formats many any other file formats. 5. Conclusions In this work, we describe a number of issues researchers have to face when dealing with standardization and integration of SNP array data. Researchers with programming skills can overcome these difficulties by programming their own methods. However, the issues described here can be a large obstacle for those researchers having low or no programming skills. Here, we present a simple, open-source, multi-platform, and user-friendly set of tools to overcome these difficulties. The tool is available at: https://github.com/nicolazzie/SNPConvert.git. Most SNPConvert functionalities address Illumina-specific issue; however, if combined with already available open-source software, the solutions presented here are able to solve the problems described for all genotyping technologies. Acknowledgments This research was financially supported by the national research project “GenHome.” The two anonymous reviewers are kindly acknowledged for their useful comments. Author Contributions Ezequiel Luis Nicolazzi and Alessandra Stella conceived and designed the experiments; Ezequiel Luis Nicolazzi developed the first version of the tool; Ezequiel Luis Nicolazzi and Gabriele Marras developed the graphical user interface; Ezequiel Luis Nicolazzi, Gabriele Marras, and Alessandra Stella wrote the paper. Conflicts of Interest The authors declare no conflict of interest. Abbreviations The following abbreviations are used in this manuscript: SNPSingle nucleotide polymorphism GUIGraphical user interface Figure 1 Flowchart of SNPConvert functionalities for files in Illumina row, matrix, or PLINK file format. Dashed lines indicate optional steps. Pointed boxes indicate interactions with third-party software. ==== Refs References 1. Goddard M.E. Hayes B.J. Mapping genes for complex traits in domestic animals and their use in breeding programmes Nat. Rev. Genet. 2009 10 381 391 10.1038/nrg2575 19448663 2. Nicolazzi E.L. Caprera A. Nazzicari N. Cozzi P. Strozzi F. Lawley C. Pirani A. Soans C. Brew F. Jorjani H. SNPchiMp v.3: Integrating and standardizing single nucleotide polymorphism data for livestock species BMC Genom. 2015 16 1 6 10.1186/s12864-015-1497-1 25881165 3. PLINK: Whole genome data analysis toolset—PEDfile Available online: http://pngu.mgh.harvard.edu/~purcell/plink/data.shtml#ped (accessed on 29 March 2016) 4. PLINK: Whole genome data analysis toolset—MAPfile Available online: http://pngu.mgh.harvard.edu/~purcell/plink/data.shtml#map (accessed on 29 March 2016) 5. Wang K. Li M. Hakonarson H. Analysing biological pathways in genome-wide association studies Nat. Rev. Genet. 2010 11 843 854 10.1038/nrg2884 21085203 6. The Aroma Project Available online: http://www.aroma-project.org/ (accessed on 25 January 2016) 7. Nicolazzi E.L. Iamartino D. Williams J.L. AffyPipe: An open-source pipeline for Affymetrix Axiom genotyping workflow Bioinformatics 2014 30 3118 3119 10.1093/bioinformatics/btu486 25028724 8. PyQt4—Python Wiki Available online: https://wiki.python.org/moin/PyQt4 (accessed on 25 January 2016) 9. Qt-Home Available online: http://www.qt.io/ (accessed on 25 January 2016) 10. Welcome to PyInstaller official website Available online: http://www.pyinstaller.org/ (accessed on 25 January 2016) 11. Genotyping Console Software|Affymetrix Available online: http://www.affymetrix.com/estore/browse/level_seven_software_products_only.jsp?productId=131535#1_1 (accessed on 25 January 2016) 12. Nicolazzi E.L. Biffani S. Biscarini F. Orozco Ter Wengel P. Caprera A. Nazzicari N. Stella A. Software solutions for the livestock genomics SNP array revolution Anim. Genet. 2015 46 343 353 10.1111/age.12295 25907889 13. Lischer H.E.L. Excoffier L. PGDSpider: An automated data conversion tool for connecting population genetics and genomics programs Bioinformatics 2012 28 298 299 10.1093/bioinformatics/btr642 22110245

PMC005xxxxxx/PMC5003504.txt

==== Front CureusCureus2168-8184Cureus2168-8184Cureus Palo Alto (CA) 10.7759/cureus.713OtolaryngologyOncologyRadiation OncologySurvival Outcomes and Patterns of Recurrence in Patients with Stage III or IV Oropharyngeal Cancer Treated with Primary Surgery or Radiotherapy Muacevic Alexander Adler John R Debenham Brock J 1Banerjee Robyn 2Warkentin Heather 1Ghosh Sunita 1Scrimger Rufus 1Jha Naresh 1Parliament Matthew 11 Department of Oncology, University of Alberta 2 Department of Oncology, University of Calgary Brock J. Debenham debenham@ualberta.ca26 7 2016 7 2016 8 7 e7137 6 2016 26 7 2016 Copyright © 2016, Debenham et al.2016Debenham et al.This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.This article is available from http://www.cureus.com/articles/4738-survival-outcomes-and-patterns-of-recurrence-in-patients-with-stage-iii-or-iv-oropharyngeal-cancer-treated-with-primary-surgery-or-radiotherapyPurpose To compare and contrast the patterns of failure in patients with locally advanced squamous cell oropharyngeal cancers undergoing curative-intent treatment with primary surgery or radiotherapy +/- chemotherapy. Methods and materials Two hundred and thirty-three patients with stage III or IV oropharyngeal squamous cell carcinoma who underwent curative-intent treatment from 2006-2012, were reviewed. The median length of follow-up for patients still alive at the time of analysis was 4.4 years. Data was collected retrospectively from a chart review. Results One hundred and thirty-nine patients underwent primary surgery +/- adjuvant therapy, and 94 patients underwent primary radiotherapy +/- chemotherapy (CRT). Demographics were similar between the two groups, except primary radiotherapy patients had a higher age-adjusted Charleston co-morbidity score (CCI). Twenty-nine patients from the surgery group recurred; 15 failed distantly only, seven failed locoregionally, and seven failed both distantly and locoregionally. Twelve patients recurred who underwent chemoradiotherapy; ten distantly alone, and two locoregionally. One patient who underwent radiotherapy (RT) alone failed distantly. Two and five-year recurrence-free survival rates for patients undergoing primary RT were 86.6% and 84.9% respectively. Two and five-year recurrence-free survival rates for primary surgery was 80.9% and 76.3% respectively (p=0.21). There was no significant difference in either treatment when they were stratified by p16 status or smoking status. Conclusions Our analysis does not show any difference in outcomes for patients treated with primary surgery or radiotherapy. Although the primary pattern of failure in both groups was distant metastatic disease, some local failures may be preventable with careful delineation of target volumes, especially near the base of skull region. oropharyngealhpvoropharyngeal cancersurgeryradiationradiotherapyrecurrencepatterns of failurefailureThe content published in Cureus is the result of clinical experience and/or research by independent individuals or organizations. Cureus is not responsible for the scientific accuracy or reliability of data or conclusions published herein. All content published within Cureus is intended only for educational, research and reference purposes. Additionally, articles published within Cureus should not be deemed a suitable substitute for the advice of a qualified health care professional. Do not disregard or avoid professional medical advice due to content published within Cureus. ==== Body Introduction Locally advanced oropharyngeal cancers are increasing in incidence. Although most centers throughout Canada and the United States of America (USA) favor treating these malignancies with an organ-preservation approach using combined chemoradiotherapy (CRT) [1], some centers, including ours, have a large experience treating with primary surgery followed by adjuvant therapy [2]. Recently our center reported outcomes of our experience from the years 1998 to 2009, which appeared to show an improved disease-free survival at two years for surgery as a primary treatment compared to CRT (73.7% vs. 57.4%) [2]. Previous studies from Stanford and others have consistently reported 3-4 year local control rates for patients treated with CRT of 90% or higher, and 3-year disease-free survival rates of approximately 80% [3-8]. Due to the large discrepancy in our outcomes compared to other large academic centers, we undertook a quality assurance study looking at stage-matched patients with locally advanced oropharyngeal cancer undergoing either primary surgery or radiotherapy with an emphasis on disease-free survival, overall survival, and patterns of recurrence. Materials and methods Ethics approval was obtained before initiating this study through the Health Research Ethics Board of Alberta – Cancer Committee (ETH#26196). The patient list was obtained from the Alberta Cancer Registry (ACR). The list was created by searching for all stage III and IV squamous cell cancers (SCC) of the oropharynx treated with primary surgery +/- adjuvant therapy or radiotherapy +/- chemotherapy. The timelines used were from January 1, 2006 to December 31, 2012 and the location was Northern Alberta. All patients had CT or PET imaging of the neck and chest prior to initiation of curative-intent therapy, as well as a formal quadroscopy for biopsy of the primary site of disease. An initial list of 333 patients was obtained from the ACR. A comprehensive chart review was undertaken, and a database was populated. A final list of 233 patients who underwent non-clinical trial, curative-intent treatment were included in the analysis. The median length of follow-up for patients still alive at the time of analysis was 4.4 years. Reasons for exclusion of the other 100 patients from the ACR were as follows: 27 patients had a non-oropharyngeal primary tumor; 27 had palliative-intent treatment (radiotherapy, chemotherapy, or best supportive care); 22 had metastases at diagnosis; 15 had their primary treatment outside of Northern Alberta; six had recurrent disease from a previous head and neck cancer (prior to 2006); two of them had Stage I or II disease; one had synchronous head and neck (H&N) primaries; one had a non-SCC cancer; and one had been included in the registry twice. Statistical analysis Patient demographics, treatment factors, follow-up dates, imaging results, and pathology results were collected and anonymized. Summary statistics were calculated, including mean and standard deviations for continuous variables, and frequency and percentages for categorical variables. Recurrence-free survival (RFS) and overall survival (OS) was measured from the date of diagnosis to the date of recurrence or death. Kaplan–Meier estimates of the median RFS & OS and 95% confidence interval (95% CI) were obtained. Logistic regression was used to explore the association between factors commonly available at the time of consultation (age, histology, PS, gender) as well treatment factors for both surgery and radiation. After univariate analysis, variables significant at the p < 0.10 level were entered into multivariate models. Final models selected variables significant at the p < 0.05 level. All analyses were conducted using SAS version 9.3, with p < 0.05 indicating statistical significance. Results Patient demographics  Patient demographics were analyzed and summarized in Table 1 below. Table 1 Patient Demographics   Primary Surgery (n=139) Primary Radiation (n=94) p-value Male/Female 121/18 82/12 p=0.88 Age-adjusted Charleston Co-morbidity Index (median) 3 (95% CI 3-4) 4 (95% CI 3-4) p=0.046 Age at Diagnosis (median) 56 (95% CI 54-57) 56 (95% CI 54-59) p=0.19 AJCC Stage III – 21 (15.1%) IVA – 104 (74.8%) IVB – 14  (10.1%) III – 14 (14.9%) IVA – 62 (65.9%) IVB – 18 (19.1%) p=0.13 Clinical T-Stage (RT) Pathologic T-Stage (Surgery) T1 (21.6%) T2 (35.3%) T3 (20.2%) T4a (21.6%) T4b (1.4%) T1 (40.4%) T2 (21.3%) T3 (20.2%) T4a (11.7%) T4b (6.4%) p=0.06 Clinical N-stage (RT) Pathologic N-Stage (Surgery) N0 (6.5%) N1 (13.8%) N2a (11.6%) N2b (33.3%) N2c (28.3%) N3 (6.5%) N0 (1.1%) N1 (14.9%) N2a (18.1%) N2b (31.9%) N2c (19.1%) ​N3 (14.9%) p=0.14 Smoking Status Lifelong non-smoker -  30 (21.6%) Former smoker – 71 (51.1%) Current smoker – 37 (26.6%) Unknown – 1 (0.7%) Lifelong non-smoker – 19 (20.2%) Former smoker – 41 (43.6%) Current smoker – 34 (36.1%) Unknown – 0 (0%) p=0.38 P16 Status Positive – 25 (18.0%) Negative – 8 (5.7%) Unknown – 106 (76.3%) Positive – 26 (27.7%)      Negative – 5 (5.3%) Unknown – 63 (67.0%) p=0.21 Time from Diagnosis to Initial Treatment (mean, days) 74.6 84.4 p=0.03 Primary surgery One hundred thirty-nine patients underwent primary surgery. Seventeen underwent surgery alone, 27 underwent surgery plus adjuvant radiotherapy (SRT), and 95 underwent surgery plus adjuvant chemoradiotherapy (SCRT). The reasons for patients who had surgery alone and did not receive any adjuvant treatment included patient refusal (n=6), patients were not offered adjuvant therapy (n=3), metastases presented after surgery but prior to starting adjuvant therapy (n=4), patient died prior to starting adjuvant therapy (n=3), or poor performance status after surgery (n=1). Patients at our center are routinely offered concurrent chemotherapy post-surgery for intermediate risk factors such as T3/T4 disease, perineural invasion (PNI), lymphovascular space invasion (LVSI), or node positive disease rather than only in patients with positive margins or extracapsular extension [9, 10]. There was not a significant difference in RFS or OS in patients who received SRT or SCRT. Patients began their adjuvant treatment, on average, 56 days (95% CI 53-59 days) post-surgery, with only 8% of patients starting within our guideline of six weeks post-surgery [11]. Twenty-nine patients from the surgery group recurred; 15 failed distantly only, seven failed locoregionally, and seven failed both distantly and locoregionally. Regression analysis was performed, and on univariate analysis, the following variables were found to be significant, as listed below in Table 2. Table 2 Univariate analysis for risk factors for recurrence in patients undergoing primary surgery. Factor Hazard Ratio p-value Nodes Positive (0, <5, >5) > 5 nodes - 5.08 (95% CI 2.31-11.1) p<0.0001 Age Adjusted CI   NS Age   NS AJCC Stage   NS Chemotherapy Type (SCRT only) Carboplatin – 3.35 (95% CI 1.29-8.64) p=0.013 Chemotherapy Schedule (Weekly vs every 3 weeks) (SCRT only) Weekly – 4.40 (95% CI 1.57-12.29) p=0.003 Radiation Dose (<6000, 6000-6600, >6600)   NS ECE status 4.23 (95% CI 1.99-9.53) p=0.0002 Gender Female 2.61 (95% CI 1.12-6.10) p=0.04 LVI status 2.15 (95% CI 1.03-4.50) p=0.04 Margin status 4.11 (95% CI 1.92-8.83) p=0.001 P16 P16 neg 4.11 (95% CI 1.42-11.80) p=0.02 pN status N2c 5.53 (95% CI 2.64-11.6) p<0.0001 pT status T3 4.09 (95% CI 1.58 – 10.55) T4a 4.68 (95% CI 1.85-11.83) T4b 55.3 (95% CI 5.67-541.61) p=0.0004 Smoking status   NS Time from diagnosis to surgery   NS Time from surgery to start of radiotherapy (> 6 weeks vs < 6 weeks)   NS Grade 3 3.07 (95% CI 1.40 – 6.73) p=0.0052 PNI status 2.30 (95% CI 1.20-4.42) p=0.013 These variables were then entered into a multivariable analysis. For SCRT patients, chemotherapy schedule was not significant in the multivariate model. For all surgery patients combined, the following variables were significant on multivariate analysis, as listed in Table 3. Table 3 Multivariate analysis for risk factors for recurrence in patients undergoing primary surgery Factor Hazard Ratio p-value Nodes Positive (0, <5, >5) > 5 nodes - 4.72 (95% CI 1.59-13.96) p=0.0054 Gender Female – 5.08 (95% CI 2.03-12.74) p=0.0005 P16 negative 4.44 (95% CI 1.92-10.24) p=0.0005 pT4b 46.98 (95% CI 4.04-546.14) p=0.0001 Chemotherapy (SCRT only) Carboplatin – 3.35 (95% CI 1.29-8.64) p=0.013 Primary radiotherapy Ninety-four patients underwent CRT (n=84) or RT alone (n=7). Patients who underwent RT alone did so for the following reasons: four refused chemotherapy, two patients were not chemotherapy candidates, and one patient was not offered a chemotherapy consultation. Our standard dose fractionation at our center is to deliver 6600 cGy over 30 daily fractions, based on RTOG 00-22 [12]. Univariate analysis was performed for risk factors for recurrence, and the results are summarized in Table 4. Table 4 Univariate analysis for risk factors for recurrence in patients undergoing primary RT Factor Hazard Ratio p-value Age Adjusted CCI   NS Age   NS AJCC Stage IVB – 5.72 (95%CI 1.93 – 16.96) p=0.0017 Chemotherapy Type (CRT only)   NS Chemotherapy Schedule (Weekly vs every 3 weeks) (CRT only)   NS Radiation Dose (<6000, 6000-6600, >6600)   NS Gender   NS Persistent disease after primary RT treatment 9.14 (95% CI 3.07-27.21) p=0.0001 P16   NS cN status N3 - 5.23 (95% CI 1.76 – 15.45) p=0.003 cT status   NS Smoking status   NS Time from diagnosis to RT   NS Grade   NS The significant variables were entered into multivariate analysis. The results are summarized below in Table 5. Table 5 Multivariate analysis for risk factors for recurrence in patients undergoing primary RT Factor Hazard Ratio p-value Stage IVB – 4.85 (95%CI 1.61 – 14.58) p=0.0051 Persistent disease after RT 7.70 (95% CI 2.55-23.22) p=0.0003 Patterns of recurrence Twelve patients recurred who underwent chemoradiotherapy; ten distantly alone, and two locoregionally. One patient who underwent RT alone failed distantly. Eighteen patients who recurred underwent surgery followed by chemoradiotherapy: 11 distantly alone, two locoregionally alone, and five locoregionally and distantly. Five patients failed who underwent surgery followed by radiation alone: one distantly, three locoregionally, and one locoregionally and distantly. Six patients failed who underwent surgery alone; three distantly, two locoregionally, and one locoregionally and distantly. For the patients that received radiotherapy as part of their treatment, and failed locally or locoregionally, we analyzed their radiotherapy plans to look at the location of recurrence versus the dose in the region. The results are summarized in Table 6. Table 6 Review of locoregional failures radiotherapy plans Case Treatment Stage/risk factors Failure Location Notes 1 SCRT T4aN2b, positive surgical margins Base of skull/pterygoid plates Patient terminated RT early, received 50.4Gy/28 to recurrent area 2 SRT T2N2c Sphenoid bone No coverage of base of skull despite level 2 nodes positive 3 SRT T2N3, positive margins Near parotid Recurrence in 60 Gy region (no RT boost or chemo (poor KPS)) 4 SRT T1N2a, positive margins, ECE High level 2 High level 2 not covered despite positive lymph node in level 2, marginal miss 5 SCRT T4aN2c, ECE Neck Only completed 48 Gy, quit RT 6 CRT T3N3 Neck In high dose RT area 7 CRT T1N3 Neck In high dose RT area As an example, the marginal miss in Case 4 is demonstrated in Figures 1-2 below. Figure 1 Radiotherapy plan, Case 4, marginal miss, poor coverage of high level 2/base of skull. The plan shows poor coverage (covered by less than the 95% isodose line) at the high level 2 neck lymph nodes.   Figure 2 PET scan of recurrence, Case 4, marginal miss, poor coverage of high level 2/base of skull. Recurrence-free survival comparison Two & five-year recurrence-free survival rates for patients undergoing primary RT was and 86.6% and 84.9% respectively. Two and five-year recurrence-free survival rates for primary surgery was 80.9% and 76.3% respectively. There was no significant difference in either treatment when stratified by p16 status or smoking status. The Kaplan-Meier estimate of recurrence-free survival is shown in Figure 3 (p = 0.21). Figure 3 Kaplan-Meier Recurrence-Free Survival for Primary Surgery vs Primary RT Overall survival comparison Two & five-year overall survival rates for primary RT was 86.6% and 73.4% respectively. Two & five-year overall survival rates for primary surgery was 83.9% and 66.5% respectively (p=0.38). There was no significant difference in either treatment when stratified by p16 status or smoking status. The Kaplan-Meier estimate of recurrence-free survival is shown in Figure 4 (p=0.38) Figure 4 Kaplan-Meier Overall Survival for Primary Surgery vs Primary RT Causes of death Twenty-four patients died in the primary RT group (25.5%). Five died of non-cancer causes (20.8%), 11 died of oropharyngeal cancer (45.8%), and eight died of new cancer primaries, the majority being biopsy-confirmed lung cancer (32.0%). Forty-six patients died in the primary-surgery group (33.1%). Sixteen (34.8%) died of non-cancer causes, 24 died of oropharyngeal cancer (52.2%), and six died of a new cancer primary (13.0%). Discussion Our results are consistent with other large academic centers in patients who undergo CRT as primary treatment for locally advanced oropharyngeal carcinoma with two and five-year RFS rates of 86.6% and 84.9%. In comparison to our centre’s previously published results, we found that the percentage of patients undergoing RT alone was not as high (18.3% in previous results vs 3% in this cohort) [2]. This likely reflects the fact that patients receiving RT alone was likely palliative, and these patients should have been removed from the previous study. Weaknesses of this study include bias in treatments, as patients who underwent primary RT compared to surgery had higher Charleston Co-Morbidity Index (CCI) [13], and a higher proportion of T4b disease. We are missing human papilloma virus (HPV) status on the majority of our patients, as our centre did not routinely test for p16 status until 2010/2011, which limits comparisons on comparing modalities when stratifying by HPV status. Additionally, we have no data in regards to functional outcomes of our patients, or the cost difference in treatment between the two groups. Although the dominant pattern of failure for patients treated with both primary surgery and radiotherapy remains a distant failure, it may have been possible to prevent some local recurrences with adjustment of the radiotherapy plans. Specifically, we had two recurrences at the base of the skull and one near the parotid gland in primary surgery patients who underwent adjuvant treatment. This phenomenon has been described before by Eisbruch et al. [14], therefore, it is important to ensure that this coverage is achieved during radiotherapy planning and QA processes. There were more local recurrences in the surgery group compared to the radiotherapy groups in our study. We do not have a good explanation for this, except perhaps that after surgery oxygenation to the tumor bed may be altered, and perhaps adjuvant radiotherapy is not as effective with the altered oxygenation to the post-surgical bed. The primary pattern of failure in both primary surgery or radiotherapy patients was distant. This pattern was in many other studies. Results from RTOG 0234 demonstrated a decreased rate of distant metastatic disease in patients receiving docetaxel chemotherapy rather than standard cisplatin chemotherapy [15]. This hypothesis is being further tested in high-risk postoperative patients in RTOG 1216, which is currently open to accrual [16]. Our standard chemotherapy offered to these patients may change in the future based on the results of RTOG 1216, and will hopefully alter the patterns of failure for these patients. Conclusions Our analysis does not show any difference in outcomes for patients treated with primary surgery or radiotherapy. Although the primary pattern of failure in both groups was distant metastatic disease, some local failures may be preventable with careful delineation of target volumes, especially near the base of skull region. The authors have declared that no competing interests exist. Human Ethics Health Research Ethics Board of Alberta – Cancer Committee issued approval 26196 Animal Ethics Animal subjects: This study did not involve animal subjects or tissue. ==== Refs References 1 Management of oropharyngeal cancer: a cross-sectional review of institutional practice at a large Canadian referral centre. Journal of Otolaryngology - Head & Neck Surgery Wilson L Enepekides D Higgins K 19 43 2014 24961273 2 Primary surgery versus chemoradiotherapy for advanced oropharyngeal cancers: a longitudinal population study Journal of Otolaryngology - Head & Neck Surgery O' Connell D Seikaly H Murphy R 31 42 2013 23663568 3 Intensity-modulated radiotherapy in the treatment of oropharyngeal cancer: an update of the Memorial Sloan-Kettering Cancer Center experience International journal of radiation oncology, biology, physics Setton J Caria N Romanyshyn J 291 298 82 2012 4 A multi-institution pooled analysis of gastrostomy tube dependence in patients with oropharyngeal cancer treated with definitive intensity-modulated radiotherapy Cancer Setton J Lee NY Riaz N 294 301 121 2015 25286832 5 Randomized phase III Trial of concurrent accelerated radiation plus cisplatin with or without cetuximab for stage III to IV head and neck carcinoma: RTOG 0522 Journal of Clinical Oncology Ang KK Zhang Q Rosenthal DI 2940 2950 32 2014 25154822 6 Long-term regional control in the observed neck following definitive chemoradiation for node-positive oropharyngeal squamous cell cancer International Journal of Cancer Goenka A Morris LG Rao SS 1214 1221 133 2013 23436584 7 Intensity-modulated chemoradiation for treatment of stage III and IV oropharyngeal carcinoma: the University of California-San Francisco experience Cancer Huang K Xia P Chuang C 497 507 113 2008 18521908 8 Intensity-modulated radiotherapy in the treatment of oropharyngeal cancer: clinical outcomes and patterns of failure International journal of radiation oncology, biology, physics Daly ME Le QT Maxim PG 1339 1346 76 2010 9 Defining risk levels in locally advanced head and neck cancers: a comparative analysis of concurrent postoperative radiation plus chemotherapy trials of the EORTC (#22931) and RTOG (# 9501) Head & Neck Bernier J Cooper JS Pajak TF 843 850 27 2005 16161069 10 Chemoradiation after surgery for high-risk head and neck cancer patients: how strong is the evidence? The Oncologist Bernier J Cooper JS 215 224 10 2005 15793225 11 The Organization and Delivery of Healthcare Services for Head and Neck Cancer Patients 6 2016 2015 http://www.albertahealthservices.ca/assets/info/hp/cancer/if-hp-cancer-guide-hn001-organization.pdf 12 Multi-institutional trial of accelerated hypofractionated intensity-modulated radiation therapy for early-stage oropharyngeal cancer (RTOG 00-22) International journal of radiation oncology, biology, physics Eisbruch A Harris J Garden AS 1333 1338 76 2010 13 Comorbidity and prognosis in head and neck cancers: Differences by subsite, stage, and human papillomavirus status Head & Neck Habbous S Harland LT La Delfa A 802 810 36 2014 23616414 14 Recurrences near base of skull after IMRT for head-and-neck cancer: implications for target delineation in high neck and for parotid gland sparing International journal of radiation oncology, biology, physics Eisbruch A Marsh LH Dawson LA 28 42 59 2004 15 Postoperative chemoradiotherapy and cetuximab for high-risk squamous cell carcinoma of the head and neck: Radiation Therapy Oncology Group RTOG-0234 Journal of Clinical Oncology Harari PM Harris J Kies MS 2486 2495 32 2014 25002723 16 Radiation therapy with cisplatin, docetaxel, or cetuximab after surgery in treating patients with stage iii-iv squamous cell head and neck cancer 6 2016 2016 http://clinicaltrials.gov/show/NCT01810913

PMC005xxxxxx/PMC5003528.txt

==== Front Prev Chronic Dis Prev Chronic Dis PCD Preventing Chronic Disease 1545-1151 Centers for Disease Control and Prevention 27560722 16_0129 10.5888/pcd13.160129 Special Topic Peer ReviewedThe “Retrofitting” Approach to Adapting Evidence-Based Interventions: A Case Study of Pediatric Asthma Care Coordination, United States, 2010–2014 Janevic Mary R. PhD MPH Stoll Shelley C. MPH Lara Marielena MD MPH Ramos-Valencia Gilberto DrPH Stephens Tyra Bryant MD Persky Victoria MD Uyeda Kimberly MD MPH Lesch Julie Kennedy MPA Malveaux Floyd J. MD PhD Author Affiliations: Shelley C. Stoll, University of Michigan School of Public Health, Ann Arbor, Michigan; Marielena Lara, RAND Corporation, Santa Monica, California; Gilberto Ramos-Valencia, University of Puerto Rico School of Public Health, San Juan, Puerto Rico; Tyra Bryant Stephens, Children’s Hospital of Philadelphia, Philadelphia, Pennsylvania; Victoria Persky, University of Illinois at Chicago School of Public Health, Chicago, Illinois; Kimberly Uyeda, Los Angeles Unified School District, Los Angeles, California; Julie Kennedy Lesch, Floyd J. Malveaux, Merck Childhood Asthma Network, Washington, DC. Corresponding Author: Mary R. Janevic, PhD, MPH, University of Michigan School of Public Health, 1425 Washington Heights, Ann Arbor, MI 48109-2029. Telephone: 734-647-3194. Email: mjanevic@umich.edu. 2016 25 8 2016 13 E114Adaptation of evidence-based interventions upon implementation into new practice settings is universal, yet poorly understood. During a cross-site evaluation of the implementation of a proven intervention for pediatric asthma care coordination into 4 resource-challenged settings, we conducted in-depth interviews with site representatives, who reported how and why they modified intervention components. Interview notes were coded for themes. We focused on a single theme from a respondent who described the adaptation process as “backing” the intervention into ongoing services; we found evidence of a similar process at other sites. We labeled this process “retrofitting” to signify adaptation that consists of altering existing services to align with intervention components, rather than modifying the intervention to fit a new setting. Advantages of retrofitting may include allowing organizations to keep what works, capitalizing on existing support for program activities, elevating the role of local knowledge, and potentially promoting the sustainability of effective innovations. ==== Body Background Practitioners are increasingly being called on to use evidence-based approaches (1), often as a prerequisite for federal or health department funding (1,2). The need to modify these evidence-based interventions (EBIs) when they are implemented in new practice settings is somewhere between common and universal (3–5), yet program adaptation is an area with significant unresolved issues (6,7). Among those issues are differing perspectives on the value of program changes (7,8). From one perspective, they are undesirable deviations from a tried-and-true formula that lead to a voltage drop in the intervention’s efficacy (9,10). From another perspective, local tailoring of an EBI can enhance program outcomes by improving the fit of an intervention for a particular population or setting, thus allowing a process of evolution to optimize program functioning and effects (11) and fostering a sense of ownership among staff (12). A common middle-ground solution is for intervention developers to designate certain program components as core (essential to achieving outcomes) (13), while leaving others as peripheral (nonessential or modifiable) (6,10). However, isolating the core elements that are both necessary and sufficient to achieve program effect across replications in diverse settings is often more art than science. For example, selecting core components is often based on the judgment of the program designers (10) rather than on resource-intensive systematic testing or experiential learning from replications (13), and some researchers suggest that even core program elements can be changed to beneficial effect in certain contexts (2,11). Clearly, much remains to be learned about optimizing the EBI adaptation process. One step in this direction is to deepen our understanding of the ways that implementing organizations adapt EBIs in the real world (8). On the basis of our observations from a cross-site evaluation of the implementation of an EBI for pediatric asthma care coordination into 4 urban community settings, we propose the notion of “retrofitting” as a type of adaptation that may be common in practice but is not fully accounted for in existing theoretical frameworks. Retrofitting applies to situations in which EBIs are implemented by health and social service organizations that already offer similar services to their patients or clients. Existing services are then altered, or retrofitted, to more closely match EBI components. In this article, we present a description of retrofitting, as observed in the cross-site evaluation described, and explore the potential practical and theoretical value of this concept. Methods In 2010, sites funded by the Merck Childhood Asthma Network, Inc (MCAN) to implement EBIs to address asthma disparities among children (14) were invited to submit a proposal for a second round of funding. They were asked to refine their program models to focus on evidence-based care coordination for children with asthma. Four sites received this funding: the Los Angeles Unified School District Asthma Program (Los Angeles, CA); the Children's Hospital of Philadelphia Asthma Care Navigator Program (Philadelphia, PA); the Federally Qualified Health Center-based La Red de Asma Infantíl de Puerto Rico (San Juan, PR); and the neighborhood-based Addressing Asthma in Englewood Project (Chicago, IL). The cross-site evaluation of the implementation process focused on the EBI Yes We Can (YWC) because it was the one EBI that all sites implemented to a significant degree. YWC is a medical-social model of care that promotes optimal clinical care while deploying community health workers to provide asthma education, link families to health and social services, and facilitate family–clinician communication (15). The goal of YWC, as described by its developers, is to “assemble a set of best practices and implement them under real-world conditions” (15), and the program has evolved over time. In a series of evaluations (all using a single-group, pre–post design), YWC is associated with many positive asthma-related outcomes, including improvements in daytime and nighttime symptoms, increased prescribing of controller medications and use of asthma action plans, reduced activity impairment, reduced school and parental work absences, and fewer emergency department visits and hospitalizations (15,16). The Centers for Disease Control and Prevention (CDC) includes a case study of YWC on its National Asthma Control Program website as a potentially effective intervention (www.cdc.gov/asthma/interventions/yes_we_can.htm). Identifying core components of Yes We Can As the cross-site evaluators studying the translation of established programs into new settings, our first challenge was to determine the core components of the programs, beginning with YWC. From the literature, we determined that no published evidence clearly demonstrated which YWC components were essential to producing the outcomes. Rather, as with most studies, the outcomes were associated with the intervention as a whole. We next examined the CDC online case study of this program. This description identifies 5 “readily distinguishable” program components, but it is unclear whether these should be considered “core” components. We then conducted separate interviews with 3 YWC developers. When asked, “What about the intervention really made a difference in the outcomes?” developers emphasized different elements; for example, one focused on characteristics and actions of the community health workers, whereas another emphasized the broader concept of pairing proper medical care with attention to social aspects that can impede asthma control. On the basis of the content of the published article (15), the CDC case study, and learnings from the developers, our evaluation team used a process of consensus to form working descriptions of YWC core components (Table 1), making subjective decisions based on the apparent weight placed on each component by the various sources cited. Review of the draft core component scheme by a leading expert in pediatric asthma intervention research (Noreen M. Clark) provided additional support for its validity. Table 1 Core Components of Yes We Can, Case Study of Pediatric Asthma Care Coordination, United States, 2010–2014 Component Key Characteristics Risk stratification: establishing levels of care Risk based on medical severity and control, psychological risk, and social risk Care pathway and intervention activities matched to risk level Asthma care coordinator (ACC) Culturally and linguistically aligned with families served Provides basic education including “how to” use spacers, reduce triggers, etc Addresses social problems as they arise: working with schools, providing assistance in finding new housing, making referrals to smoking cessation programs, obtaining refills, etc Makes the family feel like a valued member of the care team Asthma clinical care: chronic care approach Prevention-based “asthma clinic” established with set hours to see children with asthma ACC integrated into the health care team Team members reinforce colleagues’ educational efforts Routine case conferences follow clinic hours Careful planning and integration of clinic and home visits supported by telephone calls Schedule for ongoing assessment of control Designated clinical champion Clinical case management Organized, systematic tracking of patients to assess needs and match services Established network for referrals Key informant interviews We conducted 9 key informant interviews by telephone (average length, 80 minutes) with principal investigators (1 per site), project managers (1 per site), and field staff (asthma care coordinators; 1–3 per site). Interview questions were developed by the evaluators (M.R.J., S.C.S.) and incorporated implementation-related factors identified by Durlak and Dupre (9). The interviews included review of a YWC core components table; respondents were asked whether and how each component was implemented. Interview notes (verified and augmented with audio-recordings) were coded with both codes that we prespecified based on the theoretical constructs used to develop interview questions as well as on additional codes suggested by the interview data; “retrofitting” fell into the latter category. Results and Discussion Data indicated that no site implemented all YWC components, adapted or not. One site leader described their adaptation process as follows: “We didn’t really adapt the EBI so much as we backed it into what we were already doing.” A leader of another site urged us to take our study of “translation” out of the evaluation, because she did not like the idea of assessing fidelity and instead saw greater value in examining the strengths and weaknesses inherent in each setting where care coordination interventions are implemented. These comments alluded to the reality that implementing organizations often have a base of existing services and select EBI components that will 1) satisfy funder requirements for EBI implementation, 2) enhance or expand the services they already provide, or 3) both. Of the 15 core components listed in Table 1, the number that existed at project sites before funding ranged from 0 to 7 (mean, 4.8); the number added with project funding ranged from 1 to 8 (mean, 4.8); and the number that were not present at any point ranged from 2 to 10 (mean, 5.5) (Table 2). Table 3 provides examples from each site — gleaned from interviews, annual reporting forms, and conference call minutes — of how existing services were altered to better align with YWC components. For example, because children in YWC are stratified according to medical and psychosocial risk for purposes of determining intervention intensity, the school nurses in the Los Angeles Unified School District formalized a process of risk stratification based on asthma control, while continuing to informally account for psychosocial risks as they had done previously. We termed these types of changes as “retrofitting.” Table 2 Implementation Level of Yes We Can Core Components at MCAN Sites, Case Study of Pediatric Asthma Care Coordination, United States, 2010–2014a Core Component Philadelphia Los Angeles Chicago Puerto Rico Risk stratification: establishing levels of care Risk based on medical severity and control, psychological risk, and social risk 3 3 2 1 Care pathway and intervention activities matched to risk level 2 3 2 1 Asthma care coordinator Culturally and linguistically aligned with families served 2 2 2 3 Provides basic education, including how to use spacers and reduce triggers 2 2 2 3 Addresses social problems as they arise, including working with schools, providing assistance in finding new housing, making referrals to smoking cessation programs, and obtaining refills 2 2 2 1 Makes the family feel like a valued member of the care team 3 2 2 3 Asthma clinical care: chronic care approach Prevention-based asthma clinic established with set hours to see children with asthma 1 2 3 1 Asthma care coordinator integrated into the health care team 3 1 1 1 Team members reinforce colleagues’ educational efforts 3 3 1 3 Routine case conferences follow clinic hours 1 1 1 1 Careful planning and integration of clinic and home visits supported by phone calls 3 1 1 1 Schedule for ongoing assessment of control 3 3 1 1 Designated clinical champion 2 2 2 3 Clinical case management Organized and systematic tracking of patients to assess needs and match services 3 3 1 1 Established network for referrals 3 2 1 1 Abbreviation: MCAN, Merck Childhood Asthma Network. a Stage of implementation: 1 = never existed, 2 = already existed, 3 = added with MCAN funding. Table 3 Retrofitting Yes We Can Core Components, Examples From 4 Program Sites, Case Study of Pediatric Asthma Care Coordination, United States, 2010–2014 Site Yes We Can Core Component Characteristics Existing Retrofits Philadelphia ACC culturally and linguistically aligned with families, provides basic education, addresses social problems, and makes family feel like part of care team ACCs who made home visits were already culturally and linguistically aligned with families and provided basic education ACCs learned how to assess and address social problems and were better able to make families feel like a valuable part of the care team, because the ACCs themselves were integrated into the clinical care team Los Angeles Formal risk stratification process based on medical control and social and psychological risk; intervention activities matched to risk level ACCs were registered nurses who informally accounted for risk when determining the level of care Program implemented formal risk stratification based on asthma control; ACCs continued to informally account for social and psychological risks Chicago ACC integrated into the health care team ACCs were community based; they made educational home visits but were not connected to clinical care ACCs also recruited and provided education in clinic waiting rooms and communicated with clinical providers Puerto Rico Chronic care approach to asthma clinical care, including aligning educational efforts among care providers Standard asthma care; not all clinicians using an asthma action plan Clinical champion helped implement routine use of an asthma action plan, which reinforced colleagues’ educational efforts Abbreviation: ACC, asthma care coordinator. Why the term “retrofitting,” and what are its potential benefits? A dictionary defines “to retrofit” as “to install new or modified parts or equipment in something previously manufactured or constructed” (www.merriam-webster.com/dictionary/retrofit). For example, a homeowner might retrofit a house to make it more energy-efficient by adding insulation, or replacing windows. Similarly, program providers might retrofit existing services in a certain program area (the analog to a house) to make them more like those of an EBI. Just as it often makes more sense for homeowners to retrofit their existing house rather than build a new one, so might program providers modify what they are already doing rather than replace services to implement an EBI in its entirety. We believed that the term “retrofitting” was appropriate for the implementation-related phenomenon we sought to describe, as it denotes improving, via scientifically supported practices, something that already exists. When EBIs are implemented in settings where services overlap with EBI elements, the 2 must be smoothly integrated. This could take the form of “tweaking” existing services to align them with EBI elements, continuing services that have the same goals as the EBI but are not in the EBI design, and basing decisions on adding EBI elements on the degree to which they would enhance, or adversely affect, current practices. Potential benefits of retrofitting include allowing organizations to keep what already works and capitalize on political support for program activities and resources needed for implementation, such as training, physical space, and time (6). Because political support and organizational integration are associated with program sustainability (17), retrofitting may enhance an organization’s ability to maintain effective innovations following the initial funding period. Finally, the notion of retrofitting elevates practice-based and local expertise, by placing value on what an organization already does. The notion of overlapping EBI components with existing services can be found in several implementation science articles: a category from Stirman and colleagues’ program-modification typology, “Integrating the intervention into another framework — ie, selecting elements” (7); Durlak and Dupre’s (9) “Integration of new programming,” defined as “the extent to which an organization can incorporate an innovation into its existing practices and routines”; and the construct of “Organizational Fit/Compatibility,” which encompasses how well the innovation fits with existing systems, in the Consolidated Framework for Implementation Research (6). Generally, however, adaptation has been conceptualized as changing the content or process of the EBI to improve its fit (7), rather than reshaping existing services to be more similar to the EBI (ie, retrofitting). The difference between adaptation and retrofitting may merely be one of vantage point and degree. However, we suggest that recognition of the retrofitting phenomenon can inform various stages of the EBI pipeline: the design of programs and the trials to establish their efficacy; isolating core components and preparing EBIs for dissemination; conducting needs and capacity assessments of implementing organizations; training and technical support; and evaluations of the processes and outcomes of retrofitted programs and practices. We give brief examples of how this influence might occur: Program design and efficacy testing: A consideration of retrofitting may mean decreased frequency of designing programs “from scratch” (eg, in an academic setting) with increased attention to current practices and how they can be incrementally improved. Equitable community–academic partnerships may be especially valuable in this regard. Isolating core components: In a retrofitting-aware approach, current practice is seen as a foundation that can be improved with evidence-based modifications. Therefore, being able to accurately identify the “active ingredients” in a program (whether processes or principles) takes on special importance. The challenges we experienced in identifying YWC’s core components have been described (18). Although solving the thorny practical and theoretical issues surrounding core components was beyond the scope of this article, it is worth considering emerging methods such as qualitative comparative analysis (19) for improving the validity of core component identification. Needs and capacity assessment: EBI implementation frameworks (20,21) typically include an assessment phase — which examines an organization’s strengths, needs, and structure, among other attributes — with the purpose of informing how the EBI is subsequently adapted and implemented. One explicit aim of such a phase could be to identify where and how existing services can be retrofit with EBI components. Training and technical support: Training and technical support that accommodate retrofitting would place greater emphasis on smoothly integrating EBI components into existing services and less on fidelity to the procedures used in the EBI’s original trial (5). Process and outcome evaluations of retrofitted programs and practices: Miller et al (2) pose an ontological question with practical implications, “At what point does a replicated program become a new model program in its own right?” EBIs implemented via significant retrofitting, as in this study, may be considered new programs that would benefit from process and outcome evaluations. One formal approach to doing this is Rapkin et al’s Comprehensive Dynamic Trial paradigm (11). This approach prescribes careful monitoring of the performance of interventions that are implemented in new settings. This is done for purposes of quality improvement and to inform the design of embedded “mini experiments” to test the value of specific adaptations. Indeed, MCAN sites engaged in a process similar to this, as each site reported using evaluation for quality improvement (22), and 1 site (Puerto Rico) embedded a randomized experiment to compare the effects of asthma care coordination delivered in a combined clinical–home versus a clinic setting. Several limitations apply. First, the concept of retrofitting emerged from the qualitative data collected from a single evaluation, and we did not collect data specifically addressing this concept. In future studies, instruments could be developed to measure retrofitting more explicitly. Second, retrofitting may apply to only a limited number of circumstances; for example, it may be less relevant where the innovation under study changes a single existing practice as opposed to a multicomponent package of activities and processes. Third, our sample was small, consisting of 4 sites with different patient populations, recruitment approaches, and settings. Therefore, we were not able to correlate degree of retrofitting with important implementation-related outcomes such as reach, adoption, or sustainability. Nor were we able to associate retrofitting with asthma-related outcomes among participants. Notably, each program site reported improvements in asthma-related outcomes among program participants, offering tentative evidence that the retrofits contributed to, or at least did not hinder, the success of the programs (23). Conclusion We found preliminary support for the validity of the retrofitting concept, although the explanatory or practical value of this concept is yet to be determined. The key assumption underlying the notion of retrofitting is that practitioners are doing things that work and that may benefit from evidence-based modifications rather than new interventions. This perspective aligns with calls for more practice-based evidence (24) and with community-based participatory approaches to research that emphasize local knowledge, resources, and practices (25). Future research is needed to explore retrofitting in a systematic manner, using larger samples and mixed quantitative and qualitative methods to learn more about processes and effects. Acknowledgments Funding for this study came from MCAN, a nonprofit, 501(c)(3) organization funded by the Merck Foundation from 2005–2015. Two authors (J.K.L., MCAN Programs Manager, and F.J.M., MCAN Executive Director) were Merck employees and played a role in the study design, in the interpretation of data, in the writing of the manuscript, and in the decision to submit the manuscript for publication. They did not, and were not permitted to promote the commercial products of Merck. A version of this study was presented at AcademyHealth’s 7th Annual Conference on the Science of Dissemination and Implementation in December 2014. We thank Kelsey Thome for her assistance with manuscript preparation. The opinions expressed by authors contributing to this journal do not necessarily reflect the opinions of the U.S. Department of Health and Human Services, the Public Health Service, the Centers for Disease Control and Prevention, or the authors' affiliated institutions. Suggested citation for this article: Janevic MR, Stoll SC, Lara M, Ramos-Valencia G, Stephens TB, Persky V, et al. The “Retrofitting” Approach to Adapting Evidence-Based Interventions: A Case Study of Pediatric Asthma Care Coordination, United States, 2010–2014. Prev Chronic Dis 2016;13:160129. DOI: http://dx.doi.org/10.5888/pcd13.160129. ==== Refs References 1. Jacobs JA , Jones E , Gabella BA , Spring B , Brownson RC . Tools for implementing an evidence-based approach in public health practice. Prev Chronic Dis 2012;9 :E116. 22721501 2. Miller RL , Forney JC , Hubbard P , Camacho LM . Reinventing Mpowerment for black men: long-term community implementation of an evidence-based program. Am J Community Psychol 2012;49 (1-2 ):199–214. 10.1007/s10464-011-9459-5 21773862 3. Green LW , Glasgow RE . Evaluating the relevance, generalization, and applicability of research: issues in external validation and translation methodology. Eval Health Prof 2006;29 (1 ):126–53. 10.1177/0163278705284445 16510882 4. Rebchook GM , Kegeles SM , Huebner D . Translating research into practice: the dissemination and initial implementation of an evidence-based HIV prevention program. AIDS education and prevention: official publication of the International Society for AIDS Education. 2006;18(4 Suppl A):119-36. 5. Kalichman SC , Hudd K , Diberto G . Operational fidelity to an evidence-based HIV prevention intervention for people living with HIV/AIDS. J Prim Prev 2010;31 (4 ):235–45. 10.1007/s10935-010-0217-5 20582629 6. Damschroder LJ , Aron DC , Keith RE , Kirsh SR , Alexander JA , Lowery JC . Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science. Implement Sci 2009;4 (1 ):50. 10.1186/1748-5908-4-50 19664226 7. Stirman SW , Miller CJ , Toder K , Calloway A . Development of a framework and coding system for modifications and adaptations of evidence-based interventions. Implement Sci 2013;8 (1 ):65. 10.1186/1748-5908-8-65 23758995 8. Carvalho ML , Honeycutt S , Escoffery C , Glanz K , Sabbs D , Kegler MC . Balancing fidelity and adaptation: implementing evidence-based chronic disease prevention programs. J Public Health Manag Pract 2013;19 (4 ):348–56. 10.1097/PHH.0b013e31826d80eb 23462111 9. Durlak JA , DuPre EP . Implementation matters: a review of research on the influence of implementation on program outcomes and the factors affecting implementation. Am J Community Psychol 2008;41 (3-4 ):327–50. 10.1007/s10464-008-9165-0 18322790 10. Carroll C , Patterson M , Wood S , Booth A , Rick J , Balain S . A conceptual framework for implementation fidelity. Implement Sci 2007;2 (1 ):40. 10.1186/1748-5908-2-40 18053122 11. Rapkin BD , Weiss ES , Lounsbury DW , Thompson HS , Goodman RM , Schechter CB , Using the interactive systems framework to support a quality improvement approach to dissemination of evidence-based strategies to promote early detection of breast cancer: planning a comprehensive dynamic trial. Am J Community Psychol 2012;50 (3-4 ):497–517. 10.1007/s10464-012-9518-6 22618023 12. Dolcini M , Gandelman AA , Vogan SA , Kong C , Leak TN , King AJ , Translating HIV interventions into practice: community-based organizations’ experiences with the diffusion of effective behavioral interventions (DEBIs). Soc Sci Med 2010;71 (10 ):1839–46. 10.1016/j.socscimed.2010.08.011 20926169 13. Fixsen DL , Naoom SF , Blase KA , Friedman RM , Wallace F . Implementation research: a synthesis of the literature. FMHI publication no. 231. Tampa (FL): University of South Florida, Louis de la Parte Florida Mental Health Institute, The National Implementation Research Network; 2005. 14. Lara M , Bryant-Stephens T , Damitz M , Findley S , Gonzalez Gavillán J , Mitchell H , Balancing “fidelity” and community context in the adaptation of asthma evidence-based interventions in the “real world”. Health Promot Pract 2011;12 (6 Suppl ):63S–72S. 10.1177/1524839911414888 22068362 15. Thyne SM , Rising JP , Legion V , Love MB . The Yes We Can Urban Asthma Partnership: a medical/social model for childhood asthma management. J Asthma 2006;43 (9 ):667–73. 10.1080/02770900600925288 17092847 16. Thyne SM , Marmor AK , Madden N , Herrick G . Comprehensive asthma management for underserved children. Paediatr Perinat Epidemiol 2007;21 (s3 , Suppl 3 ):29–34. 10.1111/j.1365-3016.2007.00882.x 17935573 17. Calhoun A , Mainor A , Moreland-Russell S , Maier RC , Brossart L , Luke DA . Using the Program Sustainability Assessment Tool to assess and plan for sustainability. Prev Chronic Dis 2014;11 :130185. 10.5888/pcd11.130185 24456644 18. Century J , Rudnick M , Freeman C . A framework for measuring fidelity of implementation: a foundation for shared language and accumulation of knowledge. Am J Eval 2010;31 (2 ):199–218. 10.1177/1098214010366173 19. Kane H , Lewis MA , Williams PA , Kahwati LC . Using qualitative comparative analysis to understand and quantify translation and implementation. Transl Behav Med 2014;4 (2 ):201–8. 10.1007/s13142-014-0251-6 24904704 20. McKleroy VS , Galbraith JS , Cummings B , Jones P , Harshbarger C , Collins C , ; ADAPT Team. Adapting evidence-based behavioral interventions for new settings and target populations. AIDS Educ Prev 2006;18 (4 Suppl ):59–73. 10.1521/aeap.2006.18.supp.59 16987089 21. Kilbourne AM , Neumann MS , Pincus HA , Bauer MS , Stall R . Implementing evidence-based interventions in health care: application of the replicating effective programs framework. Implement Sci 2007;2 (1 ):42. 10.1186/1748-5908-2-42 18067681 22. Stoll S , Janevic M , Lara M , Ramos-Valencia G , Stephens TB , Persky V , A mixed-method application of the program sustainability assessment tool to evaluate the sustainability of 4 pediatric asthma care coordination programs. Prev Chronic Dis 2015;12 :150133. 10.5888/pcd12.150133 26632955 23. Janevic MR , Stoll SC , Wilkin M , Song P , Baptist A , Lara M , Pediatric asthma care coordination in underserved communities: a quasi-experimental study. Am J Public Health. Forthcoming. 24. Green LW . Making research relevant: if it is an evidence-based practice, where’s the practice-based evidence? Fam Pract 2008;25 (Suppl 1 ):i20–4. 10.1093/fampra/cmn055 18794201 25. Israel BA , Schulz AJ , Parker EA , Becker AB . Review of community-based research: assessing partnership approaches to improve public health. Annu Rev Public Health 1998;19 (1 ):173–202. 10.1146/annurev.publhealth.19.1.173 9611617

PMC005xxxxxx/PMC5003529.txt

==== Front Prev Chronic Dis Prev Chronic Dis PCD Preventing Chronic Disease 1545-1151 Centers for Disease Control and Prevention 27560723 16_0130 10.5888/pcd13.160130 Original Research Peer ReviewedDifferences in Fruit and Vegetable Intake by Race/Ethnicity and by Hispanic Origin and Nativity Among Women in the Special Supplemental Nutrition Program for Women, Infants, and Children, 2015 Di Noia Jennifer PhD Monica Dorothy MPH Cullen Karen Weber DrPH RD Pérez-Escamilla Rafael PhD Gray Heewon Lee PhD RD Sikorskii Alla PhD Author Affiliations: Dorothy Monica, Saint Joseph’s WIC Program, Paterson, New Jersey; Karen Weber Cullen, USDA/ARS Children’s Nutrition Research Center, Baylor College of Medicine, Houston, Texas; Rafael Pérez-Escamilla, Yale School of Public Health, New Haven, Connecticut; Heewon Lee Gray, Teachers College, Columbia University, New York, New York; Alla Sikorskii, Michigan State University, East Lansing, Michigan. Corresponding Author: Jennifer Di Noia, PhD, Department of Sociology, William Paterson University, 300 Pompton Rd, Wayne, NJ 07470. Telephone: 973-720-3714. Email: dinoiaj@wpunj.edu. 2016 25 8 2016 13 E115Introduction The objective of this exploratory study was to determine whether fruit and vegetable consumption differed by race/ethnicity, by origin and nativity among Hispanics, and by language preference (as an indicator of acculturation) among foreign-born Hispanics. Methods We recruited 723 women enrolled in the Special Supplemental Nutrition Program for Women, Infants and Children (WIC) and orally administered a questionnaire containing demographic items, validated measures of food security status and social desirability trait, and the Behavioral Risk Factor Surveillance System fruit and vegetable module. Differences in intakes of 100% fruit juice, fruit, cooked or canned beans, and dark green, orange-colored, and other vegetables were assessed by using analysis of covariance with Bonferroni post hoc tests. Analyses were controlled for age, pregnancy status, breastfeeding status, food security status, educational attainment, and social desirability trait. Results The frequency of vegetable intake differed by race/ethnicity (cooked or canned beans were consumed more often among Hispanic than non-Hispanic black and non-Hispanic white or other participants, orange-colored vegetables were consumed more often among Hispanics than non-Hispanic black participants, and other vegetables were consumed more often among non-Hispanic white or other than among non-Hispanic black and Hispanic participants), origin (other vegetables were consumed more often among Columbian and other Hispanics than Dominican participants) and nativity (orange-colored vegetables were consumed more often among foreign-born than US-born Hispanics). Fruit and vegetable intake did not differ by language preference among foreign-born Hispanics. Conclusion Differences in fruit and vegetable consumption among WIC participants by race/ethnicity and by Hispanic origin and nativity may have implications for WIC nutrition policies and nutrition education efforts. ==== Body Introduction The US population does not meet dietary guidelines (1) for intake of vegetables in any vegetable subgroup (dark green, red and orange, legumes, starchy, or other), and most (80%) do not meet recommendations for intake of fruit (whole fruit or 100% juice). Low-income populations are among those least likely to meet guidelines (2). A better understanding of fruit and vegetable intake in this population is needed, in particular, fruits and vegetables emphasized in dietary guidelines. Differences by race/ethnicity are found in total intake of fruit, fruit juice, legumes, plantains, and root crops (higher among Hispanics than non-Hispanic whites), total vegetables, potatoes, other vegetables, and green salad (higher among non-Hispanic whites than Hispanics), and non-citrus fruit (higher among Hispanic than non-Hispanic black and and non-Hispanic white women) and by nativity in intakes of fruits and legumes (consumed by more foreign-born than US-born Hispanics) (3–6). As acculturation increases among foreign-born Hispanics, intake of fruit, beans, and starchy root vegetables decreases (7,8). Differences by origin are found in intake of white potatoes. South American women eat more white potatoes than do Mexican, Central American, Caribbean, and Spanish women and women of multiple Hispanic origin and more green salad than Central American women. Mexican, Central American, Caribbean, and Spanish women eat more cooked dried beans than do South American women (9,10). Whether differences by race/ethnicity occur among low-income women and by origin and nativity among low-income Hispanic women is unclear. The objective of this exploratory study was to determine whether intake of fruits and vegetables differed by race/ethnicity, origin, and nativity among Hispanics, and by language preference among foreign-born Hispanics (as an indicator of acculturation). Methods We conducted an exploratory study of 723 low-income women enrolled in the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC). Of these, 436 were Hispanic, and of these, 244 were foreign-born. Information on foods made available to women in the WIC program is available elsewhere (11). Design and sample This cross-sectional study examined baseline data from WIC Fresh Start, a randomized controlled trial of nutrition education to promote purchases and consumption of fruits and vegetables sold at farmers’ markets among women enrolled in WIC. The trial is described elsewhere (12). Eligibility criteria for the trial were being pregnant, postpartum, or a female caregiver of an infant or child enrolled in WIC; having no known dietary restrictions; and not being at high risk for nutritional or health problems (as defined by WIC). Participants were recruited from among women in the waiting room of a large WIC agency located in a densely populated urban area of New Jersey by research assistants fluent in both English and Spanish. Of the 1,345 women approached, 64 did not meet eligibility criteria, 537 were eligible but declined to participate, and 744 were enrolled (58% consent rate). This study examined data provided by WIC Fresh Start participants reporting one race/ethnicity (N = 723). Participants provided informed written consent. The study was approved by the William Paterson University Institutional Review Board for Human Subject Research (no. 2014–368). Data were collected from June 1, 2015, through August 12, 2015. Measurement and statistical analyses We orally administered a questionnaire that consisted of closed-ended, multiple-choice questions assessing participants’ race/ethnicity, nativity (US-born, born outside of the United States [defined as outside of the United States and its territories, including Puerto Rico]) and language preference (English, Spanish, other, and, if “other,” what that preferred language was). We made Hispanic/Latina a choice of race because many Hispanic WIC participants consider their ethnicity their race (13). After reporting race/ethnicity, participants were asked about their origin (eg, “If you answered Hispanic/Latina [to the question on race/ethnicity], what is your origin or origins, for example Puerto Rican, Mexican, Dominican, Columbian and so on?”). For those reporting more than one origin (40 [6%]), origin was coded as the first origin reported. Some researchers advocate grouping Hispanic origin countries by geographic region (eg, grouping Mexican, Mexican American, and Chicano together as Mexican) because geographic regions tend to share sociopolitical history and culture (9). However, health needs and health outcomes differ significantly among Hispanics by country of origin (14). For this reason, we did not group Hispanic participants by region in this study so as not to obscure differences in fruit and vegetable intake among Hispanics from different countries within the same region. Previous work has shown that greater acculturation is associated with English language preference than with Spanish language preference (15). For this reason, foreign-born Hispanics who preferred to speak English were considered more acculturated than were those who preferred to speak Spanish. Participants reported their birthdate (used to calculate their age in years), pregnancy status, breastfeeding status, and educational attainment and completed validated measures of food security status and social desirability trait (a response set reflecting the tendency to respond in a manner consistent with perceived social norms) (16,17). Participants were classified as food secure or insecure on the basis of their response to the food security measure. On the measure of social desirability trait, scores ranged from 0 to 10, with higher scores indicating a higher social desirability trait. Analyses were adjusted for the potentially biasing effects of these variables on fruit and vegetable intake. We used the 2013 Behavioral Risk Factor Surveillance System (BRFSS) fruit and vegetable module to measure intake of 100% fruit juice, fruit, cooked or canned beans, and dark green, orange-colored, and other vegetables (18). The instrument has moderate validity (correlations with diet records, 24-hour recalls, and food frequency questionnaires range from 0.29 to 0.63) and reliability (test–retest correlations and κ values range from 0.19 to 0.77) (19). Respondents could report their intake as the number of times per day, week, or month the included foods were consumed. For the WIC Fresh Start trial, we changed the reference period from the previous month to the previous 2 weeks. Before presenting the questions, the interviewer read the following: “These next questions are about the fruit and vegetables you ate or drank during the past 2 weeks.” The original response categories were retained. As such, participants could report their intake as times per month foods were consumed because it was assumed that intake occurred uniformly (ie, a response of “2 times a month” was equivalent to a response of “once every 2 weeks” ). Interviewers followed BRFSS guidelines on which foods and juices to count in fruit and vegetable categories (Appendix). We examined BRFSS fruit and vegetable item response distributions, and all but one (the distribution for the item assessing fruit intake) were positively skewed. Square-root transformations were therefore applied to improve the normality of the distributions (transformed data are presented). Descriptive statistics were used to characterize the sample. Differences in fruit and vegetable intake by race/ethnicity were examined with analysis of covariance (ANCOVA) with Bonferroni post hoc tests adjusted for covariates. Subgroup analyses among Hispanic participants (N = 436) examined differences in intake by origin and nativity. A final set of ANCOVAs conducted among foreign-born participants (N = 244) examined whether consumption differed by language preference. To gauge the practical significance of differences found (in addition to the statistical significance of the differences), effect sizes were computed. Effect sizes were expressed as Cohen d, the difference between subgroup means divided by the adjusted standard deviation (square root of the mean-squared error). A Cohen d of 0.20 is considered small, 0.50 is considered medium, and 0.80 is considered large (20). There was 80% power to detect medium effect sizes in tests of intake differences by race/ethnicity and nativity and large effect sizes in tests of intake differences by Hispanic origin and language preference. Analyses were conducted in 2016 with SPSS for Windows, version 23, 2015 (IBM Inc). Across analyses, we used a P value of .05 to establish significance. Results The sample of 723 participants had a mean age of 29.0 years (standard deviation [SD] 6.9 y); 17% were pregnant, 22% were breastfeeding, 55% were food insecure, 60% were Hispanic, 31% were non-Hispanic black, and 9% were non-Hispanic white or other. The largest origin groups among Hispanics were Dominican (n = 159, 36%), Puerto Rican (n = 103, 24%), Mexican (n = 43, 10%), Peruvian (n = 33, 8%), and Columbian (n = 28, 6%). Nearly three-fifths (56%) of Hispanics were born outside the United States. Although 229 (53%) Hispanics preferred to speak English, most foreign-born Hispanics (n = 244) preferred to speak Spanish (n = 190, 78%). Fifty percent of the sample reported a high school or general equivalency degree or less. The mean score in the sample on the measure of social desirability trait was 7.7 (SD, 1.7) (Table 1). Relative to WIC Fresh Start participants, higher percentages of women served by the collaborating WIC agency were pregnant (44%) and breastfeeding (36%). Sixty-six percent of the WIC clinic population was Hispanic, 13% was non-Hispanic black, and 21% was non-Hispanic white or other. Table 1 Characteristics of Participants (N = 723) in Study of Differences in Intake of Fruits and Vegetables by Race/Ethnicity and by Hispanic Origin and Nativity Among Women in the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC), 2015 Characteristic Number (%)a All Participants Age, mean (SD), y 29.0 (6.9) Pregnant 124 (17) Breastfeeding 157 (22) Food insecure 396 (55) Social desirability trait, mean (SD)b 7.7 (1.7) Race/ethnicity Non-Hispanic black 221 (31) Hispanic/Latina 436 (60) Non-Hispanic white or other 66 (9) Nativity United States 431 (60) Outside United States 292 (40) Educational attainment Elementary school (grades 1–8) 27 (4) Some high school (grades 9–12), no diploma 103 (14) High school or General Equivalency Degree or equivalent 235 (32) More than high school 358 (50) Non-Hispanic Race/Ethnicity Origin Non-Hispanic blackc Jamaican 18 (52) Nigerian 4 (11) Other origin 13 (37) Non-Hispanic white or otherc Italian 18 (41) Irish 9 (21) Other origin 17 (38) Nativity Non-Hispanic black United States 188 (85) Outside United States 33 (15) Non-Hispanic white or other United States 51 (77) Outside United States 15 (23) Hispanic Race/Ethnicityd Origin Dominican 159 (36) Puerto Rican 103 (24) Mexican 43 (10) Peruvian 33 (8) Columbian 28 (6) Other Hispanic origin 70 (16) Nativity United States 192 (44) Outside United States 244 (56) Preferred language English 229 (53) Spanish 207 (47) Language preference by nativity United States English language preference 175 (91) Spanish language preference 17 (9) Outside United States English language preference 54 (22) Spanish language preference 190 (78) a Values are n (%) unless otherwise indicated. b The tendency to respond in a manner consistent with perceived social norms. Measured with a short form of the Marlowe-Crowne Social Desirability Scale (M-C 2[10]) (16); scores ranged from 0 to 10, with higher scores indicating a higher social desirability trait. Sample size = 719. c Thirty-five non-Hispanic black and 44 non-Hispanic white participants reported an origin. d Hispanic/Latina was offered as a choice of race in the study questionnaire because many Hispanic WIC participants consider their ethnicity their race (13). We found a significant association between race/ethnicity and intake of cooked or canned beans, orange-colored vegetables, and other vegetables (Table 2). Hispanic participants had a higher frequency of intake of cooked or canned beans than non-Hispanic blacks (adjusted mean difference, 0.16; 95% confidence interval [CI], 0.09–0.23]; Cohen d = 0.45, P < .001) and non-Hispanic white or other participants (adjusted mean difference, 0.18; 95% CI, 0.06–0.29; Cohen d = 0.51, P = .001) and a higher frequency of intake of orange-colored vegetables than non-Hispanic blacks (adjusted mean difference, 0.14; 95% CI, 0.07–0.21; Cohen d = 0.41; P < .001). Non-Hispanic white or other participants had a higher frequency of intake of other vegetables than did non-Hispanic blacks (adjusted mean difference, 0.16; 95% CI, 0.02–0.30; Cohen d = 0.41; P = .01) and Hispanics (adjusted mean difference, 0.16; 95% CI, –0.29 to –0.03; Cohen d = 0.41; P = .008). Table 2 Differences in Intake of Fruits and Vegetables by Race/Ethnicity Among Women in the Special Supplemental Nutrition Program for Women, Infants, and Children, 2015 Food Adjusted Mean Difference (95% Confidence Interval)a,b Hispanic (N = 221) Non-Hispanic Black (N = 221) Non-Hispanic White or Other (N = 66) 100% fruit juice Reference −0.02 (−0.13 to 0.10) 0.01 (−0.18 to 0.19) Fruit Reference 0.02 (−0.23 to 0.27) −0.11 (−0.51 to 0.29) Cooked or canned beans Reference 0.16 (0.09 to 0.23) 0.18 (0.06 to 0.29) Dark green vegetables Reference −0.05 (−0.12 to 0.02) −0.08 (−0.20 to 0.04) Orange-colored vegetables Reference 0.14 (0.07 to 0.21) 0.05 (−0.06 to 0.17) Other vegetables Reference 0.00 (−0.76 to 0.08) −0.16 (−0.29 to −0.03) a Values are reported as times per day items were consumed. Differences were examined with analysis of covariance with Bonferroni post hoc tests. Analyses were adjusted for age; pregnancy, breastfeeding, and food security status; educational attainment; and social desirability trait. Intake of other vegetables was higher among non-Hispanic white or other participants than among non-Hispanic blacks (adjusted mean difference = 0.16, 95% CI [0.02–0.30]). b Values were obtained by subtracting the mean frequency of intake of each group from the mean frequency of intake among Hispanics (reference group). Among Hispanic participants, we found a significant association between origin and intake of other vegetables (Table 3). Columbian participants more frequently ate vegetables in the “other vegetables” category than Dominican participants (adjusted mean difference, 0.25; 95% CI, –0.47 to –0.03; Cohen d = 0.71; P = .01), and participants of other Hispanic origin more frequently ate other vegetables than did Dominican participants (adjusted mean difference, 0.19; 95% CI, –0.34 to –0.03; Cohen d = 0.52; P = .008). A significant association was found between nativity and eating orange-colored vegetables. Intake of orange-colored vegetables was higher among foreign-born than US-born participants (adjusted mean difference, 0.11; 95% CI, –0.20 to –0.02; Cohen d = 0.31; P = .02) (Table 4). Fruit and vegetable intake did not differ by language preference among foreign-born Hispanics. Table 3 Differences in Intake of Fruits and Vegetables by Origin Among Hispanic Women in the Special Supplemental Nutrition Program for Women, Infants, and Children, 2015a,b Food Adjusted Mean Difference (95% Confidence Interval) Dominican (N = 159) Puerto Rican (N = 103) Mexican (N = 43) Columbian (N = 28) Peruvian (N = 33) Other Hispanic (N = 70) 100% fruit juice Reference −0.01 (−0.27 to 0.25) 0.07 (−0.25 to 0.27) 0.02 (−0.32 to 0.36) 0.11 (−0.21 to 0.43) 0.21 (−0.22 to 0.26) Fruit Reference 0.24 (−0.31 to 0.79) −0.05 (−0.72 to 0.62) 0.42 (−0.30 to 1.13) 0.43 (−0.25 to 1.12) −0.10 (−0.61 to 0.42) Cooked/canned beans Reference 0.06 (−0.09 to 0.22) 0.06 (−0.13 to 0.24) 0.21 (−0.01 to 0.40) 0.19 (−0.00 to 0.39) 0.10 (−0.05 to 0.24) Dark green vegetables Reference 0.09 (−0.09 to 0.27) −0.05 (−0.27 to 0.17) 0.05 (−0.18 0.29) 0.10 (−0.12 to 0.33) −0.04 (−0.21 to 0.13) Orange-colored vegetables Reference 0.07 (0.10 to 0.24) −0.04 (−0.24 to 0.16) −0.01 (−0.23 to 0.20) −0.04 (−0.25 to 0.16) −0.01 (−0.16 to 0.15) Other vegetables Reference −0.13 (−0.31 to 0.04) −0.06 (−0.27 to 0.15) −0.25 (−0.47 to −0.03) −0.01 (−0.23 to 0.20) −0.19 (−0.34 to −0.03) a Values are reported as times per day items were consumed. Differences were examined with analysis of covariance with Bonferroni post hoc tests. Analyses were adjusted for age; pregnancy, breastfeeding, and food security status; educational attainment; and social desirability trait. b Values were obtained by subtracting the mean frequency of intake of each group from the mean frequency of intake among Dominicans (reference group). Table 4 Differences in Intake of Fruits and Vegetables by Nativity Among Hispanic Women in the Special Supplemental Nutrition Program for Women, Infants, and Children, 2015a,b Food Adjusted Mean Difference (95% Confidence Interval) US-born, n = 192 Foreign-born, n = 244 100% fruit juice Reference −0.08 (−0.22 to 0.06) Fruit Reference −0.03 (−0.33 to 0.27) Cooked or canned beans Reference −0.01 (−0.09 to 0.07) Dark green vegetables Reference 0.02 (−0.08 to 0.11) Orange-colored vegetables Reference −0.11 (−0.20 to −0.02) Other vegetables Reference −0.07 (−0.17 to 0.02) a Values are reported as times per day items were consumed. Differences were examined with analysis of covariance with Bonferroni post hoc tests. Analyses were adjusted for age; pregnancy, breastfeeding and food security status; educational attainment; and social desirability trait. b Values were obtained by subtracting the mean frequency of intake of each group from the mean frequency of intake among US-born participants (reference group). Discussion We found differences by race/ethnicity in intake of cooked or canned beans, orange-colored vegetables, and vegetables in the other vegetables category. Among Hispanics, we found differences by origin in intakes of other vegetables and by nativity in intake of orange-colored vegetables. Language preference was unrelated to fruit and vegetable intake among foreign-born Hispanics, and intake of fruit and fruit juice did not differ by any of the demographic variables studied. The magnitude of differences in times per day vegetables were consumed ranged from 0.31 to 0.71. Differences by race/ethnicity in the frequency of eating cooked or canned beans were not surprising in light of previous findings, thereby confirming existing knowledge of acculturation, cultural practices, and food preferences (3). However, other differences between our findings and previous findings were evident. Whereas Hispanics consumed more fruit, noncitrus fruit, and fruit juice than non-Hispanic blacks and non-Hispanic whites in other studies (4,5), intake of fruit and 100% juice did not differ by race/ethnicity in our study, possibly because of the low-income population that made up the sample. One study of low-income adults found that fruit and juice intake did not differ by race/ethnicity (21). A qualitative study conducted among low-income Hispanic women showed that food choices were shaped by cultural food preferences; a meal of rice, beans, and meat was considered healthy, satisfying, and more affordable than a meal including such healthy items as fruit and fruit juice (22). Such preferences may explain why Hispanic participants in our study did not consume fruit and 100% fruit juice more often than non-Hispanic blacks or non-Hispanic whites. Why Hispanics ate orange-colored vegetables more often than non-Hispanic black participants is unclear. Previous work showed that neighborhood availability of orange-colored vegetables was associated with consumption of these foods and that convenience and corner stores in Hispanic neighborhoods were more likely to carry items such as carrots than stores in black neighborhoods (23,24). Alternatively, findings may be explained by differences between racial/ethnic groups in preferences for orange-colored vegetables such as carrots and winter squash. A study of carotenoid intake and profiles found that non-Hispanic whites obtained most α-carotene and β-carotene from carrots and mixed vegetables. Although these vegetables also were sources among Hispanics, winter squash (frequently included in the preparation of beans) was a more significant source (3). Moreover, although non-Hispanic whites obtained important amounts of lutein and zeaxanthin from leafy green vegetables (spinach and broccoli), Hispanics obtained more of these nutrients from winter squash (3). Also unclear is why the frequency of intake of other vegetables was higher among non-Hispanic white participants than non-Hispanic blacks or Hispanics. A study of adults found that non-Hispanic whites consumed more other vegetables than Hispanics (5); however, the sample did not include non-Hispanic blacks. Potatoes are counted among other vegetables assessed by the BRFSS fruit and vegetable module (18). Previous work showed that non-Hispanic white women eat potatoes more often than Hispanic women, which may explain the higher intake of other vegetables found in this group (3). Among Hispanics, Columbians had a higher frequency of intake of other vegetables than did Dominican and other Hispanic participants. Colón-Ramos et al similarly found that South American women (including Columbians) had higher intakes of other white potatoes (potatoes other than French fries, home fries, or hash browns, an item comparable to non-fried potatoes counted among other vegetables by the BRFSS fruit and vegetable module) than did Caribbean women (including Dominicans) and higher intakes of lettuce (also counted among other vegetables by the BRFSS) than did Central American women (including Salvadorians, Guatemalans, Costa Ricans, Hondurans, Nicaraguans, and Panamanians) (9). Yet, our findings are not directly comparable to the findings of Colón-Ramos et al because multiple origins were subsumed within the Hispanic groups studied by Colón-Ramos et al. In other work, total vegetable intakes differed among Cubans, Dominicans, and Puerto Ricans, and intakes of dark green/yellow and non-starchy vegetables differed between Puerto Rican and non-Puerto Rican Hispanic women (10). Together with previous findings (25), our findings suggest that vegetable intakes differ by Hispanic origin. The few studies to date underscore the need for further research of this type. Studies of vegetable intakes in diverse Hispanic origin groups may aid understanding of similarities and differences between groups. Most foreign-born Hispanics preferred to speak Spanish, suggesting that they were less acculturated than foreign-born Hispanics who preferred to speak English. The higher intake of orange-colored vegetables among foreign-born than US-born Hispanics is consistent with literature demonstrating an inverse association between acculturation and fruit and vegetable intake (7,8). The homogeneity of language preference among foreign-born Hispanics may also explain why fruit and vegetable intake did not differ by language preference in this group. Our study had limitations. Fruit and vegetable intake was assessed by self-report, which may lead to misclassification and bias (9). To minimize bias, analyses were controlled for social desirability trait, a factor associated with overestimation of self-reported intake (26). The self-selected sample and possible self-selection bias resulting from the high refusal rate limit the generalizability of findings. Findings may be generalizable only to WIC populations with characteristics similar to those in our sample (25). Findings of this study are not directly comparable to findings of previous studies because of the few similar studies conducted among low-income women in general and among women in WIC in particular. Most Hispanic participants in this study were Dominican or Puerto Rican. Intake patterns observed among Hispanics may therefore better reflect patterns found among Dominicans and Puerto Ricans than among Hispanics in general. Although this study was powered to detect moderate to large effects, observed effect sizes were small to medium by Cohen’s conventions (20). As such, there was insufficient power for some pairwise comparisons, highlighting the need for replication studies with sufficiently large subgroups of race/ethnicity, origin, and nativity. Despite these limitations, our findings add to the limited data on fruit and vegetable intake among low-income women in the WIC program. Findings advance understanding of variations in intake based on race/ethnicity and Hispanic origin and nativity. Fruit and vegetable intake was assessed with a validated measure in widespread use thereby facilitating comparisons with intake estimates derived from other similar studies reporting intake as times per day fruit and vegetables were consumed. To our knowledge, this is the first study to examine fruit and vegetable intake among the Hispanic origin groups we included. Most research to date has focused on Mexican Americans (9); therefore, our findings extend knowledge of intake differences among these groups. Analyses were controlled for several potential confounders, among them, social desirability trait. Although shown to bias self-reported dietary intake, social desirability assessments have not been included in previous studies of this type. Although reasons for the differences found in fruit and vegetable intake were considered, further research is needed to confirm factors that explain the differences. In addition to country of origin, future studies should collect data on geographic region of origin, because ingredients, preparation techniques, the use of culinary marinades and seasonings, and the names of similar foods and dishes can vary by country and from one region to another within the same country (27). This study was conducted in the summer. Fruit and vegetable availability and consumption vary seasonally (28). Such variations may have affected the results. The collection of dietary data at different times of the year is therefore recommended in future studies. BRFSS fruit and vegetable items encompass several foods. To better understand group differences in item intakes, assessments of foods counted in the definition of each are needed. We also recommend assessments of the neighborhood availability and accessibility of commonly consumed and culturally specific items because limited availability and accessibility may serve as barriers to consumption. Among WIC participants, vegetable intake differed by race/ethnicity, and among Hispanics, by origin and nativity. Differences found for some fruits and vegetables but not others underscore the need for WIC nutrition policies to focus on fruit and vegetable components (particularly those for which intakes are furthest from meeting recommendations) rather than fruit and vegetable intake overall. The success of WIC fruit and vegetable interventions may also be enhanced by adapting messages and materials to the needs of participants who differ by race/ethnicity and among Hispanics, by origin and nativity. Replication studies with large and diverse Hispanic origin groups are needed to confirm study findings. To aid understanding of components of dietary intake, the use of dietary assessment methods that allow for the collection of detailed information on food and juice intakes (eg, 24-hour recalls) is recommended. Attention to geographic region of origin in addition to country of origin also may advance understanding of differences in fruit and vegetable intake within Hispanic origin groups. Acknowledgments This article was produced under a project supported with federal funds from the US Department of Agriculture, Food and Nutrition Service, through grant WIC NEI-12-TX to Baylor College of Medicine to Jennifer Di Noia at William Paterson University. The contents of this publication do not necessarily reflect the view or policies of the US Department of Agriculture, nor does mention of trade names, commercial products, or organizations imply endorsement by the US government. Appendix. Foods Assessed by the Behavioral Risk Factor Surveillance System Fruit and Vegetable Module and Corresponding Foods and Juices Included and Excluded in the Definition of Each This file is available for download as a Microsoft Word document. The opinions expressed by authors contributing to this journal do not necessarily reflect the opinions of the U.S. Department of Health and Human Services, the Public Health Service, the Centers for Disease Control and Prevention, or the authors' affiliated institutions. Suggested citation for this article: Di Noia J, Monica D, Cullen KW, Pérez-Escamilla R, Gray HL, Sikorskii A. Differences in Fruit and Vegetable Intake by Race/Ethnicity and by Hispanic Origin and Nativity Among Women in the Special Supplemental Nutrition Program for Women, Infants, and Children, 2015. Prev Chronic Dis 2016;13:160130. DOI: http://dx.doi.org/10.5888/pcd13.160130. ==== Refs References 1. Dietary Guidelines Advisory Committee. Scientific report of the 2015 Dietary Guidelines Advisory Committee, advisory report to the Secretary of Health and Human Services and the Secretary of Agriculture. Washington (DC): US Department of Agriculture and US Department of Health and Human Services; 2015. 2. Robinson T . Applying the socio-ecological model to improving fruit and vegetable intake among low-income African Americans. J Community Health 2008;33 (6 ):395–406. 10.1007/s10900-008-9109-5 18594953 3. Bermudez OI , Ribaya-Mercado JD , Talegawkar SA , Tucker KL . Hispanic and non-Hispanic white elders from Massachusetts have different patterns of carotenoid intake and plasma concentrations. J Nutr 2005;135 (6 ):1496–502. 15930459 4. Forshee RA , Storey ML . Demographics, not beverage consumption, is associated with diet quality. Int J Food Sci Nutr 2006;57 (7-8 ):494–511. 10.1080/09637480600991240 17162328 5. Grimm KA , Blanck HM . Survey language preference as a predictor of meeting fruit and vegetable objectives among Hispanic adults in the United States, Behavioral Risk Factor Surveillance System, 2009. Prev Chronic Dis 2011;8 (6 ):A133. 22005626 6. Duffey KJ , Gordon-Larsen P , Ayala GX , Popkin BM . Birthplace is associated with more adverse dietary profiles for US-born than for foreign-born Latino adults. J Nutr 2008;138 (12 ):2428–35. 10.3945/jn.108.097105 19022968 7. Pérez-Escamilla R , Putnik P . The role of acculturation in nutrition, lifestyle, and incidence of type 2 diabetes among Latinos. J Nutr 2007;137 (4 ):860–70. 17374645 8. Ayala GX , Baquero B , Klinger S . A systematic review of the relationship between acculturation and diet among Latinos in the United States: implications for future research. J Am Diet Assoc 2008;108 (8 ):1330–44. 10.1016/j.jada.2008.05.009 18656573 9. Colón-Ramos U , Thompson FE , Yaroch AL , Moser RP , McNeel TS , Dodd KW , Differences in fruit and vegetable intake among Hispanic subgroups in California: results from the 2005 California Health Interview Survey. J Am Diet Assoc 2009;109 (11 ):1878–85. 10.1016/j.jada.2009.08.015 19857629 10. Siega-Riz AM , Sotres-Alvarez D , Ayala GX , Ginsberg M , Himes JH , Liu K , Food-group and nutrient-density intakes by Hispanic and Latino backgrounds in the Hispanic Community Health Study/Study of Latinos. Am J Clin Nutr 2014;99 (6 ):1487–98. 10.3945/ajcn.113.082685 24760972 11. US Department of Agriculture, Food and Nutrition Service. Foods for children and women. http://www.fns.usda.gov/sites/default/files/wic/SNAPSHOT-of-WIC-Child-Women-Food-Pkgs.pdf. Accessed June 26, 2016. 12. Di Noia J , Monica D , Cullen KW , Sikorskii A . A randomized controlled trial of nutrition education to promote farmers’ market fruit and vegetable purchases and consumption among women enrolled in the Special Supplemental Nutrition Program for Women, Infants and Children (WIC): rationale and design of the WIC Fresh Start program. BMC Nutr 2015;1 (1 ):33 10.1186/s40795-015-0032-8 13. US Department of Agriculture, Food and Nutrition Service, Office of Research and Analysis. National survey of WIC participants II, volume I: participant characteristics (final report). http://www.fns.usda.gov/sites/default/files/NSWP-II.pdf. Accessed December 3, 2015. 14. Aragones A , Hayes SL , Chen MH , González J , Gany FM . Characterization of the Hispanic or latino population in health research: a systematic review. J Immigr Minor Health 2014;16 (3 ):429–39. 10.1007/s10903-013-9773-0 23315046 15. Norris AE , Ford K , Bova CA . Psychometrics of a brief acculturation scale for Hispanics. Hisp J Behav Sci 1996;18 (1 ):29–38. 10.1177/07399863960181004 16. Strahan R , Gerbasi KC . Short, homogenous versions of the Marlowe-Crowne Social Desirability Scale. J Clin Psychol 1972;28 (2 ):191–3. 10.1002/1097-4679(197204)28:2<191::AID-JCLP2270280220>3.0.CO;2-G 17. Hager ER , Quigg AM , Black MM , Coleman SM , Heeren T , Rose-Jacobs R , Development and validity of a 2-item screen to identify families at risk for food insecurity. Pediatrics 2010;126 (1 ):e26–32. 10.1542/peds.2009-3146 20595453 18. Centers for Disease Control and Prevention. 2013 Behavioral Risk Factor Surveillance System questionnaire. Published December 28, 2012 http://www.cdc.gov/brfss/questionnaires/pdf-ques/2013%20BRFSS_English.pdf. Accessed March 20, 2015. 19. Moore LV . Public health surveillance of fruit and vegetable intake using the Behavioral Risk Factor Surveillance System. Published 2014 http://www.cdc.gov/brfss/data_documentation/pdf/fruits_vegetables.pdf. Accessed September 19, 2015. 20. Ivarsson A , Andersen MB , Johnson U , Lindwall M . To adjust or not to adjust: nonparametric effect sizes, confidence intervals, and real-world meaning. Psychol Sport Exerc 2013;14 (1 ):97–102. 10.1016/j.psychsport.2012.07.007 21. Leung CW , Ding EL , Catalano PJ , Villamor E , Rimm EB , Willett WC . Dietary intake and dietary quality of low-income adults in the Supplemental Nutrition Assistance Program. Am J Clin Nutr 2012;96 (5 ):977–88. 10.3945/ajcn.112.040014 23034960 22. Hromi-Fiedler A , Bermúdez-Millán A , Segura-Pérez S , Damio G , Pérez-Escamilla R . Adaptation of the US Food Security Survey Module for low-income pregnant Latinas: qualitative phase. J Hunger Environ Nutr 2009;4 (1 ):62–80. 10.1080/19320240802706841 20046909 23. Izumi BT , Zenk SN , Schulz AJ , Mentz GB , Wilson C . Associations between neighborhood availability and individual consumption of dark-green and orange vegetables among ethnically diverse adults in Detroit. J Am Diet Assoc 2011;111 (2 ):274–9. 10.1016/j.jada.2010.10.044 21272702 24. Grigsby-Toussaint DS , Zenk SN , Odoms-Young A , Ruggiero L , Moise I . Availability of commonly consumed and culturally specific fruits and vegetables in African-American and Latino neighborhoods. J Am Diet Assoc 2010;110 (5 ):746–52. 10.1016/j.jada.2010.02.008 20430136 25. Hromi-Fiedler A , Bermúdez-Millán A , Segura-Pérez S , Pérez-Escamilla R . Nutrient and food intakes differ among Latina subgroups during pregnancy. Public Health Nutr 2012;15 (2 ):341–51. 10.1017/S136898001100108X 21729472 26. Hebert JR , Hurley TG , Peterson KE , Resnicow K , Thompson FE , Yaroch AL , Social desirability trait influences on self-reported dietary measures among diverse participants in a multicenter multiple risk factor trial. J Nutr 2008;138 (1 , Suppl 1 ):226S–34S. 18156429 27. Hernandez-Garbanzo Y , Chavez-Martinez A . Food choices and healthy eating in Hispanic adults. In: Preedy VR, Hunter LA, Patel VB, editors. Diet quality: an evidenced-based approach. Vol 2. Totowa(NJ): Humana Press; 2013. p. 199–211. 28. Locke E , Coronado GD , Thompson B , Kuniyuki A . Seasonal variation in fruit and vegetable consumption in a rural agricultural community. J Am Diet Assoc 2009;109 (1 ):45–51. 10.1016/j.jada.2008.10.007 19103322

PMC005xxxxxx/PMC5003530.txt

==== Front Prev Chronic Dis Prev Chronic Dis PCD Preventing Chronic Disease 1545-1151 Centers for Disease Control and Prevention 27560721 16_0020 10.5888/pcd13.160020 Original Research Peer ReviewedThe Precarious Health of Young Mexican American Men in South Texas, Cameron County Hispanic Cohort, 2004–2015 Watt Gordon P. Vatcheva Kristina P. PhD Griffith Derek M. PhD Reininger Belinda M. DrPH Beretta Laura PhD Fallon Michael B. MD McCormick Joseph B. MD Fisher-Hoch Susan P. MD Author Affiliations: Kristina P. Vatcheva, Belinda M. Reininger, Joseph B. McCormick, Susan P. Fisher-Hoch, University of Texas School of Public Health, Brownsville Regional Campus, Brownsville, Texas; Derek M. Griffith, Institute for Research on Men’s Health, Vanderbilt University, Nashville, Tennessee; Laura Beretta, Department of Molecular and Cellular Oncology, The University of Texas, MD Anderson Cancer Center, Houston, Texas; Michael B. Fallon, Division of Gastroenterology, Hepatology, and Nutrition, The University of Texas Health Science Center at Houston Medical School, Houston, Texas. Corresponding Author: Gordon P. Watt, University of Texas School of Public Health, Brownsville Regional Campus, 1 W University Blvd, Brownsville, TX 78520. Telephone: 956-755-0628. Email: Gordon.P.Watt@uth.tmc.edu. 2016 25 8 2016 13 E113Introduction Hispanic men have higher rates of illness and death from various chronic conditions than do non-Hispanic men. We aimed to characterize the health of Mexican American men living on the US–Mexico border in South Texas and elucidate indications of chronic disease in young men. Methods We sampled all male participants from the Cameron County Hispanic Cohort, an ongoing population-based cohort of Mexican Americans in Brownsville, Texas. We calculated descriptive statistics and stratified the sample into 3 age groups to estimate the prevalence of sociodemographic, behavioral, and clinical factors by age group and evaluated differences between age groups. Results Obesity prevalence was approximately 50% across all age groups (P = .83). Diabetes prevalence was high overall (26.8%), and 16.9% (95% confidence interval [CI], 10.1%–23.8%) of men younger than 35 had diabetes. More than 70% of these young men had elevated liver enzymes, and mean values of aspartate aminotransferase were significantly higher in younger men (45.0 u/L; 95% CI, 39.5–50.6 u/L) than in both older age groups. Less than 20% of young men had any form of health insurance. Current smoking was higher in young men than in men in the other groups, and the rate was higher than the national prevalence of current smoking among Hispanic men. Conclusions We suggest a need for obesity and diabetes prevention programs and smoking cessation programs for men in this region. Opportunities exist to expand current intervention programs and tailor them to better reach this vulnerable population of young Hispanic men. Elevated liver enzymes in men younger than 35 suggest a substantial burden of liver abnormalities, a finding that warrants further study. ==== Body Introduction Hispanic men comprise a rapidly growing demographic of the United States, yet the research literature on Hispanic men’s health is sparse. Studies indicate that Hispanics in the United States face many health inequities, ranging from lack of access to care to higher rates of infectious and chronic diseases (1–4). However, few studies have addressed health inequalities in Hispanic men (3,4). Despite having a longer life expectancy than non-Hispanic white men (5), Hispanic men have higher mortality rates than do non-Hispanic men for chronic conditions such as type 2 diabetes, end-stage renal disease, colorectal cancer (6), and liver disease (7). Research suggests that a lack of access to care contributes to this increased risk of death from chronic conditions in Hispanic men (8). Hispanic men are an epidemiologically heterogeneous group. Improvements in Hispanic health surveillance data (9) and data from the Hispanic Community Health Study/Study of Latinos (HCHS/SOL) and others indicate disparate health outcomes in Hispanics by ethno-regional subgroup (3,10,11). However, to date, studies that examined the health of Hispanic ethnic subgroups in the United States neglect men’s health; of the 37 publications that emerged from the HCHS/SOL, none specifically studied men’s health (12). As an additional illustration of the dearth of health research in US Hispanic men, a 2016 MeSH term search of “men’s health” and “Hispanic Americans or Mexican Americans” on Ovid Medline yielded only 20 relevant articles since 1946. This study describes the health characteristics of Mexican American men on the southern Texas–Mexico border in Brownsville, Texas. This area is resource-poor, among the poorest cities in the United States (13). With a population of primarily Mexican descent (93.2% Hispanic, of whom 86.2% are of Mexican descent), data from this ethnically homogenous population of men are ideal for comparing with those of nationwide surveys of Mexican American men’s health (such as the National Health and Nutrition Examination Survey and HCHS/SOL). We have shown that this population has high rates of obesity, insulin resistance, diabetes, liver disease, and subclinical atherosclerosis, with significant differences by sex (14–19). Additionally, in reviewing the data from the Cameron County Hispanic Cohort (CCHC), a population-based cohort study of the region, we observed high rates of obesity in younger men (younger than 35 years). The purpose of this study is to further characterize men’s health by age group to assess the burden of chronic disease in younger men. Age is a relevant trait for risk stratification and intervention, and a precise characterization of the population is fundamental to any further work in Mexican American men. Methods We conducted a cross-sectional analysis of 945 men in the CCHC from Brownsville, Texas, recruited from 2004 through 2015. The CCHC is a representative sample of the Mexican American population in South Texas, selected by a 2-stage methodology using US Census data; socioeconomic quartiles are the sampling strata, and census blocks are the randomized sampling units. We invited all members (aged 18 years or older) of all households in the selected blocks to participate in the CCHC. Participants gave informed consent and then came to the local clinical research unit where extensive sociodemographic, clinical, and laboratory data were collected. We have detailed these protocols previously (14,15). Trained bilingual health workers conducted a clinical exam and health interview in the participants’ language of preference (English or Spanish). Participants arrived fasting for 10 hours so that blood samples could be obtained for a comprehensive metabolic panel, complete blood count with differential, and hemoglobin A1c (HbA1c) measurements. The study team measured anthropometrics and blood pressure and asked participants a range of demographic, clinical, health care use, and health behavior questions. Definitions The definition of elevated liver function tests (LFTs) was alanine transaminase (ALT) levels greater than 40 u/L and/or aspartate aminotransferase (AST) levels greater than 37 u/L. We used the 2010 American Diabetes Association diagnostic criteria for diabetes mellitus (DM) (20). A body mass index (BMI) of more than 30 kg/m2 indicated obesity. We defined hypertension as mean systolic blood pressure (SBP) at or above 130 mm Hg or a mean diastolic blood pressure (DPB) at or above 85 mm Hg, or taking antihypertensive medication. Three or more of the following constituted metabolic syndrome: hypertension, triglyceride levels over 150 mg/dL, high-density lipoprotein (HDL) cholesterol levels less than 40 mg/dL, fasting blood glucose over 100 mg/dL or taking hypoglycemic medication, or waist circumference higher than 102 cm (21). Total cholesterol of more than 200 mg/dL indicated hypercholesterolemia. A “heavy drinker” was any participant reporting drinking more than 14 alcoholic drinks per week (22). We defined history of smoking as an affirmative response to the question, “Have you ever smoked more than 100 cigarettes in your entire life?” and current smoking as an affirmative response to the question, “Do you now smoke cigarettes?” among those with a history of smoking. Statistical methods The CCHC has a 2-stage population-based sampling methodology. To address possible sampling bias, we adjusted all analyses for the probability of sampling using age- and sex-adjusted sampling weights, based on the population of Brownsville, Texas. We also accounted for the potential clustering effect among participants from the same household or census block. We considered results statistically significant at P < .05. In descriptive analyses, we calculated unweighted frequencies and weighted proportions for categorical variables. We then stratified participants into 3 age groups (younger, aged 18–34 years; middle-aged, 35–54 years; and older, aged ≥55 years) to estimate health outcomes by age group. Weighted proportions of participants in each age group were approximately equal (Rao-Scott χ2 = 1.3, P = .53). We used the Rao-Scott χ2 test to assess equality of proportions of categorical variables across age groups. For continuous variables, we used survey-weighted linear regression analysis to detect differences in mean values across age groups. For blood pressure analysis, we adjusted for current antihypertensive medication, and for analyses of triglycerides and LDL and HDL cholesterol variables, we adjusted for lipid-controlling medications. Because this population is binational, we sought to rule out confounding of place of birth on the results; using logistic and linear regression, we repeated each analysis controlling for place of birth. Where the adjustment for place of birth affected results, we have noted this effect; otherwise, crude results are presented. In the case of a significant independent overall association of age with the variable of interest, we used multiple comparisons techniques to detect significant differences in the younger age group compared with the other age groups. For categorical associations, we used logistic regression to calculate odds ratios (ORs) for the outcome of interest in the younger age group compared with the other age groups. For significant associations with continuous variables, we applied regression models to obtain t-test results with Bonferroni-adjusted P values for multiple pairwise comparisons across age groups; we reported the difference of least square means and associated 95% confidence intervals (CIs) and P values. For all analyses, we used SAS version 9.4 (SAS Institute, Inc). Results Characteristics of the study population We excluded 86 participants who were missing critical data; our final sample was 945 men from the CCHC. Mean age of participants was 44.3 years, and approximately half (50.4%) were born in Mexico (Table 1). The mean age of men born in the United States (40.4 y) was significantly lower than the mean age of men born outside the United States (48.0 y, P = .002; data not shown). Confounding by place of birth was significant for only one analysis: multiple comparisons of AST levels. The remaining results were not significantly affected when adjusting for place of birth, so crude results are presented. Most participants (63.6%) had no medical insurance (including Medicare and Medicaid), and 18.9% were unemployed (not retired) at the time of the interview. Many participants (38.5%) declined to provide income data. Approximately half of the participants had a history of smoking (54.2%), and 28.4% were current smokers. With regard to drinking habits, 7.3% reported heavy drinking, and 60.3% reported occasional alcohol consumption. We found high rates of metabolic abnormalities: 49.9% of participants were obese, 26.8% had diabetes, 44.0% met the criteria for metabolic syndrome, and 64.4% had elevated liver enzymes (Table 1). Table 1 Descriptive Statistics of Men in the Cameron County Hispanic Cohort (N = 945), 2004–2015 Participant Characteristic Valuea Mean age, y (SE) (n = 945) 44.3 (1.0) Mean years in Brownsville (SE) (n = 945) 28.2 (1.9) Mean weight, kg (SE) (n = 945) 90.3 (1.0) Mean height, cm (SE) (n = 945) 170.7 (0.3) Mean waist circumference, cm (SE) (n = 945) 105.2 (0.7) Mean waist-to-hip ratio (SE) (n = 945) 1.0 (0) Mean body mass index, kg/m2 (SE) (n = 945) 30.9 (0.3) Marital status (n = 944) Single/never married 21.6 Married 71.5 Divorced/separated 5.9 Widowed 1.1 Insuranceb (n = 944) No insurance 63.6 Insurance 36.3 Employment (n = 944) Retired 15.2 Full-time 46.9 Part-time 14.7 Unemployed 18.9 Not in workforce 4.4 Place of birth (n = 945) United States 48.0 Mexico 50.4 Other 1.6 Education level (n = 945) Completed high school 58.3 Did not complete high school 41.7 History of smokingc (n = 945) Yes 54.2 No 45.8 Current smokerd (n = 945) Yes 28.4 No 71.6 Drinking (n = 944) Never 39.7 Sometimes 60.3 Heavy drinking (drinks/wk) (n = 945) Yes (>14) 7.3 No (≤14) 92.7 Elevated liver enzymese (n = 945) No 35.6 Yes 64.4 Body mass index categories (kg/m2) (n = 945) Not obese (≤ 30) 50.1 Obese (>30) 49.9 Diabetes categoriesf (n = 945) Normal 35.9 Prediabetes 37.3 Diabetes 26.8 Hypertriglyceridemia (mg/dL) (n = 945) No (≤150) 52.5 Yes (>150) 47.5 High-density lipoprotein cholesterol (mg/dL) (n = 945) Normal (≥40) 59.8 Low (<40) 40.2 Low-density lipoprotein cholesterolg (mg/dL) (n = 945) Normal (≤160) 92.7 Elevated (>160) 7.3 Hypertensionh (n = 945) No 66.1 Yes 33.9 Metabolic syndromei (n = 945) No 56.0 Yes 44.0 Abbreviation: SE, standard error. a All statistics weighted. Percentages may not reflect the expected value due to sampling weights and design-based analyses. Values expressed as percentages, unless otherwise indicated. b “Insurance” includes both public and private coverage of any type. c Defined as an affirmative response to, “Have you ever smoked more than 100 cigarettes in your entire life?” d Defined as affirmative responses to 1) “Have you ever smoked more than 100 cigarettes in your entire life?” and 2) “Do you now smoke cigarettes?” e Defined as alanine transaminase >40 u/L and/or aspartate aminotransferase >37 u/L. f According to American Diabetes Association 2010 Diagnostic Guidelines (20). g Calculated low-density lipoprotein cholesterol levels. h Defined as systolic blood pressure ≥130 mm Hg or diastolic blood pressure ≥85 mm Hg or currently taking antihypertensive medication. i According to Adult Treatment Panel III (21). Stratified and multiple comparisons analyses Obesity prevalence was similar across age groups, at approximately 50% (P = .83) (Table 2). Mean waist circumference (P = .29) and mean BMI (P = .19) were also similar across age groups. The proportion with diabetes was significantly associated with age group (P < .001). This proportion was highest in the older age group (38.2%; 95% CI, 29.3%–47.1%), although 16.9% (95% CI, 10.1%–23.8%) of the younger age group had diabetes. Additionally, 51.3% (95% CI, 42.9%–59.8%) of men in the younger group had either prediabetes or diabetes, and this prevalence was even higher in the other 2 age groups (Figure 1). Table 2 Comparison of Variables, by Age Group, of Men in the Cameron County Hispanic Cohort (N = 945), 2004–2015a Categorical Variable Age, y P Value 18–34 35–54 ≥55 n % (95% CI) n % (95% CI) n % (95% CI) Obese BMI (>30 kg/m2) (n = 945) 129 48.9 (40.4–57.4) 207 52.4 (45.9–58.9) 114 50.2 (40.3–60.1) .83 Diabetesb (n = 933) 39 16.9 (10.1–23.8) 106 26.3 (20.8–31.8) 116 38.2 (29.3–47.1) <.001 Prediabetes or diabetesb (n = 933) 125 51.3 (42.9–59.8) 243 60.6 (54.0–67.3) 215 82.2 (75.9–88.6) <.001 Elevated LFTsc (n = 945) 218 70.2 (62.0–78.5) 282 70.4 (64.3–76.6) 127 51.2 (41.4–61.0) .001 Reduced HDLd (n = 945) 108 41.2 (32.7–49.7) 149 35.9 (29.8–41.9) 107 44.0 (34.1–53.8) .37 Elevated LDLe (n = 933) 17 5.2 (2.3–8.2) 40 9.7 (6.3–13.1) 20 7.0 (3.2–10.8) .19 Hypertriglyceridemiaf (n = 945) 104 37.5 (29.1–45.9) 226 56.9 (50.4–63.4) 137 48.1 (38.2–58.0) .005 Hypertensiong (n = 945) 49 23.8 (15.3–32.3) 97 24.2 (18.8–29.6) 144 56.0 (46.5–65.6) <.001 Metabolic syndromeh (n = 945) 81 33.7 (25.2–42.1) 171 42.5 (36.1–48.8) 155 57.3 (47.4–67.3) <.001 Smoking historyi (n = 945) 160 51.5 (43.1–59.9) 175 45.2 (38.6–51.7) 102 40.0 (29.6–50.3) .20 Current smokerj (n = 945) 95 35.4 (27.2–43.6) 120 29.5 (23.8–35.3) 57 19.4 (12.9–25.8) .006 Sometimes drink (n = 945) 159 61.3 (53.5–69.1) 278 65.0 (58.7–71.3) 146 53.8 (43.9–63.8) .15 Heavy drinking (n = 945) 35 8.2 (4.7–11.6) 15 7.5 (3.4–11.6) 17 5.7 (2.3–9.1) .61 Insurance (n = 944) 49 19.2 (11.7–26.8) 119 30.5 (24.1–36.9) 152 61.8 (53.0–70.6) <.001 Continuous Variable, Mean Value Age, y (95% Confidence Interval) P Value 18–34 35–54 ≥55 Waist circumference, cm (n = 945) 103.8 (100.8–106.8) 105.3 (103.5–107.1) 106.5 (104.7–108.3) .29 BMI, kg/m2 (n = 945) 31.2 (29.9–32.4) 31.3 (30.5–32.2) 30.2 (29.3–31.1) .19 SBP, mm Hg (n = 945) 116.0 (113.7–118.4) 116.8 (115.0–118.7) 125.2 (121.7–128.6) <.001 DBP, mm Hg (n = 945) 74.1 (72.4–75.8) 75.3 (74.0–76.7) 71.8 (69.8–73.9) .03 ALT, u/L (n = 945) 56.9 (49.5–64.3) 47.8 (45.2–50.5) 39.8 (36.5–43.0) <.001 AST, u/L (n = 945) 45.0 (39.5–50.6) 38.0 (36.0–40.0) 33.8 (30.6–37.0) .002 Triglycerides, mg/dL (n = 945) 182.5 (144.4–220.6) 230.9 (184.4–277.5) 195.9 (191.2–220.6) .01 HDL, mg/dL (n = 945) 41.7 (39.4–44.0) 42.7 (40.8–44.7) 42.1 (40.4–43.9) .71 LDL, mg/dL (n = 933) 92.2 (84.3–100.1) 108.6 (101.5–115.8) 102.2 (96.0–108.4) <.001 Abbreviations: ALT, alanine transaminase; AST, aspartate aminotransferase; BMI, body mass index; CI, confidence interval; DBP, diastolic blood pressure; HDL, high-density lipoprotein cholesterol; LDL, low-density lipoprotein cholesterol; LFTs, liver function tests; SBP, systolic blood pressure. a All statistics weighted. Percentages may not reflect the expected result due to sampling weights and design-based analyses. b According to American Diabetes Association 2010 Diagnostic Guidelines (20). c Defined as ALT >40 u/L and/or AST >37 u/L. d Defined as <40 mg/dL. e Defined as LDL >160 mg/dL. f Defined as triglyceride levels >150 mg/dL. g Defined as systolic blood pressure ≥130 mm Hg or diastolic blood pressure ≥85 mm Hg or taking antihypertensive medication. h According to Adult Treatment Panel III (21). i Defined as an affirmative response to, “Have you ever smoked more than 100 cigarettes in your entire life?” j Defined as affirmative responses to 1) “Have you ever smoked more than 100 cigarettes in your entire life?” and 2) “Do you now smoke cigarettes?” Figure 1 Proportion of male participants with diabetes, prediabetes, and prediabetes or diabetes, by age group, Cameron County Hispanic Cohort, 2004–2015. This figure shows that prevalence of prediabetes is above 30% across age groups and that more than 50% of men younger than 35 years in this population have either diabetes or prediabetes. Age Group, y Diabetes Prediabetes Prediabetes or Diabetes Prevalence, % 18–34 16.9 34.4 51.3 35–54 26.3 34.3 60.6 ≥55 38.2 44.0 82.2 The proportion of men with elevated LFTs was significantly associated with age group (P = .001). In multiple comparisons analysis, the odds of elevated LFTs was significantly lower in the older age group than in the younger age group (OR = 0.4; 95% CI, 0.3–0.8). When examining continuous values of ALT and AST, we found differences in both mean ALT (P < .001) and mean AST (P = .002) across age groups (Figure 2). In multiple comparisons (Table 3), both ALT and AST mean levels were significantly higher in the younger group than in the older group (ALT difference: 17.2 u/L; 95% CI, 7.4–27.0; P < .001; AST difference: 11.2 u/L; 95% CI, 3.7–18.7; P = .001). AST mean levels were also significantly higher in the younger group than in the middle-aged group (difference: 7.1 u/L; 95% CI, 0.02–14.1; P = .05), but this finding did not remain significant when controlling for place of birth. There was an overall association between elevated triglycerides and age group (P = .01). Multiple comparisons indicated that the odds of elevated triglycerides was significantly higher in the middle-aged group than in the younger group (OR = 2.2; 95% CI, 1.4–3.4), as were mean levels of triglycerides (difference: −48.4 mg/dL, 95% CI, −87.1 to −9.7, P = .008). We observed that the younger age group had a 37.5% prevalence of elevated triglyceride levels (95% CI, 29.1%–45.9%) (Table 2). Furthermore, 33.7% of men aged 18 to 34 years (95% CI, 25.2%–42.1%) satisfied the criteria for metabolic syndrome, compared with 57.3% of men aged 55 years or older (95% CI, 47.4%–67.3%), with a significant overall association with age group (P < .001). Figure 2 Mean levels of ALT and AST, by age group, male participants of the Cameron County Hispanic Cohort, 2004–2015. Mean levels of ALT and AST are highest in men younger than 35 and lower in older age groups. The upper limit of normal for ALT is 40 u/L, and the upper limit of normal for AST is 37 u/L (indicated by horizontal lines on graph). Abbreviations: ALT, alanine transaminase; AST, aspartate aminotransferase. Age Group, y Mean Level, u/L ALT AST 18–34 56.9 45.0 35–54 47.8 38.0 ≥55 39.8 33.8 Table 3 Multiple Comparison Analysis of Variables, by Age Group, of Men in the Cameron County Hispanic Cohort (N = 945), 2004–2015 Categorical Variable Odds Ratioa Diabetesb 18–34 1 [Reference] 35–54 1.8 (1.0 to 3.1) ≥55 3.0 (1.7 to 5.6) Prediabetes or diabetes 18–34 1 [Reference] 35–54 1.5 (0.9 to 2.3) ≥55 4.4 (2.6 to 7.4) Elevated liver function testsc 18–34 1 [Reference] 35–54 1.0 (0.6 to 1.6) ≥55 0.4 (0.3 to 0.8) Hypertriglyceridemiad 18–34 1 [Reference] 35–54 2.2 (1.4 to 3.4) ≥55 1.5 (0.9 to 2.6) Hypertensione 18–34 1 [Reference] 35–54 1.0 (0.6 to 1.8) ≥55 4.1 (2.2 to 7.7) Metabolic syndromef 18–34 1 [Reference] 35–54 1.5 (0.9 to 2.3) ≥55 2.6 (4.5 to 4.6) Health insurance 18–34 1 [Reference] 35–54 1.8 (1.1 to 3.2) ≥55 6.8 (3.8 to 12.3) Current smokerg 18–34 1 [Reference] 35–54 0.8 (0.5 to 1.2) ≥55 0.4 (0.3 to 0.8) Continuous Variable Difference (95% Confidence Interval) P Valueh Systolic blood pressure, mmHg 18–34 Reference 35–54 −0.8 (−4.5 to 2.9) .99 ≥55 −9.1 (−14.2 to −4.0) <.001 Diastolic blood pressure, mmHg 18–34 Reference 35–54 −1.2 (−3.9 to 1.5) .82 ≥55 2.3 (−1.0 to 5.5) .28 Triglycerides, mg/dL 18–34 Reference 35–54 −48.4 (−87.1 to −9.7) .008 ≥55 −13.4 (−58.4 to 31.6) .99 Low-density lipoprotein cholesterol, mg/dL 18–34 Reference 35–54 −16.4 (−25.7 to −7.2) <.001 ≥55 −10.0 (−20.1 to 0.1) .05 Alanine transaminase levels, u/L 18–34 Reference 35–54 9.1 (−0.6 to 18.9) .08 ≥55 17.2 (7.4 to 27.0) <.001 Aspartate aminotransferase levels, u/L 18–34 Reference 35–54 7.1 (0.02 to 14.1)i .05i ≥55 11.2 (3.7 to 18.7) .001 a Survey-weighted odds ratio generated from logistic regression. b According to American Diabetes Association 2010 Diagnostic Guidelines (20). c Defined as alanine transaminase >40 u/L and/or aspartate aminotransferase >37 u/L. d Defined as triglyceride levels >150 mg/dL. e Defined as systolic blood pressure ≥130 mm Hg or diastolic blood pressure ≥85 mm Hg or taking antihypertensive medication. f According to Adult Treatment Panel III (21). g Defined as affirmative responses to 1) “Have you ever smoked more than 100 cigarettes in your entire life” and 2) “Do you now smoke cigarettes?” h Bonferroni adjusted P values for multiple pairwise comparisons. i Nonsignificant after controlling for place of birth. The proportion of men with health insurance was highest in the older age group (61.8%; 95% CI, 53.0%–70.6%), and lowest in the younger age group (19.2%; 95% CI, 11.7%–26.8%), with an overall significant association (P < .001) between age group and insurance (Table 2). Having health insurance was also significantly associated with being born in the United States (OR = 2.1; 95% CI, 1.4–3.2; data not shown), but both age group and place of birth remained independently associated with insurance status in logistic regression (data not shown). Rates of current smoking were highest in the younger age group (35.4%; 95% CI, 27.2%–43.6%) and lowest in the older age group (19.4%; 95% CI, 12.9%–25.8%), with an overall significant association with age (P = .006). Multiple comparisons indicated that current smoking prevalence was significantly higher in the younger group than in the older group (OR [older vs younger] = 0.4; 95% CI, 0.2–0.7) but similar to the middle group. Overall history of smoking (past or present) and drinking behavior had no significant associations with age group. Discussion This article is among the few population-based studies of the health of Mexican American men, and it allows for evidence-based risk stratification by age in the Mexican American population. Before this study, little was known about the health needs of Mexican American men in the southern Texas–Mexico border region. Our results show strikingly adverse metabolic and behavioral outcomes in men younger than 35 years. Poor metabolic health (eg, dyslipidemia, elevated blood pressure, obesity, and prediabetes) appears to extend into the 35 to 54 years age group. Variables that tend to be associated with older age — such as hypertension, diabetes, and metabolic syndrome — were also associated with older age in this population. Mean BMI and mean waist circumference were uniform but high across all age groups. Although mounting evidence suggests that metabolic health, as opposed to obesity, is a more important indicator of cardiovascular risk (16,23,24), obesity is an important predictor of several health outcomes. In this population, obesity begins in adolescence (25) and persists through middle and older age. A relevant finding from this study was the significant burden of obesity, prediabetes, and diabetes in the younger age group. Outcomes such as obesity tend to peak during middle age in men (26), but we found a 48.9% prevalence of obesity in younger men. For comparison, the HCHS/SOL study found a 36.8% prevalence of obesity in 2,337 Mexican American men in their nationwide survey (10); our findings suggest a higher burden of obesity in Mexican American men residing on the Texas–Mexico border than in Mexican American men nationwide. We found a 34.3% prevalence of prediabetes and 16.9% prevalence of diabetes in the younger group. Although high rates of diabetes have been documented in this population (15,27), our findings indicate a high prevalence of diabetes in young men, which has not been widely addressed in the literature. Nationwide, the overall prevalence of diabetes among Mexican American men is 18.7% (28), so the prevalence of diabetes in men younger than 35 in this population is nearly as high as the nationwide average for all ages. Employing widely used cut-offs for elevated ALT and AST, we found a 70.2% prevalence of elevated LFTs in the younger group, which was similar to the prevalence for the middle-aged group and significantly higher than that for the older group. Research indicates that ALT levels may decrease slightly with age (29), but our findings were nonetheless remarkable. Given the documented high rates of nonalcoholic fatty liver disease in this cohort (30), the authors believe that this association is real and that the drivers of elevated liver enzymes in young Mexican American men warrant further study. The younger group fares worse than the older groups in 2 other important measures: lack of health insurance and high rates of current smoking. We found that less than 20% of men younger than 35 had any health care coverage. Limited resources exist for uninsured, and especially undocumented, men to obtain affordable health care, so preventive care may not be sought by this young population. Among men in the younger age group, 51.5% had a history of smoking and 35.4% identified as current smokers. The percentage of young men with any history of smoking was higher (though not significantly) than the percentage of older men with a history of smoking (51.5% vs 40.0%). This finding suggests that smoking initiation across generations is consistent. Both the rate of current smoking among younger men in the CCHC (35.4%) and rate of current smoking among all men in the CCHC (28.4%) appear to be higher than the overall rate of current smoking in US Hispanic men (17.3%) (31). Although poor health outcomes were not restricted to the younger group, the findings in this group provoke the greatest concern from a prevention perspective. The high prevalence of poor metabolic health outcomes in Mexican Americans is now well documented (14,17,26), but the high prevalence of poor metabolic health outcomes in young men has not yet been adequately studied. Given that more than 60% of the overall male population in this region lacks health insurance and that men in general are less likely to exhibit health-seeking behavior than women (4), these data support the need for aggressive chronic disease intervention programs for young Mexican American men. There were several limitations to this study. The data we used were cross-sectional and do not provide insight into changes in health over time or causality. Only longitudinal data will allow us to determine whether the differences in the age groups are cohort effects or whether we may see premature death in metabolically unhealthy young men, or both. Additionally, many participants declined to provide income information. Despite these limitations, we contribute to the characterization of Mexican American men’s health, affirming the importance of stratified analyses of health among Hispanic men of distinct ages and ethno-regional subgroups. There are active obesity and diabetes prevention programs in South Texas, but we suspect that men are missed by these efforts. For example, the Coordinated Approach to Child Health (CATCH) targets obesity in children and families, and “Salud y Vida” aims to prevent diabetes in people with prediabetes. However, less than 30% of participants in “Salud y Vida” are men (M. Zolezzi, University of Texas School of Public Health, written communication, 2015). Men rarely attend free exercise classes offered by the University of Texas School of Public Health (A. Davé, University of Texas School of Public Health, written communication, 2016). The literature corroborates these findings, suggesting that interventions should be consciously tailored toward men (32). Additionally, we suspect that vastly different approaches are needed for each age group. We believe our findings will contribute to a re-evaluation of intervention programs in the region and shape new interventions targeting men at highest risk. Our data also suggest that men in South Texas would benefit from culturally appropriate smoking cessation programs. Lower smoking rates would consequently reduce overall cardiovascular risk. By prioritizing younger men for primary prevention, we can reduce poor health behaviors (eg, poor diet, sedentary lifestyle, tobacco use) that begin in early adulthood or even before and mitigate the burden of more severe disease (eg, cardiovascular disease, diabetes, cancer) later in life. Acknowledgments We thank our cohort team, particularly Rocío Uribe and her team, who recruited and documented the participants. We also thank Pablo Sánchez and Israel Hernández for data management, Marcela Morris and other laboratory staff for their contributions, and Christina Villarreal and Norma Pérez-Olazarán for administrative support. We thank the Valley Baptist Medical Center in Brownsville for housing our Clinical Research Unit. This work was supported by the Center for Clinical and Translational Sciences, which is funded by National Institutes of Health Clinical and Translational Award no. UL1 TR000371 from the National Center for Advancing Translational Sciences. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Center for Advancing Translational Sciences or the National Institutes of Health. This research was approved by the Committee for the Protection of Human Subjects of the University of Texas Health Science Center at Houston as HSC-SPH-03-007-B. The opinions expressed by authors contributing to this journal do not necessarily reflect the opinions of the U.S. Department of Health and Human Services, the Public Health Service, the Centers for Disease Control and Prevention, or the authors' affiliated institutions. Suggested citation for this article: Watt GP, Vatcheva KP, Griffith DM, Reininger BM, Beretta L, Fallon MB, et al. The Precarious Health of Young Mexican American Men in South Texas, Cameron County Hispanic Cohort, 2004–2015. Prev Chronic Dis 2016;13:160020. DOI: http://dx.doi.org/10.5888/pcd13.160020. ==== Refs References 1. Centers for Disease Control and Prevention, Division of Diabetes Translation. National diabetes statistics report, 2014 p. 12. http://www.cdc.gov/diabetes/pubs/statsreport14/national-diabetes-report-web.pdf. Accessed March 9, 2015. 2. Henao-Martínez AF , Castillo-Mancilla JR . The Hispanic HIV epidemic. Curr Infect Dis Rep 2013;15 (1 ):46–51. 10.1007/s11908-012-0306-0 23184612 3. Ramirez AG , Suarez L , Chalela P , Talavera GA , Marti J , Trapido EJ , Cancer risk factors among men of diverse Hispanic or Latino origins. Prev Med 2004;39 (2 ):263–9. 10.1016/j.ypmed.2004.03.034 15226034 4. MacNaughton NS . Health disparities and health-seeking behavior among Latino men: a review of the literature. J Transcult Nurs 2008;19 (1 ):83–91. 10.1177/1043659607309144 18165429 5. Miniño AM . Death in the United States, 2011. National Center for Health Statistics data brief, report no. 115; 2013 p. 8. http://www.cdc.gov/nchs/data/databriefs/db115.pdf. Accessed October 8, 2015. 6. Enewold L , Horner M-J , Shriver CD , Zhu K . Socioeconomic disparities in colorectal cancer mortality in the United States, 1990-2007. J Community Health 2014;39 (4 ):760–6. 10.1007/s10900-014-9824-z 24477390 7. Smedley BD , Stith AY , Nelson AR . Unequal treatment: confronting racial and ethnic disparities in healthcare. Washington (DC): National Academies Press; 2003. 8. Aguirre-Molina M , Pond A . Latino access to primary and preventive health services: barriers, needs, and policy implications. New York (NY): Columbia University; 2003. 9. Carter-Pokras P , Fischer A . Improvements in Latino health data. In: Aguirre-Molina M, Borrel LN, Vega W, editors. Health issues in Latino males: a social and structural approach. New Brunswick (NJ): Rutgers University Press; 2010. p. 53–64. 10. Daviglus ML , Talavera GA , Avilés-Santa ML , Allison M , Cai J , Criqui MH , Prevalence of major cardiovascular risk factors and cardiovascular diseases among Hispanic/Latino individuals of diverse backgrounds in the United States. JAMA 2012;308 (17 ):1775–84. 10.1001/jama.2012.14517 23117778 11. Kallwitz ER , Daviglus ML , Allison MA , Emory KT , Zhao L , Kuniholm MH , Prevalence of suspected nonalcoholic fatty liver disease in Hispanic/Latino individuals differs by heritage. Clin Gastroenterol Hepatol 2015;13 (3 ):569–76. 10.1016/j.cgh.2014.08.037 25218670 12. University of North Carolina at Chapel Hill. Hispanic Community Health Study/Study of Latinos; 2015 https://www2.cscc.unc.edu/hchs/view/biblio/year. Accessed July 11, 2016. 13. Clark S . Census Bureau: Brownsville poorest city in U.S. Brownsville (TX): Brownsville Herald; 2013 November 7. http://www.brownsvilleherald.com/news/local/article_b630f374-475c-11e3-a86e-001a4bcf6878.html. Accessed July 11, 2016. 14. Fisher-Hoch SP , Rentfro AR , Salinas JJ , Pérez A , Brown HS , Reininger BM , Socioeconomic status and prevalence of obesity and diabetes in a Mexican American community, Cameron County, Texas, 2004-2007. Prev Chronic Dis 2010;7 (3 ):A53. 20394692 15. Fisher-Hoch SP , Vatcheva KP , Laing ST , Hossain MM , Rahbar MH , Hanis CL , Missed opportunities for diagnosis and treatment of diabetes, hypertension, and hypercholesterolemia in a Mexican American population, Cameron County Hispanic Cohort, 2003-2008. Prev Chronic Dis 2012;9 :110298. 22863308 16. Laing ST , Smulevitz B , Vatcheva KP , Rahbar MH , Reininger B , McPherson DD , Subclinical atherosclerosis and obesity phenotypes among Mexican Americans. J Am Heart Assoc 2015;4 (3 ):e001540. 10.1161/JAHA.114.001540 25787312 17. Pan J-J , Qu H-Q , Rentfro A , McCormick JB , Fisher-Hoch SP , Fallon MB . Prevalence of metabolic syndrome and risks of abnormal serum alanine aminotransferase in Hispanics: a population-based study. PLoS One 2011;6 (6 ):e21515. 10.1371/journal.pone.0021515 21720553 18. Perez A , Anzaldua M , McCormick J , Fisher-Hoch S . High frequency of chronic end-stage liver disease and hepatocellular carcinoma in a Hispanic population. J Gastroenterol Hepatol 2004;19 (3 ):289–95. 10.1111/j.1440-1746.2003.03277.x 14748876 19. Salinas J , McCormick JB , Rentfro A , Hanis C , Hossain MM , Fisher-Hoch SP . The missing men: high risk of disease in men of Mexican origin. Am J Men Health 2011;5 (4 ):332–40. 10.1177/1557988310379390 20930218 20. American Diabetes Association. Diagnosis and classification of diabetes mellitus. Diabetes Care 2010;33 (Suppl 1 ):S62–9. 10.2337/dc10-S062 20042775 21. Grundy SM , Becker D , Clark LT , Cooper RS , Denke MA , Howard WJ , Third report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III). p. 284. Report no. 02-5215; 2002 http://www.nhlbi.nih.gov/files/docs/resources/heart/atp3full.pdf. 22. National Institutes of Health, National Institute on Alcohol Abuse and Alcoholism. What’s at-risk or heavy drinking? http://rethinkingdrinking.niaaa.nih.gov/How-much-is-too-much/Is-your-drinking-pattern-risky/Whats-At-Risk-Or-Heavy-Drinking.aspx. Accessed April 18, 2016. 23. Wildman RP , McGinn AP , Lin J , Wang D , Muntner P , Cohen HW , Cardiovascular disease risk of abdominal obesity vs. metabolic abnormalities. Obesity (Silver Spring) 2011;19 (4 ):853–60. 10.1038/oby.2010.168 20725064 24. Wu S , Fisher-Hoch SP , Reininger BP , Vatcheva KP , McCormick JB . Metabolic health has greater impact on diabetes than simple overweight/obese in Mexican-Americans. J Diabetes Res 2016;2015 :2016. 25. Rentfro AR , Nino JC , Pones RM , Innis-Whitehouse W , Barroso CS , Rahbar MH , Adiposity, biological markers of disease, and insulin resistance in Mexican American adolescents, 2004-2005. Prev Chronic Dis 2011;8 (2 ):A40. 21324254 26. Kaplan RC , Avilés-Santa ML , Parrinello CM , Hanna DB , Jung M , Castañeda SF , Body mass index, sex, and cardiovascular disease risk factors among Hispanic/Latino adults: Hispanic community health study/study of Latinos. J Am Heart Assoc 2014;3 (4 ):e000923. 10.1161/JAHA.114.000923 25008353 27. Brown HS 3d , Wilson KJ , Pagán JA , Arcari CM , Martinez M , Smith K , Cost-effectiveness analysis of a community health worker intervention for low-income Hispanic adults with diabetes. Prev Chronic Dis 2012;9 :120074. 10.5888/pcd9.120074 22916995 28. Schneiderman N , Llabre M , Cowie CC , Barnhart J , Carnethon M , Gallo LC , Prevalence of diabetes among Hispanics/Latinos from diverse backgrounds: the Hispanic Community Health Study/Study of Latinos (HCHS/SOL). Diabetes Care 2014;37 (8 ):2233–9. 10.2337/dc13-2939 25061138 29. Goh GB-B , Pagadala MR , Dasarathy J , Unalp-Arida A , Pai RK , Yerian L , Age impacts ability of aspartate-alanine aminotransferase ratio to predict advanced fibrosis in nonalcoholic Fatty liver disease. Dig Dis Sci 2015;60 (6 ):1825–31. 10.1007/s10620-015-3529-8 25708897 30. Pan J-J , Fisher-Hoch SP , Chen C , Feldstein AE , McCormick JB , Rahbar MH , Burden of nonalcoholic fatty liver disease and advanced fibrosis in a Texas Hispanic community cohort. World J Hepatol 2015;7 (11 ):1586–94. 10.4254/wjh.v7.i11.1586 26085918 31. Jamal A , Agaku IT , O’Connor E , King BA , Kenemer JB , Neff L . Current cigarette smoking among adults—United States, 2005-2013. MMWR Morb Mortal Wkly Rep 2014;63 (47 ):1108–12. 25426653 32. Robertson LM , Douglas F , Ludbrook A , Reid G , van Teijlingen E . What works with men? A systematic review of health promoting interventions targeting men. BMC Health Serv Res 2008;8 (1 ):141. 10.1186/1472-6963-8-141 18598339

PMC005xxxxxx/PMC5003531.txt

==== Front Prev Chronic Dis Prev Chronic Dis PCD Preventing Chronic Disease 1545-1151 Centers for Disease Control and Prevention 27560724 15_0593 10.5888/pcd13.150593 Brief Peer Reviewed“You have the right to protect your health”: Perceptions of Secondhand Smoke and Exposure Mitigation Strategies in Low-Income Patients With Heart Disease, San Francisco, 2011–2012 Brown-Johnson Cati G. PhD Oppezzo Marily PhD MS RD Benowitz Neal L. MD Prochaska Judith J. PhD MPH Author Affiliations: Cati G. Brown-Johnson, Stanford Prevention Research Center, and Evaluation Sciences Unit, Department of Medicine, Stanford University, Stanford, California; Marily Oppezzo, Stanford Prevention Research Center, Department of Medicine, Stanford University, Stanford, California; Neal L. Benowitz, Departments of Medicine and Bioengineering and Therapeutic Sciences, Division of Clinical Pharmacology and Experimental Therapeutics, University of California, San Francisco, San Francisco, California. Corresponding Author: Judith J. Prochaska, PhD, MPH, Stanford Prevention Research Center, Department of Medicine, Stanford University School of Medicine, 1265 Welch Rd, Stanford, California 94305. Telephone: 650-724-3608. Email: jpro@stanford.edu. 2016 25 8 2016 13 E116We examined the understanding of the harms of secondhand smoke (SHS) exposure among low-income, hospitalized adults with cardiovascular disease. Participants were 15 nonsmokers reporting daily SHS exposure and 15 light or nondaily cigarette smokers. We coded responses from audiotaped semistructured interviews for themes. No participant spontaneously identified heart risks related to SHS exposure. Strategies to avoid SHS included verbal requests to not smoke and physically avoiding smoke; both smokers and nonsmokers prioritized politeness over urgency. Most participants thought a blood test quantifying SHS exposure would be clinically useful. Health education, assertiveness communication training, and protective policies (eg, smoke-free multiunit housing) also were supported. ==== Body Objective Secondhand smoke (SHS) accounts for 41,000 US deaths annually, more than 80% of which result from cardiovascular disease (CVD) (1). Although the prevalence of daily smoking has declined, any level of tobacco smoke exposure brings serious health consequences (1). With CVD, heavy SHS and light or nondaily smoking have negative and comparable levels of harm (2). However, medical systems rarely assess or provide interventions for intermittent tobacco use or SHS exposure (3). Few studies have examined interventions to reduce SHS exposure in adult CVD patients. One quasi-experimental study in middle-income nonsmokers with CVD reported increased awareness of SHS risk; the study did not report change in behavior or biomarkers of exposure (4). Our randomized study with nondaily smokers found that messages focused on SHS harms to others led to greater cotinine-confirmed abstinence compared with the traditional emphasis of tobacco’s harms to self (5). Tobacco use and SHS exposure are associated with poverty (6,7). To inform innovations to address tobacco-related health disparities, we interviewed uninsured patients with CVD recruited from a public hospital. The sample consisted of light (<5 cigarettes/d) and nondaily (<7 d/wk) cigarette smokers and nonsmokers exposed to SHS. Methods The study was conducted from April 2011 through May 2012 on the cardiology service at San Francisco General Hospital, a large, urban public hospital that serves an ethnically diverse and low-income population. We recruited 15 nonsmokers who reported daily SHS exposure before hospitalization and 15 light/nondaily cigarette smokers. Institutional review boards approved study procedures, and participants provided informed consent. Study procedures were performed in-hospital, averaged 1.5 hours, and included a half-hour structured interview on SHS risk perceptions and strategies to reduce SHS exposure. When available, we analyzed blood samples obtained at hospital admission for cotinine. We transcribed and analyzed the audiotaped interviews using a general inductive approach (8). The first 2 authors (C. G. B., M. O.) independently coded structured queries with good agreement (Cohen’s κ = 0.76). Emergent themes captured strategies to reduce SHS exposure. Two research assistants independently coded the transcripts with moderate to high interrater reliability (Cohen’s κ = 0.66–0.75). The first author resolved coding conflicts. Results Participant demographics did not differ by smoking status. The sample (N = 30) was primarily male (n = 25) and racially/ethnically diverse (African American, n = 13; Asian/Pacific Islander, n = 4; white, n = 7; multiracial/other, n = 6). Before hospitalization, most patients were unhoused (n = 8) or residing in hotels or single-room occupancy units (n = 9). Most nonsmokers were former smokers (n = 12). Most smokers (n = 12) wanted to quit; only 4 were prepared to quit within the month. Blood samples were available for 9 participants; cotinine values averaged 8.66 ng/ml (standard deviation, 11.58 ng/ml; range, 0.33–34.88 ng/ml). Most understood SHS to be harmful (Table 1); however, without prompting, not one participant identified adverse effects on the heart. After prompting, 19 linked SHS to heart outcomes, although comprehension of SHS effects on the heart was neither uniform nor always accurate. Although 8 participants considered SHS equal to or worse than smoking (27%), 3 said it was less harmful. Many smokers (n = 6) reported concern about the effects of SHS on others. All nonsmoking patients and 13 smokers (87%) thought use of a blood test to quantify recent SHS exposure would increase risk awareness, motivate self-care, and help light/nondaily smokers “cut back.” Table 1 Secondhand Smoke (SHS) Perceptions of Nonsmokers and Light/Nondaily Cigarette Smokers, San Francisco, 2011–2012 Characteristic Nonsmoker (n = 15) Light/Nondaily Smokera (n = 15) No. (%) Perceptions of SHS Spontaneously identified health risksb 13 (93) 12 (80) General illness 6 (43) 6 (40) Lung disease (emphysema, asthma, bronchitis) 4 (29) 5 (33) Cancer 4 (29) 4 (27) Death 2 (14) 2 (13) With prompting, acknowledged SHS heart risksb 9 (64) 10 (67) Nonhealth-related perceptions of SHS Nuisance 12 (80) 9 (60) Bad smell 4 (27) 0 a Light smokers smoked fewer than 5 cigarettes per day and nondaily smokers smoked cigarettes weekly but not every day. Participants gave multiple responses in a semi-structured interview tapping their perceptions of the risks of SHS exposure. Less commonly identified health risks (n<4) were throat problems (n = 2), nausea (n = 2), bad for brain/difficult to think (n = 2), toxic blood levels (n = 1), irritability (n = 1), eye irritation (n = 1), and addiction (n = 1). No participants spontaneously mentioned adverse effects of SHS exposure on the heart. b Reponses to questions about health risks provided by 14 of 15 nonsmokers. When asked how to reduce SHS exposure, participants’ suggestions focused on communication, information, physical avoidance, policy strategies, and biomarker feedback (Table 2). Both smokers and nonsmokers emphasized politeness and respect; a minority of nonsmokers suggested more forceful verbal and physical actions. Some smokers reported that a simple “please” prompted them to move elsewhere, extinguish their cigarette, or not smoke the entire day. A “please” coupled with harsh language was still perceived as respectful. Other strategies for avoiding SHS included giving personal health reasons when asking others not to smoke (n = 4) and invoking the presence of children (n = 3). One participant stated that children trumped the need for politeness. Table 2 Reported Strategies to Avoid Secondhand Smoke Exposure Among Nonsmokers and Light/Nondaily Cigarette Smokers, San Francisco, 2011–2012a Strategiesb Nonsmokers (n = 15) Light/Nondaily Smokers (n = 15) Sample Quotes No. (%) Communication approachesc 15 (100) 11 (73) — Conversational (with friends/family) 12 (80) 9 (60) Say, “Hey, well you’re married to me, and you’re part of my life and you are killing me [with SHS].” NS4 Polite/sincere 13 (87) 7 (47) They said, “Please don’t smoke near me” . . . Yeah, politely. “Please put that cigarette out.” LNDS2 I said, “I can’t tell you what to do, but can you move over away from me?” NS8 Demanding/forceful 7 (47) 6 (40) “You ain’t smoking in here.” LNDS6 “Put that f-ing cigarette out.” NS3 Threats of violence 2 (13) 0 “Threaten them [smokers].” NS7 Informationd 9 (60) 6 (40) — Invoke smoke-free rule 5 (33) 5 (33) “Look there’s no smoking in the house” . . . that was understood. It’s an unwritten rule. LNDS5 Invoke children 2 (13) 1 (7) Try to talk with reason: “You can’t smoke ’cause my son’s got asthma.” NS4 Moms can be direct . . . [they] don’t need to be kind. LNDS8 State personal reasons 2 (13) 2 (13) I just went straight and said, “Well I’m very sick, please don’t smoke near me.” LNDS7 Educate on SHS risks 3 (20) 1 (7) I just told him, “You know, even secondhand smoking kills.” LNS13 Physical strategiese 7 (47) 1 (7) — Better ventilation or filtration 2 (7) 0 Open a window. NS5 [A personal air filter] . . . handy pack on the shoulder strap to breathe oxygen. NS11 Move away from SHS 6 (40) 0 If I see somebody smoking, I try to walk far away from them. NS4 Avoid smokers 2 (13) 1 (7) Don’t hang around people that smoke. LNDS14 Policyf 4 (27) 3 (27) — Smoke-free multiunit housing 2 (13) 0 If I had the money, then I would build a building where smoking isn’t allowed — period. NS9 Smoke-free signage and designated smoking areas 2 (13) 3 (20) . . . put signs telling you where to smoke and where not to smoke. LNDS6 Smoke exposure blood test 15 (100) 13 (87) I’ll know how much I’ve been exposed to [with a blood test], so I will try to . . . quit using cigarettes. LNDS2 Abbreviations: — , not applicable; LNDS, light nondaily smoker; NS, nonsmoker; SHS, secondhand smoke. a Percentages may exceed 100% within an area, because participants could list more than 1 strategy. b Using a general inductive approach, we determined categories for strategies by describing all strategies, and exploring how those descriptions were similar or different for a final reduction to 4 thematic categories: communication, information strategy, physical strategy, and policy. c Communication approaches focused on how and in what way a person might communicate with other people to best avoid SHS exposure. All communication approaches were verbal, including threats of violence. d Information strategies focused on what information was being communicated instead of how communication happened. e Physical approaches focused on strategies to directly manipulate the physical world, including objects or bodies. f Policy strategies included reference to rules and policies that might affect more than one person. Strategies emphasizing participants’ physical agency (ie, moving away from SHS) were less common, suggested by 7 nonsmokers and 1 light smoker. SHS exposure in single-room occupancy hotel-like settings was common, regardless of participants’ personal home smoking rules; smoke-free multiunit housing policies were encouraged (n = 2), and one participant dreamed of building smoke-free apartments. Discussion The study findings indicate SHS knowledge gaps in a low-income sample of CVD patients. Although most participants in our study identified SHS exposure as a nuisance and harmful to health, not one participant spontaneously listed heart disease as an SHS risk. There was greater awareness that SHS causes lung disease and some cancers. Although these are also serious health concerns, SHS effects on the heart are more immediate, acute, and relevant to study participants’ hospitalization on a cardiology service (9). Both nonsmokers and light/nondaily smokers reported motivation to avoid SHS and use of similar techniques to protect themselves from tobacco smoke. Most thought a blood test quantifying SHS exposure would motivate assertive communications and avoidance behaviors, provide useful data to support protective policies (eg, smoke-free multiunit housing), and raise motivation to quit among light/nondaily smokers. A biomarker test would align with calls to integrate behavioral data for personalized medical care (10), and prompt intervention with light/nondaily smokers, often overlooked in cessation counseling (3). In practice, availability of SHS blood tests is influenced by cost and treatment prioritization. Participants residing in low-income multiunit housing reported low perceived control over building-level clean-air policies. A 2015 review concluded the evidence is sufficient to support multiunit housing smoke-free policies on a broad scale (11). SHS interventions should advise home smoking bans and refer to local action networks that support clean-air policies. The study sample was limited in size, geographic region, language (English-speaking), and socioeconomic status. The findings, however, are novel and informative for future treatment efforts. Patients with CVD may lack critical information connecting their light/nondaily smoking and SHS exposure to immediate heart risks. Furthermore, low-income patients may experience SHS exposure as a result of environmental and residential factors. Extended clean-air policies in public (eg, worksites) and private (eg, multiunit housing) environments are needed to protect communities at all income strata. Individualized interventions should address the immediate risk of SHS exposure for low-income patients with CVD. Acknowledgments We thank Romina Kim and Kathleen Gali for conducting the patient interviews and Amy Chieng and Samantha Lok-Yung Wong for coding the transcripts. This research was supported by the Flight Attendant Medical Research Institute, Miami, Florida (William Cahan Distinguished Professor Award to William Grossman, MD) and the National Heart, Lung, and Blood Institute (nos. T32 HL007034-39 and R01 HL117736), Bethesda, Maryland, and the State of California Tobacco-Related Disease Research Program (TRDRP no. 21BT-0018), Oakland, California. The opinions expressed by authors contributing to this journal do not necessarily reflect the opinions of the U.S. Department of Health and Human Services, the Public Health Service, the Centers for Disease Control and Prevention, or the authors' affiliated institutions. Suggested citation for this article: Brown-Johnson CG, Oppezzo M, Benowitz NL, Prochaska JJ. “You have the right to protect your health”: Perceptions of Secondhand Smoke and Exposure Mitigation Strategies in Low-Income Patients With Heart Disease, San Francisco, 2011–2012. Prev Chronic Dis 2016;13:150593. DOI: http://dx.doi.org/10.5888/pcd13.150593. ==== Refs References 1. US Department of Health and Human Services. The health consequences of smoking — 50 years of progress. A report of the Surgeon General. Atlanta (GA): US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 2014. 2. Whincup PH , Gilg JA , Emberson JR , Jarvis MJ , Feyerabend C , Bryant A , Passive smoking and risk of coronary heart disease and stroke: prospective study with cotinine measurement. BMJ 2004;329 (7459 ):200–5. 10.1136/bmj.38146.427188.55 15229131 3. Danesh D , Paskett ED , Ferketich AK . Disparities in receipt of advice to quit smoking from health care providers: 2010 National Health Interview Survey. Prev Chronic Dis 2014;11 :140053. 10.5888/pcd11.140053 25078568 4. Rigotti NA , Park ER , Streck J , Chang Y , Reyen M , McKool K , An intervention to address secondhand tobacco smoke exposure among nonsmokers hospitalized with coronary heart disease. Am J Cardiol 2014;114 (7 ):1040–5. 10.1016/j.amjcard.2014.07.017 25124185 5. Schane RE , Prochaska JJ , Glantz SA . Counseling nondaily smokers about secondhand smoke as a cessation message: a pilot randomized trial. Nicotine Tob Res 2013;15 (2 ):334–42. 10.1093/ntr/nts126 22592447 6. Jamal A , Homa DM , O’Connor E , Babb SD , Caraballo RS , Singh T , Current cigarette smoking among adults — United States, 2005–2014. MMWR Morb Mortal Wkly Rep 2015;64 (44 ):1233–40. 10.15585/mmwr.mm6444a2 26562061 7. Homa DM , Neff LJ , King BA , Caraballo RS , Bunnell RE , Babb SD , ; Centers for Disease Control and Prevention (CDC). Vital signs: disparities in nonsmokers’ exposure to secondhand smoke — United States, 1999–2012. MMWR Morb Mortal Wkly Rep 2015;64 (4 ):103–8. 25654612 8. Thomas DR . A general inductive approach for analyzing qualitative evaluation data. Am J Eval 2006;1;27(2):237-46. 9. US Department of Health and Human Services. The health consequences of involuntary exposure to tobacco smoke: a report of the Surgeon General. Atlanta (GA): US Department of Health and Human Services, Centers for Disease Control and Prevention, Coordinating Center for Health Promotion, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 2006. 10. Collins FS , Varmus H . A new initiative on precision medicine. N Engl J Med 2015;372 (9 ):793–5. 10.1056/NEJMp1500523 25635347 11. Snyder K , Vick JH , King BA . Smoke-free multiunit housing: a review of the scientific literature. Tob Control 2016;25 (1 ):9–20. 10.1136/tobaccocontrol-2014-051849 25566811

PMC005xxxxxx/PMC5003533.txt

==== Front NPJ Breast CancerNPJ Breast CancerNPJ Breast Cancer2374-4677Nature Publishing Group npjbcancer20161610.1038/npjbcancer.2016.1627583302ArticleA microchip platform for structural oncology applications Structural oncology applicationsWinton Carly E 12Gilmore Brian L 1Demmert Andrew C 13Karageorge Vasilea 1Sheng Zhi 13Kelly Deborah F 1234*1 Virginia Tech Carilion Research Institute, Roanoke, VA, USA2 School of Biomedical Engineering and Science, Virginia Tech, Blacksburg, VA, USA3 Virginia Tech Carilion School of Medicine, Roanoke, VA, USA4 Department of Biological Sciences, Virginia Tech, Blacksburg, VA, USA* (debkelly@vt.edu)C.E.W., B.L.G., Z.S., and D.F.K. conceived and designed the experiments. C.E.W., B.L.G., and A.C.D. performed the experiments. C.E.W., A.C.D., and D.F.K. performed the image processing procedures and molecular modeling. All authors contributed to the written manuscript and have given approval to the final version of the manuscript. 15 06 2016 2016 2 16016 24 08 2015 26 02 2016 05 05 2016 Copyright © 2016 Breast Cancer Research Foundation/Macmillan Publishers Limited2016Breast Cancer Research Foundation/Macmillan Publishers LimitedThis work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/Recent advances in the development of functional materials offer new tools to dissect human health and disease mechanisms. The use of tunable surfaces is especially appealing as substrates can be tailored to fit applications involving specific cell types or tissues. Here we use tunable materials to facilitate the three-dimensional (3D) analysis of BRCA1 gene regulatory complexes derived from human cancer cells. We employed a recently developed microchip platform to isolate BRCA1 protein assemblies natively formed in breast cancer cells with and without BRCA1 mutations. The captured assemblies proved amenable to cryo-electron microscopy (EM) imaging and downstream computational analysis. Resulting 3D structures reveal the manner in which wild-type BRCA1 engages the RNA polymerase II (RNAP II) core complex that contained K63-linked ubiquitin moieties—a putative signal for DNA repair. Importantly, we also determined that molecular assemblies harboring the BRCA15382insC mutation exhibited altered protein interactions and ubiquitination patterns compared to wild-type complexes. Overall, our analyses proved optimal for developing new structural oncology applications involving patient-derived cancer cells, while expanding our knowledge of BRCA1’s role in gene regulatory events. ==== Body Introduction Mutations in the breast cancer susceptibility protein (BRCA1) are known to contribute to cancer induction.1,2 At the molecular level, the intricate details of these events are poorly understood. During normal cellular activities, BRCA1 interacts with its binding partner, BARD1 (BRCA1-associated ring domain protein), to ensure genomic stability and cell survival.3 In this context, BRCA1 functions as a tumor suppressor by safeguarding genetic material.4–6 A critical opportunity to monitor for errors in DNA, and to correct them, occurs during RNA synthesis. The BRCA1–BARD1 heterodimer has an important role in this process as BRCA1-related repair proteins are found in proximity to exposed DNA during transcription.7,8 However, the precise manner in which BRCA1 works in concert with RNA polymerase II (RNAP II) is ill-defined. Currently, there is little structural information available for BRCA1 protein assemblies, despite their well-known contribution to human disease. This lack of information is due to many factors including: (1) the size of the BRCA1 protein (~208 kDa) makes it difficult to express recombinantly; (2) the inherent flexibility of full-length BRCA1 renders it problematic to crystallize; and (3) few strategies are available to isolate BRCA1 protein assemblies from human tumor cells for structural analysis. The size and flexibility of BRCA1 are intrinsic properties of the protein that shape its biological activity, and are thus not easy to modify in patient-derived cell lines. As an alternative strategy we chose to develop new tools to investigate protein complexes naturally formed in human breast cancer cells. Specifically, we have recently reported the production of the tunable microchip system, which enabled the first structural analysis of BRCA1 protein assemblies.9 As part of our work to establish the microchip system, we determined a likely scenario to explain how BRCA1 associates with the RNAP II core complex. We resolved the position of the BRCA1 C-terminal domain (BRCT) with respect to the RNAP II core, and distinguished the level of structural variability present in the biological samples. Information that was missing from these initial analyses, however, included a more detailed understanding of the BRCA1 N-terminal (RING) domain, and the manner in which ubiquitin patterns affect protein–protein interactions. Here we present biochemical and structural results that expand upon these initial findings and reveal new molecular insights for BRCA1 protein architectures. These results show the proximity of the BRCA1 RING domain in relation to DNA fragments that were bound to transcriptional assemblies. We also define regions on the RNAP II core that accommodate K63-linked ubiquitin moieties, which are known signals for DNA repair mechanisms. Equally importantly, we now illustrate that the 3D structures of wild-type and mutated BRCA1 assemblies vary considerably. Taken together, our technical advances provide a new molecular framework to study gene regulatory assemblies with and without cancer-related mutations. As such, we refer to this exciting new opportunity as ‘structural oncology.’ Results Capturing BRCA1 complexes from breast cancer cells for structural analysis We recently established a streamlined approach to isolate native BRCA1 assemblies from the nuclear contents of primary ductal carcinoma cells (HCC70 line).9 Here we employed the same strategy to examine new molecular interfaces of wild-type assemblies, and to compare how these interfaces differ among mutated complexes (summarized in Figure 1). Briefly, RNAP II, BRCA1, and BARD1 contained in the nuclear material of HCC70 cells were naturally enriched and co-eluted from Nickel–Nitrilotriacetic acid (Ni–NTA) agarose beads. In the eluted fractions we found that wild-type BRCA1 associated with BARD1 and the RNAP II large subunit (RPB1) as determined by co-immunoprecipitation (co-IP) experiments. In addition, the RNAP II complexes were post-translationally phosphorylated at pSer5/pSer2 peptide repeats, and ubiquitinated with K63-type linkages (Supplementary Figure S1). After verifying these biochemical associations, we used the microchip system to examine the molecular arrangements of the proteins that constituted the BRCA1 assemblies. 3D structures of wild-type BRCA1 assemblies reveal new molecular interfaces To gain structural insights of BRCA1–RNAP II interactions, we applied aliquots of the eluted fractions to Cryo-SiN microchips10 decorated with antibodies against either the structured BRCA1 N- or C-terminal domains. This step selected for BRCA1-associated RNAP II complexes and excluded those complexes not bound to BRCA1, as previously described.9 Tethered protein assemblies were then plunge-frozen into liquid ethane for cryo-EM imaging and downstream analysis (see Materials and Methods section for a full description of imaging procedures). We employed computational procedures11,12 to separately determine the positions of where the BRCA1 structural domains interacted with the RNAP II assemblies, based on antibody-labeling results. The extra major densities found in the experimentally determined EM density map were attributed to either the BRCA1–BARD1 N-terminal RING domains or the C-terminal BRCT domain (Figure 2). The orientation of the BRCT domain (pdbcode, 1JNX)13 was previously resolved and found to be proximal to the C-terminal region of the RNAP II core.9 This observation is consistent with other biochemical findings.7,14 Antibody-labeling results also indicated the unoccupied density in the EM map located near the DNA channel was attributed to the BRCA1–BARD1 N-terminal RING domain. This information now permits us to place the structure of the RING domain into the density map, which fit uniquely within the 3D envelope (pdbcode, 1JM7)6 (Figure 2, magenta and green). We further assigned a minor density over the DNA channel to a short strand of DNA (Figure 2, blue). This position of the DNA strand is in the same location described in other models of RNAP II engaging DNA in a ‘closed state’ during the initial stages of transcription.15,16 Additional modeling experiments guided the placement of the K63-linked ubiquitin moieties (pdbcode, 1UBQ)17 (Figure 2; orange and red). The current position of the ubiquitin moieties did not introduce atomic clashes. Other minor differences between the fit crystal structures and the experimentally determined 22-Å density map (0.5 FSC criteria; Supplementary Figures S2 and S3; Supplementary Movie S1) may be attributed to missing loops in the yeast RNAP II crystal structure (pdbcode, 4A93),18 and the fact that the complexes in our investigation were derived from human tumor cells, rather than from yeast. Similar differences were also noted in a previously determined EM structure derived from other immortalized human cell lines.19 Moreover, as the central region of BRCA1 is highly flexible, it is reasonable that we cannot fully visualize this region of the protein in the reconstruction. Collectively, these structural results indicated that phosphorylated RNAP II core complexes (1) interacted with BRCA1 N- and C-terminal domains, (2) contained K63-linked ubiquitin moieties, and (3) are likely primed for DNA repair. These findings were in good agreement with our biochemical assessments. BRCA1 directly engages DNA and the RNAP II core To test whether transient intermediate states were present in our samples, we utilized the RELION software package.12 Statistical output generated by RELION identified multiple structures were represented in the original image stack. The number of 3D classes determined by RELION was independent of user-defined starting parameters, and resulting 3D classes showed subtle variations in density (Figures 3a–c, black arrows). Comparing the composite EM map to the structures having the lowest and the highest DNA density, we noted potential differences in DNA engagement (Figure 3b,c; Supplementary Figures S4 and S5; Supplementary Movie S2 and S3). As these structures were determined from native assemblies, gently removed from the nuclear material, we posit that the observed heterogeneity may be due to differences in functional states. As such, we interpreted the low occupancy structure to represent a weakly bound DNA state, and the high occupancy structure to represent a strongly bound DNA state. Other important differences noted in the high occupancy structure included greater density for the K63-linked ubiquitins and for the BRCT domain (Figure 3; red, orange, and gray). These findings are consistent with BRCA1 complexes engaging DNA through a series of concerted steps that may be linked to DNA repair. Similar observations have been described in functional studies.8,14,20,21 The BRCT domain prefers specific phospho-peptide sequences In addition to the variability noted near the DNA-binding site, we also found differences in the region containing the BRCT domain. As previously noted, the BRCT domain is adjacent to the C-terminus of the RNAP II core (Figure 4a). This region of RNAP II emanates from residue P1455 but is disordered in the crystal structure.18 As this disordered region is highly mobile, it can conceivably interact with the BRCT domain. To provide a conceptual framework for this interaction, we examined the substrate peptide pSPTF (Figure 4b) that was co-crystalized with the BRCT domain (pdbcode, 3K0H).22 Phosphorylated peptides are highly repetitive in the RNAP II C-terminus, and include the pSer5 (pSPSY) and pSer2 (pSPTS) consensus sequences. Atomic models for the pSer5 and pSer2 peptides have also been independently crystallized (pdbcode, 4H3K).23 In the course of the present study, we performed molecular modeling experiments to test for optimal BRCT–peptide interactions. We overlaid the model for the pSer5 peptide onto the substrate peptide that was co-crystalized within the BRCT domain. The pSer5 peptide model fit within the BRCT binding cleft that is defined by residues S1655, L1701, L1705, and M1775 (Figure 4b,c; Supplementary Movie S4). However, the pSer5 peptide contained a unique tyrosine residue in the consensus sequence compared with the analogously positioned phenylalanine residue. Although mutagenesis studies have reported that phenylalanine is the preferred residue that fits within the BRCT domain,24,25 our modeling results suggest that the heptad repeats in the RNAP II C-terminal domain may also have binding potential for this region. In particular, the pSer5 peptide more closely matches the stereochemistry requirements of the BRCT binding cleft than the pSer2 peptide (Figure 4d), which contains a terminal serine residue and is not likely to fit. These interactions are important to further investigate as many mutations in the BRCT are implicated in cancer induction. Therefore, we examined BRCA1 assemblies in cancer cell lines having a naturally mutated BRCT domain and known deficiencies in transcriptional activities. Differences exist between the wild type and mutant BRCA1 protein complexes To examine the molecular architecture of mutated BRCA1-transcriptional assemblies, we implemented the same biochemical and structural approaches described for the wild-type complexes. We probed the nuclear material of breast cancer cells (HCC1937 line) that harbors a homozygous BRCA1 mutation (BRCA15382insC). This mutation in the BRCA1 gene is associated with deficiencies in transcription-coupled repair events and high incidences of breast cancer.1,2 We hypothesized that the functional differences observed in cells containing mutated BRCA15382insC may be related to its ability to form proper protein assemblies. To test this idea, we used the tunable microchip platform to isolate transcriptional assemblies containing BRCA15382insC. We collected transmission electron microscopic (TEM) images of the mutated protein assemblies under the same low-dose conditions used for wild-type complexes. Individual assemblies were selected from the images using the PARTICLE program,11 and exported to the RELION software package.12 Implementing standard reconstruction procedures in RELION, we calculated the first 3D structure of mutated BRCA15382insC transcriptional assemblies. The EM density map of the mutated BRCA1 complex was interpreted by drawing from information made available from the wild-type structure. First, we positioned the RNAP II core (pdbcode, 4A93)18 and BRCA1–BARD1 RING domain (pdbcode, 1JM7)6 into the density map (Figure 5a; Supplementary Figures S6 and S7, Supplementary Movie S5). Upon comparing the mutant and wild-type structures, there was a notable difference in orientation of the BRCA1–BARD1 RING (Figure 5a, green and magenta) domain relative to the RNAP II core (Figure 5a, yellow). Another major difference we noted was that the full-length BRCT domain did not fit well in the map of the mutated complex. We reasoned that the limited density seen in the region of the mutated BRCT was due to the frameshift mutation that imparts a stop codon, resulting in a protein truncation. Other studies have shown that truncations in the BRCT domain render it susceptible to proteolysis, whereas the full-length protein is highly resistant to cleavage.26 The fact that the mutated complexes were tethered to the microchips by polyclonal antibodies against the BRCT, indicate that some portion of this domain is intact and properly folded. Therefore, we placed a homology model of the truncated BRCA15382insC domain (Figure 5a, aqua) into the density available in this region of the EM map. The homology model of the mutated BRCT domain fit well within the given density. Proximal to the BRCA15382insC homology model, there was a small region of unoccupied density that accommodated a single ubiquitin (pdbcode, 1UBQ)23 (Figure 5a, red). A similarly sized unoccupied density was protruding from the RNAP II core complex, near the mutated BRCA1–BARD1 RING domain. This density also accommodated a single ubiquitin (pdbcode, 1UBQ)23 (Figure 5a, orange). To compliment our structural studies, and shed light on the nature of the ubiquitin moieties present in the mutated complexes, we performed co-IP experiments. Similar to the wild-type assemblies, BRCA15382insC interacted with phosphorylated RNAP II (Figure 5b). Also, the RNAP II large subunit contained K63-linked ubiquitin moieties, which is consistent with our structural findings. In contrast to wild-type BRCA1, we found that BRCA15382insC was the likely target of multiple K48-ubiqutin linkages as indicated by the smeared band present in western blots of the BRCA1 co-IPs (Figure 5c). This information suggested the extra density adjacent to the mutated BRCT domain was a potential linkage site for K48-specific ubiquitin moieties. Additional ubiquitin chains attached to BRCA1 may be flexible and hence not visible in our density map. These modifications suggested that the signal for DNA repair on the RNAP II core complex was conserved in the mutated assemblies, but that BRCA15382insC had acquired modifications to direct its degradation by the proteasome. The BRCA15382insC mutation alters protein interactions with BARD1 Biochemical experiments have shown that the BRCA15382insC mutation weakens native protein interactions in the nucleus.27 One important nuclear interaction affected by this mutation is the heterodimer formed by the BRCA1 and BARD1 RING domains. The results presented here suggest that BRCA1 may contain K48-linked ubiquitin groups proximal to the BRCA15382insC mutation. This region in the BRCA1 protein is located distal to the BARD1-binding site. How then, does a mutation in the BRCT domain affect protein–protein interactions that are primarily at the N-terminus of BRCA1? Similar to BRCA1, BARD1 contains BRCT motifs at its C-terminus. These motifs in BARD1 are also known to bind to phosphorylated peptide substrates having a pS-X-X-pF consensus sequence.27 On the basis of this information, we predicted that the BRCT of BARD1 interacts with the central domain of BRCA1, known as the serine-containing domain (SCD). The SCD region of BRCA1 contains multiple sites for ubiqutination and phosphorylation. Therefore, reduced interactions between mutated BRCA1 and BARD1 may be influenced by the addition of ubiquitin moieties in the SCD region of BRCA1. To test these ideas biochemically, we probed the enzymatic accessibility of the SCD region of wild type and mutated BRCA1 contained in the nuclear fractions of breast cancer cell lines. We used phosphatase assays to assess the extent by which wild type and mutant BRCA1 can be dephosphorylated. The same total protein concentration of nuclear material (10 μg) was incubated with either lamba phosphatase (40 μl, 16,000 units; New England Biolabs) or buffer solution lacking the enzyme as a negative control. The mixtures were then analyzed by SDS-polyacrylamide gel electrophoresis (PAGE) and western blot analysis. Western blots of the phosphatase digests and control samples were probed with antibodies against BRCA1 (AB1; Millipore). We observed a bandshift in the nuclear extracts that contained wild-type BRCA1 and lamba phosphastase (Figure 5c, top right). Control mixtures lacking the enzyme did not show this bandshift. By comparison, we observed no bandshifts in the treated and untreated nuclear material containing BRCA15382insC (Figure 5c, bottom right). These results suggested that some of the phosphorylation sites in the BRCA15382insC protein were inaccessible to phosphatase cleavage. This information also supports the idea that protein misfolding in the mutated BRCA15382insC can lead to ubiquitination in the SCD, and possibly hinder the proper binding interactions with BARD1. Discussion Here we present the first 3D comparison of wild type and mutated BRCA1 protein assemblies derived from human breast cancer cells. Employing the recently developed tunable microchip system, we could enrich for and selectively isolate BRCA1 nuclear assemblies while still maintaining native protein–protein interactions. We found that wild type and mutated BRCA15382insC interacted directly with the RNAP II core, which was modified with K63-type ubiquitin moieties. As this modification to the RNAP II core is a known signal for DNA repair,20 the structures of the BRCA1 complexes presented here are likely primed for this function. Differences between the wild type and mutated assemblies included altered positioning of the RING domains in the density maps, DNA-binding capacity, ubiquitination patterns, and biochemical interactions with BARD1. These results also complement previous biochemical studies that demonstrate how other forms of ubiqutination can lead to the degradation of transcriptional assemblies.28 On the basis of our new molecular insights, we found that ubiquitination has an important role in protein complex formation during the early stages of transcription. Ongoing investigations aimed at understanding the structural complexity of BRCA1 assemblies during various stages of RNA synthesis will help to delineate the functional relevance of these interactions. In a broader sense, our combined structural and biochemical approaches provide a unique opportunity to study native protein interactions related to both normal and diseased processes. On a technical front, the widespread use of commercially available protein adapters can enhance future microchip applications toward a variety of disease conditions. As such, our tunable approach may help shed light on the inner-workings of native proteins in a unique way that has not been fully explored in human cancer research. Materials and methods Nuclear extraction and fractionation procedures HCC70 and HCC1937 lines of human breast cancer cells (ATCC) were grown until near confluence in a 5% CO2 environment at 37 °C in RPMI-1640 medium (Mediatech, Manassas, VA, USA) supplemented with 10% fetal bovine serum (Fisher Scientific, Hanover Park, IL, USA). The cells were collected into pellets by detaching them using trypsin-EDTA (Life Technologies, Carlsbad, CA, USA), quickly centrifuging (500 g, 5 min) them and then washing them with phosphate-buffered saline (PBS). The cells were lysed using the NE-PER extraction kit (Thermo Scientific, Miami, OK, USA) and the nuclear contents were collected. After dilution to ~1 mg/ml in HEPES buffer (20 mmol/l HEPES, 2 mmol/l MgCl2, pH 7.2) supplemented with 5 mmol/l imidazole and protease inhibitor cocktail (EDTA-free, Roche, Branchburg, NJ, USA), the extracts were incubated with pre-equilibrated Ni–NTA) agarose beads (Qiagen, Hilden, Germany) to enrich for phosphorylated RNAP II. After 1-h incubation at 4 °C on the clinical rotator, the solution was added to a column and the flow-through was collected and reserved for later analysis. The column was washed three times with HEPES buffer supplemented with 140 mmol/l NaCl and 5 mmol/l imidazole. The addition of HEPES buffer with NaCl supplemented with 150 mmol/l imidiazole resulted in the elution of proteins. The Bradford Assay (Thermo Scientific) was used to estimate all protein concentrations. Phosphatase assays For phosphatase digestion, two tubes were prepared for each cell line (HCC70 & HCC1937) with one tube serving as the experimental tube and the other serving as the control sample. Approximately 10 μg of protein was loaded into each tube. The following buffers were added to the tubes (experimental & control): 10 μl of 10× MnCl2 (New England Biolabs, Ipswich, MA, USA), 10 μl of 10× PMP (New England Biolabs), 1 μl of 100× protease inhibitor cocktail (EDTA-free, Roche). Lambda phosphatase (40 μl; 16,000 units) was added to the experimental tube only while additional buffer solution was added to the control tube. Each tube was incubated in a water bath at 37 °C for 90 min. After the incubation, aliquots of the samples (3.25 μg of total protein) were prepared for western blot analysis. We used 3–8% NuPAGE Bis–Tris mini gels with Tris-Acetate running buffer to perform SDS-PAGE analysis. The separated proteins were transferred onto an Immobolin-P membrane (Millipore, Billerica, MA, USA) in a Mini-PROTEAN Tetra system (Bio-Rad, Hercules, CA, USA). Blocking solution (1% non-fat dry milk (NFDM)) was added to the blot for 1 h with gentle rocking. The blots were incubated overnight at 4 °C with primary antibody solution against the BRCA1 RING domain (Millipore, AB1, MS110) diluted to 0.001 mg/ml in 1% NFDM. After 3 washes with TBS-T solution (standard TBS containing 0.05% Tween-20 (Fisher, Hanover Park, IL, USA)), the blots were incubated for 1 h with HPRO-conjugated anti-mouse secondary antibody (Jackson ImmunoResearch, West Grove, PA, USA). Following the incubation step, the membranes were again washed 3 times with TBS-T. ECL Prime Western blotting reagent (GE Healthcare, Marlborough, MA, USA) was applied to the blot for detection using a ChemiDoc MP (Bio-Rad) for imaging purposes. Co-IP experiments The eluates from the Ni–NTA agarose beads were supplemented with protease inhibitor and phosphatase inhibitor cocktail (Thermo Scientific). Antibody (5 μg) diluted in PBS-T (0.02% Tween-20, Fisher) was combined with 0.75 mg Dynabeads Protein G (Life Technologies) and incubated with rotation for 30 min at 4 °C. Antibodies used in the immunoprecipitation were POLR2C (Abcam, Cambridge, MA, USA, ab138436), BRCA1 (Santa Cruz Biotechnology, Dallas, TX, USA (SCBT) sc-642, C-20), BARD1 (SCBT sc-11438, H-300), RPL3 (Abcam ab83098) and normal mouse IgG (SCBT sc-2025). After the antibody-coated beads were washed with HEPES buffer, the eluates were added and immunoprecipitated overnight with gentle rotation at 4 °C. HEPES buffer was used to wash the beads (three times) and the proteins were eluted with NuPAGE LDS sample buffer. A 4–12% NuPAGE Bis–Tris mini gel with MOPS running buffer was used to separate the proteins. Following separation, the proteins were transferred onto an Immobilon-P membrane (Millipore) in a Mini-PROTEAN Tetra system (Bio-Rad). Blocking solution (1% NFDM or 4% bovine serum albumin (SCBT)) was added to blots with gentle rocking for 1 h. TBS-T was applied 3 consecutive times after incubation with blocking buffer to wash the blots. The blots were incubated with primary antibody, diluted in 1% NFDM or bovine serum albumin solution, overnight at 4 °C. Other antibodies used were RNAP II (SCBT sc-9001, H-224), RNA Polymerase II H5 and H14 (Covance, Raleigh, NC, USA, MMS-129 and MMS-134), Polyubiquitin (K63-linkage-specific, Enzo, Farmingdale, NY, USA, BML-PW0600) and ubiquitin (K48-linkage-specific, Abcam ab140601). After three washes with TBS-T (0.05% Tween-20), either goat anti-rabbit or goat anti-mouse secondary antibodies conjugated to horseradish peroxide (Jackson ImmunoResearch) were added to the blots and incubated for 1 h. A ChemiDoc MP (Bio-Rad) was used for imaging and ECL Prime western blotting reagent (GE Healthcare) for detection. Preparation of tunable microchip samples Functionalized Ni–NTA (Avanti Polar Lipids, Alabaster, AL, USA) lipid monolayers were formed over 15-μl aliquots of Milli-Q water on parafilm and incubated for 1 h in a sealed petri dish. Negatively stained specimens required 5% Ni–NTA and cryo-EM specimens required 25% Ni–NTA. C-flat grids with 2-μm holes and 1-μm of spacer (2/1) between holes (Protochips, Morrisville, NC, USA) or Cryo-SiN microchips (TEMwindows, West Henrietta, NY, USA) were placed on the surface of each monolayer. Each grid or microchip was removed from the monolayer surface and incubated for 1 min with aliquots (3-μl) of His-tagged Protein A (0.01 mg/ml) (Abcam) in buffer solution containing 50 mM HEPES (pH 7.5), 150 mmol/l NaCl, 10 mmol/l MgCl2, and 10 mmol/l CaCl2. The protein-A coated chips were blotted to remove excess solution and 3-μl aliquots of IgG antibodies (0.01 mg/ml), in the same buffer as protein A, were added. Antibodies against the BRCA1 C-terminus (SCBT, sc-642, C-20) and the RING domain (Millipore, MS110, AB1) were employed for the BRCA1 labeling experiments. A Hamilton syringe was utilized to remove the antibody solution from the grid surface after a 1-min incubation. The protein fractions collected during the Ni–NTA chromatography step described above were incubated with the enhanced chips (Cryo-SiN)10 for 2 min. Following a wash with Milli-Q water, the grids were stained with 0.2% uranyl formate or plunge-frozen into liquid ethane using a Cryoplunge 3 device equipped with GentleBlot technology (Gatan, Pleasanton, CA, USA). Electron microscopy and image analysis The BRCA1-associated transcriptional complexes were viewed with a FEI BioTwin Transmission Electron Microscope (FEI Company) equipped with a LaB6 filament at 120 kV under low-dose conditions (~5 electrons per Å2) for both negatively stained and cryo-EM samples. A FEI Eagle 2 k HS CCD camera (FEI Company) recorded the images with a pixel size of 30-μm at a nominal magnification of 50,000× (for wild-type complexes) and 68,000× (for mutant complexes), for a final sampling of 6 Å per pixel and 4.4 Å per pixel, respectively. The same TEM conditions were used to collect images of negatively stained and ice-embedded specimens except for varying the defocus range. Roughly 22,000 particles of wild-type complexes bound to DNA were selected from EM images by the automated program, PARTICLE11 and the selected particle images were exported to the RELION software package.12 The RELION software package was used to refine and reconstruct the individual complexes employing an initial model for the RNAP II core complex (pdbcode, 4A93)18 low-pass filtered to a resolution of 80 Å. The initial model was only used in the first round of the refinement to assign initial orientation parameters to each particle. Later iterations were heavily dependent on the experimental data to refine the assigned angles by setting the regularization parameter to T=4. We followed standard reconstruction routines and employed a pixel size of 6 Å to produce a final composite 3D structure masked at ~250 Å with a resolution filtered to 2.2 nm. The RELION software package identified variable structures present in our image stack. Five distinct structures were output by the RELION software package independent of the user-defined starting parameters based on Bayesian statistical comparisons computed between the initial model and the experimental particle images. The degrees of DNA occupancy varied among the five structures, each of which contained ~4,400 particles and we highlighted density maps having the lowest and the highest DNA occupancy. For comparison, we selected ~3,000 particles of mutant BRCA15382insC complexes, and performed the same reconstruction routines, but using a pixel size of 4.4 Å, as images were acquired at a nominal magnification of ×68,000. RELION identified one major class during refinement. Molecular modeling The pSer5 motif found within the C-terminus of RNAP II was previously reported (pdbcode, 4H3K23). Using the Chimera software package,29 we found that the peptide fit within the binding cleft of the BRCT. The Chimera program established the most optimum fit by employing an energy minimization strategy. The energy minimization technique implemented algorithms to calculate the amount of force generated by different atom arrangements while assessing the atomic positions requiring the least amount of force. Chimera calculated the energy minimization method to find a local minimum without crossing energy barriers or searching for global minimums. This step is achieved by following an iterative optimization procedure where the force is calculated on each atom. Atoms are then moved by a computed step predicted to decrease the force. This process was iterated until the measured force falls below a set threshold. This strategy is useful in biochemical studies as the arrangement that generates the least amount of force correlates to the most likely arrangement present in nature. This research is supported by development funds from Virginia Tech, the Commonwealth Health Research Board (2080914), the Concern Foundation (303872), and NIH/NCI (R01CA193578) to D.F.K. C.W. is funded through the ICTAS Doctoral Scholar’s program at Virginia Tech and the Medical Research Scholar’s program at the Virginia Tech Carilion Research Institute. The authors declare no conflict of interest. Supplementary Information Click here for additional data file. Supplementary Movie S1 Click here for additional data file. Supplementary Movie S2 Click here for additional data file. Supplementary Movie S3 Click here for additional data file. Supplementary Movie S4 Click here for additional data file. Supplementary Movie S5 Click here for additional data file. Figure 1 The tunable microchip system captures native proteins produced in breast cancer cells. Native BRCA1 protein assemblies formed in the nucleus of hereditary breast cancer cells were tethered to tunable SiN-based microchips for 3D structural analysis. Representative 3D reconstructions (white and cyan) show variations in structural features and molecular domains as described in the present work. Figure 2 EM structure reveals BRCA1 domains directly engage the RNAP II core proximal to DNA in human breast cancer cells. The EM density map (white) is shown in different orientations. The position of the BRCT domain13 was recently determined based on antibody-labeling results.9 In the present study, the BRCA1–BARD1 RING domain6 was uniquely placed into the density map. The DNA strand (blue) was positioned over the DNA channel accordingly.15 K63-linked ubiquitins17 occupied the remaining density. The RNAP II core was localized in the EM map based on a model of the RNAP II X-ray crystal structure.18 Bar = 5 nm. Cross-sections through the density map (1–4) indicate the overall fit of the atomic models within the envelope. Also see Supplementary Figure S3 and Supplementary Movie S1. BRCT, BRCA1 C-terminal domain; BARD1, BRCA1-associated ring domain protein; EM, electron microscopy; RNAP II, RNA polymerase II. Figure 3 BRCA1 engages DNA in a variable manner while bound to the RNAP II core. (a) The composite EM structure was compared to transient intermediate structures having low and high DNA occupancies. (b) An intermediate structure having low DNA occupancy (yellow density map) accommodates a short fragment of DNA (blue) located proximal (black arrows) to the BRCA1–BARD1 RING domains.6 Limited density was present in the density map to accommodate K63-linked ubiquitins.17 (c) An intermediate structure having high DNA occupancy (cyan density map) accommodates a longer strand of DNA (blue) located proximal (black arrows) to the BRCA1–BARD1 RING domains. Bar=5 nm. Also see Supplementary Movie S2 and S3. BARD1, BRCA1-associated ring domain protein; EM, electron microscopy. Figure 4 The pSer5 peptide exhibits the optimal stereochemistry to interact with the BRCT domain. (a) The composite 3D structure highlighting the BRCT domain (gray) within the density map. (b) A close-up view of the BRCT crystal structure (pdbcode, 3K0H 22) showing that the hydrophobic binding pocket (gray rectangle) accommodates a known peptide, pSPTF. Molecular modeling experiments were performed to overlay the pSer5 and pSer2 peptides onto the pSPTF model. (c) The pSer5 peptide contains a terminal tyrosine residue (blue dashed circle) that fits within the hydrophobic binding cleft. Also see Supplementary Movie S4. (d) The pSer2 peptide contains a terminal serine residue (blue dashed circle) that does not maintain the proper stereochemistry to optimally fit within the BRCT binding site. BRCT, BRCA1 C-terminal domain. Figure 5 EM structure of mutated BRCA15382insC transcriptional complexes. (a) The EM density map (cyan) shown in different orientations was calculated using the RELION software package. Placement of the BRCA1–BARD1 (magenta, green) RING domains6 and the BRCT (aqua)13 varied compared with the wild-type structure. Models for ubiquitin modifications (red and orange)17 occupied the remaining minor density. RNAP II (yellow) was localized in the EM map based on a model of the RNAP II X-ray crystal structure.18 Bar = 5 nm. Additional cross-sections through the density map (1–4) indicate the fit of the atomic models within the envelope. Also see Supplementary Figure S7 and Supplementary Movie S5. (b) Western blot analysis of co-IP experiments showed the RNAP II core was phosphorylated at pSer5 and pSer2 peptide repeats, while interacting with mutated BRCA15382insC. (c) The RNAP II core contained K63-linked ubiquitin moieties, while K48-type linkages are likely present on BRCA15382insC. Wild-type BRCA1 shows a bandshift upon digestion with lambda phosphatase (+Ppase) in comparison with control samples lacking the enzyme (-Ppase). Mutated BRCA15382insC does not show a change in migration upon incubation with lambda phosphastase. * denotes protein interactions. DEP, unbound material; EM, electron microscopy; IB, immunoblot; IN, input material; IP, immunoprecipitated interaction; RNAP II, RNA polymerase II. ==== Refs Le Page, F. et al. BRCA1 and BRCA2 are necessary for the transcription-coupled repair of the oxidative 8-oxoguanine lesion in human cells . Cancer Res. 60 , 5548 –5552 (2000 ).11034101 King, M. C., Marks, J. H. & Mandell, J. B. Breast and ovarian cancer risks due to inherited mutations in BRCA1 and BRCA2 . Science 302 , 643 –646 (2003 ).14576434 Caestecker, K. W. & Van de Walle, G. R. The role of BRCA1 in DNA double-strand repair: past and present . Exp. Cell Res. 319 , 575 –587 (2013 ).23200932 Friedman, L. S. et al. The search for BRCA1 . Cancer Res. 54 , 6374 –6382 (1994 ).7987831 Wu, L. C. et al. Identification of a RING protein that can interact in vivo with the BRCA1 gene product . Nat. Genet. 14 , 430 –440 (1996 ).8944023 Brzovic, P. S., Rajagopal, P., Hoyt, D. W., King, M. C. & Klevit, R. E. Structure of a BRCA1-BARD1 heterodimeric RING-RING complex . Nat. Struct. Biol. 8 , 833 –837 (2001 ).11573085 Krum, S. A., Miranda, G. A., Lin, C. & Lane, T. F. BRCA1 associates with processive RNA polymerase II . J. Biol. Chem. 278 , 52012 –52020 (2003 ).14506230 Lane, T. F. BRCA1 and transcription . Cancer Biol. Ther. 3 , 528 –533 (2004 ).15254397 Gilmore, B. L. et al. A molecular toolkit to visualize native protein assemblies in the context of human disease . Sci. Rep. 5 , 14440 (2015 ).26395823 Tanner, J. R. et al. Cryo-SiN—an alternative substrate to visualize active viral assemblies . J. Anal. Mol. Tech. 1 , 1 –6 (2013 ). Chen, Z. et al. PARTICLE Available at http://image-analysis.net/EM (2012 ); Accessed on March, 2014. Scheres, S. H. A Bayesian view on cryo-EM structure determination . J. Mol. Biol. 415 , 406 –418 (2012 ).22100448 Williams, R. S., Green, R. & Glover, J. N. Crystal structure of the BRCT repeat region from the breast cancer-associated protein BRCA1 . Nat. Struct. Biol. 8 , 838 –842 (2001 ).11573086 Haile, D. T. & Parvin, J. D. Activation of transcription in vitro by the BRCA1 carboxyl-terminal domain . J. Biol. Chem. 274 , 2113 –2117 (1999 ).9890972 Kostrewa, D. et al. RNA polymerase II-TFIIB structure and mechanism of transcription initiation . Nature 462 , 323 –330 (2009 ).19820686 Grunberg, S., Warfield, L. & Hahn, S. Architecture of the RNA polymerase II preinitiation complex and mechanism of ATP-dependent promoter opening . Nat. Struct. Mol. Biol. 19 , 788 –796 (2012 ).22751016 Vijay-Kumar, S., Bugg, C. E. & Cook, W. J. Structure of ubiquitin refined at 1.8A resolution . J. Mol. Biol. 194 , 531 –544 (1987 ).3041007 Walmacq, C. et al. Mechanism of translesion transcription by RNA polymerase II and its role in cellular resistance to DNA damage . Mol. Cell 46 , 18 –29 (2012 ).22405652 Kelly, D. F., Dukovski, D. & Walz, T. Strategy for the use of affinity grids to prepare non-His-tagged macromolecular complexes for single-particle electron microscopy . J. Mol. Biol. 400 , 675 –681 (2010 ).20562026 Lee, K. B. & Sharp, P. A. Transcription-dependent polyubiquitination of RNA polymerase II requires lysine 63 of ubiquitin . Biochemistry 43 , 15223 –15229 (2004 ).15568815 Starita, L. M. et al. BRCA1/BARD1 ubiquitinate phosphorylated RNA polymerase II . J. Biol. Chem. 280 , 24498 –24505 (2005 ).15886201 Campbell, S. J., Edwards, R. A. & Glover, J. N. Comparison of the structures and peptide binding specificities of the BRCT domains of MDC1 and BRCA1 . Structure 18 , 167 –176 (2010 ).20159462 Xiang, K., Manley, J. L. & Tong, L. An unexpected binding mode for a Pol II CTD peptide phosphorylated at Ser7 in the active site of the CTD phosphatase Ssu72 . Genes Dev. 26 , 2265 –2270 (2012 ).23070812 Clapperton, J. A. et al. Structure and mechanism of BRCA1 BRCT domain recognition of phosphorylated BACH1 with implications for cancer . Nat. Struct. Mol. Biol. 11 , 512 –518 (2004 ).15133502 Rodriguez, M., Yu, X., Chen, J. & Songyang, Z. Phosphopeptide binding specificities of BRCA1 COOH-terminal (BRCT) domains . J. Biol. Chem. 278 , 52914 –52918 (2003 ).14578343 Williams, R. S. et al. Detection of protein folding defects caused by BRCA1-BRCT truncation and missense mutations . J. Biol. Chem. 278 , 53007 –53016 (2003 ).14534301 Simons, A. M. et al. BRCA1 DNA-binding activity is stimulated by BARD1 . Cancer Res. 66 , 2012 –2018 (2006 ).16489000 Kleiman, F. E. et al. BRCA1/ BARD1 inhibition of mRNA 3' processing involves targeted degradation of RNA polymerase II . Genes Dev. 19 , 1227 –1237 (2005 ).15905410 Pettersen, E. F. et al. UCSF Chimera--a visualization system for exploratory research and analysis . J. Comput. Chem. 25 , 1605 –1612 (2004 ).15264254

PMC005xxxxxx/PMC5003539.txt

==== Front Prev Chronic Dis Prev Chronic Dis PCD Preventing Chronic Disease 1545-1151 Centers for Disease Control and Prevention 27560720 16_0126 10.5888/pcd13.160126 Original Research Peer ReviewedCan States Simultaneously Improve Health Outcomes and Reduce Health Outcome Disparities? Kindig David MD PhD Lardinois Nicholas MPA Chatterjee Debanjana PhD Author Affiliations: Nicholas Lardinois, University of Wisconsin-Madison, Population Health Institute, Madison, Wisconsin; Debanjana Chatterjee, Division of General Pediatrics and Adolescent Health, University of Minnesota School of Medicine, Minneapolis, Minnesota. Corresponding Author: David Kindig, MD, PhD, University of Wisconsin-Madison, Population Health Institute, 610 Walnut St, 550 WARF, Madison, WI 53726. Telephone: 608-263-4886. Email: dakindig@wisc.edu. 2016 25 8 2016 13 E112Introduction Reducing racial health disparities is often stated as a population health goal, but specific targets for such improvement are seldom set. It is often assumed that improving overall health outcomes will be linked to disparity reduction, but this is not necessarily the case. Methods We compared the annual change from 1999 through 2013 in combined-race (black and white) mortality with the annual change in absolute and relative racial mortality disparities for US states. Results Median annual improvement in combined-race mortality was 1.08% per year. Annual overall mortality rate reductions ranged from 0.24% per year in Oklahoma to 1.83% per year in Maryland. For disparities, the median for the black–white absolute gap was 3.60% per year, and the median for the relative black-to-white ratio was 1.19% per year. There was no significant correlation between the combined-race measure and either the absolute (0.03) or relative disparity measure reductions (−0.17). Conclusion For mortality in US states over a recent period, improvement in the population mean and disparity reduction do not usually occur together. The disparity reduction rates observed may provide realistic guidance for public and private policy makers in setting goals for reducing population health disparity and creating investment priorities. As a starting point for discussion, the observed national median annual percentage improvement of 1.1 per year combined, 3.6% per year absolute gap reduction, and 1.2% per year relative gap reduction would be modest and reasonable goals. ==== Body Introduction National and state health outcome goals are often framed in terms of improving the population mean and reducing or eliminating disparities within the population. For example, in Healthy People 2020, the 2 overarching goals are 1) attain high-quality, longer lives free of preventable disease, disability, injury, and premature death, and 2) achieve health equity, eliminate disparities, and improve the health of all groups (1). However as Keppel et al pointed out with regard to Healthy People 2010, the first goal does not necessarily achieve health equity, eliminate disparities, and improve the health of all groups (2). Different strategies are often needed for these 2 goals, and innovations often have higher effect, at least initially, on well-educated or advantaged populations, which can at least temporarily increase disparities (2–4). Trends in state health outcomes show large variations over time (5–8). Satcher et al showed that the black–white gap in mortality rates changed little between 1960 and 2000 (9). Another recent study found that large racial disparities in some states are explained by higher-than-average life expectancy among whites or lower-than-average life expectancy among blacks (10). Webb et al constructed a health disparity index by race, which compared state performance but did not contrast disparity reduction with mean improvement (11). To our knowledge no jurisdiction in the United States has emphasized the potential trade-off between these 2 outcomes or given policy attention to how such trade-offs should be addressed. The objective of this study was to describe what US states recently experienced in overall mean improvement in mortality compared with the improvement in the black–white mortality gap. We tested the hypothesis that states that experience the greatest improvements in combined-race mortality also experience the greatest improvements in reducing racial disparities. We hope that such evidence will guide policy and investment planning so that the United States and the 50 individual states can set reasonable annual improvement targets to achieve in the coming decades. Methods Data on age-adjusted mortality come from the publicly available CDC Wonder’s Compressed Mortality Database (12). We extracted mortality data on people younger than 75 years for all 50 US states and Washington, DC, for all sexes, for blacks, and for white non-Hispanics for all years from 1999 through 2013. Mortality was calculated per 100,000 people and age-adjusted by using the 2000 US standard population. We calculated the annual percentage change for the combined mortality of both races as well as by 2 mortality disparity measures for each state: racial gap as an absolute disparity measure and the black-to-white ratio as a relative disparity measure. Absolute disparity refers to the simple difference in a health outcome, whereas relative disparity refers to the health outcome of one group as a ratio of the other’s health outcome. To generate annual percentage changes, we used Stata/SE 14.0 (StataCorp LP) to predict values from a linear regression of each measure on years for each state, providing a smooth linear trend (13). The absolute mortality gap by race was calculated by subtracting the white age-adjusted mortality from the black age-adjusted mortality for each year and each state. States that reported 50 or fewer deaths in 2000 in either racial category were excluded from our analysis because too few events can cause large variation in year-to-year mortality; for blacks, the states excluded were Idaho, Maine, Montana, New Hampshire, North Dakota, South Dakota, Vermont, and Wyoming. Also excluded from our analysis were 4 states with a statistically insignificant relationship between the age-adjusted mortality racial gap and year according to a linear regression model of age-adjusted mortality racial gap and year: the excluded states were Alaska, Hawaii, West Virginia, and Wisconsin. If a basic linear regression reveals statistically insignificant results, we cannot conclude that these states’ disparity reductions were statistically different from zero. Recently published research results note that decisions about measuring disparities using absolute or relative methods, such as group rankings and direction and magnitude of changes over time, has an effect on results (14–16). Therefore, we follow the recommendation to examine and report both absolute and relative disparity results. The relative disparity measure we examined was a black-to-white ratio of age-adjusted mortality for each state. Five additional states were found not to have significant annual percentage changes in relative mortality disparity and therefore were excluded from the relative-disparity portion of our analysis: Colorado, Iowa, Nebraska, New Mexico, and Utah. Results The Table describes 1) the relationship across the states examined between annual state percentage improvement in mortality and 2) the change in the absolute racial gap and relative racial disparity from 1999 through 2013. The mean combined-race mortality rate of annual improvement was 1.11%, (standard deviation [SD] 0.42%). The range was from 1.83% in Maryland to 0.24% in Oklahoma. For annual percentage change in the racial gap, the mean was a reduction of 3.64% per year (SD, 0.97%). The range was from 6.6% in Rhode Island to 2.13% in Iowa. For annual percentage change in the black-to-white ratio, the mean was an annual reduction of 1.2% (SD, 0.4%). State mortality varied from 2.47% in Rhode Island to 0.43% in California. Figure 1 displays findings for the absolute disparity measure for all states examined. Table State Annual Percentage Variation in Improvement in Combined-Race (Black and White) Mortalitya, Absolute Racial Gap, and Relative Racial Gap, 1999–2013 Category Mean Median High Low Standard Deviation Combined-race mortality (38 states) −1.11 −1.08 −1.83 −0.24 0.42 Absolute racial gap (38 states) −3.64 −3.60 −6.6 −2.13 0.97 Combined-race mortality (33 states)b −1.14 −1.16 −1.83 −0.24 0.43 Relative racial disparity (33 states)b −1.2 −1.19 −2.47 −0.43 0.4 a Mortality was calculated per 100,000 people and age-adjusted by using the 2000 standardized US population. b Only 33 states were relevant for the method used to measure relative disparity. Figure 1 Racial gap between blacks and whites versus annual percentage change in combined-race mortality: variation in 38 states’ annual percentage improvement in combined-race mortality and in absolute racial gaps, 1999–2013. State Combined-Race Mortality, Annual Percentage Change Racial Gap in Mortality, Annual Percentage Change Alabama (AL) −0.5 −3.9 Alaska (AK) −0.38 −3.64 Arizona (AZ) −1.08 −4.09 California (CA) −1.56 −2.37 Colorado (CO) −1.42 −2.62 Connecticut (CT) −1.66 −3.31 Delaware (DE) −1.16 −4.98 Florida (FL) −1.08 −5.07 Georgia (GA) −1.43 −4.68 Iowa (IA) −0.69 −2.13 Illinois (IL) −1.54 −2.61 Indiana (IN) −0.79 −3.3 Kansas (KS) −0.71 −2.71 Kentucky (KY) −0.39 −5.22 Louisiana (LA) −0.94 −3.17 Massachusetts (MA) −1.74 −4.73 Maryland (MD) −1.83 −3.96 Michigan (MI) −1.07 −2.33 Minnesota (MN) −1.37 −3.32 Missouri (MO) −0.87 −3.37 Mississippi (MS) −0.75 −3.6 North Carolina (NC) −1.29 −4.25 Nebraska (NE) −1.07 −2.2 New Jersey (NJ) −1.75 −3.4 New Mexico (NM) −0.64 −3.93 Nevada (NV) −1.02 −4.5 New York (NY) −1.82 −3.37 Ohio (OH) −0.8 −2.43 Oklahoma (OK) −0.24 −3.22 Oregon (OR) −1.23 −3.64 Pennsylvania (PA) −1.22 −2.94 Rhode Island (RI) −1.22 −6.6 South Carolina (SC) −1.39 −3.94 Tennessee (TN) −0.77 −4.51 Texas (TX) −1.06 −3.96 Utah (UT) −0.84 −2.56 Virginia (VA) −1.49 −3.59 Washington (WA) −1.39 −4.12 The extent of improvement in combined mortality was not correlated with the reduction in the racial mortality gap (correlation coefficient = 0.03). Some states (eg, Massachusetts, Georgia) improved substantially on both combined-race mortality and absolute racial gap outcomes while other states’ (eg, Oklahoma, Iowa) improvement on both was not as great. Similarly, states (eg, California, Illinois) did well on combined-race mortality improvement but not as well on absolute racial gap improvement. Conversely, some states (eg, Kentucky, Tennessee) failed to have great improvements in combined-race mortality but saw a large reduction in the absolute racial gap from 1999–2013. In every state there was improvement in both black and white mortality, but improvement in black mortality was always greater (data not shown). Figure 2 displays a similar relationship between relative disparity and annual percentage changes in combined-race mortality across states. Again, the extent of state improvement in combined mortality is not correlated with the reduction in the relative black-to-white ratio disparity measure (correlation coefficient = −0.17). Massachusetts and Maryland improved substantially on both combined-race mortality and relative racial gap outcomes while other states’ (eg, Oklahoma, Kansas) improvement on both was not as great. California and Illinois improved greatly on the combined-race mortality rate but not as well on the relative black-to-white ratio; the opposite was true for Kentucky and Tennessee. Figure 2 Relative racial disparity versus annual percentage change in combined-race mortality: variation in 33 states’ annual percentage improvement in combined-race (black and white) mortality and relative racial disparities, 1999–2013. State Combined-Race Mortality, Annual Percentage Change Relative Racial Disparity in Mortality, Annual Percentage Change Alabama (AL) −0.5 −1.37 Alaska (AK) −0.38 −1.37 Arizona (AZ) −1.08 −1.23 California (CA) −1.56 −0.43 Connecticut (CT) −1.66 −0.82 Delaware (DE) −1.16 −1.80 Florida (FL) −1.08 −1.83 Georgia (GA) −1.43 −1.61 Illinois (IL) −1.54 −0.67 Indiana (IN) −0.79 −1.10 Kansas (KS) −0.71 −0.95 Kentucky (KY) −0.39 −1.58 Louisiana (LA) −0.94 −1.09 Massachusetts (MA) −1.74 −1.30 Maryland (MD) −1.83 −1.25 Michigan (MI) −1.07 −0.72 Minnesota (MN) −1.37 −0.96 Missouri (MO) −0.87 −1.19 Mississippi (MS) −0.75 −1.22 North Carolina (NC) −1.29 −1.53 New Jersey (NJ) −1.75 −1.04 Nevada (NV) −1.02 −0.95 New York (NY) −1.82 −0.83 Ohio (OH) −0.8 −0.72 Oklahoma (OK) −0.24 −0.96 Oregon (OR) −1.23 −0.90 Pennsylvania (PA) −1.22 −0.98 Rhode Island (RI) −1.22 −2.47 South Carolina (SC) −1.39 −1.34 Tennessee (TN) −0.77 −1.67 Texas (TX) −1.06 −1.35 Virginia (VA) −1.49 −1.14 Washington (WA) −1.39 −1.19 Discussion The lack of a strong relationship between the combined improvement in mortality and improvement in the disparity gaps answers the question posed in the title and in our hypothesis; most states have not achieved these 2 outcomes simultaneously. Some states, such as Massachusetts, did well on both mean mortality improvement and disparity reduction, while others, such as Oklahoma, have had difficulty with both. However, states often perform well with one dimension but struggle with the other. We believe our data provide some guidance about what is possible for any state to achieve. At least one state, Maryland, experienced an annual improvement in combined mortality by 1.83% from 1999 through 2013; at least one state, Rhode Island, experienced an annual improvement of 6.6% per year in the black–white mortality absolute gap; at least one state, Rhode Island, had an annual improvement of 2.47% per year in the relative black-to-white mortality ratio. These are not theoretical targets; they are results that were achieved by at least one state during the past decade. Massachusetts performed the best in simultaneous overall improvement (1.74%), absolute disparity reduction (4.73%), and relative disparity reduction (1.3%). More work could also model and project realistic overall and disparity reduction targets that the United States and the individual US states could each achieve by using the highest performing states as guides. As a starting point for discussion, the observed national median annual percentage of 1.1% combined improvement, 3.6% absolute gap reduction, and 1.2% relative gap reduction could be future state continuous improvement benchmarks. Some states might want to use these baselines to develop targets, while others might want to use as baselines what peer states accomplished. Our results reinforce the importance of reporting disparity results in both absolute and relative terms. Although neither measure showed an overall correlation with combined mortality improvement, there were differences across the states depending on the disparity measure used. Although each disparity measure has slightly different results, the annual percentage change for absolute and relative disparity are strongly linked (correlation coefficient = 0.91). Although in this study both measures showed improvement, there are examples where one measure improves but the other does not (17). Since neither is intrinsically preferable from a policy perspective, policy makers should continue to measure and target reductions in both. Several limitations should be considered when interpreting our results. This analysis is limited to all-cause mortality; additional research is needed to look into age-specific mortality as well as non-mortality outcomes such as morbidity and health-related quality of life measures (18). In addition, specific-age groups should be examined to see whether there are life stage differences in these findings. We limited our analysis to state changes, so we do not know the extent of variation across counties in improvements or declines in trends. Nor did we examine other disparity domains such as socioeconomic status, which should be investigated since health-related socioeconomic disparities are large and are seen in every state (19). It would also be useful to examine other periods to determine whether the range in improvement we found was achieved in other periods, since what occurred in our study period might be an imperfect guide to future possibilities; the mortality experience of the previous cohorts that produced these results may be different in either direction for more recent cohorts. Of course, these results do not indicate how any state achieved the results that we show here. We use the term “experienced” explicitly rather than “produced,” since it is unclear and probably unlikely that any states produced such results intentionally in response to explicit mean improvement-disparity reduction targets, although Maryland set mean and disparity targets for a variety of health outcomes and determinants (20). Nor do we know the most cost-effective way of achieving the improvement some states achieved, either in any one of the measures or both together. Although we are beginning to collect evidence on effective programs and policies, such evidence often shows relationships for mean improvement rather than disparity reduction and overall effectiveness instead of cost effectiveness (21). Examining the states that performed well and states that performed poorly on both dimensions may reveal clues about the most effective policy packages to use for large improvement. If a public or a private policy maker were interested in trying to determine what would produce optimal results, some standard of what “optimal” means would need to be defined. For this purpose, we believe it would be useful to have some summary metric of mean and disparity gap improvement such as the achievement index suggested by Wagstaff (22). As Wagstaff indicated, such a metric would have to reflect a value choice of the relative importance of mean improvement in mortality versus disparity reduction. Such a metric could be more complicated than the one he proposed, since several disparity domains also need to be considered. Since different states or communities would probably value each component (or each disparity domain) differently, a useful tool would be one that allows different weights to be used for each component so that achievement progress could be assessed component by component, perhaps beginning with a default standard that weighted components equally. Despite these challenges, we believe that our results are useful now in beginning to set benchmarks for what is possible and to identify programs and policies that are most closely related to improved performance. We encourage public and private entities to 1) review what many states achieved in both general improvement and disparity reduction and 2) set policy and investment priorities in accordance with their own values and perspectives (23). Acknowledgments This work was partly supported by the Robert Wood Johnson Foundation Health and Society Scholars program at the University of Wisconsin-Madison as well as a Robert Wood Johnson Foundation research grant to the University of Wisconsin Population Health Institute. We appreciate methods advice from Yukiko Asada, PhD, at Dalhousie University. The opinions expressed by authors contributing to this journal do not necessarily reflect the opinions of the U.S. Department of Health and Human Services, the Public Health Service, the Centers for Disease Control and Prevention, or the authors' affiliated institutions. Suggested citation for this article: Kindig D, Lardinois N, Chatterjee D. Can States Simultaneously Improve Health Outcomes and Reduce Health Outcome Disparities? Prev Chronic Dis 2016;13:160126. DOI: http://dx.doi.org/10.5888/pcd13.160126. ==== Refs References 1. Aungst RB . Healthy people 2020. Perspectives on Audiology 2011;7 (1 ):29–33. 10.1044/poa7.1.29 2. Keppel K , Bilheimer L , Gurley L . Improving population health and reducing health care disparities. Health Aff (Millwood) 2007;26 (5 ):1281–92. 10.1377/hlthaff.26.5.1281 17848438 3. Benach J , Malmusi D , Yasui Y , Martínez JM , Muntaner C . Beyond Rose’s strategies: a typology of scenarios of policy impact on population health and health inequalities. Int J Health Serv 2011;41 (1 ):1–9. 10.2190/HS.41.1.a 21319717 4. Deaton A . Policy implications of the gradient of health and wealth. Health Aff (Millwood) 2002;21 (2 ):13–30. 10.1377/hlthaff.21.2.13 11900153 5. Remington PL , Catlin BB , Kindig DA . Monitoring progress in population health: trends in premature death rates. Prev Chronic Dis 2013;10 :E214. 10.5888/pcd10.130210 24370109 6. Orsi JM , Margellos-Anast H , Whitman S . Black-white health disparities in the United States and Chicago: a 15-year progress analysis. Am J Public Health 2010;100 (2 ):349–56. 10.2105/AJPH.2009.165407 20019299 7. Drewette-Card RJ , Landen MG . The disparity change score: a new methodology to examine health disparities in New Mexico. J Public Health Manag Pract 2005;11 (6 ):484–92. 10.1097/00124784-200511000-00003 16224282 8. Pearcy JN , Keppel KG . Monitoring change in health disparity. J Public Health Manag Pract 2008;14 (5 ):481–6. 10.1097/01.PHH.0000333884.30023.d6 18708893 9. Satcher D , Fryer GE Jr , McCann J , Troutman A , Woolf SH , Rust G . What if we were equal? A comparison of the black-white mortality gap in 1960 and 2000. Health Aff (Millwood) 2005;24 (2 ):459–64. 10.1377/hlthaff.24.2.459 15757931 10. Bharmal N , Tseng CH , Kaplan R , Wong MD . State-level variations in racial disparities in life expectancy. Health Serv Res 2012;47 (1 Pt 2 ):544–55. 10.1111/j.1475-6773.2011.01345.x 22092060 11. Webb BC , Simpson SL , Hairston KG . From politics to parity: using a health disparities index to guide legislative efforts for health equity. Am J Public Health 2011;101 (3 ):554–60. 10.2105/AJPH.2009.171157 21233445 12. Underlying cause of death database. Atlanta (GA): National Center for Health Statistics, Centers for Disease Control and Prevention; 1999–2013. http://wonder.cdc.gov/controller/datarequest/D76. Accessed April 8, 2015. 13. State disparity trends. Madison (WI): University of Wisconsin-Madison Population Health Institute. http://uwphi.pophealth.wisc.edu/programs/match/recent-projects/StateDisparityTrends.htm 14. Harper S , King NB , Meersman SC , Reichman ME , Breen N , Lynch J . Implicit value judgments in the measurement of health inequalities. Milbank Q 2010;88 (1 ):4–29. 10.1111/j.1468-0009.2010.00587.x 20377756 15. Asada Y . On the choice of absolute or relative inequality measures. Milbank Q 2010;88 (4 ):616–22, discussion 623–7. 10.1111/j.1468-0009.2010.00614.x 21166871 16. Kjellsson G , Gerdtham UG , Petrie D . Lies, damned lies, and health inequality measurements: understanding the value judgments. Epidemiology 2015;26 (5 ):673–80. 10.1097/EDE.0000000000000319 26133019 17. Canadian Institute for Health Information. Trends in income-related health inequalities in Canada: technical report; 2015 https://secure.cihi.ca/free_products/trends_in_income_related_inequalities_in_canada_2015_en.pdf. Accessed March 31, 2016. 18. Kindig DA . Which outcomes should we improve? University of Wisconsin-Madison, Department of Population Health Sciences; 2011 http://www.improvingpopulationhealth.org/blog/2011/02/which_outcomes_improve.html. Accessed March 31, 2016. 19. Asada Y , Yoshida Y , Whipp AM . Summarizing social disparities in health. Milbank Q 2013;91 (1 ):5–36. 10.1111/milq.12001 23488710 20. Maryland state health improvement process. Baltimore (MD): Department of Health and Mental Hygiene, Office of Population Health Improvement, State of Maryland. http://dhmh.maryland.gov/ship/Pages/home.aspx. Accessed May 22, 2016. 21. Rankings: roadmaps to county health. Madison (WI): Robert Wood Johnson Foundation and the University of Wisconsin-Madison Population Health Institute. http://www.countyhealthrankings.org/roadmaps. Accessed December 14, 2015. 22. Wagstaff A . Inequality aversion, health inequalities and health achievement. J Health Econ 2002;21 (4 ):627–41. 10.1016/S0167-6296(02)00006-1 12146594 23. Rigby E , Soss J , Booske BC , Rohan AM , Robert SA . Public responses to health disparities: how group cues influence support for government intervention. Soc Sci Q 2009;90 (5 ):1321–40. 10.1111/j.1540-6237.2009.00646.x

PMC005xxxxxx/PMC5003542.txt

==== Front Small GTPasesSmall GTPasesKSGTksgt20Small GTPases2154-12482154-1256Taylor & Francis 27104658117376710.1080/21541248.2016.1173767ReviewDeregulation of Rho GTPases in cancer A. P. Porter et al.Small GTPasesPorter Andrew P. a#Papaioannou Alexandra ab#Malliri Angeliki a#a Cell Signaling Group, Cancer Research UK Manchester Institute, The University of Manchester, Manchester, UKb “Cellular and Genetic Etiology, Diagnosis and Treatment of Human Disease” Graduate Program, Medical School, University of Crete, Heraklion, Greece# These authors contributed equally to this review. Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/ksgt. Angeliki Malliri Angeliki.Malliri@cruk.manchester.ac.ukCell Signaling Group, Cancer Research UK Manchester Institute, The University of Manchester, Manchester, UK, M20 4BXJul-Sep 2016 22 4 2016 22 4 2016 7 3 123 138 29 12 2015 18 3 2016 28 3 2016 © 2016 The Author(s). Published by Taylor & Francis2016The Author(s)This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The moral rights of the named author(s) have been asserted.ABSTRACT In vitro and in vivo studies and evidence from human tumors have long implicated Rho GTPase signaling in the formation and dissemination of a range of cancers. Recently next generation sequencing has identified direct mutations of Rho GTPases in human cancers. Moreover, the effects of ablating genes encoding Rho GTPases and their regulators in mouse models, or through pharmacological inhibition, strongly suggests that targeting Rho GTPase signaling could constitute an effective treatment. In this review we will explore the various ways in which Rho signaling can be deregulated in human cancers. Keywords cancerGAPsGDIGEFsRho GTPases mutationstumorigenesis ==== Body Introduction Rho GTPases bind to a wide range of effector proteins and play central roles in the regulation of the actin and microtubule cytoskeletons and gene transcription.1 Through these effects, Rho family proteins influence many normal cellular functions such as adhesion, polarity, motility and invasion, as well as cell cycle progression and survival.2,3 Rho, Rac and Cdc42 were initially characterized as regulators of the actin cytoskeleton1 with a typical pattern being Rho activation leading to the formation of contractile actin, Rac activation controlling peripheral actin structures such as lamellipodia and membrane ruffles, and Cdc42 actin structures such as filopodia.1 However, it has long been clear that these proteins have roles far beyond direct regulation of the actin cytoskeleton. For instance, Cdc42 is a master regulator of polarity in organisms from yeast to mammals, while Rac regulates phagocytosis in the immune system, including production of reactive oxygen species.1,4 They are involved in many essential physiological processes including embryonic development, neuronal differentiation and neurite formation and maintenance of stem cells in the bone marrow, skin and intestine.2,3,5 Conversely, deregulation of Rho GTPases is linked to many of the “hallmarks of cancer,” including oncogenic transformation, cell survival and tumor metabolism as well as metastasis (reviewed in ref. 2). While some consequences of deregulated Rho family signaling can be considered pro-tumorigenic, a number of cellular processes stimulated by Rho family proteins—such as the role of Rac1 in apoptosis and maintenance of apicobasal polarity—can be considered to antagonize tumor formation and progression.6 The anti-tumorigenic effects of Rho family proteins must be sufficiently differentiated from those pro-oncogenic functions to avoid undermining the therapeutic benefits to be achieved by pharmacologically antagonizing Rho GTPases. The Rho GTPase cycle Rho GTPases are molecular switches which cycle between an inactive GDP-bound form and an active GTP-bound form (see Fig. 1). The GTPase cycle is largely regulated by guanine nucleotide exchange factors (GEFs) and GTPase-activating proteins (GAPs). 7 GEFs displace the GDP bound in the active site allowing GTP binding. GTP binding alters the conformation of the GTPase, allowing it to interact with downstream effector molecules (Fig. 1).7 GEFs have also been thought to contribute to signaling specificity through scaffolding upstream and downstream interactors;8 this was recently demonstrated with the GEFs Tiam1 and P-Rex1 driving different behaviors via the same small GTPase, Rac1.9 Figure 1. The Rho GTPase cycle GTPase regulation occurs in a number of distinct stages. Guanine nucleotide exchange factors (GEFs) are able to bind to inactive GTPases, displacing the bound GDP, which is then replaced by GTP from the cytoplasm. In their active form Rho GTPases bind to a wide variety of effectors, mediating a large number of cellular processes, including migration, cell-cell adhesion, transcription and proliferation. GEFs also may act to direct signaling by scaffolding particular effectors. To end signaling, GTPase activating proteins (GAPs) bind to the GTPase and enhance their weak intrinsic GTPase activity. Bound GTP is converted to GDP, changing the conformation of the GTPase and rendering it unable to bind effector proteins. Inactive GTPases are mainly found in the cytoplasm, where they can be degraded, or stabilised by binding to Rho GDIs, which act as molecular chaperones and prevent activation by sequestering the GTPases away from GEFs. Conversely, GAPs activate the weak intrinsic GTPase activity of Rho proteins leading to the hydrolysis of bound GTP, switching the GTPase to an inactive conformation (Fig. 1).7 The abundance of GEFs (at least 80) and GAPs (over 70) indicates the importance of tightly controlling Rho GTPase signaling. Guanine nucleotide dissociation inhibitors (GDIs) are a third class of regulators of Rho proteins. They sequester inactive GTPases in the cytoplasm by masking their C-terminal lipid moieties that mediate plasma membrane localization, which can inhibit their activation7 (Fig. 1). They can also protect GTPases from degradation10 and also have more subtle effects, such as directing activation of Rho GTPases to specific membrane compartments.11 Rho GTPases are also known to be modulated by a host of post-translational modifications, including phosphorylation, ubiquitylation, SUMOylation, ADP-ribosylation, glycosylation, adenylation, and transglutamination/deamidation. Given the wide variety of these modifications, detailed analysis is outside the scope of this review; for more details see refs. 12-14. At the simplest conceptual level, anything which increases the abundance of the active form should increase signaling, while anything decreasing the abundance of the active state, or actively stabilizing the inactive state, should decrease signaling. Disruption of this balance—by direct activation of Rho GTPases or indirectly through changes in regulators as described above—is increasingly being linked to oncogenesis (see Fig. 2). In this review we will focus on the variety of ways in which Rho signaling has been shown to be disrupted in cancer: alterations in protein levels of the GTPases, disruption to regulators of GTPases, changes in post-translational modifications of GTPases and finally we review the emerging literature on direct mutation of GTPases. Figure 2. Rho GTPase signaling can be deregulated in cancer by a wide range of mechanisms. (1) Evidence is emerging of many direct mutations of GTPases, such as the Rac1 P29S mutation which is a novel driver in melanoma. (2) GEFs are found overexpressed in many different cancer types, consistent with aberrant Rho GTPase signaling driving transformation and oncogenic progression. (3) Negative regulators of Rho GTPases, such as Rho GAPs and Rho GDIs, have been shown to be tumour suppressors, and lost in human cancers. (4) GTPases are often found to be overexpressed in human cancers, where they drive a variety of oncogenic processes. (5) Post-translational modifications of GTPases, such as changes in ubiquitylation or sumoylation, can alter their signaling. (6) The Rac1b splice form of Rac1 is found in multiple cancers including breast, colon and lung. Copy number alterations and misexpression of Rho GTPases in cancer Before the finding of direct mutations of Rho GTPases, the main way they were thought to be misregulated in cancer was through changes in expression levels (see Fig. 2). Increased expression of Rho proteins is often associated with tumor formation, growth and progression, an indication of a positive contribution of increased Rho GTPase activity to tumorigenesis.2 The interesting exception is RhoB which, as discussed below, appears to more commonly play a tumor suppressor role, and is accordingly found at reduced levels in tumor samples. Rac1 has been found to be overexpressed in testicular, breast and prostate cancer, as well as gastric and lung cancers.15-19 In recent studies, its overexpression in gastric cancer was correlated with the aggressiveness of the tumors, greater invasion and lymph node metastasis, as well as poor patient survival.19 Rac1 is also overexpressed in acute myeloid leukemia cells, where it enhances migration and cell growth, and is linked to chemoresistance.20 Animal experiments support a requirement for Rac1 in tumor formation and growth in many different tumor models. Mice with Rac1 deletion specifically from keratinocytes are resistant to developing Ras-induced skin cancer21 while those with a Rac1 deletion in pancreatic progenitor cells are protected from development of pancreatic ductal adenocarcinoma (PDAC).22 Rac1 is also required for K-Ras-induced lung tumors in mice,23 and cooperates with APC loss in a mouse model of colorectal cancer, driving a stem-cell like signature in the developing cancer cells.24 Another recent study showed that Rac1 affects stem cell behavior to drive oncogenic progression, by reducing the differentiation of tumor cells.25 A splice variant of Rac1, Rac1b, was found at elevated levels in colon and breast tumors.26 Rac1b includes an additional 14 amino acids compared with wild-type Rac1 and it is mainly found in its active GTP-bound state. Rac1b has reduced affinity for GDIs, meaning it is not sequestered in the cytoplasm, which could explain its increased activity and ability for cellular transformation.27 Rac1b alone is insufficient to drive tumor formation in a non-small cell lung cancer mouse model, but it enhances the activity of K-ras mutations.28 It is highly expressed in stages 1 and 2 of human lung adenocarcinoma, making it a candidate target for preventing progression to more aggressive stages.29 Rac1b is also overexpressed in papillary thyroid carcinoma (PTC), where it is associated with the BRAF V600E mutations and subsequently with poor clinical outcomes.30 Rac2, which shares a high degree of sequence conservation with Rac1, is restricted to expression in the haematapoetic cell lineage. Although no aberrations of Rac2 have been directly linked to oncogenesis, Rac2 is emerging as a therapeutic target, as abrogation of Rac2 signaling slows the growth of AML and CML tumors (reviewed in ref. 31). Rac3 activity was found to be increased in highly proliferative breast cancer cell lines, although this does not correspond to increases at the protein level32 suggesting other mechanisms of activation. RhoA and RhoC have been found overexpressed in a wide range of tumors, particularly those with epithelial origins,2 and in some instances have been linked to oncogenic progression, such as in testicular cancer15 and to poor prognosis, such as in esophageal squamous cell carcinoma.33 In contrast to overexpression, loss or reduced expression of RhoB was observed in lung cancer and head and neck squamous cell carcinoma34,35 suggesting that loss of function of RhoB can contribute to oncogenic progression. However, in a contradictory finding, RhoB is found overexpressed in breast cancer,16 which suggests possible cell- or cancer-type specific roles for this GTPase which may result from differential expression of downstream effectors and/or upstream scaffolding proteins, or the balance between other Rho GTPases. Analysis of gene expression data from the SAGE database reveals changes in Cdc42 levels in cancer tissue, both increased and decreased, compared to normal tissue.36 Cdc42 is overexpressed in testicular and breast cancer,15,16,37 in non-small cell lung cancer,38 and in colorectal adenocarcinoma and cutaneous melanoma.39,40 Finally a less well-studied GTPase, Rnd3/RhoE is downregulated in HCC (hepatocellular carcinoma) and its downregulation is correlated with poor prognosis and tumor progression,41,42 while it is upregulated in gastric cancer cells under hypoxic conditions promoting EMT,43 again highlighting the signaling complexity of these GTPases and their downstream targets. The evidence for altered expression of the above mentioned GTPases is indicative of a role in tumor initiation and/or progression. It should also be considered that lack of data for some of the lesser-studied members of the Rho GTPase family may in part be due to fewer reagents being available with which to look for alternations in these proteins. More unbiased screening, and particularly genome-level sequencing for activating mutations (see below), may help to reverse some of this historical bias. Indirect regulation of Rho GTPases in cancer Modulation of Rho family regulators An alternative mechanism by which many tumors upregulate Rho GTPase signaling is by changing the levels or activities of GTPase regulators, including GEFs, GAPs and Rho GDIs (Fig. 2).44,45 While the general trend is toward overexpression of GEFs, and reduced expression of GAPs and GDIs (indicative of a positive contribution of Rho GTPase signaling to tumorigenesis) the detailed picture emerging is of much more complex regulation, seemingly dependent on tumor type and level of progression. Τhe GEFs Ect2, MyoGEF, P-Rex1, Tiam1, LARG, Dock180, Vav1, Vav2, Vav3 and β-PIX are overexpressed in a variety of human tumors.46 Ect2, which has activity for multiple members of the Rho GTPase family including RhoA, Rac1 and Cdc42, has been recognized as an oncogene in human cancer since 2010, being aberrantly overexpressed and mislocalised in various types of tumors.47 Activation of MyoGEF – a RhoA and RhoC GEF - regulates the invasion of breast cancer cells.48 Overexpression of the Rac1 GEF P-Rex1 promotes metastasis of prostate cancer49 and mutations have been identified in PREX2 (a Rac GEF) in melanoma.50 Tiam1, another Rac1 GEF, was initially identified as being important for invasion in T-cell lymphoma.51 Tiam1 displays high levels of expression in breast cancer where it is associated with grade and metastatic potential52 and is a marker for poor prognosis.53 Overexpression of Tiam1 has also been observed in prostate cancer.17 Furthermore, overexpression of Tiam1 in lung adenocarcinomas as well as in squamous-cell carcinoma of the head and neck (SCCHN) is associated with disease progression and poor patient survival.54 In lung cancer, levels of Tiam1 inversely correlate with expression of the E3 ubiquitin ligase HUWE1, which degrades Tiam1 specifically from cell-cell adhesions, indicating that localized regulation of GEF abundance may play a role in cancer.55 The Tiam1 ortholog STEF/Tiam2 was found to promote proliferation and invasion in liver cancer when overexpressed, and is therefore implicated in the pathogenesis of HCC.56 β-PIX has also been found overexpressed in many breast cancers.57 The haematopoietic specific GEF Vav1 is ectopically expressed in pancreatic adenocarcinoma with a positive correlation to reduced patient survival58 and its presence in a subset of neuroblastoma tumors indicates its involvement in the tumorigenesis process.59 Moreover, high levels of expression of Vav1 are a marker for poor prognosis in breast cancer.53 The Vav1 orthologues Vav2 and Vav3 have also been shown to be deregulated in human tumors. Vav3 is overexpressed in gastric cancer60 as well as in prostate cancer where a novel nuclear function was found to be responsible for its role in malignant progression.61 Moreover, both Vav2 and Vav3 regulate a lung-metastasis specific transcriptome that leads to breast cancer progression.62 Finally, the bipartite Rac1 GEF composed of Dock180 and ELMO1 is overexpressed in malignant gliomas, where it contributes to invasion,63 whereas LARG (leukemia-associated Rho GEF) is found fused with the MLL locus in acute myeloid leukemia (AML)64 leading to aberrant expression. While not exhaustive, this list is highly indicative of an oncogenic function for upregulated Rho GTPase signaling. This data from human tumors is supported by evidence from transgenic mouse models highlighting the importance of a number of GEFs in oncogenic progression. Tiam1 has been shown to be important for Ras-mediated skin65 and intestinal tumorigenesis.66 Interestingly, Tiam1 deficient mice develop fewer tumors, but those which do grow are more invasive, suggesting both positive and negative roles for Tiam1 in oncogenesis. Loss of P-Rex1 leads to a reduction in the invasive potential of melanoma cells in a mouse model of the disease, consistent with work in vitro showing that P-Rex1 can regulate invasion.67 P-Rex2 is also frequently mutated in melanoma, and a truncating mutant, E824*, has recently been shown to cooperates with NRAS to accelerate melanoma development in a mouse model.68 Mice deficient for the Rac1/Cdc42 GEFs Asef1 and Asef2, which are downstream of APC and are overexpressed in colorectal tumors, show reduced spontaneous formation of intestinal adenomas.69 Mice transplanted with leukemic B-cell progenitors expressing the p190-BCR-ABL transgene develop tumors at high frequency; however if these cells are deficient for Vav3 then tumor formation is significantly decreased, and survival time increased.70 Both Vav2 and Vav3 are required for initiation and promotion of skin tumorigenesis.71 The GAP DLC1 (deleted in liver cancer) is a tumor suppressor frequently downregulated in many cancer types either by deletion or epigenetic silencing. Loss of DLC1 leads to an activation of RhoA, and cooperates with oncogenic Myc in a mouse model of liver cancer.72 DLC2 was also found downregulated in hepatocellular carcinoma,73 and more recently was shown to be required to regulate Cdc42 activity for faithful chromosome segregation during mitosis.74 P190RhoGAP is another RhoGAP thought to act as a tumor suppressor; it is frequently deleted in gliomas, and its overexpression is able to suppress tumor formation in a mouse model of the disease.75 However not all GAPs are found downregulated in human tumors; ARHGAP8 is found overexpressed in colon cancer.76 The picture for Rho GDIs is relatively complex, possibly due to their ability to target multiple Rho GTPases and their roles in regulating Rho GTPase activity, stability and trafficking.11 For instance, Rho GDI1 is found downregulated in some breast cancer studies,77 but overexpressed in others.78 Downregulation of Rho GDI2 in bladder cancer is associated with decreased patient survival79 whereas overexpression in pancreatic cancer is associated with invasion. 80 Post-translational modifications As discussed earlier, Rho GTPases are regulated by a whole host of post-translational modifications, many of which are now being linked to inappropriate Rho GTPase function in human cancers and a few of which we will discuss here as illustrative examples. Ubiquitylation of Rac1, RhoA and Cdc42 can be deregulated in cancer cell lines, a fact that could indicate a link between Rho GTPase protein ubiquitylation and cancer.14 For instance, the E3 ligase SMURF1 targets RhoA for degradation at the leading edge of migrating cells, affecting tumor cell migration.81 PIAS3 SUMOylates Rac1 stabilizing the active form of the protein following HGF stimulation and therefore promoting cell migration and invasion, suggesting a possible role in cancer progression.13 Conversely, Rac1 can be ubiquitylated by the E3 ligase HACE1, resulting in its proteasomal degradation, reducing Rac1 mediated migration.82 Ubiquitylation of RhoA has also been reported to be impaired following FBXL19 downregulation in lung cancer epithelial cells.83 FBXL19 ligase also ubiquitylates Rac1 and Rac3, with degradation impairing esophageal cancer cell EMT.84 Finally, phosphorylation of Rho GTPases has also been shown to regulate their transforming ability; for instance phosphorylation of Cdc42 by the Src tyrosine kinase modulates its interaction with Rho GDI which is necessary for cellular transformation.85 These examples from the literature demonstrate some of the great diversity of mechanisms by which cancer cells can indirectly disrupt upstream signals which lead to Rho GTPase activation. Direct mutations of GTPases in human cancers Early studies had identified mutations in RhoH such as the rearrangement of RhoH/TTF gene and the mutation of the 5′-UTR of RhoH gene in some haematopoietic malignancies.86,87 However, mutations within Rho GTPases, except for RhoH, were believed to be rare in cancer until recently. This led to the speculation that Rho GTPases were not direct drivers of oncogenic progression, but merely downstream players in a disease more directly modulated by upstream signaling pathways. With the development of faster and cheaper deep sequencing technology this idea has been challenged, as Rho GTPases have now been found mutated in a wide variety of cancer types (see Table 1).88 In particular, the discovery of a recurrent Rac1 mutation in melanoma has significantly altered the perception of the role of Rho GTPases as drivers of oncogenic progression. For this review, we gathered data on published mutations in the Rho GTPases Rac1, Rac2, Rac3, Cdc42, RhoA, RhoB, RhoC, RhoH and RhoT1 using the cBio portal (http://www.cbioportal.org/), a database that collects cancer genomics data sets from tumor samples across cancer studies,89,90 and IntOGen (https://www.intogen.org/search), which assesses mutational data across multiple tumor types to identify potential driver mutations.91 Both databases are user-friendly, regularly updated, and include additional information such as expression levels, amplifications and deletions (see Table 1). While any table of this kind becomes quickly outdated, it nonetheless serves to highlight the remarkable impact of sequencing technology on the discovery of mutations in human cancers in recent years, as well as the range of cancer types harboring mutations in Rho GTPases. The following section will focus on the emerging literature around these newly-identified mutations and other identified deregulations of Rho GTPases in human cancers. Table 1. A search was conducted using the cBioportal and IntoGen databases for mutations in the Rho GTPases Rac1, Rac2, Rac3, Cdc42, RhoA, RhoB, RhoC, RhoH and RhoT1 which occurred in samples from human patients, and which have been published in the literature. We did not include mutations from cancer cell lines, or provisional data uploaded on the sites. Rho GTPase Acute Myeloid Leukemia Bladder Urothelial Carcinoma Breast Invasive Carcinoma Clear Cell Renal Cell Carcinoma Colorectal Adenocarcinoma Cutaneous squamous cell carcinoma Esophageal Adenocarcinoma/Esophageal Squamous Cell Carcinoma/Nasopharyngeal Carcinoma Pediatric Ewing Sarcoma Glioma and Glioblastoma Head and Neck Squamous Cell Carcinoma Liver Hepatocellular Carcinoma Lung Adenocarcinoma Lung Squamous Cell Carcinoma and Small Cell Lung Cancer Multiple Myeloma Malignant Peripheral Nerve Sheath Tumors Non-renal clear cell carcinoma Ovarian Serous Cystadeno-carcinoma Papillary Thyroid Carcinoma Prostate adenocarcinoma Skin Cutaneous Melanoma Stomach Adenocarcinoma Uterine Corpus Endometrial Carcinoma Cdc42 S30L F110L   S185C R68*         A41T   Q116E E95Q         D63V   G12V (2) L70P E62D, E127D           R186H             D122N                 K166E K150N Rac1   G30E P69S Y40S R68H     A159V   G15S   A159V V46G N92K       N111I Q61R P29S (12) P34H P29S     L53V               C18F, C18Y     R102L             D65N                         P29S, P29T                                             N39S                                             K116N, K116R                                             G142S, A159V (2)                         Rac2       X12_splice V36A   V168M   R102W I21M   C189S K147M           V93I A27V, P29L (2) G15D C18R                     D124E                 R187H E62K, R102Q V168M D65N                                         R174W P136H F82L Rac3       A95V           P34L         Y23H       S89Pfs*64 X76_splice (2) V44M, E100D                       S158*                 A88T A135T R102W, P185L       R5W (3)     R5Q   R5L   I80T E40Q (5)   E40*         G17V   D59G P75S R5W (2) V24F     G17A (2)     T37I   R5W     Y42I   D59N         D76N     S85F G17E, L22R A148T     L22P, S26R     A61V   Y34C                           Y34C, F39C T175M     E40Q (2)     L69P   G184E                           E40K   RhoA   Q63K, P75R     S85Ffs*6                               N41K       F106L, E125Q     F154C                               Y42C (7),Y42S (2)       E142K, D146H                                     L57V (4), D59Y       A161V (2)                                     T60K, A61D       R168T, E172K                                     G62E (2), G62R       E172K, S188Ifs*30                                     Y74D       G14S, E47K (2) D13Y (2) S26T (2) F30L       G12D E165K A15S D13Y A2S       V127G     L81F V9L S88N     Y66_R68del D13Y Y34C               P108R                 R133S       P75S (2) W58C                                       RhoB   P75T, P75L D59N                                           K135Q, E158K E125K                                           K162N, E172K (2) E169K                                           Q180P                                               Y42C   D120N     D146E   E125Q   R145W S73A     L22H D59E   E64K   G178D R68Q RhoC     E142K   R150W               K162N                 S73Lfs*5                                             R182S     E135K     D58G, R69Q S129F S155G   V167I     C7*, F65L A32T, E39K       R23H     P35S, G49S R23C S53T RhoH         R177Q             R121L, R127M Y83C, S84Y             E101K, R168Q         R104K, Q213E E39Q (2)   E12K, E353*   X110_splice   E505K   R104K C377F S156L, R234G       D106H   N422D P30L, P43S R50Sfs*15 K230N, L307V RhoTI   E300K, S479L D91N (2), D91N   I407Dfs*16   P326L           R261S, D317Y       A305V     R191C (2) Y82H (2) V418M, R450C     E505Q K230N, E285G (2)   K412Nfs*12               E484A, T543A       V418L     P220L, K405* R263T A458V (2)           N428Tfs*16                               I407Dfs*16             K469I                                   Notes. * = Stop codon; fs = Frame shift; Mutations in bold = Hotspot mutation; Solid boxes = Identified as a driver mutation by IntOGen algorithms. Numbers in brackets indicate number of times a mutation identified across data sets. Rac1 mutations One early study aiming to detect Rac1 mutations in human brain tumors identified deletions, frame shift and point mutations in 12 out of 45 samples from human patients with brain tumors,92 suggestive of a role for Rac1 in brain tumor development. Now, next generation sequencing has identified a number of cancer-associated mutations along the length of the Rac1 protein, with Rac1 being identified as a driver mutation in head and neck squamous cell carcinoma and cutaneous melanoma (see Table). Among these, P29 is a hot-spot for Rac1 mutation. It was first identified by 2 groups in 2012.93,94 Whole-exome sequencing was performed in melanoma samples and 5% of them were found to harbour missense mutations in the Rac1 gene, making Rac1 the third most highly mutated gene in melanoma (after BRaf and NRas).93 The functional effect of the P29S recurrent mutation is to induce a ‘fast cycling’ form of Rac1, as opposed to the more common gain-of-function mutations used in a laboratory setting which are modeled on activating Ras mutations found at high frequency in human cancers. These mutations, found at G12 and Q61, block GTPase activity and so trap the GTPase in its active, GTP-bound form. In contrast, the P29 residue lies in a hydrophobic pocket in the switch I region of the Rac1 GDP-bound form, and the substitution of the proline residue for a serine enhances the exchange of GDP for GTP,95 while still maintaining the ability to hydrolyse GTP back to GDP. Overall this enhances the interaction of Rac1 with effectors, such as the Pak family of kinases. P29S is therefore considered a gain-of-function mutation that likely promotes oncogenic events during melanoma through mechanisms thought to include altered cell proliferation, adhesion, migration and invasion.93 Expression of the mutant form of Rac1 in melanocytes leads to enhanced cell proliferation and migration,94 and the Rac1 P29S mutant form is able to transform mouse fibroblasts and immortalised breast epithelial MCF10A cells.96 Subsequently 2 other fast-cycling mutants of Rac1 have been identified, N92I and C157Y.96 The ability to cycle from the off-state to the on-state may render these fast-cycling mutants more efficient at driving transformation than the constitutively active mutants, possibly because they more closely mimic normal signaling by being able to associate and dissociate from effectors, or potentially by still associating with GEFs acting also as scaffolding proteins. Rac1 N92I was able to efficiently transform mouse fibroblasts and MCF10A cells, whereas the C157Y mutation was less effective.96 Interestingly, Rac1 P29S (which has also been found as a somatic mutation in a breast cancer cell line) transformed MCF10A cells more efficiently than fibroblasts, whereas the opposite was true for the Rac1 N92I mutation (known as a somatic mutation in a fibrosarcoma cell line),96 suggesting that there are further subtleties to the effects of these different activating mutations still to be uncovered. A serious clinical problem in the treatment of melanoma is the swift development of resistance to the front line treatments of RAF and MEK inhibition. A 2014 study revealed that Rac1 P29S expression in melanoma cell lines and in mouse tumor models conferred resistance to RAF and MEK inhibitors97 with overexpression of Rac1 P29S decreasing apoptosis after RAF and MEK inhibitor treatment. A further clinical study suggested the potential of Rac1 P29S as a predictive biomarker for resistance to therapy in melanoma patients under treatment with these inhibitors.98 Further histological and clinical evidence showed that this hot spot mutation may be responsible for the early metastatic progression of BRAF mutant and BRAF wild-type melanoma.99 A more recent biological insight into the P29S mutation showed increased expression of PD-L1 in Rac1 P29S melanoma patients compared to Rac1 wild type or other Rac1 mutants.100 PD-L1 is a suppressor of the immune system thus its upregulation may allow cancers to evade the host immune system and therefore oncogenic Rac1 P29S may promote the reduction of anti-tumor immune response. As PD-L1 is a candidate biomarker for increased benefit from treatment with anti-PD1 or anti-PD-L1 antibodies, this finding could also have implications in the clinic. Rac2 and Rac3 mutations In the 2012 study conducted by Hodis et al., a homolog to the Rac1 P29 residue was found to be mutated in Rac2, substituting Proline (P) with Leucine (L) (P29L mutation).93 Two Rac2 mutations - P29L and P29Q – similar to the P29S mutation, were confirmed as transforming mutations of Rac2.96 Additional mutations have since been identified in Rac2 (see Table 1) and among these the R102Q was found as a hot spot mutation. Mutations in the Rac3 gene have been identified from a range of cancers, including melanoma, stomach and prostate, but none has yet been studied functionally. Rac2-KO and Rac3-KO mice showed slightly increased survival in a CML and ALL background, respectively,101,102 suggesting a possible oncogenic role for these genes; further experimentation will be required to determine the functional significance of these cancer-associated mutations. RhoA mutations As with Rac1, no mutations in RhoA had been detected in human cancers until very recently. RhoA mutations were identified by several groups in 2014,103 with an IntOGen search indicating a driver role for RhoA in stomach adenocarcinoma (see Table 1), as well as a general pan-cancer driver role. RhoA mutations have been identified in 25% of diffuse-type gastric carcinoma cases studied.104 Recurrent mutations were R5Q, G17E and Y42C. Expression of both RhoA G17E and Y42C were able to rescue growth defects of SW948 colon cancer cells grown in 3D culture following knockdown of endogenous RhoA in contrast to re-expression of wild-type RhoA which was unable to rescue.104 Several groups have found frequent RhoA mutations, specifically the G17V mutation, in angioimmunoblastic T cell lymphoma and peripheral T cell lymphomas.105-107 Interestingly this mutation appears to act similarly to well-characterized dominant negative mutations of RhoA, rather than as an activating mutation. Expression of this mutant form of RhoA increases proliferation in Jurkat cells, an effect also observed with expression of dominant negative RhoA. This fits well with work showing that inactivation of RhoA promotes tumor formation in colorectal cancer models.108 Silencing of RhoA in colon cancer cell lines promoted proliferation, largely through activation of the Wnt/β-catenin pathway and subsequent upregulation of Myc signaling, and this led to increased metastasis. In a mouse model of colorectal cancer, metastatic sites were found to have lower RhoA signaling than the primary tumors, and this held for samples from human tumors as well.108 Another example of inactivating RhoA mutations are those found recurrently in Burkitt Lymphoma, the most common type of childhood B-cell lymphoma. Translocations of the MYC locus leading to deregulated Myc signaling are necessary but not sufficient to drive disease progression, and both whole genome studies109 and exome sequencing110 identified RhoA mutations as additional drivers of the disease. 8.5% of cases had RhoA mutations, and molecular modeling of these mutations suggested that they would reduce RhoA activity, or reduce binding to RhoA effectors.110 Another study conducted with gastric adenocarcinoma samples103 added a number of additional mutations including Y34C, F39C, E40K, N41K, Y42S/C/I, L57V, D59Y, T60K, A61D and G62R/E (and see Table 1). These were accumulated in regions that participate in the interaction of RhoA protein with effector molecules; for instance mutations at Y42 reduce downstream activation of PKN but not mDia or ROCK1.111 This indicates that distinct mutants may have different alterations in effector binding/activation with some of them leading to reduced interaction of RhoA with specific effector proteins. Depending on the target affected, this altered RhoA activity could account for the increased cell spread and the absence of cell cohesion observed in this kind of tumors. These studies suggest either that wild-type RhoA, in the cells of origin for these cancer types, is acting in a tumor suppressive capacity, or that inactivation of RhoA in some way leads to hyperactivation of an oncogenic pathway. C3 toxin-mediated inactivation of RhoA, B and C causes the development of aggressive malignant thymic lymphomas in mice.112 Such findings support a tumor suppressor role for these members of the Rho family. It will require further experiments to reconcile data from these mutational studies with earlier work showing that overexpression of RhoA promotes tumorigenesis. This could be due to differences in the expression of downstream effectors in different tissue types, or different requirements for RhoA throughout the life-cycle of a tumor. RhoB, RhoC and RhoT1 mutations RhoB has been found to be mutated in 5% of bladder cancer cases from a sample of 131 high grade tumors not treated with chemotherapy (with more than 200 additional samples still to be sequenced at the time of writing) making it one of 9 genes mutated in this disease.113 Our cBio search for published RhoB mutations (see Table 1) indicates that P75S/T/L is a hot spot mutation, though it has not yet been studied functionally. In a model of Ras-driven skin cancer, Liu and colleagues showed that the RhoB-null mice had increased skin tumors compared to the heterozygote mice and that RhoB-deficient MEFs transformed with E1A and Ras showed greater resistance to DNA-damage induced apoptosis,114 which suggests that, if functional, these might be inactivating mutations. Two other family members, RhoC and RhoT1, present a number of published mutations in cancer samples and cell lines, with the S73 residue a hotspot in RhoC, while mutations in RhoT1 include a P30L mutation, which by homology may have similar effects to the Rac1 P29S mutation. Deletion of RhoC from mice has been observed to reduce the frequency and growth of tumors,115 which might suggest that activating mutations might promote tumor formation, but further analysis of mutations in these family members is required to determine their functional relevance. Cdc42 mutations The classical activating mutation G12D (equivalent to the G12V activation mutation of Ras) has been found in Cdc42 in melanoma cells in the same study which identified the Rac1 P29S mutation,93 although this mutation was present in only a single patient sample, and has not been functionally characterized. Table lists 14 different published mutations in Cdc42, although no function has yet been ascribed to them. However, given the evidence for a role for Cdc42 in cellular transformation,36 we conclude that it is highly likely that at least some of these mutations will be functionally active. It is also possible that some of these may be inactivating mutations, as in vivo evidence, such as deletion of Cdc42 from hepatocytes which lead to spontaneous tumor formation,116 suggests that Cdc42 might also play a role as a tumor suppressor. Pharmacological inhibition of Rho GTPases Given the long-standing in vitro and in vivo data showing Rho protein involvement in malignant transformation, observed changes in Rho protein expression levels or changes in their regulators and post-translational modifications, and now direct mutation of Rho GTPases, in human cancers, targeting Rho protein signaling is an increasingly attractive target for new cancer therapeutics. Small molecule inhibitors of many Rho proteins are currently being developed and tested. Two different small molecule inhibitors of Rac are currently in use, utilizing 2 different strategies for inhibition. NSC23766 works by inhibiting the interaction between Rac1 and its GEF Tiam1, reducing the activation of Rac1.117 EHT 1864 is a pan-Rac inhibitor which directly targets the Rac GTPase itself, by displacing GTP from the active site.118 NSC23766 can halt the proliferation, anchorage-independent growth and invasion of prostate cancer cells.117 Rac1 inhibition can additionally reduce growth of non-small-cell lung cancer tumors in a mouse model that present resistance to (EGFR)-tyrosine kinase inhibitors such as gefitinib, making it attractive as a potential combination therapy to help circumvent the resistance mechanisms.119 Moreover, Rac1 inhibition impedes the growth, invasion and metastasis of gastric tumors.19 However, while both these inhibitors do indeed target Rac activity, they also have significant off-target effects, as demonstrated by assays using wild-type and Rac1-deficient mouse platelets.120 This emphasizes the need to develop better versions of these drugs, or find other ways of targeting Rac, and other small GTPases. One approach is to use in silico screening to predict potential binding partners which might block GTPase-GEF interactions.121 It is worth noting that this strategy of targeting the interaction between GEFs and GTPases is predicated on the function of the GEFs regulating tumorigenesis via their ability to activate the GTPases. However, this is not always the case. For instance, the activation of the PI3K/Akt pathway by the GEF P-Rex2 does not depend on the GEF activity of the protein.122 Also, a recent paper from our lab demonstrates that different GEFs can have differential effects on cell behavior, despite activating the same GTPase to similar levels,9 most likely by scaffolding different downstream effectors of the GTPase; therefore it will be important to target the correct GEF-GTPase activity for the specific cancer type. Given that Rac1 and Cdc42 are highly expressed and active in ovarian cancer,123 inhibitors of these 2 GTPases have been tested in immortalized and primary human ovarian cancer cells.124 The R enantiomer of ketorolac, (ketorolac is given as an anti-inflammatory drug), can inhibit Rac1 and Cdc42 and was shown to improve patient outcomes in treatment for ovarian cancer.124 Another Rac1 and Cdc42 dual-inhibitor, AZA1, identified from a screen of molecules based on modifying the structure of NSC23766, has been used in in vitro studies to target prostate cancer cells.125 This synthetic compound reduced cancer cell migration and proliferation and succeeded in increasing the survival of xenograft mouse models of prostate cancer by targeting Rac1 and Cdc42 but not RhoA.125 An additional 3 inhibitors of Cdc42 have been developed. Secramine has been identified as a small molecule inhibitor that perturbs Cdc42 activity in a RhoGDI1-dependent manner,126 although is likely to affect other GTPases in the same manner. ZCL278 is a small molecule inhibitor of Cdc42, designed to block the interaction of Cdc42 with the GEF Intersectin. It is thought to disrupt both GEF interactions and GTP binding127 and was shown to inhibit actin-based motility and migration in a metastatic prostate cancer cell line.127 Finally, AZA197, another recently developed Cdc42 inhibitor which appears not to inhibit Rac1 activity has shown some efficacy in reducing tumor size in a xenograft model of colon cancer.128 Reducing signaling through the Rho pathway is often achieved by targeting the Rho target ROCK.129 The ROCK inhibitor Y-27632 retards migration of human prostate cancer cells in mice130 and blocks the invasive activity of cultured rat hepatoma cells.131 Moreover, inhibiting the Rho/ROCK signaling pathway in NSCLC using the ROCK inhibitor fasudil, when combined with inhibition of the proteasome, effectively reduced the viability of mutant K-Ras cells compared with wild-type cells.132 It is likely that further structural modification of these compounds, or further high-throughput compound-screening, will lead to more specific inhibitors, and that as we further our understanding of both normal and abnormal Rho GTPase signaling we will be better placed to deploy them therapeutically. Conclusion In conclusion, Rho GTPase signaling is frequently seen to be modified in human cancers through a variety of mechanisms, and work is continuing to understand the consequences of this aberrant signaling. Understanding the wider landscape of Rho GTPase signaling in a tumor type is likely to be important for making the correct, clinically-relevant interventions. Modifications occur from the level of mutation of the GTPases to under or overexpression of their regulating proteins, which both generates a highly complex signaling network that needs further work to be untangled and also suggests many fertile avenues for therapeutic intervention. Disclosure of potential conflicts of interest No potential conflicts of interest were disclosed. ==== Refs References [1] Hall A . Rho family GTPases . Biochem Soc Trans 2012 ; 40 :1378 -82 ; PMID:23176484; http://dx.doi.org/10.1042/BST20120103 23176484 [2] Orgaz JL , Herraiz C , Sanz-Moreno V . Rho GTPases modulate malignant transformation of tumor cells . Small GTPases 2014 ; 5 :e29019 ; PMID:25036871; http://dx.doi.org/10.4161/sgtp.29019 25036871 [3] Sadok A , Marshall CJ . Rho GTPases: masters of cell migration . Small GTPases 2014 ; 5 :e29710 ; PMID:24978113; http://dx.doi.org/10.4161/sgtp.29710 24978113 [4] Knaus UG , Heyworth PG , Evans T , Curnutte JT , Bokoch GM . Regulation of phagocyte oxygen radical production by the GTP-binding protein Rac 2 . Science 1991 ; 254 :1512 -5 ; PMID:1660188; http://dx.doi.org/10.1126/science.1660188 1660188 [5] Jaffe AB , Hall A . Rho GTPases: biochemistry and biology . Annu Rev Cell Dev Biol 2005 ; 21 :247 -69 ; PMID:16212495; http://dx.doi.org/10.1146/annurev.cellbio.21.020604.150721 16212495 [6] Mack NA , Whalley HJ , Castillo-Lluva S , Malliri A . The diverse roles of Rac signaling in tumorigenesis . Cell Cycle 2011 ; 10 :1571 -81 ; PMID:21478669; http://dx.doi.org/10.4161/cc.10.10.15612 21478669 [7] Cherfils J , Zeghouf M . Regulation of small GTPases by GEFs, GAPs, and GDIs . Physiol Rev 2013 ; 93 :269 -309 ; PMID:23303910; http://dx.doi.org/10.1152/physrev.00003.2012 23303910 [8] Rossman KL , Der CJ , Sondek J . GEF means go: turning on RHO GTPases with guanine nucleotide-exchange factors . Nat Rev Mol Cell Biol 2005 ; 6 :167 -80 ; PMID:15688002; http://dx.doi.org/10.1038/nrm1587 15688002 [9] Marei H , Carpy A , Woroniuk A , Vennin C , White G , Timpson P , Macek B , Malliri A . Differential Rac1 signalling by guanine nucleotide exchange factors implicates FLII in regulating Rac1-driven cell migration . Nat Commun 2016 ; 7 :10664 ; PMID:26887924; http://dx.doi.org/10.1038/ncomms10664 26887924 [10] Boulter E , Garcia-Mata R , Guilluy C , Dubash A , Rossi G , Brennwald PJ , Burridge K . Regulation of Rho GTPase crosstalk, degradation and activity by RhoGDI1 . Nat Cell Biol 2010 ; 12 :477 -83 ; PMID:20400958; http://dx.doi.org/10.1038/ncb2049 20400958 [11] Garcia-Mata R , Boulter E , Burridge K . The ‘invisible hand’: regulation of RHO GTPases by RHOGDIs . Nat Rev Mol Cell Biol 2011 ; 12 :493 -504 ; PMID:21779026; http://dx.doi.org/10.1038/nrm3153 21779026 [12] Nethe M , Hordijk PL . The role of ubiquitylation and degradation in RhoGTPase signalling . J Cell Sci 2010 ; 123 :4011 -8 ; PMID:21084561; http://dx.doi.org/10.1242/jcs.078360 21084561 [13] Castillo-Lluva S , Tatham MH , Jones RC , Jaffray EG , Edmondson RD , Hay RT , Malliri A . SUMOylation of the GTPase Rac1 is required for optimal cell migration . Nat Cell Biol 2010 ; 12 :1078 -85 ; PMID:20935639; http://dx.doi.org/10.1038/ncb2112 20935639 [14] Visvikis O , Maddugoda MP , Lemichez E . Direct modifications of Rho proteins: deconstructing GTPase regulation . Biol Cell 2010 ; 102 :377 -89 ; PMID:20377524; http://dx.doi.org/10.1042/BC20090151 20377524 [15] Kamai T , Yamanishi T , Shirataki H , Takagi K , Asami H , Ito Y , Yoshida K . Overexpression of RhoA, Rac1, and Cdc42 GTPases is associated with progression in testicular cancer . Clin Cancer Res 2004 ; 10 :4799 -805 ; PMID:15269155; http://dx.doi.org/10.1158/1078-0432.CCR-0436-03 15269155 [16] Fritz G , Brachetti C , Bahlmann F , Schmidt M , Kaina B . Rho GTPases in human breast tumours: expression and mutation analyses and correlation with clinical parameters . Br J Cancer 2002 ; 87 :635 -44 ; PMID:12237774; http://dx.doi.org/10.1038/sj.bjc.6600510 12237774 [17] Engers R , Ziegler S , Mueller M , Walter A , Willers R , Gabbert HE . Prognostic relevance of increased Rac GTPase expression in prostate carcinomas . Endocr Relat Cancer 2007 ; 14 :245 -56 ; PMID:17639041; http://dx.doi.org/10.1677/ERC-06-0036 17639041 [18] Pan Y , Bi F , Liu N , Xue Y , Yao X , Zheng Y , Fan D . Expression of seven main Rho family members in gastric carcinoma . Biochem Biophys Res Commun 2004 ; 315 :686 -91 ; PMID:14975755; http://dx.doi.org/10.1016/j.bbrc.2004.01.108 14975755 [19] Ji J , Feng X , Shi M , Cai Q , Yu Y , Zhu Z , Zhang J . Rac1 is correlated with aggressiveness and a potential therapeutic target for gastric cancer . Int J Oncol 2015 ; 46 :1343 -53 ; PMID:2558579525585795 [20] Wang JY , Yu P , Chen S , Xing H , Chen Y , Wang M , Tang K , Tian Z , Rao Q , Wang J . Activation of Rac1 GTPase promotes leukemia cell chemotherapy resistance, quiescence and niche interaction . Mol Oncol 2013 ; 7 :907 -16 ; PMID:23726395; http://dx.doi.org/10.1016/j.molonc.2013.05.001 23726395 [21] Wang Z , Pedersen E , Basse A , Lefever T , Peyrollier K , Kapoor S , Mei Q , Karlsson R , Chrostek-Grashoff A , Brakebusch C . Rac1 is crucial for Ras-dependent skin tumor formation by controlling Pak1-Mek-Erk hyperactivation and hyperproliferation in vivo . Oncogene 2010 ; 29 :3362 -73 ; PMID:20383193; http://dx.doi.org/10.1038/onc.2010.95 20383193 [22] Heid I , Lubeseder-Martellato C , Sipos B , Mazur PK , Lesina M , Schmid RM , Siveke JT Early requirement of Rac1 in a mouse model of pancreatic cancer . Gastroenterology 2011 ; 141 :719 -30 , 30 e1–7 .21684285 [23] Kissil JL , Walmsley MJ , Hanlon L , Haigis KM , Bender Kim CF , Sweet-Cordero A , Eckman MS , Tuveson DA , Capobianco AJ , Tybulewicz VL , et al. Requirement for Rac1 in a K-ras induced lung cancer in the mouse . Cancer Res 2007 ; 67 :8089 -94 ; PMID:17804720; http://dx.doi.org/10.1158/0008-5472.CAN-07-2300 17804720 [24] Myant KB , Cammareri P , McGhee EJ , Ridgway RA , Huels DJ , Cordero JB , Schwitalla S , Kalna G , Ogg EL , Athineos D , et al. ROS production and NF-kappaB activation triggered by RAC1 facilitate WNT-driven intestinal stem cell proliferation and colorectal cancer initiation . Cell Stem Cell 2013 ; 12 :761 -73 ; PMID:23665120; http://dx.doi.org/10.1016/j.stem.2013.04.006 23665120 [25] Frances D , Sharma N , Pofahl R , Maneck M , Behrendt K , Reuter K , Krieg T , Klein CA , Haase I , Niemann C . A role for Rac1 activity in malignant progression of sebaceous skin tumors . Oncogene 2015 ; 34 :5505 -12 ; PMID:25659584; http://dx.doi.org/10.1038/onc.2014.471 25659584 [26] Schnelzer A , Prechtel D , Knaus U , Dehne K , Gerhard M , Graeff H , Harbeck N , Schmitt M , Lengyel E . Rac1 in human breast cancer: overexpression, mutation analysis, and characterization of a new isoform, Rac1b . Oncogene 2000 ; 19 :3013 -20 ; PMID:10871853; http://dx.doi.org/10.1038/sj.onc.1203621 10871853 [27] Matos P , Collard JG , Jordan P . Tumor-related alternatively spliced Rac1b is not regulated by Rho-GDP dissociation inhibitors and exhibits selective downstream signaling . J Biol Chem 2003 ; 278 :50442 -8 ; PMID:14506233; http://dx.doi.org/10.1074/jbc.M308215200 14506233 [28] Zhou C , Licciulli S , Avila JL , Cho M , Troutman S , Jiang P , Kossenkov AV , Showe LC , Liu Q , Vachani A , et al. The Rac1 splice form Rac1b promotes K-ras-induced lung tumorigenesis . Oncogene 2013 ; 32 :903 -9 ; PMID:22430205; http://dx.doi.org/10.1038/onc.2012.99 22430205 [29] Stallings-Mann ML , Waldmann J , Zhang Y , Miller E , Gauthier ML , Visscher DW , Downey GP , Radisky ES , Fields AP , Radisky DC . Matrix metalloproteinase induction of Rac1b, a key effector of lung cancer progression . Sci Transl Med 2012 ; 4 :142ra95 ; PMID:2278668022786680 [30] Silva AL , Carmo F , Bugalho MJ . RAC1b overexpression in papillary thyroid carcinoma: a role to unravel . Eur J Endocrinol 2013 ; 168 :795 -804 ; PMID:23482591; http://dx.doi.org/10.1530/EJE-12-0960 23482591 [31] Thomas EK , Cancelas JA , Zheng Y , Williams DA . Rac GTPases as key regulators of p210-BCR-ABL-dependent leukemogenesis . Leukemia 2008 ; 22 :898 -904 ; PMID:18354486; http://dx.doi.org/10.1038/leu.2008.71 18354486 [32] Mira JP , Benard V , Groffen J , Sanders LC , Knaus UG . Endogenous, hyperactive Rac3 controls proliferation of breast cancer cells by a p21-activated kinase-dependent pathway . Proc Natl Acad Sci U S A 2000 ; 97 :185 -9 ; PMID:10618392; http://dx.doi.org/10.1073/pnas.97.1.185 10618392 [33] Faried A , Faried LS , Usman N , Kato H , Kuwano H . Clinical and prognostic significance of RhoA and RhoC gene expression in esophageal squamous cell carcinoma . Ann Surg Oncol 2007 ; 14 :3593 -601 ; PMID:17896152; http://dx.doi.org/10.1245/s10434-007-9562-x 17896152 [34] Mazieres J , Antonia T , Daste G , Muro-Cacho C , Berchery D , Tillement V , Pradines A , Sebti S , Favre G . Loss of RhoB expression in human lung cancer progression . Clin Cancer Res 2004 ; 10 :2742 -50 ; PMID:15102679; http://dx.doi.org/10.1158/1078-0432.CCR-03-0149 15102679 [35] Sato N , Fukui T , Taniguchi T , Yokoyama T , Kondo M , Nagasaka T , Goto Y , Gao W , Ueda Y , Yokoi K , et al. RhoB is frequently downregulated in non-small-cell lung cancer and resides in the 2p24 homozygous deletion region of a lung cancer cell line . Int J Cancer 2007 ; 120 :543 -51 ; PMID:17096327; http://dx.doi.org/10.1002/ijc.22328 17096327 [36] Arias-Romero LE , Chernoff J . Targeting Cdc42 in cancer . Expert Opin Ther Targets 2013 ; 17 :1263 -73 ; PMID:23957315; http://dx.doi.org/10.1517/14728222.2013.828037 23957315 [37] Fritz G , Just I , Kaina B . Rho GTPases are over-expressed in human tumors . Int J Cancer 1999 ; 81 :682 -7 ; PMID:10328216; http://dx.doi.org/10.1002/(SICI)1097-0215(19990531)81:5%3c682::AID-IJC2%3e3.0.CO;2-B 10328216 [38] Liu Y , Wang Y , Zhang Y , Miao Y , Zhao Y , Zhang PX , Jiang GY , Zhang JY , Han Y , Lin XY , et al. Abnormal expression of p120-catenin, E-cadherin, and small GTPases is significantly associated with malignant phenotype of human lung cancer . Lung Cancer 2009 ; 63 :375 -82 ; PMID:19162367; http://dx.doi.org/10.1016/j.lungcan.2008.12.012 19162367 [39] Gomez Del Pulgar T , Valdes-Mora F , Bandres E , Perez-Palacios R , Espina C , Cejas P , Garcia-Cabezas MA , Nistal M , Casado E , Gonzalez-Baron M , et al. Cdc42 is highly expressed in colorectal adenocarcinoma and downregulates ID4 through an epigenetic mechanism . Int J Oncol 2008 ; 33 :185 -93 ; PMID:1857576518575765 [40] Tucci MG , Lucarini G , Brancorsini D , Zizzi A , Pugnaloni A , Giacchetti A , Ricotti G , Biagini G . Involvement of E-cadherin, beta-catenin, Cdc42 and CXCR4 in the progression and prognosis of cutaneous melanoma . Br J Dermatol 2007 ; 157 :1212 -6 ; PMID:17970806; http://dx.doi.org/10.1111/j.1365-2133.2007.08246.x 17970806 [41] Grise F , Sena S , Bidaud-Meynard A , Baud J , Hiriart JB , Makki K , Dugot-Senant N , Staedel C , Bioulac-Sage P , Zucman-Rossi J , et al. Rnd3/RhoE Is down-regulated in hepatocellular carcinoma and controls cellular invasion . Hepatology 2012 ; 55 :1766 -75 ; PMID:22234932; http://dx.doi.org/10.1002/hep.25568 22234932 [42] Luo H , Dong Z , Zou J , Zeng Q , Wu D , Liu L . Down-regulation of RhoE is associated with progression and poor prognosis in hepatocellular carcinoma . J Surg Oncol 2012 ; 105 :699 -704 ; PMID:22213123; http://dx.doi.org/10.1002/jso.23019 22213123 [43] Zhou J , Li K , Gu Y , Feng B , Ren G , Zhang L , Wang Y , Nie Y , Fan D . Transcriptional up-regulation of RhoE by hypoxia-inducible factor (HIF)-1 promotes epithelial to mesenchymal transition of gastric cancer cells during hypoxia . Biochem Biophys Res Commun 2011 ; 415 :348 -54 ; PMID:22037464; http://dx.doi.org/10.1016/j.bbrc.2011.10.065 22037464 [44] Vigil D , Cherfils J , Rossman KL , Der CJ . Ras superfamily GEFs and GAPs: validated and tractable targets for cancer therapy? Nat Rev Cancer 2010 ; 10 :842 -57 ; PMID:21102635; http://dx.doi.org/10.1038/nrc2960 21102635 [45] Barrio-Real L , Kazanietz MG . Rho GEFs and cancer: linking gene expression and metastatic dissemination . Sci Signal 2012 ; 5 :pe43 ; PMID:23033535; http://dx.doi.org/10.1126/scisignal.2003543 23033535 [46] Cook DR , Rossman KL , Der CJ . Rho guanine nucleotide exchange factors: regulators of Rho GTPase activity in development and disease . Oncogene 2014 ; 33 :4021 -35 ; PMID:24037532; http://dx.doi.org/10.1038/onc.2013.362 24037532 [47] Fields AP , Justilien V . The guanine nucleotide exchange factor (GEF) Ect2 is an oncogene in human cancer . Adv Enzyme Regul 2010 ; 50 :190 -200 ; PMID:19896966; http://dx.doi.org/10.1016/j.advenzreg.2009.10.010 19896966 [48] Wu D , Asiedu M , Wei Q . Myosin-interacting guanine exchange factor (MyoGEF) regulates the invasion activity of MDA-MB-231 breast cancer cells through activation of RhoA and RhoC . Oncogene 2009 ; 28 :2219 -30 ; PMID:19421144; http://dx.doi.org/10.1038/onc.2009.96 19421144 [49] Qin J , Xie Y , Wang B , Hoshino M , Wolff DW , Zhao J , Scofield MA , Dowd FJ , Lin MF , Tu Y . Upregulation of PIP3-dependent Rac exchanger 1 (P-Rex1) promotes prostate cancer metastasis . Oncogene 2009 ; 28 :1853 -63 ; PMID:19305425; http://dx.doi.org/10.1038/onc.2009.30 19305425 [50] Berger MF , Hodis E , Heffernan TP , Deribe YL , Lawrence MS , Protopopov A , Ivanova E , Watson IR , Nickerson E , Ghosh P , et al. Melanoma genome sequencing reveals frequent PREX2 mutations . Nature 2012 ; 485 :502 -6 ; PMID:2262257822622578 [51] Habets GG , Scholtes EH , Zuydgeest D , van der Kammen RA , Stam JC , Berns A , Collard JG . Identification of an invasion-inducing gene, Tiam-1, that encodes a protein with homology to GDP-GTP exchangers for Rho-like proteins . Cell 1994 ; 77 :537 -49 ; PMID:7999144; http://dx.doi.org/10.1016/0092-8674(94)90216-X 7999144 [52] Adam L , Vadlamudi RK , McCrea P , Kumar R . Tiam1 overexpression potentiates heregulin-induced lymphoid enhancer factor-1/beta -catenin nuclear signaling in breast cancer cells by modulating the intercellular stability . J Biol Chem 2001 ; 276 :28443 -50 ; PMID:11328805; http://dx.doi.org/10.1074/jbc.M009769200 11328805 [53] Lane J , Martin TA , Mansel RE , Jiang WG . The expression and prognostic value of the guanine nucleotide exchange factors (GEFs) Trio, Vav1 and TIAM-1 in human breast cancer . Int Semin Surg Oncol 2008 ; 5 :23 ; PMID:18925966; http://dx.doi.org/10.1186/1477-7800-5-23 18925966 [54] Liu S , Li Y , Qi W , Zhao Y , Huang A , Sheng W , Lei B , Lin P , Zhu H , Li W , et al. Expression of Tiam1 predicts lymph node metastasis and poor survival of lung adenocarcinoma patients . Diagn Pathol 2014 ; 9 :69 ; PMID:24661909; http://dx.doi.org/10.1186/1746-1596-9-69 24661909 [55] Vaughan L , Tan CT , Chapman A , Nonaka D , Mack NA , Smith D , Booton R , Hurlstone AF , Malliri A . HUWE1 ubiquitylates and degrades the RAC activator TIAM1 promoting cell-cell adhesion disassembly, migration, and invasion . Cell Rep 2015 ; 10 :88 -102 ; PMID:25543140; http://dx.doi.org/10.1016/j.celrep.2014.12.012 25543140 [56] Chen JS , Su IJ , Leu YW , Young KC , Sun HS . Expression of T-cell lymphoma invasion and metastasis 2 (TIAM2) promotes proliferation and invasion of liver cancer . Int J Cancer 2012 ; 130 :1302 -13 ; PMID:21469146; http://dx.doi.org/10.1002/ijc.26117 21469146 [57] Ahn SJ , Chung KW , Lee RA , Park IA , Lee SH , Park DE , Noh DY . Overexpression of betaPix-a in human breast cancer tissues . Cancer Lett 2003 ; 193 :99 -107 ; PMID:12691829; http://dx.doi.org/10.1016/S0304-3835(03)00004-1 12691829 [58] Fernandez-Zapico ME , Gonzalez-Paz NC , Weiss E , Savoy DN , Molina JR , Fonseca R , Smyrk TC , Chari ST , Urrutia R , Billadeau DD . Ectopic expression of VAV1 reveals an unexpected role in pancreatic cancer tumorigenesis . Cancer Cell 2005 ; 7 :39 -49 ; PMID:15652748; http://dx.doi.org/10.1016/j.ccr.2004.11.024 15652748 [59] Hornstein I , Pikarsky E , Groysman M , Amir G , Peylan-Ramu N , Katzav S . The haematopoietic specific signal transducer Vav1 is expressed in a subset of human neuroblastomas . J Pathol 2003 ; 199 :526 -33 ; PMID:12635144; http://dx.doi.org/10.1002/path.1314 12635144 [60] Lin KY , Wang LH , Hseu YC , Fang CL , Yang HL , Kumar KJ , Tai C , Uen YH . Clinical significance of increased guanine nucleotide exchange factor Vav3 expression in human gastric cancer . Mol Cancer Res 2012 ; 10 :750 -9 ; PMID:22544459; http://dx.doi.org/10.1158/1541-7786.MCR-11-0598-T 22544459 [61] Rao S , Lyons LS , Fahrenholtz CD , Wu F , Farooq A , Balkan W , Burnstein KL . A novel nuclear role for the Vav3 nucleotide exchange factor in androgen receptor coactivation in prostate cancer . Oncogene 2012 ; 31 :716 -27 ; PMID:21765461; http://dx.doi.org/10.1038/onc.2011.273 21765461 [62] Citterio C , Menacho-Marquez M , Garcia-Escudero R , Larive RM , Barreiro O , Sanchez-Madrid F , Paramio JM , Bustelo XR . The rho exchange factors vav2 and vav3 control a lung metastasis-specific transcriptional program in breast cancer cells . Sci Signal 2012 ; 5 :ra71 ; PMID:23033540; http://dx.doi.org/10.1126/scisignal.2002962 23033540 [63] Jarzynka MJ , Hu B , Hui KM , Bar-Joseph I , Gu W , Hirose T , Haney LB , Ravichandran KS , Nishikawa R , Cheng SY . ELMO1 and Dock180, a bipartite Rac1 guanine nucleotide exchange factor, promote human glioma cell invasion . Cancer Res 2007 ; 67 :7203 -11 ; PMID:17671188; http://dx.doi.org/10.1158/0008-5472.CAN-07-0473 17671188 [64] Kourlas PJ , Strout MP , Becknell B , Veronese ML , Croce CM , Theil KS , Krahe R , Ruutu T , Knuutila S , Bloomfield CD , et al. Identification of a gene at 11q23 encoding a guanine nucleotide exchange factor: evidence for its fusion with MLL in acute myeloid leukemia . Proc Natl Acad Sci U S A 2000 ; 97 :2145 -50 ; PMID:10681437; http://dx.doi.org/10.1073/pnas.040569197 10681437 [65] Malliri A , van der Kammen RA , Clark K , van der Valk M , Michiels F , Collard JG . Mice deficient in the Rac activator Tiam1 are resistant to Ras-induced skin tumours . Nature 2002 ; 417 :867 -71 ; PMID:12075356; http://dx.doi.org/10.1038/nature00848 12075356 [66] Malliri A , Rygiel TP , van der Kammen RA , Song JY , Engers R , Hurlstone AF , Clevers H , Collard JG . The rac activator Tiam1 is a Wnt-responsive gene that modifies intestinal tumor development . J Biol Chem 2006 ; 281 :543 -8 ; PMID:16249175; http://dx.doi.org/10.1074/jbc.M507582200 16249175 [67] Lindsay CR , Lawn S , Campbell AD , Faller WJ , Rambow F , Mort RL , Timpson P , Li A , Cammareri P , Ridgway RA , et al. P-Rex1 is required for efficient melanoblast migration and melanoma metastasis . Nat Commun 2011 ; 2 :555 ; PMID:22109529; http://dx.doi.org/10.1038/ncomms1560 22109529 [68] Lissanu Deribe Y , Shi Y , Rai K , Nezi L , Amin SB , Wu CC , Akdemir KC , Mahdavi M , Peng Q , Chang QE , et al. Truncating PREX2 mutations activate its GEF activity and alter gene expression regulation in NRAS-mutant melanoma . Proc Natl Acad Sci U S A 2016 ; 113 :E1296 -305 ; PMID:26884185; http://dx.doi.org/10.1073/pnas.1513801113 26884185 [69] Kawasaki Y , Tsuji S , Muroya K , Furukawa S , Shibata Y , Okuno M , Ohwada S , Akiyama T . The adenomatous polyposis coli-associated exchange factors Asef and Asef2 are required for adenoma formation in Apc(Min/+)mice . EMBO Rep 2009 ; 10 :1355 -62 ; PMID:19893577; http://dx.doi.org/10.1038/embor.2009.233 19893577 [70] Chang KH , Sanchez-Aguilera A , Shen S , Sengupta A , Madhu MN , Ficker AM , Dunn SK , Kuenzi AM , Arnett JL , Santho RA , et al. Vav3 collaborates with p190-BCR-ABL in lymphoid progenitor leukemogenesis, proliferation, and survival . Blood 2012 ; 120 :800 -11 ; PMID:22692505; http://dx.doi.org/10.1182/blood-2011-06-361709 22692505 [71] Menacho-Marquez M , Garcia-Escudero R , Ojeda V , Abad A , Delgado P , Costa C , Ruiz S , Alarcon B , Paramio JM , Bustelo XR . The Rho exchange factors Vav2 and Vav3 favor skin tumor initiation and promotion by engaging extracellular signaling loops . PLoS Biol 2013 ; 11 :e1001615 ; PMID:23935450; http://dx.doi.org/10.1371/journal.pbio.1001615 23935450 [72] Xue W , Krasnitz A , Lucito R , Sordella R , Vanaelst L , Cordon-Cardo C , Singer S , Kuehnel F , Wigler M , Powers S , et al. DLC1 is a chromosome 8p tumor suppressor whose loss promotes hepatocellular carcinoma . Genes Dev 2008 ; 22 :1439 -44 ; PMID:18519636; http://dx.doi.org/10.1101/gad.1672608 18519636 [73] Ching YP , Wong CM , Chan SF , Leung TH , Ng DC , Jin DY , Ng IO . Deleted in liver cancer (DLC) 2 encodes a RhoGAP protein with growth suppressor function and is underexpressed in hepatocellular carcinoma . J Biol Chem 2003 ; 278 :10824 -30 ; PMID:12531887; http://dx.doi.org/10.1074/jbc.M208310200 12531887 [74] Vitiello E , Ferreira JG , Maiato H , Balda MS , Matter K . The tumour suppressor DLC2 ensures mitotic fidelity by coordinating spindle positioning and cell-cell adhesion . Nat Commun 2014 ; 5 :5826 ; PMID:25518808; http://dx.doi.org/10.1038/ncomms6826 25518808 [75] Wolf RM , Draghi N , Liang X , Dai C , Uhrbom L , Eklof C , Westermark B , Holland EC , Resh MD . p190RhoGAP can act to inhibit PDGF-induced gliomas in mice: a putative tumor suppressor encoded on human chromosome 19q13.3 . Genes Dev 2003 ; 17 :476 -87 ; PMID:12600941; http://dx.doi.org/10.1101/gad.1040003 12600941 [76] Johnstone CN , Castellvi-Bel S , Chang LM , Bessa X , Nakagawa H , Harada H , Sung RK , Pique JM , Castells A , Rustgi AK . ARHGAP8 is a novel member of the RHOGAP family related to ARHGAP1/CDC42GAP/p50RHOGAP: mutation and expression analyses in colorectal and breast cancers . Gene 2004 ; 336 :59 -71 ; PMID:15225876; http://dx.doi.org/10.1016/j.gene.2004.01.025 15225876 [77] Jiang WG , Watkins G , Lane J , Cunnick GH , Douglas-Jones A , Mokbel K , Mansel RE . Prognostic value of rho GTPases and rho guanine nucleotide dissociation inhibitors in human breast cancers . Clin Cancer Res 2003 ; 9 :6432 -40 ; PMID:1469514514695145 [78] Fritz G , Lang P , Just I . Tissue-specific variations in the expression and regulation of the small GTP-binding protein Rho . Biochim Biophys Acta 1994 ; 1222 :331 -8 ; PMID:8038201; http://dx.doi.org/10.1016/0167-4889(94)90038-8 8038201 [79] Theodorescu D , Sapinoso LM , Conaway MR , Oxford G , Hampton GM , Frierson HF Jr . Reduced expression of metastasis suppressor RhoGDI2 is associated with decreased survival for patients with bladder cancer . Clin Cancer Res 2004 ; 10 :3800 -6 ; PMID:15173088; http://dx.doi.org/10.1158/1078-0432.CCR-03-0653 15173088 [80] Abiatari I , DeOliveira T , Kerkadze V , Schwager C , Esposito I , Giese NA , Huber P , Bergman F , Abdollahi A , Friess H , et al. Consensus transcriptome signature of perineural invasion in pancreatic carcinoma . Mol Cancer Ther 2009 ; 8 :1494 -504 ; PMID:19509238; http://dx.doi.org/10.1158/1535-7163.MCT-08-0755 19509238 [81] Sahai E , Garcia-Medina R , Pouyssegur J , Vial E . Smurf1 regulates tumor cell plasticity and motility through degradation of RhoA leading to localized inhibition of contractility . J Cell Biol 2007 ; 176 :35 -42 ; PMID:17190792; http://dx.doi.org/10.1083/jcb.200605135 17190792 [82] Castillo-Lluva S , Tan CT , Daugaard M , Sorensen PH , Malliri A . The tumour suppressor HACE1 controls cell migration by regulating Rac1 degradation . Oncogene 2013 ; 32 :1735 -42 ; PMID:22614015; http://dx.doi.org/10.1038/onc.2012.189 22614015 [83] Wei J , Mialki RK , Dong S , Khoo A , Mallampalli RK , Zhao Y , Zhao J . A new mechanism of RhoA ubiquitination and degradation: roles of SCF(FBXL19) E3 ligase and Erk2 . Biochim Biophys Acta 2013 ; 1833 :2757 -64 ; PMID:23871831; http://dx.doi.org/10.1016/j.bbamcr.2013.07.005 23871831 [84] Dong S , Zhao J , Wei J , Bowser RK , Khoo A , Liu Z , Luketich JD , Pennathur A , Ma H , Zhao Y . F-box protein complex FBXL19 regulates TGFbeta1-induced E-cadherin down-regulation by mediating Rac3 ubiquitination and degradation . Mol Cancer 2014 ; 13 :76 ; PMID:24684802; http://dx.doi.org/10.1186/1476-4598-13-76 24684802 [85] Tu S , Wu WJ , Wang J , Cerione RA . Epidermal growth factor-dependent regulation of Cdc42 is mediated by the Src tyrosine kinase . J Biol Chem 2003 ; 278 :49293 -300 ; PMID:14506284; http://dx.doi.org/10.1074/jbc.M307021200 14506284 [86] Preudhomme C , Roumier C , Hildebrand MP , Dallery-Prudhomme E , Lantoine D , Lai JL , Daudignon A , Adenis C , Bauters F , Fenaux P , et al. Nonrandom 4p13 rearrangements of the RhoH/TTF gene, encoding a GTP-binding protein, in non-Hodgkin's lymphoma and multiple myeloma . Oncogene 2000 ; 19 :2023 -32 ; PMID:10803463; http://dx.doi.org/10.1038/sj.onc.1203521 10803463 [87] Pasqualucci L , Neumeister P , Goossens T , Nanjangud G , Chaganti RS , Kuppers R , Dalla-Favera R . Hypermutation of multiple proto-oncogenes in B-cell diffuse large-cell lymphomas . Nature 2001 ; 412 :341 -6 ; PMID:11460166; http://dx.doi.org/10.1038/35085588 11460166 [88] Alan JK , Lundquist EA . Mutationally activated Rho GTPases in cancer . Small GTPases 2013 ; 4 :159 -63 ; PMID:24088985; http://dx.doi.org/10.4161/sgtp.26530 24088985 [89] Cerami E , Gao J , Dogrusoz U , Gross BE , Sumer SO , Aksoy BA , Jacobsen A , Byrne CJ , Heuer ML , Larsson E , et al. The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data . Cancer Discov 2012 ; 2 :401 -4 ; PMID:22588877; http://dx.doi.org/10.1158/2159-8290.CD-12-0095 22588877 [90] Gao J , Aksoy BA , Dogrusoz U , Dresdner G , Gross B , Sumer SO , Sun Y , Jacobsen A , Sinha R , Larsson E , et al. Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal . Sci Signal 2013 ; 6 :pl1 ; PMID:23550210; http://dx.doi.org/10.1126/scisignal.2004088 23550210 [91] Gonzalez-Perez A , Perez-Llamas C , Deu-Pons J , Tamborero D , Schroeder MP , Jene-Sanz A , Santos A , Lopez-Bigas N . IntOGen-mutations identifies cancer drivers across tumor types . Nat Methods 2013 ; 10 :1081 -2 ; PMID:24037244; http://dx.doi.org/10.1038/nmeth.2642 24037244 [92] Hwang SL , Hong YR , Sy WD , Lieu AS , Lin CL , Lee KS , Howng SL . Rac1 gene mutations in human brain tumours . Eur J Surg Oncol 2004 ; 30 :68 -72 ; PMID:14736526; http://dx.doi.org/10.1016/j.ejso.2003.10.018 14736526 [93] Hodis E , Watson IR , Kryukov GV , Arold ST , Imielinski M , Theurillat JP , Nickerson E , Auclair D , Li L , Place C , et al. A landscape of driver mutations in melanoma . Cell 2012 ; 150 :251 -63 ; PMID:22817889; http://dx.doi.org/10.1016/j.cell.2012.06.024 22817889 [94] Krauthammer M , Kong Y , Ha BH , Evans P , Bacchiocchi A , McCusker JP , Cheng E , Davis MJ , Goh G , Choi M , et al. Exome sequencing identifies recurrent somatic RAC1 mutations in melanoma . Nat Genet 2012 ; 44 :1006 -14 ; PMID:22842228; http://dx.doi.org/10.1038/ng.2359 22842228 [95] Davis MJ , Ha BH , Holman EC , Halaban R , Schlessinger J , Boggon TJ . RAC1P29S is a spontaneously activating cancer-associated GTPase . Proc Natl Acad Sci U S A 2013 ; 110 :912 -7 ; PMID:23284172; http://dx.doi.org/10.1073/pnas.1220895110 23284172 [96] Kawazu M , Ueno T , Kontani K , Ogita Y , Ando M , Fukumura K , Yamato A , Soda M , Takeuchi K , Miki Y , et al. Transforming mutations of RAC guanosine triphosphatases in human cancers . Proc Natl Acad Sci U S A 2013 ; 110 :3029 -34 ; PMID:23382236; http://dx.doi.org/10.1073/pnas.1216141110 23382236 [97] Watson IR , Li L , Cabeceiras PK , Mahdavi M , Gutschner T , Genovese G , Wang G , Fang Z , Tepper JM , Stemke-Hale K , et al. The RAC1 P29S hotspot mutation in melanoma confers resistance to pharmacological inhibition of RAF . Cancer Res 2014 ; 74 :4845 -52 ; PMID:25056119; http://dx.doi.org/10.1158/0008-5472.CAN-14-1232-T 25056119 [98] Van Allen EM , Wagle N , Sucker A , Treacy DJ , Johannessen CM , Goetz EM , Place CS , Taylor-Weiner A , Whittaker S , Kryukov GV , et al. The genetic landscape of clinical resistance to RAF inhibition in metastatic melanoma . Cancer Discov 2014 ; 4 :94 -109 ; PMID:24265153; http://dx.doi.org/10.1158/2159-8290.CD-13-0617 24265153 [99] Mar VJ , Wong SQ , Logan A , Nguyen T , Cebon J , Kelly JW , Wolfe R , Dobrovic A , McLean C , McArthur GA . Clinical and pathological associations of the activating RAC1 P29S mutation in primary cutaneous melanoma . Pigment Cell Melanoma Res 2014 ; 27 :1117 -25 ; PMID:25043693; http://dx.doi.org/10.1111/pcmr.12295 25043693 [100] Vu HL , Rosenbaum S , Purwin TJ , Davies MA , Aplin AE . RAC1 P29S regulates PD-L1 expression in melanoma . Pigment Cell Melanoma Res 2015 ; 28 :590 -8 ; PMID:26176707; http://dx.doi.org/10.1111/pcmr.12392 26176707 [101] Thomas EK , Cancelas JA , Chae HD , Cox AD , Keller PJ , Perrotti D , Neviani P , Druker BJ , Setchell KD , Zheng Y , et al. Rac guanosine triphosphatases represent integrating molecular therapeutic targets for BCR-ABL-induced myeloproliferative disease . Cancer Cell 2007 ; 12 :467 -78 ; PMID:17996650; http://dx.doi.org/10.1016/j.ccr.2007.10.015 17996650 [102] Cho YJ , Zhang B , Kaartinen V , Haataja L , de Curtis I , Groffen J , Heisterkamp N . Generation of rac3 null mutant mice: role of Rac3 in Bcr/Abl-caused lymphoblastic leukemia . Mol Cell Biol 2005 ; 25 :5777 -85 ; PMID:15964830; http://dx.doi.org/10.1128/MCB.25.13.5777-5785.2005 15964830 [103] Cancer Genome Atlas Research N . Comprehensive molecular characterization of gastric adenocarcinoma . Nature 2014 ; 513 :202 -9 ; PMID:25079317; http://dx.doi.org/10.1038/nature13480 25079317 [104] Kakiuchi M , Nishizawa T , Ueda H , Gotoh K , Tanaka A , Hayashi A , Yamamoto S , Tatsuno K , Katoh H , Watanabe Y , et al. Recurrent gain-of-function mutations of RHOA in diffuse-type gastric carcinoma . Nat Genet 2014 ; 46 :583 -7 ; PMID:24816255; http://dx.doi.org/10.1038/ng.2984 24816255 [105] Palomero T , Couronne L , Khiabanian H , Kim MY , Ambesi-Impiombato A , Perez-Garcia A , Carpenter Z , Abate F , Allegretta M , Haydu JE , et al. Recurrent mutations in epigenetic regulators, RHOA and FYN kinase in peripheral T cell lymphomas . Nat Genet 2014 ; 46 :166 -70 ; PMID:24413734; http://dx.doi.org/10.1038/ng.2873 24413734 [106] Yoo HY , Sung MK , Lee SH , Kim S , Lee H , Park S , Kim SC , Lee B , Rho K , Lee JE , et al. A recurrent inactivating mutation in RHOA GTPase in angioimmunoblastic T cell lymphoma . Nat Genet 2014 ; 46 :371 -5 ; PMID:24584070; http://dx.doi.org/10.1038/ng.2916 24584070 [107] Manso R , Sanchez-Beato M , Monsalvo S , Gomez S , Cereceda L , Llamas P , Rojo F , Mollejo M , Menarguez J , Alves J , et al. The RHOA G17V gene mutation occurs frequently in peripheral T-cell lymphoma and is associated with a characteristic molecular signature . Blood 2014 ; 123 :2893 -4 ; PMID:24786457; http://dx.doi.org/10.1182/blood-2014-02-555946 24786457 [108] Rodrigues P , Macaya I , Bazzocco S , Mazzolini R , Andretta E , Dopeso H , Mateo-Lozano S , Bilic J , Carton-Garcia F , Nieto R , et al. RHOA inactivation enhances Wnt signalling and promotes colorectal cancer . Nat Commun 2014 ; 5 :5458 ; PMID:25413277; http://dx.doi.org/10.1038/ncomms6458 25413277 [109] Richter J , Schlesner M , Hoffmann S , Kreuz M , Leich E , Burkhardt B , Rosolowski M , Ammerpohl O , Wagener R , Bernhart SH , et al. Recurrent mutation of the ID3 gene in Burkitt lymphoma identified by integrated genome, exome and transcriptome sequencing . Nat Genet 2012 ; 44 :1316 -20 ; PMID:23143595; http://dx.doi.org/10.1038/ng.2469 23143595 [110] Rohde M , Richter J , Schlesner M , Betts MJ , Claviez A , Bonn BR , Zimmermann M , Damm-Welk C , Russell RB , Borkhardt A , et al. Recurrent RHOA mutations in pediatric Burkitt lymphoma treated according to the NHL-BFM protocols . Genes Chromosomes Cancer 2014 ; 53 :911 -6 ; PMID:25044415; http://dx.doi.org/10.1002/gcc.22202 25044415 [111] Sahai E , Alberts AS , Treisman R . RhoA effector mutants reveal distinct effector pathways for cytoskeletal reorganization, SRF activation and transformation . EMBO J 1998 ; 17 :1350 -61 ; PMID:9482732; http://dx.doi.org/10.1093/emboj/17.5.1350 9482732 [112] Cleverley SC , Costello PS , Henning SW , Cantrell DA . Loss of Rho function in the thymus is accompanied by the development of thymic lymphoma . Oncogene 2000 ; 19 :13 -20 ; PMID:10644975; http://dx.doi.org/10.1038/sj.onc.1203259 10644975 [113] Cancer Genome Atlas Research N . Comprehensive molecular characterization of urothelial bladder carcinoma . Nature 2014 ; 507 :315 -22 ; PMID:24476821; http://dx.doi.org/10.1038/nature12965 24476821 [114] Liu AX , Rane N , Liu JP , Prendergast GC . RhoB is dispensable for mouse development, but it modifies susceptibility to tumor formation as well as cell adhesion and growth factor signaling in transformed cells . Mol Cell Biol 2001 ; 21 :6906 -12 ; PMID:11564874; http://dx.doi.org/10.1128/MCB.21.20.6906-6912.2001 11564874 [115] Hakem A , Sanchez-Sweatman O , You-Ten A , Duncan G , Wakeham A , Khokha R , Mak TW . RhoC is dispensable for embryogenesis and tumor initiation but essential for metastasis . Genes Dev 2005 ; 19 :1974 -9 ; PMID:16107613; http://dx.doi.org/10.1101/gad.1310805 16107613 [116] van Hengel J , D'Hooge P , Hooghe B , Wu X , Libbrecht L , De Vos R , Quondamatteo F , Klempt M , Brakebusch C , van Roy F . Continuous cell injury promotes hepatic tumorigenesis in cdc42-deficient mouse liver . Gastroenterology 2008 ; 134 :781 -92 ; PMID:18325391; http://dx.doi.org/10.1053/j.gastro.2008.01.002 18325391 [117] Gao Y , Dickerson JB , Guo F , Zheng J , Zheng Y . Rational design and characterization of a Rac GTPase-specific small molecule inhibitor . Proc Natl Acad Sci U S A 2004 ; 101 :7618 -23 ; PMID:15128949; http://dx.doi.org/10.1073/pnas.0307512101 15128949 [118] Onesto C , Shutes A , Picard V , Schweighoffer F , Der CJ . Characterization of EHT 1864, a novel small molecule inhibitor of Rac family small GTPases . Methods Enzymol 2008 ; 439 :111 -29 ; PMID:18374160; http://dx.doi.org/10.1016/S0076-6879(07)00409-0 18374160 [119] Kaneto N , Yokoyama S , Hayakawa Y , Kato S , Sakurai H , Saiki I . RAC1 inhibition as a therapeutic target for gefitinib-resistant non-small-cell lung cancer . Cancer Sci 2014 ; 105 :788 -94 ; PMID:24750242; http://dx.doi.org/10.1111/cas.12425 24750242 [120] Dutting S , Heidenreich J , Cherpokova D , Amin E , Zhang SC , Ahmadian MR , Brakebusch C , Nieswandt B . Critical off-target effects of the widely used Rac1 inhibitors NSC23766 and EHT1864 in mouse platelets . J Thromb Haemost 2015 ; 13 :827 -38 ; PMID:25628054; http://dx.doi.org/10.1111/jth.12861 25628054 [121] Cardama GA , Comin MJ , Hornos L , Gonzalez N , Defelipe L , Turjanski AG , Alonso DF , Gomez DE , Menna PL . Preclinical development of novel Rac1-GEF signaling inhibitors using a rational design approach in highly aggressive breast cancer cell lines . Anticancer Agents Med Chem 2014 ; 14 :840 -51 ; PMID:24066799; http://dx.doi.org/10.2174/18715206113136660334 24066799 [122] Fine B , Hodakoski C , Koujak S , Su T , Saal LH , Maurer M , Hopkins B , Keniry M , Sulis ML , Mense S , et al. Activation of the PI3K pathway in cancer through inhibition of PTEN by exchange factor P-REX2a . Science 2009 ; 325 :1261 -5 ; PMID:19729658; http://dx.doi.org/10.1126/science.1173569 19729658 [123] Guo Y , Kenney SR , Cook L , Adams SF , Rutledge T , Romero E , Oprea TI , Sklar LA , Bedrick E , Wiggins CL , et al. A Novel Pharmacologic Activity of Ketorolac for Therapeutic Benefit in Ovarian Cancer Patients . Clin Cancer Res 2015 ; 21 :5064 -72 ; PMID:26071482; http://dx.doi.org/10.1158/1078-0432.CCR-15-0461 26071482 [124] Guo Y , Kenney SR , Muller CY , Adams S , Rutledge T , Romero E , Murray-Krezan C , Prekeris R , Sklar LA , Hudson LG , et al. R-Ketorolac Targets Cdc42 and Rac1 and Alters Ovarian Cancer Cell Behaviors Critical for Invasion and Metastasis . Mol Cancer Ther 2015 ; 14 :2215 -27 ; PMID:26206334; http://dx.doi.org/10.1158/1535-7163.MCT-15-0419 26206334 [125] Zins K , Lucas T , Reichl P , Abraham D , Aharinejad S . A Rac1/Cdc42 GTPase-specific small molecule inhibitor suppresses growth of primary human prostate cancer xenografts and prolongs survival in mice . PLoS One 2013 ; 8 :e74924 ; PMID:24040362; http://dx.doi.org/10.1371/journal.pone.0074924 24040362 [126] Pelish HE , Peterson JR , Salvarezza SB , Rodriguez-Boulan E , Chen JL , Stamnes M , Macia E , Feng Y , Shair MD , Kirchhausen T . Secramine inhibits Cdc42-dependent functions in cells and Cdc42 activation in vitro . Nat Chem Biol 2006 ; 2 :39 -46 ; PMID:16408091; http://dx.doi.org/10.1038/nchembio751 16408091 [127] Friesland A , Zhao Y , Chen YH , Wang L , Zhou H , Lu Q . Small molecule targeting Cdc42-intersectin interaction disrupts Golgi organization and suppresses cell motility . Proc Natl Acad Sci U S A 2013 ; 110 :1261 -6 ; PMID:23284167; http://dx.doi.org/10.1073/pnas.1116051110 23284167 [128] Zins K , Gunawardhana S , Lucas T , Abraham D , Aharinejad S . Targeting Cdc42 with the small molecule drug AZA197 suppresses primary colon cancer growth and prolongs survival in a preclinical mouse xenograft model by downregulation of PAK1 activity . J Transl Med 2013 ; 11 :295 ; PMID:24279335; http://dx.doi.org/10.1186/1479-5876-11-295 24279335 [129] Rath N , Olson MF . Rho-associated kinases in tumorigenesis: re-considering ROCK inhibition for cancer therapy . EMBO Rep 2012 ; 13 :900 -8 ; PMID:22964758; http://dx.doi.org/10.1038/embor.2012.127 22964758 [130] Somlyo AV , Bradshaw D , Ramos S , Murphy C , Myers CE , Somlyo AP . Rho-kinase inhibitor retards migration and in vivo dissemination of human prostate cancer cells . Biochem Biophys Res Commun 2000 ; 269 :652 -9 ; PMID:10720471; http://dx.doi.org/10.1006/bbrc.2000.2343 10720471 [131] Itoh K , Yoshioka K , Akedo H , Uehata M , Ishizaki T , Narumiya S . An essential part for Rho-associated kinase in the transcellular invasion of tumor cells . Nat Med 1999 ; 5 :221 -5 ; PMID:9930872; http://dx.doi.org/10.1038/5587 9930872 [132] Kumar MS , Hancock DC , Molina-Arcas M , Steckel M , East P , Diefenbacher M , Armenteros-Monterroso E , Lassailly F , Matthews N , Nye E , et al. The GATA2 transcriptional network is requisite for RAS oncogene-driven non-small cell lung cancer . Cell 2012 ; 149 :642 -55 ; PMID:22541434; http://dx.doi.org/10.1016/j.cell.2012.02.059 22541434

PMC005xxxxxx/PMC5003690.txt

==== Front Biochem JBiochem. JppbiochemjBJBiochemical Journal0264-60211470-8728Portland Press Ltd. BCJ2016000210.1042/BCJ20160002Review ArticlesReview Article10244049The emerging role of AMPK in the regulation of breathing and oxygen supply AMPK and oxygen supplyA.M. Evans and othersEvans A. Mark *1Mahmoud Amira D. *Moral-Sanz Javier *Hartmann Sandy ** Centre for Integrative Physiology, College of Medicine and Veterinary Medicine, Hugh Robson Building, University of Edinburgh, Edinburgh EH8 9XD, U.K.1 To whom correspondence should be addressed (email mark.evans@ed.ac.uk).30 8 2016 1 9 2016 473 17 172561 2572 18 1 2016 20 4 2016 3 5 2016 © 2016 The Author(s)2016This is an open access article published by Portland Press Limited on behalf of the Biochemical Society and distributed under the Creative Commons Attribution Licence 4.0 (CC BY).Regulation of breathing is critical to our capacity to accommodate deficits in oxygen availability and demand during, for example, sleep and ascent to altitude. It is generally accepted that a fall in arterial oxygen increases afferent discharge from the carotid bodies to the brainstem and thus delivers increased ventilatory drive, which restores oxygen supply and protects against hypoventilation and apnoea. However, the precise molecular mechanisms involved remain unclear. We recently identified as critical to this process the AMP-activated protein kinase (AMPK), which is key to the cell-autonomous regulation of metabolic homoeostasis. This observation is significant for many reasons, not least because recent studies suggest that the gene for the AMPK-α1 catalytic subunit has been subjected to natural selection in high-altitude populations. It would appear, therefore, that evolutionary pressures have led to AMPK being utilized to regulate oxygen delivery and thus energy supply to the body in the short, medium and longer term. Contrary to current consensus, however, our findings suggest that AMPK regulates ventilation at the level of the caudal brainstem, even when afferent input responses from the carotid body are normal. We therefore hypothesize that AMPK integrates local hypoxic stress at defined loci within the brainstem respiratory network with an index of peripheral hypoxic status, namely afferent chemosensory inputs. Allied to this, AMPK is critical to the control of hypoxic pulmonary vasoconstriction and thus ventilation–perfusion matching at the lungs and may also determine oxygen supply to the foetus by, for example, modulating utero-placental blood flow. AMP-activated protein kinase (AMPK)apnoeaCa2+–calmodulin-activated kinase kinase-β (CaMKK-β)hypoxialiver kinase B1 (LKB1)pulmonaryventilation ==== Body INTRODUCTION Regulated oxygen supply is key to the maintenance of oxidative phosphorylation and thus cellular energy status in mammals, not least because of the limited capacity for cellular oxygen storage relative to the extensive reserves of other substrates. It was proposed, therefore, that natural selection may have employed AMP-activated protein kinase (AMPK) to co-ordinate system-level adjustments of whole-body function in response to oxygen deficits in animals [1]. Consistent with this view, recent studies on high-altitude Andean populations have shown that the gene for the AMPK-α1 subunit (PRKAA1) has been influenced by natural selection through single nucleotide polymorphisms [2]. Confirmation of a role for AMPK in oxygen delivery has now been provided by conclusive experimental evidence that, in addition to its well-recognized capacity as a regulator of cell-autonomous pathways of energy supply [3], AMPK is essential to the regulation of breathing during hypoxia and thus oxygen and energy distribution to the body [4]. FRAGMENTS FROM THE LIBRARIES OF BABYLON The AMP-activated protein kinase AMPK is a cellular energy sensor that acts to maintain energy homoeostasis. It exists as heterotrimers comprising one of two catalytic α subunits, in combination with one each of the two β and three γ regulatory subunits, which together may form at least 12 different heterotrimeric subunit combinations [5,6]. In this respect it is important to note that evidence is now emerging to suggest that different subunit combinations may be selected by a given cell type, that each combination may exhibit different sensitivities to activation by AMP and ADP and thus metabolic stresses, and that each may selectively phosphorylate and regulate a different spectrum of target proteins [8]. AMPK activities are exquisitely coupled to mitochondrial metabolism through changes in the cellular AMP/ATP and ADP/ATP ratios (Figure 1). There are four nucleotide-binding sites (CBS repeats) on the γ subunit, of which only sites designated 1, 3 and 4 may ever be occupied [8]. Binding of AMP to the γ subunit causes a 10-fold increase in AMPK activity by allosteric activation, with further activation of up to 100-fold generated by binding of either AMP or ADP through their promotion of phosphorylation and inhibition of dephosphorylation at Thr172 on the α subunit; each of these effects is opposed by ATP [9,10]. Thr172 is primarily phosphorylated by the tumour-suppressor kinase liver kinase B1 (LKB1), which appears to be constitutively active but phosphorylates AMPK more rapidly when AM(D)P is bound to the γ subunit [11]. There is also an alternative Ca2+-dependent activation mechanism, the calmodulin-dependent protein kinase Ca2+–calmodulin-activated kinase kinase-β (CaMKK-β), which phosphorylates Thr172 and thus activates AMPK in an AMP-independent manner [5,6,12]. Contrary to previous proposals [13], however, there is little evidence to support the view that AMPK is directly activated by reactive oxygen species (ROS) [14,15]. Once activated the classical action of AMPK is to phosphorylate targets that switch off non-essential anabolic processes that consume ATP and switch on catabolic pathways that generate ATP [12], thereby compensating for deficits in ATP supply via, for example, reductions in mitochondrial oxidative-phosphorylation. Figure 1 Regulation of the AMP-activated protein kinase (1) AMPK is constitutively phosphorylated (AMPK-P) by LKB1. However when ATP is bound to AMPK, dephosphorylation by protein phosphatase 2C (PP2C) is promoted and AMPK remains deactivated (AMPK). Metabolic stresses, such as hypoxia, increase the AM(D)P/ATP ratio and promote displacement of ATP by AMP, and to a lesser extent by ADP, from three sites on the AMPK γ subunit. Binding of AMP or ADP to the γ subunit may promote phosphorylation by LKB1 and at the same time (2) inhibit dephosphorylation by PP2C. (3) AMP, but not ADP, binding also promotes further allosteric activation of AMPK. These three mechanisms deliver AMPK activation in response to metabolic stresses. In addition, AMPK can be activated in a Ca2+-dependent manner through CaMKK-β, which phosphorylates the same γ subunit site as LKB1. Figure adapted from [179]: Hardie, D.G., Salt, I.P., Hawley, S.A. and Davies, S.P. (1999) AMP-activated protein kinase: an ultrasensitive system for monitoring cellular energy charge. Biochem. J. 338, 717–722. Intriguingly, in the context of the present discussion, the genes encoding the α and γ subunits of the AMPK orthologue of yeast Saccharomyces cerevisiae (SNF1 and SNF4) support colony-level metabolic adaptation [16–19]. For example, in a high glucose environment, yeast initially grow rapidly using glycolytic metabolism to generate ATP, but when glucose runs low the growth rate decreases as yeast undergo diauxic shift towards greater reliance on mitochondrial oxidative phosphorylation. This adaptation to deficits in substrate supply is blocked in yeast with snf1 or snf4 mutations that are unable to support the diauxic shift [16,17,20]; i.e. they can only grow on a source of glucose. In an evolutionary context, this observation raised the possibility that natural selection may have deployed AMPK to govern the adaptation of animals to deficits in oxygen and thus energy supply at both the cellular and whole-body level. Moreover, the fact that AMPK is a serine/threonine kinase suggested the capacity for regulation of processes outside of metabolism such as ion channel activity, which our findings [21–24] and those of others have since confirmed. For example, AMPK may phosphorylate and ‘inactivate’ the pore-forming α subunit of multiple calcium-activated potassium channels (KCa1.1 and KCa3.1) [22,25], the voltage-gated potassium channel Kv1.5 [24,26,27] and the ATP-inhibited KATP channel (Kir6.2) [28], or may phosphorylate and ‘activate’ the α subunit of the voltage-gated potassium channel Kv2.1 [21]. AMPK has the potential to thus increase or decrease cell excitability, in a manner determined by the cell-specific expression of members of the ion channel superfamily, and thereby deliver system-level control of whole-body metabolic status [1]. We have now provided conclusive evidence that the LKB1/AMPK signalling pathway does indeed play a critical role in modulating the delivery of oxygen to the body [4,29], in addition to its well-recognized role in regulating cell-autonomous pathways of energy supply [3]. Perhaps most significantly, our data suggest that LKB1/AMPK signalling pathways act not only to optimize ventilation during hypoxia, but also to oppose respiratory depression during hypoxia and may thus resist hypoventilation and apnoea [4]. However, the locus at which AMPK co-ordinates the hypoxic ventilatory response was not as one would have predicted. Regulation of rhythmic ventilation That ventilatory adjustments are critical to the body's capacity to accommodate variations in oxygen demand and supply during sleep and ascent to altitude is exemplified by the fact that adaptation of mammals to hypoxia at altitude is initially characterized by progressive increases in ventilatory drive, which partially restore arterial PO2 and protect against apnoea [30]. Ventilatory movements are delivered by motor neuronal pathways that are informed by respiratory central pattern generators (rCPGs), which are distributed bilaterally in the pons and ventral medulla of the brainstem (Figure 2) [31]. These semi-autonomous neural networks comprise core circuits of excitatory and inhibitory interneurons that deliver rhythmic patterns of activity [32], and confer a set-point about which respiratory rhythm is continuously modulated through the integration of inputs from those central [32,33] and peripheral chemosensors [34] which monitor oxygen, carbon dioxide and pH. It is generally accepted that the carotid bodies, which reside at the bifurcation of the common carotid artery, represent the primary peripheral chemoreceptors [34] and that the acute hypoxic ventilatory response is delivered by increased afferent discharge from the carotid bodies to the rCPGs via, in great part, catecholaminergic networks within the caudal brainstem (Figure 3) [35,36]. Figure 2 Functional compartments of the brainstem ventilatory respiratory columns Dorsal view of the brainstem illustrating the functional compartments within the ventilatory respiratory column. KF, Kölliker-Fuse nucleus; PB, parabrachial nuclei; NA, noradrenergic A5 area; RTN, retrotrapezoid nucleus; PGi, paragigantocellular reticular nucleus; BötC, Bötzinger complex; preBotC, pre-Bötzinger complex; rVRG, rostral ventral respiratory group; cVRG, caudal ventral respiratory group. Image adapted from [180]: Rekling, J.C. and Feldman, J.L. (1998) PreBotzinger complex and pacemaker neurons: hypothesized site and kernel for respiratory rhythm generation. Annu. Rev. Physiol. 60, 385–405. Figure 3 The hypoxia-responsive respiratory network from carotid body to brainstem The hypoxia-responsive respiratory network spans the catecholaminergic cells of the carotid body type I cells, dorsal A2, C2 and ventral A1 and C1 neurons of the caudal brainstem, which are located at the AP, NTS and the ventrolateral medulla. The respiratory central pattern generators comprise: RTN, retrotrapezoid nucleus; BötC, Bötzinger complex; preBotC, pre-Bötzinger complex; rVRG, rostral ventral respiratory group; cVRG, caudal ventral respiratory group. To assess the role of LKB1 and AMPK in this process, we used the tyrosine hydroxylase promoter to drive deletion of AMPK-α1 and -α2 genes in all catecholaminergic cells [4], including therein type I cells of the carotid and aortic bodies [34,37], and downstream neurons within the brainstem respiratory network that relay afferent inputs to the rCPGs [38]. Both LKB1 and AMPK deletion precipitated pronounced ventilatory dysfunction during hypoxia [4,29] that was characterized by marked attenuation of the hypoxic ventilatory response, and which ultimately led to hypoventilation rather than hyperventilation and frequent prolonged apnoeas. Upon hypoxia at altitude or during sleep, activation of LKB1/AMPK signalling pathways may therefore aid appropriate ventilatory adjustments and thus ‘protect’ against acute ventilatory instability [30], although deficiency of either may confer greater susceptibility to disordered breathing. In this respect it is notable that, of the two available α subunits, selective loss of the AMPK-α1 catalytic subunit was the primary precipitant of ventilatory dysfunction during hypoxia [4]; consistent with the finding that natural selection in high-altitude (Andean) populations has led to single nucleotide polymorphisms in PRKAA1 [2]. Given that it is widely accepted that the carotid bodies drive the entire ventilatory response to a fall in arterial PO2, we had always presumed that this organ would be the primary site of AMPK action in this respect. Not for the first time, however, the order in which evolution may have influenced the development and thus organization of body systems appears, it now seems, counterintuitive. A PINCH OF PTOLEMY–AMPK AND THE CAROTID BODY The carotid bodies were identified as sensory organs by De Castro in 1928 [39], after which Heymans and Bouckaert [40] established that they mediated hyperventilation in response to a fall in arterial PO2 and thus defined these organs as the primary peripheral arterial chemoreceptors. The carotid body type I (glomus) cells underpin chemosensory activity [41], when upon exposure to hypoxia and/or hypercapnia they release a variety of neurotransmitters which elicit increases in afferent fibre discharge along the carotid sinus nerve and thereby govern cardiorespiratory reflexes that elicit corrective changes in ventilation [42–45]. Recent evidence now suggests that the aortic bodies, which are located at the aortic arch, are similarly activated during hypoxia and/or hypercapnia and may also contribute to the hypoxic ventilatory response [37]. Type I cells of the carotid and aortic bodies therefore define a class of oxygen-sensing cells, in which the PO2 at which mitochondrial oxidative phosphorylation is inhibited during hypoxia (≤60 mmHg oxygen) is higher than in other cell types [46–48]. Once this threshold is breached hypoxia-induced changes in cell activity increase in a manner related to the degree of hypoxia [46,49], that is until these activities begin to fail under near anoxic conditions (<2% oxygen) [50]; at this point mitochondrial oxidative phosphorylation is inhibited in cells that do not function to monitor oxygen supply [47]. In short, all oxygen-sensing cells function to respond to deficits in oxygen supply over the physiological range of PO2. Mitochondria underpin hypoxia-response coupling in carotid body type I cells A significant body of evidence now argues in favour of the view that type I cell activation during hypoxia is consequent to the inhibition of mitochondrial function. In retrospect the initial clue to this fact was provided by the seminal work of Heymans and Bouckaert [40], in that they demonstrated that cyanide mimicked and occluded the activation by hypoxia of the carotid body. However, the first direct evidence was obtained through the analysis of the respiratory chain redox status [51]. By relating outcomes to afferent sinus nerve discharge during hypoxia, it was shown that an increase in the NAD(P)H/NAD(P)+ ratio correlated with changes in afferent nerve activity over the physiological range of oxygen levels. At the time it was proposed that mitochondria of most cells may utilize a high-affinity (i.e. normal) cytochrome a3, whereas the cytochrome a3 incorporated in mitochondria of oxygen-sensing cells may have a low affinity for oxygen. Consistent with this hypothesis, recent investigations have demonstrated that NDUFA4L2 [52] and COX4I2 [53,54], two nuclear-encoded atypical subunits of the mitochondrial electron transport chain, are constitutively expressed in carotid body type I cells under normoxia [55]. This contrasts with a number of other cell types where NDUFA4L2 and COX4I2 expression is ordinarily low, but is increased during prolonged hypoxia [52–54]. Both NDUFA4L2 and COX4I2 reduce the capacity for mitochondrial oxygen consumption and act to limit mitochondrial ROS production during hypoxia, by reducing the activity of complex I and cytochrome c oxidase respectively. In this respect it is interesting to note that allosteric modulation of cytochrome c oxidase (COX) is delivered by COX4 in a subtype-specific manner, with COX4I1 but not COX4I2 conferring COX inhibition by ATP [54,56], i.e. in carotid body type I cells it seems unlikely that the rate of oxygen consumption and thus ATP supply via mitochondrial oxidative phosphorylation will increase during hypoxia as ATP levels fall [53,56–58]. It has been suggested, therefore, that constitutive expression of NDUFA4L2 and COX4I2 by carotid body type I cells might determine the affinity of their mitochondria for oxygen and thus confer, in part, the capacity of these cells to monitor changes in arterial oxygen supply. Intriguingly, COX4I2 is also constitutively expressed by pulmonary arterial myocytes [58,59] and neurons of the central nervous system [56], which may in some instances also function to monitor oxygen supply (see below). That mitochondria may be the site of oxygen-sensing within type I cells of the carotid body is supported by the fact that, in addition to cyanide, all inhibitors and uncouplers of mitochondrial electron transport both mimic and occlude the effects of hypoxia [60]. Moreover, recent studies have shown that conditional deletion in type I cells of Ndufs2, a mitochondrial complex I gene that participates in ubiquinone binding, blocks carotid body activation during hypoxia [61]. ATP, LKB1, AMPK and hypoxia-response coupling in carotid body type I cells What remains open to debate is the precise nature of the signalling pathway(s) which couples inhibition by hypoxia of mitochondrial oxidative phosphorylation to the activation of oxygen-sensing cells, such as type I cells, and whether or not all oxygen-sensing cells utilize a common signalling pathway. At the very least one would expect an initial fall in ATP supply and associated ADP accumulation that would be compensated for, in the immediate term, by the adenylate kinase reaction, leading to consequent increases in the AMP/ATP ratio [62,63]. When one considers this and the fact that AMPK is intimately coupled to mitochondrial metabolism via both increases in the AM(D)P/ATP ratio and LKB1, the possibility that AMPK may contribute to hypoxia-response coupling is immediately apparent. If this were the case, then one would naturally expect any contribution of AMPK to ventilatory control to be delivered at the level of the carotid body type I cell and through the consequent inhibition of those ‘oxygen-sensing’ potassium channels known to underpin their chemosensory response. Not least because thereafter the generally held viewpoint is that carotid body afferent inputs to the brainstem activate subordinate relays that modulate rCPG activities and thus increase ventilation. Our preliminary investigations into the role of the LKB1/AMPK signalling pathway appeared entirely consistent with this view, in that conditional deletion of LKB1 virtually abolished the capacity for type I cell activation during hypoxia, increases in afferent discharge and, like AMPK deletion, attenuated the hypoxic ventilatory response [64,65]. Contrary to these findings and against our expectations, however, AMPK deletion failed to attenuate afferent discharge from the carotid body, yet caused even greater attenuation of the hypoxic ventilatory response [4] when compared with LKB1 deletion (unpublished work). This runs counter to our previous pharmacological studies, which suggested that 5-amino-4-imidazolecarboxamide riboside (AICAR), an AMPK agonist [66], activated carotid body type I cells and increased afferent discharge [67], and that this action was inhibited by the AMPK antagonist compound C. However, compound C is a very non-selective kinase inhibitor, which in a screen of 70 protein kinases was shown to inhibit at least ten other kinases more potently than AMPK [68]. Moreover, off-target effects of other pharmacological tools have also been identified, such as inhibition by AICAR of adenosine transporters [69] (adenosine receptors being key modulators of type I cell activity [70]) and/or AICAR-mediated reductions in the adenylate pool and ATP [71,72]. One must therefore conclude that AMPK is not necessary for type I cell activation by hypoxia. Consistent with this view, recent studies on the actions of two different AMPK activators, AICAR and A769662 [73], suggest that these agents neither precisely mimic the effects of hypoxia on nor induce pronounced activation of carotid body type I cells [74,75], and our own most recent investigations now support this view (unpublished work). Nevertheless it would appear that we have inadvertently uncovered a split in the dependency on LKB1 and AMPK respectively of carotid body activation during hypoxia on the one hand and the hypoxic ventilatory response on the other. The reasons for this remain to be resolved, but experimental outcomes perhaps point to hierarchical control of the respiratory network by LKB1, AMPK and one or more of the 12 AMPK-related kinases [76]. Given that afferent discharge is, in great part, triggered by exocytotic release of ATP from type I cells [77], it is quite plausible that LKB1 may maintain, in an AMPK-independent manner, the capacity for ATP synthesis and/or exocytosis within type I cells, and thus afferent discharge from the carotid body. This is entirely in keeping with the fact that LKB1 may govern glucose homoeostasis [78,79] and mitochondrial function [80,81] independently of AMPK, perhaps via constitutive phosphorylation of an AMPK-related kinase [76,82,83], given that LKB1 deletion has been shown to decrease mitochondrial membrane potential and basal ATP levels in other cell types [80,81,84]. It is therefore possible that any cell lacking LKB1, such as carotid body type I cells, may be unable to sustain appropriate cellular energy charge and activity due to defective mitochondrial function, either at rest or during exposure to metabolic stresses such as hypoxia. So where does this leave us? Well one backward look takes us to ATP, ADP and AMP levels and the inhibition during hypoxia of type I cell K+ channels, which ultimately triggers exocytosis [85–87]. The principal players in this respect are the large conductance voltage- and Ca2+-activated K+ current (BKCa) [88,89] and the voltage-independent TASK-like leak K+ current [90–92], although it should be noted that variations in channel expression may confer identified species differences [93,94] and contribute to changes of oxygen sensitivity during postnatal maturation [95,96]. It is now clear that hypoxia (and hypercapnia) principally acts to depolarize type I cells by inhibiting TASK1/3 K+ channels [74], leading to Ca2+ entry through voltage-gated Ca2+ channels, consequent exocytosis and ATP release. Moreover, in the absence of a determining role for AMPK [4,75], substantial evidence now supports the view that TASK K+ channels directly monitor the adenylate pool [97], and close when ATP levels fall consequent to the inhibition by hypoxia of mitochondrial oxidative phosphorylation [60]. AMPK does, however, phosphorylate and, like hypoxia, inhibit BKCa channels of carotid body type I cells [22], the archetypal oxygen-sensing potassium channel [85,89]. This action will clearly have functional consequences with respect to transmitter release, conceivably by modulating the transition to ‘bursting’ patterns of action potential firing [98], but these remain to be resolved. In short, type I cell activation during hypoxia is probably precipitated by changes in the adenylate pool and ATP [99], and membrane depolarization due to subsequent inhibition of K+ currents carried by TASK1/3 heterodimers [91,100]. However, the primacy of this view has recently been challenged by three alternative hypotheses: It has been suggested that type I cell activation may be triggered by increases in hydrogen sulfide production consequent to a fall in carbon monoxide synthesis during hypoxia [101], although the findings of others suggest that activation of carotid body type I cells by exogenous hydrogen sulfide results from direct inhibition of mitochondrial oxidative phosphorylation [102]. The effects on type I cells of hydrogen sulfide may not, therefore, be inconsistent with the conclusion drawn above. This perspective has recently received support from single-cell transcriptome analysis of mouse type I cells which identified few to no reads of the enzymes responsible for generating either carbon monoxide or hydrogen sulfide [55], respectively, haem oxygenase-2, or cystathionine-γ-lyase and cystathionine-β-synthase. As mentioned previously, conditional deletion in tyrosine hydroxylase-positive cells of Ndufs2, a mitochondrial complex I gene which encodes a protein that participates in ubiquinone binding, has also been shown to selectively block carotid body activation during hypoxia (but not hypercapnia or hypoglycaemia) and thus the hypoxic ventilatory response [61]. The authors concluded that this probably results from loss, during hypoxia, of the capacity for signalling via increased generation of mitochondrial ROS. However this study did not address the impact of Ndufs2 deletion on oxidative phosphorylation in type I cells, the capacity for inhibition of type I cell mitochondrial oxidative phosphorylation during hypoxia and consequent modulation of TASK-like potassium currents by alterations in the adenylate pool (see also [103]). Furthermore, and as discussed above, NDUFA4L2 and COX4I2 are constitutively expressed by type I cells and act to limit mitochondrial ROS production during hypoxia [53,54]. That aside, it is important to note that conditional deletion of Ndufs2 in catecholaminergic cells blocked the hypoxic ventilatory response even though the capacity for both basal and activated transmitter release was retained by type I cells (see below for further discussion). Most recently a novel chemosensory signalling pathway has been proposed to be a prerequisite for type I cell activation during hypoxia, namely lactate-dependent activation of olfactory receptor 78 (Olfr78) [104]. In this study global deletion of Olfr78 was found to block carotid body activation during hypoxia and thus the hypoxic ventilatory response of mice. By virtue of a requirement for lactate production and release consequently to induction of anaerobic glycolysis, the proposed model for lactate-dependent activation of Olfr78 during hypoxia is consistent with the mitochondrial hypothesis, but is inconsistent with a mechanism in which type I cell activation is determined by TASK K+ channel inhibition through alterations in the adenylate pool [74]. That is unless, of course, these two pathways converge. Once again, however, it may be worthy of note that the hypoxic ventilatory response was blocked by global Olfr78 deletion despite the fact that basal afferent discharge from the carotid body was retained (see below for further discussion). Putting due scrutiny of the aforementioned signalling pathways to one side, it is clear from our own findings that all pathways key to carotid body type I cell activation during hypoxia must be, in some way, dependent on the continued expression of LKB1, but not AMPK, and a sufficiency of mitochondrial function and/or ATP supply. So how can it be that both LKB1 and AMPK deletion block the hypoxic ventilatory response, when deletion of the latter does not adversely affect carotid body activation during hypoxia [4,29,64]? For such a proposal runs contrary to the generally held view that increased afferent discharge from carotid body to brainstem determines the ventilatory response to a fall in arterial PO2 [34]. Well there is substantial evidence to support an alternative yet inclusive perspective, namely that the hypoxic ventilatory response is determined by the co-ordinated action of the carotid body and a hypoxia-responsive circuit within the brainstem. We will see that this must now be borne in mind when drawing conclusions from all studies described above that employed either global knockout strategies or conditional gene deletion in catecholaminergic cells. A DASH OF COPERNICUS–AMPK AND THE BRAIN-CENTRED CHEMOSENSORY NETWORK From here on in our aim is to be a little more provocative if not heretical, at least in the eyes of some respiratory physiologists, by giving emphasis to a matter that has long been quietly considered by a minority of the field. In actual fact, our investigation is merely the latest in a long line to have described experimental observations that run counter to the standard model for the control of ventilation by peripheral chemosensors, and the pre-eminence of the carotid bodies in this respect. Not surprisingly, in retrospect, the possibility that peripheral chemosensors may not be the sole arbiters of the hypoxic ventilatory response has been suggested by investigations on the evolution of ventilatory control systems, most notably with respect to the demonstration that oxygen-sensing occurs and a component of the hypoxic ventilatory response arises at the level of the caudal brainstem in amphibians, with both the location and influence of the primary peripheral chemosensors changing during the ascent from gill-breathing tadpole to lung-assisted air-breathing adult [105,106]. In fact one could quite reasonably argue that evolutionary pressures have periodically led to the reconfiguration of peripheral chemoreceptor inputs [106] about a common ancestral hypoxia-sensor within the caudal brainstem, that underpins signal integration and thus acts as the ‘gatekeeper’ of respiratory adjustments during hypoxia. That said, the possibility that neural networks within the brainstem of mammals might respond to central hypoxia was first raised over 35 years ago by the work of Dampney and Moon [107], during their investigations on the central ischaemic vasomotor response. Thereafter, during their investigations on Cushing's reflex [108], Sun and Reiss demonstrated that both cyanide and hypoxia activated neurons within the rostral ventrolateral medulla [109,110], mirroring Heymans and Bouckaert's earlier work on the carotid body. Moreover extensive evidence has been provided in support of the view that increases in ventilation are initiated by brainstem hypoxia in the presence of only basal normoxic afferent input from the carotid bodies [111,112], and it has been suggested that different aspects of the brainstem respiratory network may exhibit different sensitivities to hypoxia [113]. To date, however, little emphasis has been placed on the role of hypoxia-sensing at the brainstem, perhaps because the hypoxic ventilatory response is so effectively abolished by resection of the carotid sinus nerve in humans [114]. Yet extensive investigations have demonstrated that following carotid body resection, hypoxia-responsive catecholaminergic neurons of the caudal brainstem may underpin partial recovery of the hypoxic ventilatory response [115], at least in rodents, and it is recognized that loss of these neurons underpins ventilatory dysfunctions associated with Rett syndrome, including hypoventilation and apnoea, which are exacerbated during hypoxia [116]. Consistent with outcomes of the aforementioned studies, our findings strongly suggest that AMPK governs the activation of previously identified hypoxia-responsive nuclei within the caudal brainstem [110,117], and thus supports the delivery of increased respiratory drive during hypoxia that is required to protect against hypoventilation and apnoea. The most convincing evidence of this was provided by examination of brainstem function in AMPK knockout mice by functional magnetic resonance imaging (fMRI), which identified reduced activation during hypoxia of discrete dorsal and ventral nuclei of the caudal brainstem, despite the fact that carotid body afferent discharge was retained [4]. This was corroborated by analysis of immediate early gene (c-fos) expression. The caudal location relative to Bregma of the dorsal active region is consistent with areas of the nucleus tractus solitarius (NTS) that are activated by hypoxia and which represent the primary site of receipt of carotid body afferent input [35,117,118]. Here AMPK deletion selectively attenuated c-fos expression during hypoxia by mixed subpopulations of C2 neurons and A2 neurons (SubP; SolM) within the medial subnucleus proximal to the midline and the area postrema (AP) [4], which have been previously shown to be activated during hypoxia [38]. A2 neurons of the AP/NTS provide afferent input to and determine, together with the carotid body, activation by hypoxia of A1/C1 neurons within the ventrolateral medulla [38,119], the position of which [119] aligns well with the ventral active region identified by fMRI analysis [4]; by contrast projections of the NTS mostly avoid key components of the rCPGS [119], namely the Bötzinger and pre-Bötzinger complexes [120]. Analysis of c-fos expression at the level of the ventrolateral medulla suggested that AMPK deletion selectively reduced the activation of A1 neurons during hypoxia, although it should be noted that there is significant overlap between the most caudal C1 and the most rostral A1 neurons [121]. Our findings therefore suggest that the hypoxic ventilatory response, including that provided by afferent inputs from peripheral chemosensors, is attenuated by loss of AMPK function at the level of the caudal brainstem, within a neuronal circuit spanning the C2/A2 neurons of the NTS and A1 neurons of the ventrolateral medulla. This is consistent with optogenetic and pharmacological interventions at the level of the NTS [117,122], and the proposal that NTS neurons lie on the sensory side of the central respiratory network [123,124]. We cannot rule out the possibility that suppression of the hypoxic ventilatory response in AMPK knockouts may be allied to exacerbation of the Cushing reflex [35,108]. However, this reflex is only elicited under anaesthesia and by ischaemic hypoxia (∼1% O2), and is maintained or enhanced by hypercapnia [35,108,125]. By contrast, hypoxic ventilatory depression was evident in conscious AMPK knockouts during mild and severe hypoxia, as were deficits in brainstem activity, and was reversed rather than exacerbated by hypercapnia. Surprisingly, we observed pronounced right–left asymmetry of brainstem activation during hypoxia, which may provide for specialization sufficient to prevent delays in respiratory responses to hypoxic stress by limiting conflicting outputs from each side of the brain [126], as has been proposed previously with respect to cognitive performance [127]. Further investigation will be required to determine how right–left asymmetry may be orchestrated by the complex interplay of neurotransmitters deployed during hypoxia and the role of AMPK in such processes of selection. In this respect it is notable that C2 and A2 neurons are both catecholaminergic and glutamatergic [123,128], and that 6–10% of tyrosine hydroxylase-positive C2, A2 and A1 neurons also express neuronal nitric oxide synthase, which supports the hypoxic ventilatory response by synthesizing NO [129] and/or S-nitrosothiols [130], and in a manner that may be facilitated by AMPK [131]. It could be argued that AMPK deletion in catecholaminergic cells simply leads to the failure of central integration and transduction of peripheral chemoafferent input and consequent failure of the hypoxic ventilatory response, due to the inability of affected neurons to maintain appropriate levels of activity when exposed to metabolic stress [132]. However, following AMPK deletion, carotid body afferent discharge remained exquisitely sensitive to a fall in PO2 and ventilatory responses to hypercapnia remained unaffected even during severe (8%) hypoxia, which clearly demonstrates that AMPK deletion does not compromise the capacity during hypoxia for activation of chemosensory catecholaminergic neurons, exocytosis nor effective delivery of increased respiratory drive. This is consistent with the observation that neuronal integrity during hypoxia may be preserved, in part, by AMPK-independent mechanisms [133] that maintain ATP supply by accelerating glycolysis and in a manner supported by mobilization of astrocyte glycogen stores [134]. If one accepts this position, then AMPK must aid the modulation by hypoxia of discrete nuclei within the caudal brainstem that deliver increased drive to breathe via neural networks that modulate the rCPGs [36], and which may also co-ordinate functional hyperaemia [135]. THE CASE FOR SIGNAL INTEGRATION AT AN OXYGEN-SENSING NUCLEUS WITHIN THE BRAINSTEM The phrase ‘to say more would be pure speculation’ is often uttered and rightly so, at least when interpreting experimental outcomes. In the present context, however, we are happy to invite ridicule and scorn for the sake of greater debate and experimental inquisition, and to achieve this goal we bring to centre stage the possibility that a cluster of hypoxia-responsive neurons proximal to the NTS form a nucleus that acts as the ‘gatekeeper’ of the hypoxic ventilatory response. If this nucleus does indeed exist, then why has it not been located by the extensive efforts of so many specialists in the field? Perhaps we are dealing with an interdependent circuit mechanism, with multiple points of signal integration? When it comes down to hand waving, either a single node or multi-nodal system of signal integration appears plausible, i.e. there may be no discrete nucleus to find. In this context and in light of all things above, we need now consider why: The degree of block by AMPK deletion of the hypoxic ventilatory response is increased in a manner directly related to the severity of hypoxia [4]. The hypoxic ventilatory response can be triggered by central nervous system hypoxia alone, providing there is continued receipt of basal (normoxic) afferent input from the carotid bodies [136]. The hypoxic ventilatory response may be blocked by interference at any point within this circuit, e.g. carotid body resection [114] or AMPK deletion [4]. We propose (Figure 4) that LKB1/AMPK signalling pathways support coincidence detection and thus signal integration at either a single node or multiple nodes within and thus activation of a hypoxia-responsive circuit that encompasses, at the very least, C2/A2 neurons within the NTS and ventrolateral A1 neurons, due to the capacity for AMPK activation by increases in the AM(D)P/ATP ratio and LKB1 [3] that may be determined by ‘local hypoxic stress’ (decreased ATP supply) and in a manner that is coupled to ‘applied metabolic stress’ (increased ATP usage) delivered via afferent inputs from peripheral chemoreceptors to the NTS, and, in turn, to ventrolateral A1 neurons and perhaps also to downstream aspects of the cardiorespiratory network. Afferent input and brainstem hypoxia could thereby determine, each in part, the set-point about which AMPK and thus the brainstem respiratory network are activated during hypoxia. Thereafter AMPK-dependent modulation of cellular metabolism [3], ion channels and thus neuronal firing frequency [21], and/or transmitter release [130,131] may facilitate efferent output and thereby deliver increased drive to breathe, in a manner that may be attenuated or augmented by appropriate regulation of AMPK expression. Figure 4 Schematic description of the new hypothesis on the integration by AMPK of local and applied metabolic stresses (A) Minimal model describes a single node for integration of local and applied metabolic stress by AMPK. (B) Extended model describes the possibility that there is capacity for signal integration, of local and applied metabolic stress, at multiple nodes within the hypoxia-responsive respiratory network. Adenosine (Aden). In essence then, our proposal is that the LKB1/AMPK signalling pathway monitors changes in adenylate charge centrally as an index of local hypoxic stress and integrates with this applied metabolic stresses delivered by afferent chemosensory inputs, which are in turn providing an index of peripheral hypoxic (metabolic) status. If so, then perhaps we can garner more from our considerations on the regulation of afferent output from the peripheral chemoreceptors, namely the carotid and aortic bodies, in terms of their role in monitoring changes in adenylate charge and thus in the provision of an index of peripheral hypoxic stress. As discussed in detail previously, hypoxia depolarizes type I cells through inhibition of TASK1/3 K+ channels, leading to voltage-gated Ca2+ entry, exocytosis and ultimately ATP release. Subsequently ATP stimulates postsynaptic P2X2/3 receptors on afferent (petrosal) nerve terminals causing excitation, but at the same time activates P2Y2 receptors on adjacent glial-like type II cells [77,137]. P2Y2 receptor activation then triggers further ATP release from type II cells into the synaptic cleft, where ATP (from both type II and type I cells) is broken down by extracellular 5′-ectonucleotidase into adenosine, which primarily activates adenosine A2A receptors on type I cells [77]. Activation of A2A receptors leads to further inhibition of TASK1/3 channels and enhanced type I cell depolarization [138], and further augments ATP release during hypoxia [77]; a similar system probably operates at the aortic bodies (C. Nurse, personal communication). It seems quite possible, therefore, that LKB1 may govern a set-point for metabolic homoeostasis about which both carotid and aortic bodies monitor adenylate charge as an index of hypoxic stress, by integrative inhibition of TASK1/3 channels consequent to deficits in mitochondrial ATP production that are allied to purinergic cross-talk between type I and type II cells. Via their tripartite synapse with afferent petrosal neurons [77], type I and type II cells may therefore act in concert to relay information on changes in the ‘peripheral adenylate pool’ (ATP, ADP, AMP and adenosine) to the brainstem. During the transit of re-oxygenated blood from the heart to the brainstem, the NTS may thereby co-ordinate the integration of information on adenylate charge, as an index of arterial oxygen saturation, via at least four separate and highly vascularized nodes, namely the aortic and carotid bodies, the AP/NTS and the ventrolateral medulla, in order to appropriately co-ordinate cardiorespiratory function (Figure 4). IN ELLIPTICAL ORBIT–AMPK AND THE REGULATION OF BLOOD FLOW AND GASEOUS EXCHANGE Hypoxia without hypercapnia induces pulmonary vasoconstriction, and thus assists ventilation–perfusion matching by diverting blood from oxygen-deprived to oxygen-rich areas of the lung [139,140]. By contrast, systemic arteries dilate in response to tissue hypoxemia, in order to match local perfusion to local metabolism [141]. Whichever we consider, it is now evident that AMPK may be key to the regulation of vascular reactivity during metabolic stress [23,142] and may thus facilitate gaseous exchange across the body. AMPK and ventilation–perfusion matching at the lung Quite unlike the hypoxic ventilatory response, hypoxic pulmonary vasoconstriction is governed locally and is mediated by mechanisms intrinsic to pulmonary arterial smooth muscles and endothelial cells. This is evident from the fact that neither central nor local regulation of the autonomic nervous system contributes to hypoxic pulmonary vasoconstriction [143–145], which remains unaffected following denervation in humans [146]. However, here too the nature of the principal signalling pathway(s) involved remains open to debate [103], although it is clear that this response relies on the modulation by hypoxia of mitochondrial metabolism [48]; pulmonary arterial smooth muscle cells depleted, by ethidium bromide, of mitochondrial DNA and thus of functional mitochondria do not respond to hypoxia [147], although inhibitors of mitochondrial oxidative phosphorylation either mimic or occlude hypoxic pulmonary vasoconstriction [148,149]. As mentioned previously and consistent with findings on carotid body type I cells [55], COX4I2 is constitutively expressed by pulmonary arterial myocytes [59] and may also act here to limit mitochondrial oxygen consumption and ROS production during hypoxia and confer, in part, the capacity of these cells to monitor oxygen supply [55]. In the light of the evidence in support of a role for mitochondria in hypoxia-response coupling, it was therefore proposed that the LKB1/AMPK signalling pathway might couple inhibition by hypoxia of mitochondrial metabolism to hypoxic pulmonary vasoconstriction [1,23,150]. Consistent with this view, AMPK-α1 activity was found to be greater in pulmonary than systemic (mesenteric) arterial smooth muscles [23] and this may in its own right afford a degree of pulmonary selectivity in terms of the capacity and nature of the response to physiological levels of hypoxia, over and above that which might be conferred by COX4I2 expression. Indeed exposure of pulmonary arterial smooth muscle to hypoxia (15–20 mmHg oxygen) precipitated increases in the AMP/ATP ratio, marked activation of AMPK and phosphorylation of acetyl-CoA carboxylase [23]; which may go some way to explain why cellular ATP levels remain remarkably stable in the presence of hypoxia [148]. Inhibition of mitochondrial oxidative phosphorylation by phenformin [151] and AICAR [66] precipitated AMPK activation and acetyl-CoA carboxylase phosphorylation within pulmonary arterial myocytes [23]. Regardless of their respective mechanism of action, hypoxia, phenformin and AICAR also induced an increase in the intracellular calcium concentration in and contraction of acutely isolated pulmonary arterial myocytes, and did so by mobilizing sarcoplasmic reticulum stores via ryanodine receptors. Most significantly AICAR evoked a sustained and reversible constriction of pulmonary artery rings, which exhibited characteristics strikingly similar to hypoxic pulmonary vasoconstriction; not least clearly defined contributions from both smooth muscles and the endothelium. Furthermore, hypoxic pulmonary vasoconstriction was inhibited by compound C [152]. In this instance it would appear that the pharmacology held true, for our most recent studies on knockout mice suggest that LKB1 and AMPK, but not CaMKK-β, are indeed required for hypoxic pulmonary vasoconstriction [153] and that dysfunction within the AMPK signalling pathway may precipitate pulmonary hypertension. Further support for this view has recently been provided by our demonstration that upon inhibition of mitochondrial oxidative phosphorylation, AMPK directly phosphorylates Kv1.5 channels, and inhibits K+ currents carried by Kv1.5 in pulmonary arterial myocytes [24]. This is evident from the fact that down-regulation of Kv1.5 expression and activity is a hallmark not only of hypoxic pulmonary vasoconstriction but also of pulmonary hypertension [154–162], and may contribute to increased survival of smooth muscle cells due to attenuation of K+ channel-dependent apoptosis [163–165] and also facilitate the phenotypic switch from a contractile to a proliferative state [166,167]. Consistent with the above, Zhou and co-workers have suggested that AMPK activation promotes survival of pulmonary arterial myocytes during hypoxia and thus cell proliferation by a dual mechanism, incorporating activation of autophagy by AMPK-α1 and reductions in cell death conferred by AMPK-α2 acting to reduce apoptosis via different pathways [168]. Contrary to this latter proposal, however, up-regulation of mTORC2 signalling has been proposed to underpin smooth muscle proliferation and the progression of both idiopathic and hypoxic pulmonary arterial hypertension [169], by promoting smooth muscle cell survival in a manner, at least in part, dependent on down-regulation of AMPK and consequent activation of mammalian target of rapamycin complex 1 (mTORC1). One possible explanation for these contrary prepositions could be that AMPK action is context-dependent and/or that the progression of pulmonary hypertension at different stages is governed by temporal fluctuations in AMPK activity. Regulation of utero-placental blood flow during hypoxia AMPK has most recently been implicated in the regulation of uterine artery reactivity during hypoxia [170]. AMPK may, therefore, link maternal metabolic and cardiovascular responses during pregnancy and govern oxygen and nutrient supply to the foetus, thus determining foetal growth. Consistent with this view, PRKAA1 variants most common to Andeans are positively associated with birth weight, uterine artery diameter and to alterations in the expression of genes in the mammalian target of rapamycin (mTOR) pathway that have been previously implicated in altitude-associated foetal growth restriction [2]. CONCLUSION In summary a growing body of evidence now supports the proposal that AMPK is key to oxygen and thus energy (ATP) supply to the body as a whole, through its contribution to the governance of the hypoxic ventilatory response, ventilation–perfusion matching at the lung and local regulation of blood and thus oxygen supply to the body systems. Aberrant AMPK expression or activity may therefore compromise system responses to hypoxia or other metabolic stressors and precipitate, for example, pulmonary hypertension [171], sleep-disordered breathing [172], hypertension [173] or foetal growth restriction [170], which are associated with either ascent to altitude [2,30] and/or metabolic syndrome-related disorders [172,174,175]. Therefore, further investigations on the role of AMPK in the regulation of ventilatory and vascular function in health and disease are warranted, in order that we may identify new therapeutic strategies allied to our growing understanding of the potential for development of subunit-selective small-molecule regulators of AMPK [73,176–178]. FUNDING This work was supported by the Wellcome Trust [grant number WT081195MA]; and the British Heart Foundation [grant number RG/12/14/29885]. Abbreviations AICAR5-amino-4-imidazolecarboxamide riboside AMPKAMP-activated protein kinase AParea postrema COXcytochrome c oxidase CaMKK-βCa2+–calmodulin-activated kinase kinase-β LKB1liver kinase B1 NTSnucleus tractus solitarius Olfr78olfactory receptor 78 rCPGrespiratory central pattern generator ROSreactive oxygen species ==== Refs 1 Evans A.M. AMP-activated protein kinase and the regulation of Ca2+ signalling in O2 -sensing cells J. Physiol. 2006 574 113 123 10.1113/jphysiol.2006.108381 16709639 2 Bigham A.W. Julian C.G. Wilson M.J. Vargas E. Browne V.A. Shriver M.D. Moore L.G. Maternal PRKAA1 and EDNRA genotypes are associated with birth weight, and PRKAA1 with uterine artery diameter and metabolic homeostasis at high altitude Physiol. Genomics 2014 46 687 697 10.1152/physiolgenomics.00063.2014 25225183 3 Hardie D.G. AMPK-sensing energy while talking to other signaling pathways Cell Metab. 2014 20 939 952 10.1016/j.cmet.2014.09.013 25448702 4 Mahmoud A.D. Lewis S. Juricic L. Udoh U.A. Hartmann S. Jansen M.A. Ogunbayo O.A. Puggioni P. Holmes A.P. Kumar P. AMP-activated protein kinase deficiency blocks the hypoxic ventilatory response and thus precipitates hypoventilation and apnea Am. J. Respir. Crit. Care Med. 2016 193 1032 1043 10.1164/rccm.201508-1667OC 26669206 5 Hardie D.G. AMPK-sensing energy while talking to other signaling pathways Cell Metab. 2014 20 939 952 10.1016/j.cmet.2014.09.013 25448702 6 Hardie D.G. AMPK: positive and negative regulation, and its role in whole-body energy homeostasis Curr. Opin. Cell Biol. 2015 33 1 7 10.1016/j.ceb.2014.09.004 25259783 7 Reference deleted 8 Ross F.A. MacKintosh C. Hardie D.G. AMP-activated protein kinase: a cellular energy sensor that comes in twelve flavours FEBS J. 2016 10.1111/febs.13698 9 Gowans G.J. Hawley S.A. Ross F.A. Hardie D.G. AMP is a true physiological regulator of AMP-activated protein kinase by both allosteric activation and enhancing net phosphorylation Cell Metab. 2013 18 556 566 10.1016/j.cmet.2013.08.019 24093679 10 Ross F.A. Jensen T.E. Hardie D.G. Differential regulation by AMP and ADP of AMPK complexes containing different gamma subunit isoforms Biochem. J. 2016 473 189 199 10.1042/BJ20150910 26542978 11 Sakamoto K. Goransson O. Hardie D.G. Alessi D.R. Activity of LKB1 and AMPK-related kinases in skeletal muscle: effects of contraction, phenformin, and AICAR Am. J. Physiol. Endocrinol. Metab. 2004 287 E310 E317 10.1152/ajpendo.00074.2004 15068958 12 Hardie D.G. AMP-activated/SNF1 protein kinases: conserved guardians of cellular energy Nat. Rev. Mol. Cell Biol. 2007 8 774 785 10.1038/nrm2249 17712357 13 Emerling B.M. Weinberg F. Snyder C. Burgess Z. Mutlu G.M. Viollet B. Budinger G.R. Chandel N.S. Hypoxic activation of AMPK is dependent on mitochondrial ROS but independent of an increase in AMP/ATP ratio Free Radic. Biol. Med. 2009 46 1386 1391 10.1016/j.freeradbiomed.2009.02.019 19268526 14 Auciello F.R. Ross F.A. Ikematsu N. Hardie D.G. Oxidative stress activates AMPK in cultured cells primarily by increasing cellular AMP and/or ADP FEBS Lett. 2014 588 3361 3366 10.1016/j.febslet.2014.07.025 25084564 15 Hawley S.A. Ross F.A. Chevtzoff C. Green K.A. Evans A. Fogarty S. Towler M.C. Brown L.J. Ogunbayo O.A. Evans A.M. Hardie D.G. Use of cells expressing gamma subunit variants to identify diverse mechanisms of AMPK activation Cell Metab. 2010 11 554 565 20519126 16 Celenza J.L. Carlson M. A yeast gene that is essential for release from glucose repression encodes a protein kinase Science 1986 233 1175 1180 10.1126/science.3526554 3526554 17 Celenza J.L. Eng F.J. Carlson M. Molecular analysis of the SNF4 gene of Saccharomyces cerevisiae: evidence for physical association of the SNF4 protein with the SNF1 protein kinase Mol. Cell. Biol. 1989 9 5045 5054 10.1128/MCB.9.11.5045 2481228 18 Woods A. Munday M.R. Scott J. Yang X. Carlson M. Carling D. Yeast SNF1 is functionally related to mammalian AMP-activated protein kinase and regulates acetyl-CoA carboxylase in vivo J. Biol. Chem. 1994 269 19509 19515 7913470 19 Mitchelhill K.I. Stapleton D. Gao G. House C. Michell B. Katsis F. Witters L.A. Kemp B.E. Mammalian AMP-activated protein kinase shares structural and functional homology with the catalytic domain of yeast Snf1 protein kinase J. Biol. Chem. 1994 269 2361 2364 7905477 20 Haurie V. Boucherie H. Sagliocco F. The Snf1 protein kinase controls the induction of genes of the iron uptake pathway at the diauxic shift in Saccharomyces cerevisiae J. Biol. Chem. 2003 278 45391 45396 10.1074/jbc.M307447200 12960168 21 Ikematsu N. Dallas M.L. Ross F.A. Lewis R.W. Rafferty J.N. David J.A. Suman R. Peers C. Hardie D.G. Evans A.M. Phosphorylation of the voltage-gated potassium channel Kv2.1 by AMP-activated protein kinase regulates membrane excitability Proc. Natl. Acad. Sci. 2011 108 18132 18137 10.1073/pnas.1106201108 22006306 22 Ross F.A. Rafferty J.N. Dallas M.L. Ogunbayo O. Ikematsu N. McClafferty H. Tian L. Widmer H. Rowe I.C. Wyatt C.N. Selective expression in carotid body type I cells of a single splice variant of the large conductance calcium- and voltage-activated potassium channel confers regulation by AMP-activated protein kinase J. Biol. Chem. 2011 286 11929 11936 10.1074/jbc.M110.189779 21209098 23 Evans A.M. Mustard K.J.W. Wyatt C.N. Peers C. Dipp M. Kumar P. Kinnear N.P. Hardie D.G. Does AMP-activated protein kinase couple inhibition of mitochondrial oxidative phosphorylation by hypoxia to calcium signaling in O2-sensing cells? J. Biol. Chem. 2005 280 41504 41511 10.1074/jbc.M510040200 16199527 24 Moral-Sanz J. Mahmoud A.D. Ross F.A. Eldstrom J. Fedida D. Hardie D.G. Evans A.M. AMP-activated protein kinase inhibits K 1.5 channel currents in pulmonary arterial smooth muscle and HEK 293 cells J. Physiol. doi: 10.1113/JP272032 2016 25 Klein H. Garneau L. Trinh N.T. Prive A. Dionne F. Goupil E. Thuringer D. Parent L. Brochiero E. Sauve R. Inhibition of the KCa3.1 channels by AMP-activated protein kinase in human airway epithelial cells Am. J. Physiol. Cell Physiol. 2009 296 C285 C295 10.1152/ajpcell.00418.2008 19052260 26 Andersen M.N. Skibsbye L. Tang C. Petersen F. MacAulay N. Rasmussen H.B. Jespersen T. PKC and AMPK regulation of Kv1.5 potassium channels Channels 2015 9 121 128 10.1080/19336950.2015.1036205 26043299 27 Mia S. Munoz C. Pakladok T. Siraskar G. Voelkl J. Alesutan I. Lang F. Downregulation of Kv1.5 K channels by the AMP-activated protein kinase Cell. Physiol. Biochem. 2012 30 1039 1050 10.1159/000341480 23221389 28 Chang T.J. Chen W.P. Yang C. Lu P.H. Liang Y.C. Su M.J. Lee S.C. Chuang L.M. Serine-385 phosphorylation of inwardly rectifying K+ channel subunit (Kir6.2) by AMP-dependent protein kinase plays a key role in rosiglitazone-induced closure of the K(ATP) channel and insulin secretion in rats Diabetologia 2009 52 1112 1121 10.1007/s00125-009-1337-4 19357830 29 Mahmoud A.D. Evans A.M. LKB1 expression in carotid body type I cells is required for the ventilatory response of mice to hypoxia but not hypercapnia Proc. Physiol. Soc. 2012 27 C70 30 Ainslie P.N. Lucas S.J. Burgess K.R. Breathing and sleep at high altitude Respir. Physiol. Neurobiol. 2013 188 233 256 10.1016/j.resp.2013.05.020 23722066 31 Spyer K.M. To breathe or not to breathe? That is the question Exp. Physiol. 2009 94 1 10 10.1113/expphysiol.2008.043109 19042981 32 Smith J.C. Abdala A.P. Borgmann A. Rybak I.A. Paton J.F. Brainstem respiratory networks: building blocks and microcircuits Trends Neurosci. 2013 36 152 162 10.1016/j.tins.2012.11.004 23254296 33 Day T.A. Wilson R.J. Brainstem PCO2 modulates phrenic responses to specific carotid body hypoxia in an in situ dual perfused rat preparation J. Physiol. 2007 578 843 857 10.1113/jphysiol.2006.119594 17082232 34 Nurse C.A. Synaptic and paracrine mechanisms at carotid body arterial chemoreceptors J. Physiol. 2014 592 3419 3426 10.1113/jphysiol.2013.269829 24665097 35 Guyenet P.G. Neural structures that mediate sympathoexcitation during hypoxia Respir. Physiol. 2000 121 147 162 10.1016/S0034-5687(00)00125-0 10963771 36 Guyenet P.G. Regulation of breathing and autonomic outflows by chemoreceptors Compr. Physiol. 2014 4 1511 1562 10.1002/cphy.c140004 25428853 37 Piskuric N.A. Nurse C.A. Effects of chemostimuli on [Ca2+ ]i responses of rat aortic body type I cells and endogenous local neurons: comparison with carotid body cells J. Physiol. 2012 590 2121 2135 10.1113/jphysiol.2012.229468 22431340 38 Hirooka Y. Polson J.W. Potts P.D. Dampney R.A. Hypoxia-induced Fos expression in neurons projecting to the pressor region in the rostral ventrolateral medulla Neuroscience 1997 80 1209 1224 10.1016/S0306-4522(97)00111-5 9284071 39 De Castro F. Sur la structure et l'innervation du sinus carotidien de l'homme et des mammiferes: nouveau faits sur l'innervation et la fonction du glomus caroticum Trab. Lab. Invest. Biol. Univ. Madrid 1928 24 330 380 40 Heymans C. Bouckaert J.J. Sinus caroticus and respiratory reflexes: I. Cerebral blood flow and respiration. Adrenaline apnoea J. Physiol. 1930 69 254 266 10.1113/jphysiol.1930.sp002648 16994101 41 Verna A. Roumy M. Leitner L.M. Loss of chemoreceptive properties of the rabbit carotid body after destruction of the glomus cells Brain Res. 1975 100 13 23 10.1016/0006-8993(75)90239-5 1182506 42 Gonzalez C. Almaraz L. Obeso A. Rigual R. Carotid body chemoreceptors: from natural stimuli to sensory discharges Physiol. Rev. 1994 74 829 898 7938227 43 Nurse C.A. Neurotransmitter and neuromodulatory mechanisms at peripheral arterial chemoreceptors Exp. Physiol. 2010 95 657 667 10.1113/expphysiol.2009.049312 20360424 44 Iturriaga R. Alcayaga J. Neurotransmission in the carotid body: transmitters and modulators between glomus cells and petrosal ganglion nerve terminals Brain Res. Brain Res. Rev. 2004 47 46 53 10.1016/j.brainresrev.2004.05.007 15572162 45 Zhang M. Zhong H. Vollmer C. Nurse C.A. Co-release of ATP and ACh mediates hypoxic signalling at rat carotid body chemoreceptors J. Physiol. 2000 525 143 158 10.1111/j.1469-7793.2000.t01-1-00143.x 10811733 46 Mills E. Jobsis F.F. Simultaneous measurement of cytochrome a3 reduction and chemoreceptor afferent activity in the carotid body Nature 1970 225 1147 1149 10.1038/2251147a0 4313870 47 Duchen M.R. Biscoe T.J. Mitochondrial function in type I cells isolated from rabbit arterial chemoreceptors J. Physiol. 1992 450 13 31 10.1113/jphysiol.1992.sp019114 1432706 48 Sommer N. Pak O. Schorner S. Derfuss T. Krug A. Gnaiger E. Ghofrani H.A. Schermuly R.T. Huckstorf C. Seeger W. Mitochondrial cytochrome redox states and respiration in acute pulmonary oxygen sensing Eur. Respir. J. 2010 36 1056 1066 10.1183/09031936.00013809 20516051 49 Dipp M. Evans A.M. Cyclic ADP-ribose is the primary trigger for hypoxic pulmonary vasoconstriction in the rat lung in situ Circ. Res. 2001 89 77 83 10.1161/hh1301.093616 11440981 50 Dipp M. Thomas J.M. Galione A. Evans A.M. A PO2 window for smooth muscle cADPR accumulation and constriction by hypoxia in rabbit pulmonary artery smooth muscle Proc. Phys. Soc. 2003 547P C72 51 Mills E. Jobsis F.F. Mitochondrial respiratory chain of carotid body and chemoreceptor response to changes in oxygen tension J. Neurophysiol. 1972 35 405 428 4338562 52 Tello D. Balsa E. Acosta-Iborra B. Fuertes-Yebra E. Elorza A. Ordonez A. Corral-Escariz M. Soro I. Lopez-Bernardo E. Perales-Clemente E. Induction of the mitochondrial NDUFA4L2 protein by HIF-1alpha decreases oxygen consumption by inhibiting Complex I activity Cell Metab. 2011 14 768 779 10.1016/j.cmet.2011.10.008 22100406 53 Fukuda R. Zhang H. Kim J.W. Shimoda L. Dang C.V. Semenza G.L. HIF-1 regulates cytochrome oxidase subunits to optimize efficiency of respiration in hypoxic cells Cell 2007 129 111 122 10.1016/j.cell.2007.01.047 17418790 54 Huttemann M. Kadenbach B. Grossman L.I. Mammalian subunit IV isoforms of cytochrome c oxidase Gene 2001 267 111 123 10.1016/S0378-1119(01)00385-7 11311561 55 Zhou T. Chien M.S. Kaleem S. Matsunami H. Single cell transcriptome analysis of mouse carotid body glomus cells J. Physiol. doi: 10.1113/JP271936 2016 56 Horvat S. Beyer C. Arnold S. Effect of hypoxia on the transcription pattern of subunit isoforms and the kinetics of cytochrome c oxidase in cortical astrocytes and cerebellar neurons J. Neurochem. 2006 99 937 951 10.1111/j.1471-4159.2006.04134.x 16981895 57 Kocha K.M. Reilly K. Porplycia D.S. McDonald J. Snider T. Moyes C.D. Evolution of the oxygen sensitivity of cytochrome c oxidase subunit 4 Am. J. Physiol. Regul. integr. Comp. Physiol. 2015 308 R305 R320 10.1152/ajpregu.00281.2014 25519729 58 Aras S. Pak O. Sommer N. Finley R. Jr Huttemann M. Weissmann N. Grossman L.I. Oxygen-dependent expression of cytochrome c oxidase subunit 4–2 gene expression is mediated by transcription factors RBPJ, CXXC5 and CHCHD2 Nucleic Acids Res. 2013 41 2255 2266 10.1093/nar/gks1454 23303788 59 Huttemann M. Lee I. Gao X. Pecina P. Pecinova A. Liu J. Aras S. Sommer N. Sanderson T.H. Tost M. Cytochrome c oxidase subunit 4 isoform 2-knockout mice show reduced enzyme activity, airway hyporeactivity, and lung pathology FASEB J. 2012 26 3916 3930 10.1096/fj.11-203273 22730437 60 Wyatt C.N. Buckler K.J. The effect of mitochondrial inhibitors on membrane currents in isolated neonatal rat carotid body type I cells J. Physiol. 2004 556 175 191 10.1113/jphysiol.2003.058131 14724184 61 Fernandez-Aguera M.C. Gao L. Gonzalez-Rodriguez P. Pintado C.O. Arias-Mayenco I. Garcia-Flores P. Garcia-Perganeda A. Pascual A. Ortega-Saenz P. Lopez-Barneo J. Oxygen sensing by arterial chemoreceptors depends on mitochondrial complex i signaling Cell Metab. 2015 22 825 837 10.1016/j.cmet.2015.09.004 26437605 62 Dzeja P.P. Terzic A. Phosphotransfer networks and cellular energetics J. Exp. Biol. 2003 206 2039 2047 10.1242/jeb.00426 12756286 63 Panayiotou C. Solaroli N. Karlsson A. The many isoforms of human adenylate kinases Int. J. Biochem. Cell Biol. 2014 49 75 83 10.1016/j.biocel.2014.01.014 24495878 64 Evans A.M. The LKB1-AMPK signalling pathway is required for regulation of breathing by hypoxia and thereby energy supply to the whole body Proc. Physiol. Soc. 2012 27 SA51 65 Mahmoud A.D. Lewis S. Juričić L. Foretz M. Viollet B. Marshall I. Evans A.M. AMPK couples oxygen to energy supply at the whole-body level by delivering increased drive to breathe during hypoxia and thus protects against apnoea Proc. Physiol. Soc. 2015 34 PC041 10.1113/jphysiol.2003.058131 14724184 66 Corton J.M. Gillespie J.G. Hawley S.A. Hardie D.G. 5-Aminoimidazole-4-carboxamide ribonucleoside. A specific method for activating AMP-activated protein kinase in intact cells? Eur. J. Biochem. 1995 229 558 565 10.1111/j.1432-1033.1995.tb20498.x 7744080 67 Wyatt C.N. Mustard K.J. Pearson S.A. Dallas M.L. Atkinson L. Kumar P. Peers C. Hardie D.G. Evans A.M. AMP-activated protein kinase mediates carotid body excitation by hypoxia J. Biol. Chem. 2007 282 8092 8098 10.1074/jbc.M608742200 17179156 68 Bain J. Plater L. Elliott M. Shpiro N. Hastie C.J. McLauchlan H. Klevernic I. Arthur J.S. Alessi D.R. Cohen P. The selectivity of protein kinase inhibitors: a further update Biochem. J. 2007 408 297 315 10.1042/BJ20070797 17850214 69 Gadalla A.E. Pearson T. Currie A.J. Dale N. Hawley S.A. Sheehan M. Hirst W. Michel A.D. Randall A. Hardie D.G. Frenguelli B.G. AICA riboside both activates AMP-activated protein kinase and competes with adenosine for the nucleoside transporter in the CA1 region of the rat hippocampus J. Neurochem. 2004 88 1272 1282 10.1046/j.1471-4159.2003.02253.x 15009683 70 Murali S. Nurse C.A. Purinergic signaling mediates bidirectional crosstalk between chemoreceptor type I and glial-like type II cells of the rat carotid body J. Physiol. 2016 594 391 406 10.1113/JP271494 26537220 71 Lantier L. Fentz J. Mounier R. Leclerc J. Treebak J.T. Pehmoller C. Sanz N. Sakakibara I. Saint-Amand E. Rimbaud S. AMPK controls exercise endurance, mitochondrial oxidative capacity, and skeletal muscle integrity FASEB J. 2014 28 3211 3224 10.1096/fj.14-250449 24652947 72 Hasenour C.M. Ridley D.E. Hughey C.C. James F.D. Donahue E.P. Shearer J. Viollet B. Foretz M. Wasserman D.H. 5-Aminoimidazole-4-carboxamide-1-beta-D-ribofuranoside (AICAR) effect on glucose production, but not energy metabolism, is independent of hepatic AMPK in vivo J. Biol. Chem. 2014 289 5950 5959 10.1074/jbc.M113.528232 24403081 73 Goransson O. McBride A. Hawley S.A. Ross F.A. Shpiro N. Foretz M. Viollet B. Hardie D.G. Sakamoto K. Mechanism of action of A-769662, a valuable tool for activation of AMP-activated protein kinase J. Biol. Chem. 2007 282 32549 32560 10.1074/jbc.M706536200 17855357 74 Buckler K.J. TASK channels in arterial chemoreceptors and their role in oxygen and acid sensing Pflugers Arch. 2015 467 1013 1025 10.1007/s00424-015-1689-1 25623783 75 Kim D. Kang D. Martin E.A. Kim I. Carroll J.L. Effects of modulators of AMP-activated protein kinase on TASK-1/3 and intracellular Ca(2+) concentration in rat carotid body glomus cells Respir. Physiol. Neurobiol. 2014 195 19 26 10.1016/j.resp.2014.01.020 24530802 76 Lizcano J.M. Goransson O. Toth R. Deak M. Morrice N.A. Boudeau J. Hawley S.A. Udd L. Makela T.P. Hardie D.G. Alessi D.R. LKB1 is a master kinase that activates 13 kinases of the AMPK subfamily, including MARK/PAR-1 EMBO J. 2004 23 833 843 10.1038/sj.emboj.7600110 14976552 77 Murali S. Nurse C.A. Purinergic signalling mediates bidirectional crosstalk between chemoreceptor type I and glial-like type II cells of the rat carotid body J. Physiol. 2016 594 391 406 10.1113/JP271494 26537220 78 Koh H.J. Arnolds D.E. Fujii N. Tran T.T. Rogers M.J. Jessen N. Li Y. Liew C.W. Ho R.C. Hirshman M.F. Skeletal muscle-selective knockout of LKB1 increases insulin sensitivity, improves glucose homeostasis, and decreases TRB3 Mol. Cell. Biol. 2006 26 8217 8227 10.1128/MCB.00979-06 16966378 79 Shaw R.J. Lamia K.A. Vasquez D. Koo S.H. Bardeesy N. Depinho R.A. Montminy M. Cantley L.C. The kinase LKB1 mediates glucose homeostasis in liver and therapeutic effects of metformin Science 2005 310 1642 1646 10.1126/science.1120781 16308421 80 Gan B. Hu J. Jiang S. Liu Y. Sahin E. Zhuang L. Fletcher-Sananikone E. Colla S. Wang Y.A. Chin L. Depinho R.A. Lkb1 regulates quiescence and metabolic homeostasis of haematopoietic stem cells Nature 2010 468 701 704 10.1038/nature09595 21124456 81 Gurumurthy S. Xie S.Z. Alagesan B. Kim J. Yusuf R.Z. Saez B. Tzatsos A. Ozsolak F. Milos P. Ferrari F. The Lkb1 metabolic sensor maintains haematopoietic stem cell survival Nature 2010 468 659 663 10.1038/nature09572 21124451 82 Patel K. Foretz M. Marion A. Campbell D.G. Gourlay R. Boudaba N. Tournier E. Titchenell P. Peggie M. Deak M. The LKB1-salt-inducible kinase pathway functions as a key gluconeogenic suppressor in the liver Nat. Commun. 2014 5 4535 10.1038/ncomms5535 25088745 83 Choi S. Lim D.S. Chung J. Feeding and fasting signals converge on the LKB1-SIK3 pathway to regulate lipid metabolism in drosophila PLoS Genet. 2015 11 e1005263 10.1371/journal.pgen.1005263 25996931 84 Swisa A. Granot Z. Tamarina N. Sayers S. Bardeesy N. Philipson L. Hodson D.J. Wikstrom J.D. Rutter G.A. Leibowitz G. Loss of liver kinase B1 (LKB1) in beta cells enhances glucose-stimulated insulin secretion despite profound mitochondrial defects J. Biol. Chem. 2015 290 20934 20946 10.1074/jbc.M115.639237 26139601 85 Lopez-Barneo J. Lopez-Lopez J.R. Urena J. Gonzalez C. Chemotransduction in the carotid body: K+ current modulated by PO2 in type I chemoreceptor cells Science 1988 241 580 582 10.1126/science.2456613 2456613 86 Stea A. Nurse C.A. Whole-cell and perforated-patch recordings from O2 -sensitive rat carotid body cells grown in short- and long-term culture Pflugers Arch. 1991 418 93 101 10.1007/BF00370457 2041730 87 Delpiano M.A. Hescheler J. Evidence for a PO2-sensitive K+ channel in the type-I cell of the rabbit carotid body FEBS Lett. 1989 249 195 198 10.1016/0014-5793(89)80623-4 2737279 88 Hescheler J. Delpiano M.A. Acker H. Pietruschka F. Ionic currents on type-I cells of the rabbit carotid body measured by voltage-clamp experiments and the effect of hypoxia Brain Res. 1989 486 79 88 10.1016/0006-8993(89)91280-8 2720436 89 Peers C. Hypoxic suppression of K+ currents in type I carotid body cells: selective effect on the Ca2(+)-activated K+ current Neurosci. Lett. 1990 119 253 256 10.1016/0304-3940(90)90846-2 1704113 90 Buckler K.J. A novel oxygen-sensitive potassium current in rat carotid body type I cells J. Physiol. 1997 498 649 662 10.1113/jphysiol.1997.sp021890 9051577 91 Kim D. Cavanaugh E.J. Kim I. Carroll J.L. Heteromeric TASK-1/TASK-3 is the major oxygen-sensitive background K+ channel in rat carotid body glomus cells J. Physiol. 2009 587 2963 2975 10.1113/jphysiol.2009.171181 19403596 92 Ortega-Saenz P. Levitsky K.L. Marcos-Almaraz M.T. Bonilla-Henao V. Pascual A. Lopez-Barneo J. Carotid body chemosensory responses in mice deficient of TASK channels J. Gen. Physiol. 2010 135 379 392 10.1085/jgp.200910302 20351062 93 Perez-Garcia M.T. Colinas O. Miguel-Velado E. Moreno-Dominguez A. Lopez-Lopez J.R. Characterization of the Kv channels of mouse carotid body chemoreceptor cells and their role in oxygen sensing J. Physiol. 2004 557 457 471 10.1113/jphysiol.2004.062281 15034123 94 Lopez-Lopez J.R. De Luis D.A. Gonzalez C. Properties of a transient K+ current in chemoreceptor cells of rabbit carotid body J. Physiol. 1993 460 15 32 10.1113/jphysiol.1993.sp019456 8387583 95 Hatton C.J. Carpenter E. Pepper D.R. Kumar P. Peers C. Developmental changes in isolated rat type I carotid body cell K+ currents and their modulation by hypoxia J. Physiol. 1997 501 49 58 10.1111/j.1469-7793.1997.049bo.x 9174993 96 Wasicko M.J. Breitwieser G.E. Kim I. Carroll J.L. Postnatal development of carotid body glomus cell response to hypoxia Respir. Physiol. Neurobiol. 2006 154 356 371 10.1016/j.resp.2006.01.003 16466972 97 Varas R. Wyatt C.N. Buckler K.J. Modulation of TASK-like background potassium channels in rat arterial chemoreceptor cells by intracellular ATP and other nucleotides J. Physiol. 2007 583 521 536 10.1113/jphysiol.2007.135657 17615104 98 Duncan P.J. Sengul S. Tabak J. Ruth P. Bertram R. Shipston M.J. Large conductance Ca(2)(+)-activated K(+) (BK) channels promote secretagogue-induced transition from spiking to bursting in murine anterior pituitary corticotrophs J. Physiol. 2015 593 1197 1211 10.1113/jphysiol.2015.284471 25615909 99 Buckler K.J. Turner P.J. Oxygen sensitivity of mitochondrial function in rat arterial chemoreceptor cells J. Physiol. 2013 591 3549 3563 10.1113/jphysiol.2013.257741 23671162 100 Turner P.J. Buckler K.J. Oxygen and mitochondrial inhibitors modulate both monomeric and heteromeric TASK-1 and TASK-3 channels in mouse carotid body type-1 cells J. Physiol. 2013 591 5977 5998 10.1113/jphysiol.2013.262022 24042502 101 Yuan G. Vasavda C. Peng Y.J. Makarenko V.V. Raghuraman G. Nanduri J. Gadalla M.M. Semenza G.L. Kumar G.K. Snyder S.H. Prabhakar N.R. Protein kinase G-regulated production of H2S governs oxygen sensing Sci. Signal. 2015 8 ra37 10.1126/scisignal.2005846 25900831 102 Buckler K.J. Effects of exogenous hydrogen sulphide on calcium signalling, background (TASK) K channel activity and mitochondrial function in chemoreceptor cells Pflugers Arch. 2012 463 743 754 10.1007/s00424-012-1089-8 22419174 103 Evans A.M. Hardie D.G. Peers C. Mahmoud A. Hypoxic pulmonary vasoconstriction: mechanisms of oxygen-sensing Curr. Opin. Anaesthesiol. 2011 24 13 20 10.1097/ACO.0b013e3283421201 21157304 104 Chang A.J. Ortega F.E. Riegler J. Madison D.V. Krasnow M.A. Oxygen regulation of breathing through an olfactory receptor activated by lactate Nature 2015 527 240 244 10.1038/nature15721 26560302 105 Jia X. Burggren W. Developmental changes in chemoreceptive control of gill ventilation in larval bullfrogs (Rana catesbeiana). II. Sites of O2-sensitive chemoreceptors J. Exp. Biol. 1997 200 2237 2248 9320153 106 Porteus C. Hedrick M.S. Hicks J.W. Wang T. Milsom W.K. Time domains of the hypoxic ventilatory response in ectothermic vertebrates J. Comp. Physiol. B 2011 181 311 333 10.1007/s00360-011-0554-6 21312038 107 Dampney R.A. Moon E.A. Role of ventrolateral medulla in vasomotor response to cerebral ischemia Am. J. Physiol. 1980 239 H349 H358 7435582 108 Paton J.F. Dickinson C.J. Mitchell G. Harvey Cushing and the regulation of blood pressure in giraffe, rat and man: introducing 'Cushing's mechanism' Exp. Physiol. 2009 94 11 17 10.1113/expphysiol.2008.043455 18820004 109 Sun M.K. Jeske I.T. Reis D.J. Cyanide excites medullary sympathoexcitatory neurons in rats Am. J. Physiol. 1992 262 R182 R189 1539725 110 Sun M.K. Reis D.J. Differential responses of barosensitive neurons of rostral ventrolateral medulla to hypoxia in rats Brain Res. 1993 609 333 337 10.1016/0006-8993(93)90892-Q 8508315 111 Curran A.K. Rodman J.R. Eastwood P.R. Henderson K.S. Dempsey J.A. Smith C.A. Ventilatory responses to specific CNS hypoxia in sleeping dogs J. Appl. Physiol. 2000 88 1840 1852 10797149 112 Smith C.A. Engwall M.J. Dempsey J.A. Bisgard G.E. Effects of specific carotid body and brain hypoxia on respiratory muscle control in the awake goat J. Physiol. 1993 460 623 640 10.1113/jphysiol.1993.sp019490 8487210 113 Hill A.A. Garcia, 3rd A.J. Zanella S. Upadhyaya R. Ramirez J.M. Graded reductions in oxygenation evoke graded reconfiguration of the isolated respiratory network J. Neurophysiol. 2011 105 625 639 10.1152/jn.00237.2010 21084689 114 Wade J.G. Larson C.P. Jr Hickey R.F. Ehrenfeld W.K. Severinghaus J.W. Effect of carotid endarterectomy on carotid chemoreceptor and baroreceptor function in man N. Engl. J. Med. 1970 282 823 829 10.1056/NEJM197004092821501 5418544 115 Roux J.C. Pequignot J.M. Dumas S. Pascual O. Ghilini G. Pequignot J. Mallet J. Denavit-Saubie M. O2-sensing after carotid chemodenervation: hypoxic ventilatory responsiveness and upregulation of tyrosine hydroxylase mRNA in brainstem catecholaminergic cells Eur. J. Neurosci. 2000 12 3181 3190 10.1046/j.1460-9568.2000.00208.x 10998102 116 Roux J.C. Villard L. Biogenic amines in Rett syndrome: the usual suspects Behav. Genet. 2010 40 59 75 10.1007/s10519-009-9303-y 19851857 117 King T.L. Heesch C.M. Clark C.G. Kline D.D. Hasser E.M. Hypoxia activates nucleus tractus solitarii neurons projecting to the paraventricular nucleus of the hypothalamus Am. J. Physiol. Regul. Integr. Comp. Physiol. 2012 302 R1219 R1232 10.1152/ajpregu.00028.2012 22403798 118 Koshiya N. Guyenet P.G. NTS neurons with carotid chemoreceptor inputs arborize in the rostral ventrolateral medulla Am. J. Physiol. 1996 270 R1273 R1278 8764294 119 Alheid G.F. Jiao W. McCrimmon D.R. Caudal nuclei of the rat nucleus of the solitary tract differentially innervate respiratory compartments within the ventrolateral medulla Neuroscience 2011 190 207 227 10.1016/j.neuroscience.2011.06.005 21704133 120 Smith J.C. Ellenberger H.H. Ballanyi K. Richter D.W. Feldman J.L. Pre-Botzinger complex: a brainstem region that may generate respiratory rhythm in mammals Science 1991 254 726 729 10.1126/science.1683005 1683005 121 Guyenet P.G. Stornetta R.L. Bochorishvili G. Depuy S.D. Burke P.G. Abbott S.B. C1 neurons: the body's EMTs Am. J. Physiol. Regul. Integr. Comp. Physiol. 2013 305 R187 R204 10.1152/ajpregu.00054.2013 23697799 122 Yamamoto K. Lalley P. Mifflin S. Acute intermittent optogenetic stimulation of nucleus tractus solitarius neurons induces sympathetic long-term facilitation Am. J. Physiol. Regul. Integr. Comp. Physiol. 2015 308 R266 R275 10.1152/ajpregu.00381.2014 25519734 123 Vardhan A. Kachroo A. Sapru H.N. Excitatory amino acid receptors in commissural nucleus of the NTS mediate carotid chemoreceptor responses Am. J. Physiol. 1993 264 R41 50 8381618 124 Aicher S.A. Saravay R.H. Cravo S. Jeske I. Morrison S.F. Reis D.J. Milner T.A. Monosynaptic projections from the nucleus tractus solitarii to C1 adrenergic neurons in the rostral ventrolateral medulla: comparison with input from the caudal ventrolateral medulla J. Comp. Neurol. 1996 373 62 75 10.1002/(SICI)1096-9861(19960909)373:1<62::AID-CNE6>3.0.CO;2-B 8876463 125 Harris A.P. Helou S. Traystman R.J. Jones M.D. Jr Koehler R.C. Efficacy of the cushing response in maintaining cerebral blood flow in premature and near-term fetal sheep Pediatr. Res. 1998 43 50 56 10.1203/00006450-199801000-00008 9432112 126 Vallortigara G. Rogers L.J. Bisazza A. Possible evolutionary origins of cognitive brain lateralization Brain Res. Brain Res. Rev. 1999 30 164 175 10.1016/S0165-0173(99)00012-0 10525173 127 Dadda M. Zandona E. Agrillo C. Bisazza A. The costs of hemispheric specialization in a fish Proc. Biol. Sci. 2009 276 4399 4407 10.1098/rspb.2009.1406 19793754 128 Stornetta R.L. Sevigny C.P. Guyenet P.G. Vesicular glutamate transporter DNPI/VGLUT2 mRNA is present in C1 and several other groups of brainstem catecholaminergic neurons J. Comp. Neurol. 2002 444 191 206 10.1002/cne.10141 11840474 129 Gozal D. Gozal E. Torres J.E. Gozal Y.M. Nuckton T.J. Hornby P.J. Nitric oxide modulates ventilatory responses to hypoxia in the developing rat Am. J. Respir. Crit. Care Med. 1997 155 1755 1762 10.1164/ajrccm.155.5.9154888 9154888 130 Lipton A.J. Johnson M.A. Macdonald T. Lieberman M.W. Gozal D. Gaston B. S-nitrosothiols signal the ventilatory response to hypoxia Nature 2001 413 171 174 10.1038/35093117 11557982 131 Murphy B.A. Fakira K.A. Song Z. Beuve A. Routh V.H. AMP-activated protein kinase and nitric oxide regulate the glucose sensitivity of ventromedial hypothalamic glucose-inhibited neurons Am. J. Physiol. Cell Physiol. 2009 297 C750 C758 10.1152/ajpcell.00127.2009 19570894 132 Culmsee C. Monnig J. Kemp B.E. Mattson M.P. AMP-activated protein kinase is highly expressed in neurons in the developing rat brain and promotes neuronal survival following glucose deprivation J. Mol. Neurosci. 2001 17 45 58 10.1385/JMN:17:1:45 11665862 133 Cheng F. Xie S. Guo M. Fang H. Li X. Yin J. Lu G. Li Y. Ji X. Yu S. Altered glucose metabolism and preserved energy charge and neuronal structures in the brain of mouse intermittently exposed to hypoxia J. Chem. Neuroanat. 2011 42 65 71 10.1016/j.jchemneu.2011.06.004 21718782 134 Almeida A. Moncada S. Bolanos J.P. Nitric oxide switches on glycolysis through the AMP protein kinase and 6-phosphofructo-2-kinase pathway Nat. Cell. Biol. 2004 6 45 51 10.1038/ncb1080 14688792 135 Bucher E.S. Fox M.E. Kim L. Kirkpatrick D.C. Rodeberg N.T. Belle A.M. Wightman R.M. Medullary norepinephrine neurons modulate local oxygen concentrations in the bed nucleus of the stria terminalis J. Cereb. Blood Flow Metab. 2014 34 1128 1137 10.1038/jcbfm.2014.60 24714037 136 Smith C.A. Forster H.V. Blain G.M. Dempsey J.A. An interdependent model of central/peripheral chemoreception: evidence and implications for ventilatory control Respir. Physiol. Neurobiol. 2010 173 288 297 10.1016/j.resp.2010.02.015 20206717 137 Conde S.V. Monteiro E.C. Rigual R. Obeso A. Gonzalez C. Hypoxic intensity: a determinant for the contribution of ATP and adenosine to the genesis of carotid body chemosensory activity J. Appl. Physiol. 2012 112 2002 2010 10.1152/japplphysiol.01617.2011 22500005 138 Xu F. Xu J. Tse F.W. Tse A. Adenosine stimulates depolarization and rise in cytoplasmic [Ca2+] in type I cells of rat carotid bodies Am. J. Physiol. Cell Physiol. 2006 290 C1592 C1598 10.1152/ajpcell.00546.2005 16436472 139 von Euler U.S. Liljestrand G. Observations on the pulmonary arterial blood pressure in the cat Acta Physiol. Scand. 1946 12 301 320 10.1111/j.1748-1716.1946.tb00389.x 140 Bradford J.R. Dean H.P. The pulmonary circulation J. Physiol. 1894 16 34–158 125 10.1113/jphysiol.1894.sp000493 141 Roy C.S. Sherrington C.S. On the regulation of the blood-supply of the brain J. Physiol. 1890 11 85 117 158 10.1113/jphysiol.1890.sp000321 16991945 142 Goirand F. Solar M. Athea Y. Viollet B. Mateo P. Fortin D. Leclerc J. Hoerter J. Ventura-Clapier R. Garnier A. Activation of AMP kinase alpha1 subunit induces aortic vasorelaxation in mice J. Physiol. 2007 581 1163 1171 10.1113/jphysiol.2007.132589 17446219 143 Nisell O. The influence of blood gases on the pulmonary vessels of the cat Acta Physiol. Scand. 1951 23 85 90 10.1111/j.1748-1716.1951.tb00797.x 14856797 144 Naeije R. Lejeune P. Leeman M. Melot C. Closset J. Pulmonary vascular responses to surgical chemodenervation and chemical sympathectomy in dogs J. Appl. Physiol. 1989 66 42 50 2917946 145 Lejeune P. Vachiery J.L. Leeman M. Brimioulle S. Hallemans R. Melot C. Naeije R. Absence of parasympathetic control of pulmonary vascular pressure-flow plots in hyperoxic and hypoxic dogs Respir. Physiol. 1989 78 123 133 10.1016/0034-5687(89)90046-7 2609023 146 Robin E.D. Theodore J. Burke C.M. Oesterle S.N. Fowler M.B. Jamieson S.W. Baldwin J.C. Morris A.J. Hunt S.A. Vankessel A. Hypoxic pulmonary vasoconstriction persists in the human transplanted lung Clin. Sci. (Lond.) 1987 72 283 287 10.1042/cs0720283 3545645 147 Waypa G.B. Chandel N.S. Schumacker P.T. Model for hypoxic pulmonary vasoconstriction involving mitochondrial oxygen sensing Circ. Res. 2001 88 1259 1266 10.1161/hh1201.091960 11420302 148 Leach R.M. Hill H.M. Snetkov V.A. Robertson T.P. Ward J.P. Divergent roles of glycolysis and the mitochondrial electron transport chain in hypoxic pulmonary vasoconstriction of the rat: identity of the hypoxic sensor J. Physiol. 2001 536 211 224 10.1111/j.1469-7793.2001.00211.x 11579170 149 Weissmann N. Ebert N. Ahrens M. Ghofrani H.A. Schermuly R.T. Hanze J. Fink L. Rose F. Conzen J. Seeger W. Grimminger F. Effects of mitochondrial inhibitors and uncouplers on hypoxic vasoconstriction in rabbit lungs Am. J. Respir. Cell Mol. Biol. 2003 29 721 732 10.1165/rcmb.2002-0217OC 12791676 150 Evans A.M. Hardie D.G. Galione A. Peers C. Kumar P. Wyatt C.N. AMP-activated protein kinase couples mitochondrial inhibition by hypoxia to cell-specific Ca2+ signalling mechanisms in oxygen-sensing cells Novartis Found. Symp. 2006 272 234 252 10.1002/SERIES1767 16686439 151 Owen M.R. Doran E. Halestrap A.P. Evidence that metformin exerts its anti-diabetic effects through inhibition of complex 1 of the mitochondrial respiratory chain Biochem. J. 2000 348 607 614 10.1042/bj3480607 10839993 152 Robertson T.P. Mustard K.J. Lewis T.H. Clark J.H. Wyatt C.N. Blanco E.A. Peers C. Hardie D.G. Evans A.M. AMP-activated protein kinase and hypoxic pulmonary vasoconstriction Eur. J. Pharmacol. 2008 595 39 43 10.1016/j.ejphar.2008.07.035 18703047 153 Moral-Sanz J. Lewis S. Thomson A. Moran C. Viollet B. Foretz M. Evans A.M. AMP-activated protein kinase is necessary for hypoxic pulmonary vasoconstriction Proc. Physiol. Soc. 2015 34 PC265 154 Yuan J.X. Aldinger A.M. Juhaszova M. Wang J. Conte J.V. Gaine S.P. Jr. Orens J.B. Rubin L.J. Dysfunctional voltage-gated K+ channels in pulmonary artery smooth muscle cells of patients with primary pulmonary hypertension Circulation 1998 98 1400 1406 10.1161/01.CIR.98.14.1400 9760294 155 Morales-Cano D. Menendez C. Moreno E. Moral-Sanz J. Barreira B. Galindo P. Pandolfi R. Jimenez R. Moreno L. Cogolludo A. The flavonoid quercetin reverses pulmonary hypertension in rats PLoS One 2014 9 e114492 10.1371/journal.pone.0114492 25460361 156 Lv Y. Tang L.L. Wei J.K. Xu X.F. Gu W. Fu L.C. Zhang L.Y. Du L.Z. Decreased Kv1.5 expression in intrauterine growth retardation rats with exaggerated pulmonary hypertension Am. J. Physiol. Lung Cell. Mol. Physiol. 2013 305 L856 L865 10.1152/ajplung.00179.2013 24077947 157 Remillard C.V. Tigno D.D. Platoshyn O. Burg E.D. Brevnova E.E. Conger D. Nicholson A. Rana B.K. Channick R.N. Rubin L.J. Function of Kv1.5 channels and genetic variations of KCNA5 in patients with idiopathic pulmonary arterial hypertension Am. J. Physiol. Cell Physiol. 2007 292 C1837 C1853 10.1152/ajpcell.00405.2006 17267549 158 Burg E.D. Platoshyn O. Tsigelny I.F. Lozano-Ruiz B. Rana B.K. Yuan J.X. Tetramerization domain mutations in KCNA5 affect channel kinetics and cause abnormal trafficking patterns Am. J. Physiol. Cell Physiol. 2010 298 C496 C509 10.1152/ajpcell.00464.2009 20018952 159 Michelakis E.D. McMurtry M.S. Wu X.C. Dyck J.R. Moudgil R. Hopkins T.A. Lopaschuk G.D. Puttagunta L. Waite R. Archer S.L. Dichloroacetate, a metabolic modulator, prevents and reverses chronic hypoxic pulmonary hypertension in rats: role of increased expression and activity of voltage-gated potassium channels Circulation 2002 105 244 250 10.1161/hc0202.101974 11790708 160 Guignabert C. Izikki M. Tu L.I. Li Z. Zadigue P. Barlier-Mur A.M. Hanoun N. Rodman D. Hamon M. Adnot S. Eddahibi S. Transgenic mice overexpressing the 5-hydroxytryptamine transporter gene in smooth muscle develop pulmonary hypertension Circ. Res. 2006 98 1323 1330 10.1161/01.RES.0000222546.45372.a0 16614302 161 Young K.A. Ivester C. West J. Carr M. Rodman D.M. BMP signaling controls PASMC KV channel expression in vitro and in vivo Am. J. Physiol. Lung Cell. Mol. Physiol. 2006 290 L841 L848 10.1152/ajplung.00158.2005 16339782 162 Bonnet S. Michelakis E.D. Porter C.J. Andrade-Navarro M.A. Thebaud B. Bonnet S. Haromy A. Harry G. Moudgil R. McMurtry M.S. An abnormal mitochondrial-hypoxia inducible factor-1alpha-Kv channel pathway disrupts oxygen sensing and triggers pulmonary arterial hypertension in fawn hooded rats: similarities to human pulmonary arterial hypertension Circulation 2006 113 2630 2641 10.1161/CIRCULATIONAHA.105.609008 16735674 163 Brevnova E.E. Platoshyn O. Zhang S. Yuan J.X. Overexpression of human KCNA5 increases IK V and enhances apoptosis Am. J. Physiol. Cell Physiol. 2004 287 C715 C722 10.1152/ajpcell.00050.2004 15140747 164 Krick S. Platoshyn O. Sweeney M. Kim H. Yuan J.X. Activation of K+ channels induces apoptosis in vascular smooth muscle cells Am. J. Physiol. Cell Physiol. 2001 280 C970 C979 11245614 165 Moudgil R. Michelakis E.D. Archer S.L. The role of k+ channels in determining pulmonary vascular tone, oxygen sensing, cell proliferation, and apoptosis: implications in hypoxic pulmonary vasoconstriction and pulmonary arterial hypertension Microcirculation 2006 13 615 632 10.1080/10739680600930222 17085423 166 Cidad P. Jimenez-Perez L. Garcia-Arribas D. Miguel-Velado E. Tajada S. Ruiz-McDavitt C. Lopez-Lopez J.R. Perez-Garcia M.T. Kv1.3 channels can modulate cell proliferation during phenotypic switch by an ion-flux independent mechanism Arterioscler. Thromb. Vasc. Biol. 2012 32 1299 1307 10.1161/ATVBAHA.111.242727 22383699 167 Cidad P. Miguel-Velado E. Ruiz-McDavitt C. Alonso E. Jimenez-Perez L. Asuaje A. Carmona Y. Garcia-Arribas D. Lopez J. Marroquin Y. Kv1.3 channels modulate human vascular smooth muscle cells proliferation independently of mTOR signaling pathway Pflugers Arch. 2014 467 1711 1722 25208915 168 Ibe J.C. Zhou Q. Chen T. Tang H. Yuan J.X. Raj J.U. Zhou G. Adenosine monophosphate-activated protein kinase is required for pulmonary artery smooth muscle cell survival and the development of hypoxic pulmonary hypertension Am. J. Respir. Cell Mol. Biol. 2013 49 609 618 10.1165/rcmb.2012-0446OC 23668615 169 Goncharov D.A. Kudryashova T.V. Ziai H. Ihida-Stansbury K. DeLisser H. Krymskaya V.P. Tuder R.M. Kawut S.M. Goncharova E.A. Mammalian target of rapamycin complex 2 (mTORC2) coordinates pulmonary artery smooth muscle cell metabolism, proliferation, and survival in pulmonary arterial hypertension Circulation 2014 129 864 874 10.1161/CIRCULATIONAHA.113.004581 24270265 170 Skeffington K.L. Higgins J.S. Mahmoud A.D. Evans A.M. Sferruzzi-Perri A.N. Fowden A.L. Yung H.W. Burton G.J. Giussani D.A. Moore L.G. Hypoxia, AMPK activation and uterine artery vasoreactivity J. Physiol. 2016 594 1357 1369 10.1113/JP270995 26110512 171 Lahm T. Tuder R.M. Petrache I. Progress in solving the sex hormone paradox in pulmonary hypertension Am. J. Physiol. Lung Cell. Mol. Physiol. 2014 307 L7 L26 10.1152/ajplung.00337.2013 24816487 172 Chau E.H. Lam D. Wong J. Mokhlesi B. Chung F. Obesity hypoventilation syndrome: a review of epidemiology, pathophysiology, and perioperative considerations Anesthesiology 2012 117 188 205 10.1097/ALN.0b013e31825add60 22614131 173 Schneider H. Schubert K.M. Blodow S. Kreutz C.P. Erdogmus S. Wiedenmann M. Qiu J. Fey T. Ruth P. Lubomirov L.T. Pfitzer G. Mederos Y.S.M. Hardie D.G. Gudermann T. Pohl U. AMPK dilates resistance arteries via activation of SERCA and BKCa channels in smooth muscle Hypertension 2015 66 108 116 10.1161/HYPERTENSIONAHA.115.05514 26034200 174 Ruderman N.B. Carling D. Prentki M. Cacicedo J.M. AMPK, insulin resistance, and the metabolic syndrome J. Clin. Invest. 2013 123 2764 2772 10.1172/JCI67227 23863634 175 Vgontzas A.N. Bixler E.O. Chrousos G.P. Sleep apnea is a manifestation of the metabolic syndrome Sleep Med. Rev. 2005 9 211 224 10.1016/j.smrv.2005.01.006 15893251 176 Rajamohan F. Reyes A.R. Frisbie R.K. Hoth L.R. Sahasrabudhe P. Magyar R. Landro J.A. Withka J.M. Caspers N.L. Calabrese M.F. Probing the enzyme kinetics, allosteric modulation and activation of alpha-1 and alpha-2 subunit containing AMP-activated protein kinase (AMPK) heterotrimeric complexes by pharmacological and physiological activators Biochem. J. 2016 473 581 592 10.1042/BJ20151051 26635351 177 Gomez-Galeno J.E. Dang Q. Nguyen T.H. Boyer S.H. Grote M.P. Sun Z. Chen M. Craigo W.A. van Poelje P.D. MacKenna D.A. A potent and selective AMPK activator that inhibits de novo lipogenesis ACS Med. Chem. Lett. 2010 1 478 482 10.1021/ml100143q 24900234 178 Scott J.W. Ling N. Issa S.M. Dite T.A. O'Brien M.T. Chen Z.P. Galic S. Langendorf C.G. Steinberg G.R. Kemp B.E. Oakhill J.S. Small molecule drug A-769662 and AMP synergistically activate naive AMPK independent of upstream kinase signaling Chem. Biol. 2014 21 619 627 10.1016/j.chembiol.2014.03.006 24746562 179 Hardie D.G. Salt I.P. Hawley S.A. Davies S.P. AMP-activated protein kinase: an ultrasensitive system for monitoring cellular energy charge Biochem. J. 1999 338 717 722 10.1042/bj3380717 10051444 180 Rekling J.C. Feldman J.L. PreBotzinger complex and pacemaker neurons: hypothesized site and kernel for respiratory rhythm generation Annu. Rev. Physiol. 1998 60 385 405 10.1146/annurev.physiol.60.1.385 9558470

PMC005xxxxxx/PMC5003692.txt

==== Front Biochem JBiochem. JppbiochemjBJBiochemical Journal0264-60211470-8728Portland Press Ltd. BCJ2016024010.1042/BCJ20160240Review ArticlesReview Article5551852Targeting protein function: the expanding toolkit for conditional disruption Tools for conditional disruption of protein functionA.E. Campbell and D. BennettCampbell Amy E. *Bennett Daimark *1* Department of Biochemistry, Institute of Integrative Biology, University of Liverpool, Crown Street, Liverpool L69 7ZB, U.K.1 To whom correspondence should be addressed (email daimark.bennett@liverpool.ac.uk).30 8 2016 1 9 2016 473 17 172573 2589 18 3 2016 16 5 2016 20 5 2016 © 2016 The Author(s)2016This is an open access article published by Portland Press Limited on behalf of the Biochemical Society and distributed under the Creative Commons Attribution Licence 4.0 (CC BY).A major objective in biological research is to understand spatial and temporal requirements for any given gene, especially in dynamic processes acting over short periods, such as catalytically driven reactions, subcellular transport, cell division, cell rearrangement and cell migration. The interrogation of such processes requires the use of rapid and flexible methods of interfering with gene function. However, many of the most widely used interventional approaches, such as RNAi or CRISPR (clustered regularly interspaced short palindromic repeats)-Cas9 (CRISPR-associated 9), operate at the level of the gene or its transcripts, meaning that the effects of gene perturbation are exhibited over longer time frames than the process under investigation. There has been much activity over the last few years to address this fundamental problem. In the present review, we describe recent advances in disruption technologies acting at the level of the expressed protein, involving inducible methods of protein cleavage, (in)activation, protein sequestration or degradation. Drawing on examples from model organisms we illustrate the utility of fast-acting techniques and discuss how different components of the molecular toolkit can be employed to dissect previously intractable biochemical processes and cellular behaviours. biochemical techniques and resourcescellular targetingchemical biologyprotein dynamicsoptogenetics ==== Body INTRODUCTION Genetic manipulation, which operates at the level of the gene or its transcribed product, has proven to be indispensable for the identification of molecular components required for biological processes and understanding how these components act together to construct functional cells, tissues and organisms. There are now a myriad of tools for mutational analysis that have been accumulated for over a century, fuelling gene discovery through forward genetic screens and facilitating reverse genetics to probe gene function [1–3]. Recent developments, in particular CRISPR (clustered regularly interspaced short palindromic repeats)-based approaches and RNAi, promise to further transform our understanding by facilitating high-throughput reverse genetics and gene editing with nucleotide-level precision [4,5]. Gene overexpression technologies have also become extremely advanced, featuring a high level of spatial and temporal control that has enabled a range of developmentally targeted gain-of-function studies. Heterologous gene expression systems, such as yeast Gal4/UAS (GAL4 upstream activating sequence), have proven to be particularly versatile in this regard [6]. Yet, quite often, the genetic tools used to perturb gene function are not able to keep pace with dynamic biological events that can act over different timescales from less than a second to many hours or days, depending on the process. Processes acting over short times are particularly recalcitrant to genetic analysis because there is a considerable delay between perturbations at the transcriptional or post-transcriptional level and the corresponding effect on the encoded protein. Consequently, genetic approaches are typically incapable of selectively disrupting the encoded protein of interest at the time when the effects of experimental manipulation are measured. This is particularly a problem for analysing the function of proteins required at multiple points of the same process. A partial solution to this problem might be to monitor the process under investigation in real-time, so as to capture more information about the biological effects than can be revealed at a fixed time point [7]. Yet significant limitations remain, especially for the analysis of protein function in vivo. For instance, mutation of a gene required for development might result in early lethality making later processes impossible to analyse. Alternatively in the case of transgenic RNAi, which typically results in partial loss-of-function, it may not be possible to drive expression of the transgene early enough for knockdown to occur before the process has already taken place. This is particularly a problem early in development, where it can take a considerable time for the maternal contribution of RNA and protein to be exhausted. Another major consideration is the existence of compensatory homoeostatic mechanisms that may circumvent the requirement for the protein under investigation. Although this is often a reason cited for lack of knockout phenotypes in mice, the underlying mechanisms are frequently not well described. Studies employing fast-acting methods have the potential to resolve such issues, as illustrated by a study of the cell-surface glycoprotein CD44 [8]. CD44 isoforms act as co-receptors for the receptor tyrosine kinases c-Met and VEGFR-2 (vascular endothelial growth factor receptor 2), but do not produce overt phenotypes when knocked out in mice [8]. Using blocking antibodies, it was shown that acute disruption of CD44v6 inhibited cell proliferation and c-Met activation in wild-type mice, but that ICAM-1 (intercellular adhesion molecule 1) compensated for the CD44v6 isoform in CD44-null mice [8]. This study illustrates that rapid blockade of protein function can be a powerful way of resolving problems associated with slow-acting or constitutive methods of gene disruption. However, only a minority of proteins are currently open to pharmacological manipulation and the development of specific blocking reagents for every protein of interest based on their intrinsic properties is a long way away. A successful strategy that has been adopted by the research community over the last few years to increase the range of targets that can be manipulated pharmacologically has been to take well-characterized ligand-interaction domains from heterologous systems, and genetically engineer them into proteins of interest. In parallel to this chemical genetics approach, researchers have also found ways of incorporating domains responsive to other triggers, such as light, temperature and pH. This has spawned a new generation of tools that act directly at the level of the expressed protein and have the potential to provide insight into acute perturbations, give access to analysis over short times and allow reversible switching. Such tools can be broadly categorized according to their mode of action: those that disrupt protein activity through complex (de)formation, (in)activate proteins by induced splicing or cleavage, or directly target a protein for degradation through the endogenous cell machinery. In the present review, we focus primarily on tools for conditional control of protein function that fall into one of these three categories and are fast acting, providing examples of their application as a guide to researchers considering their use. COMPLEX FORMATION The promotion of interactions between proteins of interest represents a powerful strategy for the conditional control of protein activity. There are a number of different mechanisms by which such interactions can be engineered to spatiotemporally regulate protein activity in response to different stimuli, in a way that is both precise and fast acting (Figure 1, Table 1). Table 1 Summary of methods for conditional control of protein complex formation Can be induced by Technique Protein disruption Timescale Small molecule/hormone Light Temperature pH Chemically induced dimerization (CID) Sequestration/(in)activation min/h ✓ e.g. FK1012, rapamycin ✗ ✗ ✗ Knocksideways (KS) Sequestration min ✓ Rapamycin ✗ ✗ ✗ Light-activated dimerization Sequestration/(In)activation s/min ✗ ✓ ✗ ✗ Light-activated reversible inhibition by assembled trap (LARIAT) (In)activation s ✗ ✓ Blue light: 450–495 nm ✗ ✗ Light-oxygen-voltage (LOV) domains (In)activation s/min ✗ ✓ Blue light: 450–495 nm ✗ ✗ Figure 1 Schematic representation of methods for the control of protein activity by induced complex formation Methods described here can be broadly categorized by their method of induction. (A) Generalized mechanism for small-molecule-mediated approaches involving small-molecule CIDs to sequester protein activity by promoting (dis)aggregation, altering transcription or by changing the subcellular localization of target proteins. (B) A specific example of this approach is illustrated by the knocksideways (KS) method, which involves the rapid sequestration of target proteins to the mitochondria through rapamycin-mediated interaction between the FKBP-tagged target protein and the mitochondrially localized mito-RFP-FRB trap. (C) The generalized mechanism for light-based approaches induced by exposure to a specific wavelength of light. More specific examples of light-based methods are also illustrated in (D) and (E). (D) Schematic depiction of the LARIAT method, whereby target proteins tagged with CRY2 are sequestered into large protein complexes upon a light-induced conformational change to produce the photoactivated form of CRY2. This interacts with CIB1-bound multimeric complexes to aggregate the target protein and sequester activity. (E) Representation of LOV-domain-based approaches, which involve a light-induced conformational change leading to unravelling of the Jα helix and the loss of its interaction with the LOV domain leaving the target protein free to bind interactors and perform its function. Chemically induced dimerization One of the first mechanisms involving engineered protein complex formation utilized small molecules, referred to as chemical inducers of dimerization (CIDs), which simultaneously bind domains engineered into two proteins of interest, bringing them into close proximity and promoting their interaction (Figure 1A, Table 1). Methods involving CIDs can influence protein activity by promoting (dis)aggregation, altering transcription or by changing the sub cellular localization of target proteins. The application of CIDs to control protein activity and study protein function spans the last two decades, with the majority of applications utilizing naturally occurring CIDs, found to dimerize specific protein–ligand pairs [9–11]. For a small molecule to be successful in CID approaches it must have the ability to simultaneously bind two proteins, and therefore must have two high-affinity highly specific protein-binding domains, joined in a way that allows both target proteins to bind and interact [12]. FK1012 The first, naturally occurring CID, FK1012, was reported in 1993 [14]. FK1012 is a derivative of the immunosuppressant drug FK506, which was found in 1991 to be capable of binding calcineurin and FK506-binding protein [12] (FKBP12) with high affinity [13]. FK1012 is a synthetic dimer of FK506 that lacks the intrinsic biological activity of FK506 and has since been utilized as a CID to bind multiple FKBP12 domains [14], bringing target proteins together in a defined reversible fashion and demonstrating the subtleties required for successful CID design [12]. The initial CID concept was demonstrated through a system in which addition of FK1012 activated the endogenous T-cell signalling cascade, via fusion of FKBP12 to the proximity-regulated ζ-chain of the T-cell receptor, leading to receptor aggregation and subsequent activation [14]. FK1012 and other CIDs capable of dimerizing a single protein domain, discovered in the years following, have since been applied to the study of many important cellular processes to, for example, induce apoptosis via aggregation of the Fas membrane signalling protein [15] or regulate transcription through ligand-dependent (dis)association of transcriptional activators with promoter regions [16]. Rapamycin Although the first CIDs were only capable of homodimerization, these approaches could, in theory, be used to generate heterodimers if two proteins of interest were tagged with the same domain. The result would, however, be a mixture of heterodimers and homodimers of the two individual proteins of interest. The development of methods involving naturally occurring heterodimerizers therefore followed, with the most notable heterodimerizer rapamycin dominating the field since its discovery [17]. Rapamycin is an immunosuppressant drug that selectively binds both FKBP12 and FKBP–rapamycin) associated protein (FRAP)/mammalian target of rapamycin (mTOR) [18]. The FKBP and FKBP–rapamycin-binding (FRB) domains of these proteins, respectively, are sufficient for binding and retain the binding affinity of the full-length proteins [19]. A key step in the application of rapamycin as a CID was the production of rapamycin derivatives, known as ‘rapalogs’, which have a much lower affinity for endogenous proteins, thereby circumventing rapamycin's immunosuppressive activity. In parallel, the rapamycin-binding regions from FKBP12 and FRAP/mTOR were remodelled to bind the rapalog at nanomolar affinity, providing an orthologous rapamycin system for CID applications [20–22]. These rapalogs have since been used to conditionally dimerize proteins to interrogate many different biological processes. Notable examples include the study of mitosis, in which rapamycin-induced binding of the endoplasmic reticulum and Golgi membranes showed that these structures remain segregated during mitosis in mammalian cells [23], and also the study of phosphoinositides and their roles in endocytosis and intracellular trafficking [24,25]. One specific application of rapamycin-mediated control in mammalian cells is the knocksideways (KS) method (Table 1), which acutely sequesters protein activity through a change in subcellular localization. The KS method is capable of rapidly re-routing target proteins containing an FKBP domain to the mitochondria on a timescale of seconds, through rapamycin-induced binding to an FRB-containing protein with a mitochondrial targeting signal (mito-FFP-FRB) (Figure 1B) [26]. This method utilizes the principle of mitochondrial re-routing, whereby the protein of interest accumulates on the outer mitochondrial membrane in a way that, providing the new localization is not compatible with protein function, sequesters protein activity, but remains tolerable to cells [27]. In an initial proof-of-principle study, the KS method was used to study the role of two subunits of the adaptor protein (AP) complexes of clathrin-coated vesicles AP-1 and AP-2 [26]. Robinson et al. [26] used rapamycin-induced re-routing of AP-1 and AP-2 to the mitochondria, in combination with siRNA knockdown of the endogenous protein, to demonstrate the requirement for both proteins in the endocytosis pathway. Although the phenotype of AP-2 sequestration was similar to that resulting from siRNA approaches alone, the corresponding phenotype observed for AP-1 was distinct from that of the siRNA knockdown and is accredited to more rapid depletion achieved in the KS approach [26]. The effectiveness of the KS approach for rapid changes in protein activity has since been demonstrated in a number of varied applications to fast-acting processes in mammalian cells. For example, Cheeseman et al. [28] used the rapamycin-mediated approach to specifically remove TACC3-ch-TOG-clathrin complexes from the mitotic spindle within a timescale of 5 min following rapamycin addition. By re-routing these complexes to the mitochondria and away from mitotic spindles at defined stages in mitosis they were able to deduce their role in maintaining tension in kinetochore fibres, which are essential for correct segregation of chromosomes. Again, this phenotype was distinct from that observed with siRNA alone, demonstrating the utility of the KS approach. CIDs offer an efficient way to control the dynamics of processes reliant on oligomerization. However, the same principles can also be used for the opposite mode of control, in which processes are inhibited by oligomerization and activated upon addition of a ligand that dissociates the complexes. To enable this mode of control, Rollins and colleagues identified an FKBP12 mutant F36M-FKBP (FM) with the ability to form discrete dimers that can be dissociated rapidly upon addition of ligand [22]. Using these tools, Al-Bassam et al. [29] were able to develop a novel pulse–chase system in which exogenous FM- tagged membrane proteins were accumulated gradually in the endoplasmic reticulum, and sequestered by the formation of aggregates. Within minutes of small-molecule ligand addition, the FM domains dissociated and the accumulated membrane proteins could be simultaneously released for synchronous continuation along the secretory pathway [29]. Through this method, Al-Bassam et al. [29] were able to study proteins in a specific phase of the secretory pathway without interference from proteins in other phases of the pathway and thus overcome a major problem associated with studying this dynamic process. In order for CID to be successful, target proteins must be considered on a case-by-case basis and prior knowledge of protein function is usually required in order to achieve thorough inactivation. With nanomolar affinities between ligand–protein pairs, CID approaches have high specificity and high efficiency. However, this puts them at a disadvantage in terms of reversibility, as they often require an additional ligand that competes for binding to relieve protein inactivation/sequestration. Effects are often irreversible [30]. Although CID approaches were initially developed and demonstrated in vitro, these applications have since been developed to allow in vivo studies [31]. However, the requirement for exogenous small ligand addition and resulting potential for off-target effects somewhat limit the practicality of such applications. Also, although the KS approach demonstrates the ability for CID approaches to operate on a timescale of minutes, generally CID based methods range from minutes to hours, limited by the requirement for efficient uptake of the chemical inducer, and are consequently not among the fastest acting tools for temporal control of protein dynamics. Light-induced dimerization (LID) Another way in which protein dimerization can be induced is via light-based methods, which take advantage of naturally occurring photosensitive protein domains that dimerize upon exposure to a certain wavelength of light (Figure 1C, Table 1). Although maintaining the flexibility of CID approaches in terms of the response elicited and the many ways in which protein function can be disrupted, light-based methods generally overcome many of the limitations of CID. In particular, they provide improved spatiotemporal precision, mitigate the requirement for exogenous small molecule addition and operate on a timescale of seconds. Genetically encoded light-based (optogenetic) approaches have vastly expanded within the last decade from just a few applications to a whole toolbox of techniques with which to control protein activity [32]. Like CID, the first methods involving light-induced dimerization (LID) took advantage of naturally occurring photosensitive proteins, often discovered initially in plants, known as phytochromes and cryptochromes. Phytochromes Phytochromes are photoreceptive pigments encoded by small multigene families in plants and bacteria where they monitor red/far-red wavelengths of light [33,34]. The most thoroughly investigated phytochromes are those from Arabidopsis thaliana, which normally function to modulate seed germination and shade avoidance [35]. One such protein is phytochrome B (PhyB) which undergoes a conformational change upon exposure to light of visible red wavelengths (∼650–670 nm) to heterodimerize with the transcription factor phytochrome-interacting factor 3 (PIF3). Unlike other photosensitive proteins, this dimerization can be reverted through exposure to longer wavelengths of light (∼700–750 nm), which induces monoisomerization of PhyB and releases PIF3, allowing for very precise control of protein activity [32,36]. Cryptochromes Also commonly found in plants, cryptochromes (Cry) are plant photosensors that absorb blue light, the most well studied of which, Cry2, heterodimerizes with the cryptochrome-interacting basic helix–loop–helix 1 (CIB1) transcription factor. Cryptochrome proteins have a C-terminal domain required for signal transduction and, like phytochromes, require a flavin adenine dinucleotide (FAD) chromophore cofactor, which binds to an N-terminal DNA photolyase homology region (PHR) [37,38]. Since the discovery of Cyr2 and its ability to heterodimerize with CIB1, this system has been adapted to circumvent the need for exogenous chromophore addition [37]. With this improved system, Kennedy et al. [37] induced dimerization of Cry2–CIB1 on a sub-millisecond timescale (in under 300 μs), although the reverse process took minutes to complete. Nevertheless, this improved system has since been used in the study of a number of different cellular processes in model organisms. One field in which the Cry2–CIB1 system has been used successfully both in vitro [39] and in vivo [40] is the study of phosphoinositide signalling. This was achieved by Cry2–CIB1-mediated recruitment of a phosphoinositide phosphatase catalytic subunit responsible for the conversion of PI(4,5)P2 into PI(4)P to the plasma membrane in a light-dependent manner. Using this approach, Guglielmi et al. [40] were able to study complex morphological changes and interactions that occur within defined timescales during Drosophila embryogenesis. The recruitment of the catalytic subunit to the plasma membrane within seconds of blue light illumination was sufficient for quick depletion of PI(4,5)P2 which, given the role of phosphoinositides in regulating actin polymerization, allowed control over cell contractility and facilitated the study of cell–cell interactions, force transmission and changes in tissue geometry [40]. The use of cryptochromes for conditional dimerization has since spawned a host of methods utilizing the interaction between Cry2 and binding partners such as CIB1. One such method, known as light-activated reversible inhibition by assembled trap (LARIAT), utilizes light-mediated heterodimerization to reversibly sequester target proteins into multimeric complexes in mammalian cells, by engineered interactions with multimeric proteins (Figure 1D, Table 1) [41]. Lee et al. [41] developed the LARIAT technique by fusing Ca2+/calmodulin-dependent protein kinase IIα (CaMKIIα) protein, which self-assembles into a 12-subunit oligomer, to CIB1. Upon blue light stimulation, CIB1 interacts with Cry2 and forms clusters through interconnections between CIB1-conjugated CaMKIIα multimeric proteins. Through this method of optogenetic trapping, Lee et al. [41] were able to induce cluster formation with high spatiotemporal precision in HeLa cells within 30 s of illumination, with cluster disassembly occurring within minutes of light withdrawal. Lee et al. [41] also found that the extent of clustering was correlated with the intensity or number of light pulses administered, suggesting that it may be possible to quantitatively control clustering simply by varying light conditions for more intricate control of protein dynamics. This approach can also be used to inactivate GFP-tagged proteins, without the need to add an additional protein tag, through the use of anti-GFP nanobodies. To demonstrate this approach, Lee et al. [41] trapped a number of different GFP-tagged proteins into complexes through interactions with a CIB1-conjugated anti-GFP nanobody to acutely disrupt proteins involved in fast-acting processes such as membrane retention and spindle formation. A recent example that displays the potential of the LARIAT approach is its application to the study of intracellular membrane trafficking. Here, Nguyen et al. [42] developed a system whereby intracellular membranes can be rapidly and reversibly sequestered into complexes via Cry2-induced aggregation of CIB1-conjugated GTPases. Using diverse Rab GTPases as membrane markers, it was possible to access specific intracellular membrane compartments such as the Golgi and endoplasmic reticulum [42]. This approach makes it possible to dissect the spatiotemporal functions of intracellular membranes in a variety of processes such as receptor transport, intracellular signalling from endosomes, protein sorting and secretion. It is also known that many plant photosensors, including Cry2, are capable of forming aggregates upon light stimulation [43,44]. For example, Wend et al. [45] demonstrated the ability of Cry2 to dimerize C-Raf and activate its kinase activity in a light-inducible manner, functionally separating C-Raf from upstream growth factor signalling, enabling a more controlled approach to study dynamic downstream effects on target protein phosphorylation and cell signalling. Interestingly Wend et al. [45] also tested the ability of C-Raf-Cry2 to dimerize with CIB1-bound C-Raf and found a weaker activation of C-Raf, which they suggest may be due to a difference in stoichiometry when the larger Cry2 molecule binds to the much smaller CIB1 domain. Another use of Cry2 dimerization is in a technique called clustering indirectly using cryptochrome 2 (CLICR), which involves the clustering of transmembrane receptors to activate signal transduction (Figure 2). This is achieved by indirect clustering of Cry2 bound to a receptor-binding domain (BD); high local concentrations of the BD then serve to cluster endogenous receptors leading to signal activation [46]. An N-terminal src-homology 2 (SH2) domain, which binds receptor tyrosine kinases (RTKs) and the phosphotyrosine-binding-like F3 domain from talin, which binds β3-integrin were shown to be effective BDs [46], suggesting the method could be modified to target a wide range of transmembrane proteins. However, the selectivity of such tools needs to be empirically validated for each system. Figure 2 Clustering indirectly using cryptochrome 2 (CLICR) In the dark un, induced state, monomers of Cry2 fused to a receptor targeting BD (Cry2–BD) exist in an unclustered state and therefore have a weak affinity for the target receptor. Upon blue light stimulation, Cry2–BD molecules oligomerize, increasing the local concentration of BD and conferring a high avidity for the target receptor. These oligomers undergo membrane translocation and cluster target transmembrane receptors. Light–oxygen–voltage domains An alternative approach to light-induced dimerization involves the use of light–oxygen–voltage (LOV) domains from Avena sativa phototropin 1 (Table 1). LOV domains contain a C-terminal α-helix (Jα helix) which, upon light illumination and excitation of a flavin cofactor within the LOV domain, undergoes a large conformational change and unwinds [47]. This light-induced structural change allows for the control of protein activity through allosteric regulation of proteins containing these LOV domains (Figure 1E). One example of how LOV domains have aided the study of protein function is the application to the study of cell motility [48,49]. Wu et al. [48] fused Rac1 to a LOV domain, which, in its native α-helix state, blocked Rac1 interactions. This photo activatable Rac1 (PA-Rac1) could then be reversibly and repeatedly activated in precise cellular locations by illumination with blue light, producing localized cell ruffling and protrusions. Localized Rac1 activation was also able to promote directed cell motility [48]. PA-Rac analogues have since been used in further in vivo cell migration studies, showing for instance that Rac activation is sufficient for polarization of the border cells in Drosophila oogenesis and that the directionality of the subsequent migration of these cells during egg chamber development is dependent on Rac levels [50]. Although PA-Rac has been used successfully in a number of studies, the shift between the wound and unwound Jα helix states upon illumination is less than ideal, with at best a 10-fold shift towards the unfolded state upon light irradiation [51]. Through the identification of mutations that stabilize both the wound and unwound Jα helix states, Strickland et al. [52] modified the LOV system and reduced the proportion of unwound Jα helix in the dark state to make the switch between light and dark states more defined and increase the dynamic range of the system as a whole, with up to a 70-fold shift in Jα helix state after exposure to light. One example that takes advantage of the high spatial and temporal precision that can be achieved using LOV domains is the control of RTK activation. RTKs are a family of cell-surface receptors that respond to growth factor and hormone signals to regulate a variety of cell behaviours, and have previously proven difficult to study due to the rapid rates of receptor biosynthesis and degradation that can occur. Grusch and colleagues [54] used LOV-domain-mediated dimerization of mutant RTKs, insensitive to endogenous ligands, to induce transphosphorylation and therefore receptor activation on a timescale shorter than that of receptor synthesis/degradation. Through this approach they were able to mimic the cell behaviours induced by endogenous growth factors to provide control over cell signalling on the minute timescale, with diverse cellular responses in different cell types pointing to the involvement of different adapter proteins or feedback mechanisms [54]. Light-based methods for induced protein complex formation and control of protein activity offer a powerful solution to many of the drawbacks that come with chemical-based approaches while maintaining versatility. Although light-based methods require laser excitation to stimulate photoactivatable protein modules, the wavelengths of light used generally fall within the same range as those used for conventional fluorescence imaging, meaning cytotoxic effects are minimal and these approaches have therefore been applied successfully to both in vivo and in vitro studies [55]. The benefits of optogenetic approaches over the more traditional well-studied small-molecule approaches suggest that, with further development, these tools will be invaluable in the use of complex formation for protein inactivation or sequestration in the study of fast-acting cellular processes. PROTEIN CLEAVAGE/SPLICING Another common strategy for the inducible control of protein activity is to induce physical changes in protein sequence through the endogenous process of protein cleavage or splicing. As with methods for inducible protein complex formation, protein cleavage/splicing can be engineered to allow induction via a number of different mechanisms including both small-molecule-based and light-based approaches (Figure 3, Table 2). However, the mechanisms used to induce protein cleavage/splicing are often interchangeable, allowing these methods to be adapted to a wider range of systems and biological questions. Figure 3 Illustration of methods for conditional protein splicing or protein cleavage to (in)activate target proteins (A) Illustration of conditional protein cis-splicing induced by activation of an intein through a change in redox state or via a trigger, which may be addition of a small molecule, as depicted here, a change in pH, temperature or irradiation with a specific wavelength of light. Protein trans-splicing is also possible, whereby dimerization domains can be used to reassociate split intein fragments upon addition of a trigger. (B) Illustration of TEV protease-mediated cleavage of a TEV recognition site engineered within a protein of interest leading to protein inactivation upon induction of TEV protease expression. (C) Schematic representation of CALI/FALI induced by addition of a dye-conjugated ligand or antibody, or via genetically encoded methods involving photosensitizers such as KillerRed, eGFP, miniSOG or SuperNova. Upon irradiation with a specific wavelength of light these produce ROS (1O2) leading to inactivation of proteins in the immediate vicinity. Table 2 Summary of methods for conditional control of protein splicing/cleavage Can be induced by Technique Protein disruption Timescale Small molecule/hormone Light Temperature pH Intein-mediated protein splicing Inactivation via protein splicing h ✓ ✓ ✓ ✓ Tobacco etch virus (TEV) protease cleavage Inactivation via protein cleavage min Promoter-dependent Chromophore-assisted light inactivation (CALI) Inactivation by ROS Often <1 s ✗ ✓ e.g. Malachite Green: 616–624 nm KillerRed: 540–580 nm ✗ ✗ Fluorophore-assisted light inactivation (FALI) Inactivation by ROS <10 min ✗ ✓ e.g. FITC 493–518 nm ✗ ✗ Intein-mediated protein splicing One method that allows inducible control of protein activity uses the endogenous post-translational mechanism known as intein-mediated protein splicing (Table 2). With this method, intervening polypeptides known as inteins are used to catalyse their own removal from the flanking polypeptides, or exteins, which are subsequently joined back together. Inteins are typically removed in a four-step process involving conversion of the peptide bond linking the N-terminal extein to an ester or thioester bond and transfer of the N-extein to the C-extein by transesterification. The resulting branched ester is then resolved by asparagine cyclization followed by conversion of the newly formed ester bond linking the two exteins into an amide bond and hydrolysis of the C-terminal aminosuccinimide of the excised intein [56,57]. Inteins are used in biotechnology for a number of different applications, including the control of protein expression or modification, post-translational processing and also protein labelling [57], but perhaps the most valuable application in terms of studying protein function is to facilitate the control of protein activity. Since inteins are extensively reviewed elsewhere [57–59], so we will not discuss their use further here, except to say that they have been engineered to allow conditional protein splicing (CPS), such that the splicing process is induced by the activation of an intein through reduction or the addition of a trigger such as light, temperature, pH or the addition of a small molecule (Figure 3A) [57,60]. These systems have been used successfully both in cultured cells and in living animals to interrogate protein function, although they have not been widely adopted for this purpose perhaps because of their intrinsic lack of reversibility. TEV cleavage One common method for inducible protein cleavage as a mechanism to control protein activity exploits the ability of the tobacco etch virus (TEV) protease to cleave a highly specific seven-amino-acid recognition sequence (E-X-X-Y-X-Q-G/S) with high efficiency (Figure 3B, Table 2) [61–63]. TEV is commonly used as a mechanism for the cleavage of fusion proteins to remove protein affinity tags prior to further protein analysis [64]; however, this system has also been applied to the control of protein activity both in vivo and in vitro. Through genetic modification, the TEV recognition sequence can be engineered into a protein of interest to allow inducible protein cleavage and inactivation when in the presence of TEV. TEV techniques have previously been demonstrated in budding yeast to provide evidence of a role for separin in anaphase initiation [65] and has since been applied to the study of proteins in both Drosophila cell culture and live embryos [66,67]. For example, to show that TEV was able to effectively and specifically cleave a protein containing the recognition sequence in live Drosophila embryos, Harder et al. [66] expressed the protein Megatrachea (Mega), a Drosophila claudin protein localized to membrane compartments of ectodermal cells, containing an artificial TEV protease cleavage site (TEVpcs) and YFP (Mega-TEVpcs-YFP). Upon TEV expression using the Gal4/UAS system, Mega-TEVpcs-YFP no longer showed the correct YFP localization indicating that the YFP had been cleaved from the Mega fusion protein. Harder et al. [66] went on to adapt this system to allow induction of TEV expression at different stages of embryo development by putting TEV under the control of the heat-shock protein 70 (hsp70), thus generating a mechanism for the temporal control of TEV-mediated protein cleavage. Using this temporally controlled system, they were then able to induce cleavage of a Mega-TEVpcs construct leading to truncation of the Mega protein and subsequent degradation of the truncated protein [66]. Clearly, a key factor in determining the timescale of TEV-mediated cleavage is the promoter from which TEV is expressed. Changes in temperature using the heat-shock protein are capable of inducing TEV expression and protein cleavage in Drosophila pupae ∼3 h after a 45 min heat shock [67], whereas rapamycin-induced expression in mammalian cells induces cleavage within 150 min [68]. This may make the approach unsuitable for the study of some processes that operate on a short timescale. One disadvantage of TEV is that it readily cleaves itself at a specific site to yield a truncated enzyme with greatly reduced activity [69,70]. There have therefore been a number of iterative changes made to TEV protease to adapt the protease for more diverse applications, for example a TEV mutant has recently been designed specifically to be active in the secretory pathway [71]; various other TEV mutants offer the same recognition site cleavage, but an increased stability and reduced auto-cleavage activity [69]. Although TEV offers highly specific and efficient cleavage, a priori knowledge about protein composition is required to choose a position where the TEV recognition site can tolerably be inserted that will inactivate the target protein while reducing the possibility that the resulting protein fragments will retain function or even have novel functions of their own. There is also an optimum level of TEV protease expression at which cleavage occurs but background activity is minimalized; this level is likely to depend on both the variation of TEV protease and the system in which it is applied so would need to be considered during experimental design [68]. The timescales and spatiotemporal resolution of both intein-mediated protein splicing and TEV cleavage are dependent on the engineered mechanism of induction. Although the ability to customize these techniques allows them to be applied to a wide range of systems and cellular processes, the fastest acting methods of protein cleavage/splicing with the highest spatiotemporal resolution are again those induced by light. CALI/FALI One method for light-inducible protein cleavage applied to the study of protein function is chromophore-assisted light inactivation (CALI) (Table 2). Chromophores are photosensitive groups, often responsible for the colour of organic molecules, which produce highly reactive free radicals such as reactive oxygen species (ROS) upon illumination with a specific wavelength of light. Using the CALI approach, a chromophore-tagged protein of interest is inactivated through mild illumination for a period of time sufficient to induce generation of ROS and induce protein cleavage of proximate proteins, but short enough to ensure that the ROS act within a defined radius [30–40 Å (1 Å=0.1 nm)] to minimize off-target effects (Figure 3C) [72]. The specificity of CALI approaches is determined by the short half-life of the free radical species, which ensures that only proteins within a radius of 1.5–6 nm relative to the chromophore are affected [72–74]. After free radical generation, proteins are typically inactivated within 1 s, which, combined with laser irradiation of micrometre accuracies, allows for high spatial resolution and highly controlled protein inactivation [74]. Originally, CALI approaches used an antibody-based mechanism to attach a chromophore, such as the dye Malachite Green, to a protein of interest (Figure 3C). Fluorophores such as fluorescin isothiocyanate (FITC), which are more efficient at ROS production, were later employed in a similar approach called fluorophore-assisted light inactivation (FALI) [75]. However, the need for microinjection of a dye-labelled non-function-blocking antibody specific to the protein of interest limited the widespread application of these approaches. Subsequent methods made use of genetic modification to label proteins with a generic tag that can then be fluorescently labelled through extracellular addition of a specific reagent (Figure 3C). Proteins tagged with one or two small tetracysteine (TC) motifs will specifically bind to the membrane permeable biarsenical dye resorufin-based arsenical hairpin binder (ReAsH) or the fluorescein-based arsenical hairpin binder (FlAsH). For example, Marek and Davis [76] used FlAsH labelling to visualize synaptotagmin I (Syt I) at the neuromuscular junction (NMJ) in late-stage live Drosophila larvae. Through photo-inactivation they were able to inactivate Syt I within seconds and provide supporting evidence for a model, previously based on genetic data alone, in which Syt I plays a role post-vesicle docking to mediate vesicle fusion and calcium-dependent transmitter release [76]. Although dyes are added extracellularly in ReAsH/FlAsH based CALI/FALI approaches, the application of these techniques in vivo is limited by the difficult task of achieving sufficient uptake in live animals and also the inability to spatially control the production of ROS and limit it to particular cells or subcellular compartments [77]. There is also the problem of non-specific binding of the membrane-permeant dyes. For CALI/FALI to become a more widely used technique, there was therefore a need for a system that sidestepped the requirement for exogenous addition and could be encoded completely through genetic manipulation. There has been some limited success using eGFP, a tag commonly used to study protein localization and function. For example, CALI illumination of GFP–myosin II was shown to result in unequal-size daughter cells during asymmetric cell divisions in a Caenorhabditis elegans Q neuroblast cell lineage [78]. It is, however, believed that the chromophore within the GFP structure is protected by the outer shell meaning the generation of free radical species upon illumination is restricted and therefore GFP holds a low phototoxicity [77,79]. KillerRed The first example of a successful genetically encoded CALI reagent, with a 1000-fold increase in phototoxicity compared with GFP, is the GFP-related red fluorescent protein KillerRed, developed from the non-fluorescent red chromoprotein anm2CP of Hydrozoa jellyfish (Figure 3C) [79]. Although it was originally unclear why there was such an improvement in phototoxicity, a study into the structure–function relationship of KillerRed by Pletnev et al. [80] provided crystallographic data revealing unique structural features that may facilitate ROS generation. KillerRed has been used in a number of applications in biological research, such as the control of protein activity in the study of neuronal development in model organisms [81], and has been used to induce cell-specific killing of eukaryotic cells in culture via mitochondrial or membrane-targeted KillerRed [79]. In this regard, KillerRed has also proved to have uses in medicine; one recent and exciting application of KillerRed was to the experimental cancer treatment known as photodynamic therapy (PDT), which aims to use photosensitizers to selectively kill tumour cells through ROS generation upon laser illumination [53,82]. Following subcutaneaous injection of KillerRed-expressing Escherichia coli (KR-E. coli) into mouse xenograft models of human carcinoma cell lines the Terekawa laboratory [82] monitored the intensity and spread of fluorescence through the tumour cells. After 24 h the KR-E. coli spread throughout the whole tumour and were subsequently irradiated with orange light (540–580 nm) to induce ROS production. The generation of ROS led to necrosis and tumours gradually disappeared to leave healed skin after just 1 week, demonstrating the ability of KR-E. coli to kill cancer cells originating from humans. Although there are many questions still to be answered before this technique is applicable to humans, these results provide an insight into the capabilities of genetically encoded CALI approaches. KillerRed has proved to be an exciting solution to the difficulties associated with previous CALI approaches. However, one limitation is its tendency to homodimerize, which can potentially interfere with protein function [77,79]. There have been a number of other novel photosensitizers discovered since, such as SuperNova, a monomeric form of KillerRed [83], and the fluorescent flavoprotein mini Singlet Oxygen Generator (miniSOG), which can also be used to generate an insoluble deposit of singlet oxygen species that can be stained for visualization using electron microscopy [84]. More recently, the toolkit of phototoxic proteins was expanded further with the addition of KillerOrange, an orange mutant version of KillerRed that results in ROS formation upon illumination with either blue or green light, meaning it can be used in combination with KillerRed or other photosensitizers activated by different wavelengths of light [85]. CALI/FALI approaches enable specific protein inactivation through phototoxicity and can act with high spatial resolution through the ability to express tagged proteins in specific cells and trigger inactivation at a subcellular level (Table 2). The expansion of the phototoxic protein toolbox should also allow for the creation of more intricately controlled systems in which different proteins can be inactivated at different time points or in different cell populations, which will form a useful tool both for fundamental research and for potential medical applications. However, CALI/FALI-based methods suffer from a number of limitations, including the requirement for either exogenous ligand addition or the inclusion of a fairly large protein tag that may interfere with protein function. Importantly, there is also potential for off-target effects on proteins in close vicinity to the ROS generator, challenging the specificity of these approaches. For example, Guo et al. [86] found that the inhibition of calcium ion currents, mediated by the class C G-protein-coupled receptor (GPCR) mGluR8a, was greatly attenuated following FALI inactivation. Although initial results were consistent with acute inactivation of mGluR8a, Guo et al. [86] also reported collateral damage to proximal proteins with no overt link to pathways of GPCR signalling. These factors have limited the adoption of such methods to study protein function and cannot be overlooked when using techniques involving phototoxicity for protein inactivation. PROTEIN STABILITY/DEGRADATION Although protein cleavage or splicing can often lead indirectly to protein instability and degradation, it is likely that this will occur after some delay depending on the half-life of the protein. It is also possible for the resulting protein fragments to retain function or bind other proteins and perform independent functions of their own, potentially generating a more severe phenotype than simple protein knockout alone. It therefore follows that a more thorough and interpretable approach for complete removal of proteins from cells is to target them directly for degradation by the endogenous cellular degradation pathways (Figure 4, Table 3). Although the direct degradation of target proteins means these methods are technically irreversible and somewhat limited in their application, in most cases protein levels return to normal following relief of the degradation stimulus and thus these methods can still provide a useful tool for studying protein function [9,10,87]. Methods for inducible protein degradation generally involve the proteasome pathway for protein degradation, in which ubiquitin is transferred from the E1 ubiquitin-activating enzyme to the E2 ubiquitin-conjugating enzyme and subsequently to a lysine residue within the target protein in a transfer facilitated by the E3 ubiquitin ligase [88]. This process is repeated to polyubiquitinate the target protein until it has a sufficient number of ubiquitin molecules for targeted degradation by the proteasome [88]. Figure 4 Illustration of methods for inducible protein degradation divided into those involving the N-end rule or F-box-based pathways (A) N-end degron method, involving the exposure of an unstable N-terminal amino acid which can be induced by a number of mechanisms including temperature- or small-molecule-based mechanisms. This unstable end is subsequently targeted for polyubiquitination by an E3 ligase, such as Ubr1P in yeast, and degraded via the proteasome pathway. (B) TIPI which utilizes TEV protease-mediated cleavage of a seven-amino-acid TEV recognition site to reveal an unstable N-terminal amino acid, subsequently targeted for proteasome-mediated degradation as for the N-end degron approach. The efficiency of TEV cleavage is increased by the inclusion of a short SF3b155381–424 domain downstream of the TEV recognition site, which binds to a mutant version of the human spliceosome subunit 14 (p14*) bringing p14*-TEV to the recognition site. (C) Illustration of the mechanism for auxin-induced degradation of target proteins tagged with an AID. Upon addition of auxin, the AID-tagged protein is recruited to an engineered E3 ubiquitin ligase SCF complex, containing the TIR1 F-box protein from plants, which binds target proteins in the presence of the plant hormone auxin. Expressing TIR1 in non-plant cells is enough to result in formation of the SCF complex which then binds AID-tagged proteins in an auxin-dependent manner, leading to polyubiquitination and degradation via the proteasome. (D) Schematic illustration of the deGradFP method for inducible protein degradation. The GFP-tagged target protein is recruited to an engineered SCF complex containing the F-box protein NSlmb conjugated to an anti-GFP single-chain antibody (vhhGFP4). Target protein is polyubiquitinated via the recruited E2 ubiquitin ligase and subsequently degraded via the endogenous proteasome machinery. Table 3 Summary of methods for conditional control of protein degradation Can be induced by Technique Protein disruption Timescale Small molecule/hormone Light Temperature pH N-end degron Unstable N-end amino acid ubiquitination min/h, sometimes <30 min ✓ e.g. Methotrexate ✗ ✓ e.g. 37–42°C ✗ TEV protease-mediated induction of protein instability (TIPI) Unstable N-end amino acid ubiquitination min/h Promoter-dependent Auxin-induced degradation (AID) F-box-induced ubiquitination <30 min ✓ Auxin ✗ ✗ ✗ Proteolysis-Targeting Chimaeras (PROTACs) Direct targeting to E3 ligase complexes min/h ✓ PROTAC ✗ ✗ ✗ deGradFP F-box induced ubiquitination min/h Promoter-dependent N-end degron A common strategy for the direct induction of protein degradation utilizes the UBR1 E3 ligase pathway and the N-end rule (Table 3), which states that the half-life of a protein is determined by both the accessibility of lysine residues, for ubiquitination, and the identity of the amino acid at the N-terminus [89,90]. Varshavsky [90] demonstrated this principle through cleavage of ubiquitin, via a yeast deubiquitinating protease, from a fusion protein expressed in yeast, containing the 5′ end of a lacI linker followed by β-galactosidase (β-gal), resulting in the exposure of a new N-terminal amino acid. The half-life of β-gal following cleavage could then be vastly altered by simply changing the exposed amino acid residue. For example, β-gal with an N-terminal arginine or phenylalanine residue had a half-life of ∼3 min, whereas N-terminal methionine or valine resulted in a half-life greater than 20 h [89]. This method of degradation is conserved from bacteria to higher eukaryotes, and means proteins tagged with the unstable lacI degron can be targeted for degradation within minutes [90]. However, this strategy is not inherently inducible and therefore requires modification for the conditional control of protein degradation. One of the first examples of an inducible N-end degron involved the use of a temperature-sensitive dihydrofolate reductase variant (tsDHFR) where an N-terminal destabilizing arginine was only exposed at non-permissive temperatures [91]. By fusing the tsDHFR to the N-terminus of a target protein, degradation can be induced by a switch to the non-permissive temperature of 35°C, exposing the N-terminus at which point the N-end rule takes effect (Figure 4A). For example, this system has been used successfully in Drosophila to inducibly polyubiquitinate an eGFP reporter protein at the neuromuscular synapse following a 30 min heat shock at 35°C, in order to track the degradation of polyubiquitinated proteins [92]. By tracking the degradation of the eGFP, Speese et al. [92] showed that ubiquitinated presynaptic proteins are not removed from the synaptic terminal but rather undergo local proteasome-mediated degradation at pre-synaptic sites. In addition to examples from Drosophila, this technique has also been used successfully to characterize many essential proteins in budding yeast [93]. However, such techniques are generally limited to systems that can survive the required temperature changes and also to proteins that retain function with the required N-terminal modification. Despite this, the approach has since been used successfully in chicken DT40 cells, in which the method was first tested using a tsDHFR degron fused to eGFP. Upon transfer of the cells to the non-permissive temperature of 42°C, the protein was rapidly depleted to ∼10% of initial levels within 90 min leading to undetectable levels after 120 min [94]. Moving cells back to the permissive temperature of 35°C resulted in an efficient recovery to the pre-depletion level within 150 min. Su et al. [94] then used the approach to deplete RAD51, finding that RAD51, which plays an important role in homologous DNA recombination (HDR), does not stop DNA synthesis but causes cell cycle arrest in G2, suggesting HDR becomes important at G2. This, along with the many other applications of tsDHFR-based approaches, show that, although this method is limited in its potential applications, tsDHFR can still provide a useful tool in the study of protein function. In addition to tsDHFR, a small- molecule-controlled version of dihydrofolate reductase (DHFR) has also been engineered, for which the drug methotrexate (MTX) regulates stability of the N-terminus. Although the presence of MTX fails to inhibit recognition and therefore polyubiquitination of tsDHFR by E3 ligase, the stable high-affinity interaction between MTX and DHFR impedes protein unfolding and prevents degradation by the proteasome [95,96]. This system was demonstrated in both yeast and mammalian cells in culture, although the occupancy of the proteome by the MTX–DHFR complex is likely to inhibit degradation of other cellular proteins leading to off-target effects [97]. Also, DHFR is required for the production of tetrahydrofolate, which is subsequently required for the synthesis of purines, thymidylate and several amino acids [98]. The inhibition of DHFR by MTX therefore interferes with the synthesis of DNA, RNA and even proteins, meaning MTX is undesirable as a regulatory small molecule for exogenous addition. TEV protease-mediated induction of protein instability (TIPI) An alternative way in which the N-end degron system can be made inducible, and more widely applicable, is through the use of TEV in a technique called TEV protease-mediated induction of protein instability (TIPI) (Table 3). This technique combines TEV with the N-end rule, whereby TEV cleaves a recognition sequence engineered into a cryptic N-degron, attached to the N-terminus of a target protein, to reveal an unstable N-end amino acid (Figure 4B) [99]. According to the N-end rule, this unstable amino acid targets the protein for polyubiquitination and degradation via the UBR1 E3 ligase pathway [89,90]. TEV protease has previously been shown to allow degeneracy within its recognition sequence and is particularly flexible to changes at position 7, the amino acid residue that forms the N-end following TEV cleavage [63]. TEV protease can therefore cope with the incorporation of an amino acid that induces degradation following cleavage via the N-end rule. Taxis et al. [99] first developed the TIPI approach by designing a construct containing a reporter followed by a TEV protease recognition site, N-degron and SF3b155381–424 termed Reporter-TDegX-tag, where X represents the amino acid at position 7 which becomes the new N-terminal amino acid upon cleavage. The inclusion of a relatively short SF3b155381–424 domain allowed for more efficient cleavage as its binding to a mutant version of the human spliceosome subunit 14 (p14*) recruited p14*-TEV to the recognition site (Figure 4B). Taxis et al. [99] initially demonstrated this system in yeast using a GFP-TDegX-Don1p fusion protein, with p14*-TEV expression driven by the Gal1 promoter, monitoring cleavage via the release of GFP and testing the effect of the amino acid at position X on TEV cleavage efficiency and protein half-life. Phenylalanine or asparagine was found to provide optimal conditions for both TEV cleavage and rapid degradation of the target protein Don1p. The effectiveness of TIPI was shown by using the approach to deplete several different proteins in yeast, obtaining phenotypes correlating to those observed via genetic knockdowns [99]. The potential of TIPI as an approach for conditional degradation is yet to be realized; however, its power and versatility has recently been demonstrated in some alternative applications. For instance, TIPI has been further modified (mTIPI) to facilitate production of recombinant proteins; it does this by blocking endocytosis in yeast and combating a common problem whereby highly active endocytosis in protein expression systems reduces the overall protein yield [100]. For conditional control this method simply requires expression of the p14*-TEV fusion protein, which could be induced via the same methods as for TEV cleavage, including temperature, pH, small molecule addition or the Gal4/UAS system, making it a versatile tool for inducible degradation and the study of protein function both in vivo and in vitro. Auxin-induced degron Another way in which proteins can be directly targeted for degradation by the ubiquitin–proteasome machinery is via F-box proteins, which bind target proteins and recruit the Cullin–RING complex, also called the SCF complex consisting of Skp1, Cullin and F-box, to ubiquitinate the target protein [101]. Eukaryotes contain multiple forms of SCF, whereby the F-box protein conveys specificity towards different target proteins. One F-box particularly suited to small-molecule-induced control of protein degradation is the transport inhibitor response 1 (TIR1) protein, which binds target proteins in the presence of the plant hormone auxin and has a highly conserved interaction with the E3 ligase protein Skp1 (Figure 4C) [102,103]. This is commonly referred to as an auxin-inducible degron (AID) system (Table 3), and was initially shown to be applicable to most eukaryotes (excluding plants), including budding yeast and cell lines derived from human, mouse, hamster, monkey and chicken [104]. The AID degron consists of IAA17, also known as AXR3, from Arabidopsis thaliana and when expressed at either the C- or N-terminus of GFP in budding yeast also expressing AtTIR1, under the control of the galactose-inducible GAL promoter, the SCF–TIR1 complex was able to assemble and degrade GFP to less than 3% of initial levels within 30 min of auxin addition [104]. This approach was successfully used to degrade several essential nuclear or cytoplasmic proteins in yeast. However, to apply the system to mammalian cells, it was first necessary to modify TIR1 to convey a higher thermostability and allow use at 37°C. This was achieved by sourcing TIR1 from the rice plant Oryza sativa (OsTIR1), which also provided an improvement over AtTIR1 for use in yeast [104]. This method has since been used successfully to degrade both nuclear and cytoplasmic proteins in C. elegans [105] and mammalian cells [106] to help identify the function of several different target proteins at specific time points of the cell cycle. Holland et al. [106] tested five differentially localized proteins, some of which were known to be incorporated into protein complexes. Four out of the five AID–YFP-tagged proteins expressed under doxycycline control (Plk4, CENP-A, TFR2 and cyclin B1) showed quantitative protein degradation within 80 min of auxin addition. Degradation of the other protein studied, H2B, occurred more slowly, within 3 h [106]. These results show that AID is capable of rapidly depleting proteins involved in stable complexes with relatively long half-lives; however, the time taken for depletion following induction can vary. Holland et al. [106] demonstrated the ability of AID to induce proteolysis with the same or very similar degradation kinetics at all phases of the cell cycle and also found that proteins reappeared almost immediately upon removal of auxin stimulus. To demonstrate the ability of AID to study protein function, Holland et al. [106] achieved rapid functional inactivation of BubR1, an essential component of the mitotic checkpoint, by depleting endogenous BubR1 protein with siRNA, replacing it with siRNA-resistant GFP-AID-BubR1 and inducing mitotic arrest through nocodazole addition. GFP-AID-BubR1 rescued the function of the depleted endogenous BubR1 and this rescue could be rapidly reverted through auxin-induced degradation of the GFP-AID-BubR1 fusion protein to produce a more complete null phenotype than mRNA depletion alone [106]. The AID approach has since been developed further to increase versatility through minimization of the degron size and the inclusion of a series of epitope tags to allow detection using fluorescence microscopy or commercially-available antibodies [107]. Morawska and Ulrich [107] developed a series of vectors for PCR-based genomic tagging strategies containing different iterations of the AID degron with epitope tag, allowing for both C- or N-terminal tagging and providing a range of selection markers which they then demonstrated through application to a series of different yeast proteins. Although these vectors increase the versatility and facilitate the use of the AID approach, individual proteins must still be considered on a case by case basis to design the most effective degron; it may even be necessary to test multiple iterations to ensure proteins retain function. PROTACs Another chemical-based method for the specific degradation of target proteins by the endogenous ubiquitin–proteasome machinery is through the use of heterobifunctional small molecules known as Proteolysis-Targeting Chimaeras (PROTACs) (Table 3). PROTACs consist of one moiety that binds the target protein linked to an E3 ligase to directly recruit the protein for proteasome-mediated degradation. Initially developed to target disease-causing proteins for destruction, the first generation of PROTACs were based on large peptide motifs derived from known ubiquitin ligase substrates [108]. However, these were limited by high molecular mass, poor cellular uptake and potential metabolic instability [109,110]. Following a switch to small-molecule-based PROTACs, the last decade has seen a series of improvements to PROTAC technology, aided by the development of small ligands for a number of E3 ligases, including MDM2 (murine double minute 2), cIAP1 (cellular inhibitor of apoptosis protein 1), CRBN (cereblon) and VHL (von Hippel–Lindau protein) (as reviewed in [110]). Although these improved small- molecule PROTACs were able to successfully degrade target proteins, the overall uptake of this technique for the conditional control of protein degradation has been limited by a number of uncertainties, including PROTAC stability and E3 ligase binding affinity [109,110]. However, more recent advances in the field have provided a new generation of highly specific high-affinity low-molecular-mass PROTACs with the potential to expand the use of PROTAC technology [111–114]. For example, Bondeson et al. [111] used structure-guided approaches to develop low-molecular-mass (∼450 Da) high-affinity ligands for the Cullin-RING ligase 2 VHL E3 complex (CRL2VHL). Linked to small molecules that bind specific cellular targets, Bondeson et al. [111] were able to efficiently degrade specific proteins in cultured cell lines, including the serine/threonine kinase RIPK2, which is involved in innate immune signalling, and the oestrogen-related receptor α (ERRα), which is implicated in the regulation of various cellular metabolism pathways, with dose-dependent degradation and maximal degradation levels of >95 and 86% respectively. Bondeson et al. [111] also demonstrated this approach in vivo using a PROTAC targeting ERRα, reducing its levels by ∼50% and significantly reducing mouse heart and kidney tumours by >40%. Using a similar approach, Zengerle et al. [114] successfully designed potent PROTACs using optimized drug-like VHL ligands [115] and bromo- and extra-terminal (BET) bromodomain protein ligands to selectively degrade certain members of the BET protein family, including the epigenetic regulator BRD4 previously identified as a potential therapeutic target for acute myeloid leukaemia and ovarian cancer [114]. In addition to targeting proteins to the CRL2VHL E3 complex, potent PROTACs have also been developed to utilize the interaction between immunomodulatory drugs (IMiDs), such as thalidomide, and the CRL4CRBN E3 ligase. CRL4CRBN, together with an IMiD, forms a tertiary complex with the transcription factor Ikaros, resulting in its ubiquitination and degradation. This approach has since been used for the efficient and specific degradation of BRD2, BRD3 and BRD4 by attaching BET bromodomain protein ligands to an IMiD [112,113]. Although the principle behind PROTAC technology is not novel, there has been a recent surge of developments to generate a newer generation of more potent PROTACs, which address the limitations of previous iterations. These newer PROTACs offer greatly increased potency while retaining high specificity to their target proteins both in vitro and in vivo. It is also possible to modify this specificity through manipulation of the linker between the two PROTAC moieties [111,114]. The diversity of recently described examples shows how the PROTAC approach can work on different protein targets in a number of different systems. However, compared with small molecules used in other approaches to conditionally control protein dynamics, PROTACS are larger and more complex molecules and so may suffer limitations with respect to their pharmacokinetic properties. deGradFP The F-box/SCF complex-based approach has also been utilized in the degrade GFP (deGradFP) method to specifically degrade GFP-tagged fusion proteins via an anti-GFP nanobody/F-box chimaera (Figure 4D, Table 3). A method to allow specific degradation of proteins tagged with GFP is desirable as GFP-tagged constructs already exist for many proteins and degradation can be easily monitored by the loss of fluorescence. The deGradFP method involves the engineered F-box fusion protein NSlmb-vhhGFP4, consisting of an F-box domain derived from the Drosophila protein Slmb and the single-domain anti-GFP antibody fragment vhhGFP4 which recognizes GFP and its close derivatives (Figure 4D) [116]. This method was initially demonstrated in Drosophila, where, with NSlmb-vhhGFP4 expression restricted to the posterior of early stage embryos by the Gal4/UAS system, an EYFP-tagged histone H2A variant (His2Av–EYFP) was rapidly depleted by the deGradFP system [116]. Caussinus et al. [116] used the engrailed-Gal4 driver to express both NSlmb-vhhGFP4 and nuclear mCherry in embryos ubiquitously expressing His2Av–EYFP. Using mCherry levels as a reporter for the expression of NSlmb-vhhGFP4, His2Av–EYFP started to be degraded after ∼30 min following NSlmb-vhhGFP4 expression, with less than 10% of the maximum EGFP intensity remaining after ∼3 h [116]. Caussinus et al. [116] then went on to show the versatility of the deGradFP approach through the successful depletion of the cytoplasmic protein Spaghetti squash (Sqh), nuclear protein Apterous (Ap) and the transmembrane protein Crumbs (Crb) all of which were tagged with GFP, expressed in a null background and degraded upon induction of NSlmb-vhhGFP4 expression via the Gal4/UAS system. There were, however, a couple of cases in which the deGradFP was not effective against GFP-tagged target proteins. For example, E-cadherin/Shotgun (Shg) could not be degraded using this method, possibly as it exists in a large protein complex which may mean the GFP tag is not accessible to the vhhGFP4 antibody [116]. Also, NSlmb-vhhGFP4 was unable to induce degradation of GFP alone, perhaps as the small size of GFP prevents exposure to the SCF-recruited E2 enzyme and thus prevents poly-ubiquitination. It was, however, possible to degrade GFP containing a small nuclear localization signal via this method, so although it is possible that a minimum size limit exists, below which degradation does not occur, this limit must be very close to the size of GFP alone and should not greatly limit the versatility of the approach [116]. deGradFP has been proven to be a useful approach for the induced degradation of target proteins particularly in combination with RNAi knockdowns in order to generate a more effective depletion of protein levels [117,118]. More recently, a similar approach involving the modification of the E3 ubiquitin ligase adapter protein SPOP to alter target protein specificity was proposed [119]. By fusing an anti-GFP nanobody directly to a truncated SPOP adapter protein completely lacking its substrate-binding domain, Shin et al. [119] claim to have developed an approach that is more efficient than deGradFP, which simply involves an NSlmb deletion mutant for which the binding domain has been modified. This remains to be proved in terms of biological applications, but may offer an alternative in cases where deGradFP is not effective. deGradFP, and related methods, can in theory be adapted to allow knockdown of many endogenous proteins, providing high-affinity antibodies are available for target proteins [87]. Ubiquitin-independent Ubiquitin-mediated protein degradation is by far the most common strategy for control of protein degradation; however, it has previously been shown that localization to the proteasome is sufficient for degradation [120] and so it is worth mentioning here that there are also a handful of methods for the conditional control of ubiquitin-independent protein degradation. The most common of these is the C-degron; consisting of a 36 amino acid sequence from ornithine decarboxylase (ODC), this forms a bridged association with the proteasome, acting as both the recognition and degradation initiation signal [87]. An exciting use of this technique allows for light-induced protein degradation via the use of LOV2 domains [121]. Renicke et al. [121] designed a system in which the C-degron, fused to the C-terminus of the LOV domain, can be masked by the Jα helix under dark conditions, but exposed upon illumination with blue light via Jα helix unfolding, leading to ubiquitin-independent protein degradation. Although this provides an exciting alternative, ubiquitin-dependent methods remain the most widely used and well-studied methods for inducible protein degradation. CONCLUDING COMMENTS AND FUTURE PERSPECTIVES Manipulation of genes, at the level of DNA or RNA, has proven to be a specific and immensely powerful way of understanding the roles of encoded proteins in their native cellular environment. However, distinguishing between the initial and steady-state consequences of gene disruption, especially in vivo, is often problematic. The new generation of tools and methods that are emerging to meet this challenge address the issue by offering both rapid and specific control of protein function. Different sets of tools can be employed across a range of timescales to challenge biological processes operating at the subcellular, cellular and multicellular level. Methods for conditional control of protein complex formation (Table 1), in particular, open the door to in vivo analysis of biochemical processes operating over short times (s/min), such as intracellular signalling cascades, which are initiated within seconds of receiving the stimulus. Furthermore, the ability to reversibly switch activities on and off enables the systematic perturbation of biochemical pathways, thereby revealing how information is processed from upstream stimulus to downstream effectors at each step and providing insights into rate-limiting components and feedback control [122]. Repetitive perturbation at different times has particular value in the dissection of biological systems where frequency variation, including oscillatory behaviours, encodes information [123,124]. Other techniques, including those that conditionally control protein splicing/cleavage or degradation, typically operate over minutes/hour timescales (Tables 2 and 3), but have proven utility in studying the molecular mechanisms of downstream events, such as changes in cell proliferation, differentiation, migration or adhesion, which operate over longer times. It is important to note, as discussed in the sections above, that current methodologies for conditional perturbation of protein function are not without their technical limitations and researchers must weigh up whether the available tools offer the appropriate flexibility and precision for the desired experiment. One of the main considerations for researchers wishing to utilize the methods we have described is to decide which inducer they should use. Chemically induced methods have been the mainstay of the field for many years, but the use of chemical inducers is often restricted because of their promiscuous binding profiles, which can lead to off-target effects and cytotoxicity. As we have discussed, higher-affinity ligands might mitigate these effects, because the compounds can be used at lower concentrations, but typically suffer from not being reversible. More fundamentally, however, the inability to target some chemicals to specific subcellular localizations, combined with their relatively slow uptake in cells, with effects occurring in minutes to hours, make these methods unsuitable for the study of certain biological processes and in vivo models. Recently, there has been a surge of activity to develop methods of induction using light. One of the key attractions of this approach is that perturbations can be both rapid and reversible, in a spatially resolved manner. Consequently, with light-dependent systems, should an activated protein diffuse out of the area in which it received the activating input, it will then switch off, preventing phenotypic outputs from losing their spatial resolution [55]. Correspondingly, there has been a great deal of focus on improving the properties of the light-sensitive domains used in these approaches, much in the same way that there have been iterative improvements made to fluorescent proteins for use in cell imaging. This will make it possible to tailor the perturbation dynamics. For instance, in the case of light-induced complex formation, a derivative of the Cry2 domain (Cry2-olig) with altered off kinetics may make it more suitable for sequestration and inhibition of protein activity [125]. A challenge to future efforts for improvement to such domains will be not just to identify variants that confer beneficial properties in isolation, but ones that retain additive effects of multiple genetic changes; this will require screening procedures that simultaneously optimize constructs against multiple parameters. The ultimate goal of the approaches we have described is to understand the role of molecules in biological phenomena with quantitative precision. Quantitative in vivo biochemistry, however, requires measurement not only of the effects of perturbation on a given process, but also of the magnitude of the perturbation to the target protein in time and space. Although reliable molecular readouts might be available for the former, the latter may be somewhat harder to measure. Fluorescently labelled target proteins offer an attractive solution for methods relying on protein degradation, since fluorescence intensity can provide a measure of protein concentration with spatiotemporal resolution [126]. However, although quantification of other perturbations, e.g. protein cleavage or activation, may be straightforward in cell populations at fixed time points, measurements with single-cell resolution or in real time will be much more challenging to achieve. A number of advanced cell imaging approaches may make this possible, but these techniques are themselves technically demanding and may not be widely available to researchers who do not have access to specialized equipment. For instance, measurements of protein dynamics (e.g. with fluorescence cross-correlation spectroscopy, or raster image correlation spectroscopy) and protein proximity (e.g. with FRET and fluorescence lifetime imaging) can be employed to determine effects on protein complex formation [127], while specialist (e.g. FRET-based) reporters can measure, e.g., the activity of enzymes responsible for post-translational modifications [128]. Future developments will therefore have to consider appropriate strategies, not just to perturb protein function, but to simultaneously measure the extent of that perturbation along with the biological effect(s). Concurrently, an ongoing challenge for technology developers is to make the techniques universally appealing and easy to employ. Ultimately, whether a technique is proven to be robust and fit for purpose will depend on uptake and testing by user communities. From a practical perspective, it may still take a significant investment of time to tailor an approach to the experimental system under investigation, despite efforts to make the available tools universally applicable. Prior knowledge of protein structure–function relationships may be required, for example, to guide the production of fusion proteins that retain normal activity and respond effectively to regulatory stimuli. These issues have inevitably limited the uptake of many of these techniques. Methods that utilize tools that are already widely employed, such as GFP-tagged proteins, are likely to be among the most popular in the short term because the methodologies can be rapidly deployed. For instance, deGradFP will no doubt be of particular interest to the Drosophila and zebrafish communities, which are creating transgenic libraries in which endogenous genes have been tagged with GFP [129,130]. Indeed, such collections may be the starting point for well-designed temporally controlled screens, to identify novel genes involved in developmentally regulated biological processes. As the field continues to mature, an area of future development is likely to be how multiple techniques might be combined for improved spatiotemporal control of a single protein, or for the induction of more than one protein, which has application in the engineering and study of artificial networks. Importantly, conditional control of protein function is not exclusive of genetic manipulation. Ultimately therefore, as gene-editing ‘knockin’ strategies mature [131], it may become routine to incorporate any number of different conditional domains on a genome-wide scale to facilitate such studies. FUNDING This work was supported by the Biotechnology and Biological Sciences Research Council (BB/J014516/1) and a RCUK block grant to the University of Liverpool for open access charges. Abbreviations AIDauxin-inducible degron APadaptor protein BDbinding domain BETbromo- and extra-terminal CALIchromophore-assisted light inactivation CaMKIIαCa2+/calmodulin-dependent protein kinase IIα CIB1cryptochrome-interacting basic helix–loop–helix 1 CIDchemical inducers of dimerization CLICRclustering indirectly using cryptochrome 2 Crycryptochrome DHFRdihydrofolate reductase ERRαoestrogen-related receptor α FALIfluorophore-assisted light inactivation FKBPFK506-binding protein FlAsHfluorescein-based arsenical hairpin binder FRAPFKBP-rapamycin-associated protein FRBFKBP-rapamycin-binding β-galβ-galactosidase Gal4/UASGal4 upstream activating sequence GPCRG-protein-coupled receptor HDRhomologous DNA recombination IMiDimmunomodulatory drug KSknocksideways LARIATlight-activated reversible inhibition by assembled trap LIDlight-induced dimerization LOVlight–oxygen–voltage mTORmammalian target of rapamycin MTXmethotrexate PA-Racphotoactivatable Race PhyBphytochrome B PIF3phytochrome-interacting factor 3 PROTACProteolysis-Targeting Chimaera ReAsHresorufin-based arsenical hairpin binder ROSreactive oxygen species RTKreceptor tyrosine kinase Syt Isynaptotagmin I TEVtobacco etch virus TIPITEV protease-mediated induction of protein instability TIR1transport inhibitor response 1 tsDHFRtemperature-sensitive dihydrofolate reductase VHLvon Hippel-Lindau protein ==== Refs 1 Venken K.J. Bellen H.J. Chemical mutagens, transposons, and transgenes to interrogate gene function in Drosophila melanogaster Methods 2014 68 15 28 10.1016/j.ymeth.2014.02.025 24583113 2 Lin C.Y. Chiang C.Y. Tsai H.J. Zebrafish and Medaka: new model organisms for modern biomedical research J. Biomed. Sci. 2016 23 19 10.1186/s12929-016-0236-5 26822757 3 Boulin T. Hobert O. From genes to function: the C. elegans genetic toolbox Wiley Interdiscip. Rev. Dev. Biol. 2012 1 114 137 10.1002/wdev.1 23801671 4 Boettcher M. McManus M.T. Choosing the right tool for the Job: RNAi, TALEN, or CRISPR Mol. Cell 2015 58 575 585 10.1016/j.molcel.2015.04.028 26000843 5 Sternberg S.H. Doudna J.A. Expanding the biologist's toolkit with CRISPR-Cas9 Mol. Cell 2015 58 568 574 10.1016/j.molcel.2015.02.032 26000842 6 Duffy J.B. GAL4 system in Drosophila : a fly geneticist's Swiss army knife Genesis 2002 34 1 15 10.1002/gene.10150 12324939 7 Neumann B. Walter T. Heriche J.K. Bulkescher J. Erfle H. Conrad C. Rogers P. Poser I. Held M. Liebel U. Phenotypic profiling of the human genome by time-lapse microscopy reveals cell division genes Nature 2010 464 721 727 10.1038/nature08869 20360735 8 Olaku V. Matzke A. Mitchell C. Hasenauer S. Sakkaravarthi A. Pace G. Ponta H. Orian-Rousseau V. c-Met recruits ICAM-1 as a coreceptor to compensate for the loss of CD44 in Cd44 null mice Mol. Biol. Cell 2011 22 2777 2786 10.1091/mbc.E11-02-0134 21680714 9 Fegan A. White B. Carlson J.C.T. Wagner C.R. Chemically controlled protein assembly: Techniques and applications Chem. Rev. 2010 110 3315 3336 10.1021/cr8002888 20353181 10 Rakhit R. Navarro R. Wandless T.J. Chemical biology strategies for post-translational control of protein function Chem. Biol. 2014 21 1238 1252 10.1016/j.chembiol.2014.08.011 25237866 11 Wijdeven R.H. Neefjes J. Ovaa H. How chemistry supports cell biology: the chemical toolbox at your service Trends Cell Biol 2014 24 751 760 10.1016/j.tcb.2014.07.002 25108565 12 Schultz L.W. Clardy J. Chemical inducers of dimerization: the atomic structure of FKBP12-FK1012A-FKBP12 Bioorg. Med. Chem. Lett. 1998 8 1 6 10.1016/S0960-894X(97)10195-0 9871618 13 Liu J. Farmer J.D. Lane W.S. Friedman J. Weissman I. Schreiber S.L. Calcineurin is a common target of cyclophilin-cyclosporin A and FKBP-FK506 complexes Cell 1991 66 807 815 10.1016/0092-8674(91)90124-H 1715244 14 Spencer D.M. Wandless T.J. Schreiber S.L. Crabtree G.R. Controlling signal transduction with synthetic ligands Science 1993 262 1019 1024 10.1126/science.7694365 7694365 15 Belshaw P.J. Spencer D.M. Crabtree G.R. Schreiber S.L. Controlling programmed cell death with a cyclophilin-cyclosporin-based chemical inducer of dimerization Chem. Biol. 1996 3 731 738 10.1016/S1074-5521(96)90249-5 8939689 16 Ho S.N. Biggar S.R. Spencer D.M. Schreiber S.L. Crabtree G.R. Dimeric ligands define a role for transcriptional activation domains in reinitiation Nature 1996 382 822 826 10.1038/382822a0 8752278 17 Putyrski M. Schultz C. Protein translocation as a tool: the current rapamycin story FEBS Lett. 2012 586 2097 2105 10.1016/j.febslet.2012.04.061 22584056 18 Brown E.J. Albers M.W. Shin T.B. Ichikawa K. Keith C.T. Lane W.S. Schreiber S.L. A mammalian protein targeted by G1-arresting rapamycin-receptor complex Nature 1994 369 756 758 10.1038/369756a0 8008069 19 Chen J. Zheng X.F. Brown E.J. Schreiber S.L. Identification of an 11-kDa FKBP12-rapamycin-binding domain within the 289-kDa FKBP12-rapamycin-associated protein and characterization of a critical serine residue Proc. Natl. Acad. Sci. U.S.A. 1995 92 4947 4951 10.1073/pnas.92.11.4947 7539137 20 Choi J. Chen J. Schreiber S.L. Clardy J. Structure of the KKBP12-rapamycin complex interactign with the binding domain of human FRAP Science 1996 273 239 242 10.1126/science.273.5272.239 8662507 21 Liberles S.D. Diver S.T. Austin D.J. Schreiber S.L. Inducible gene expression and protein translocation using nontoxic ligands identified by a mammalian three-hybrid screen Proc. Natl. Acad. Sci. U.S.A. 1997 94 7825 7830 10.1073/pnas.94.15.7825 9223271 22 Rollins C.T. Rivera V.M. Woolfson D.N. Keenan T. Hatada M. Adams S.E. Andrade L.J. Yaeger D. van Schravendijk M.R. Holt D.A. A ligand-reversible dimerization system for controlling protein–protein interactions Proc. Natl. Acad. Sci. U.S.A. 2000 97 7096 7101 10.1073/pnas.100101997 10852943 23 Pecot M.Y. Malhotra V. Golgi membranes remain segregated from the endoplasmic reticulum during mitosis in mammalian cells Cell 2004 116 99 107 10.1016/S0092-8674(03)01068-7 14718170 24 Zoncu R. Perera R.M. Balkin D.M. Pirruccello M. Toomre D. De Camilli P. A phosphoinositide switch controls the maturation and signaling properties of APPL endosomes Cell 2009 136 1110 1121 10.1016/j.cell.2009.01.032 19303853 25 Zoncu R. Perera R.M. Sebastian R. Nakatsu F. Chen H. Balla T. Ayala G. Toomre D. De Camilli P.V. Loss of endocytic clathrin-coated pits upon acute depletion of phosphatidylinositol 4,5-bisphosphate Proc. Natl. Acad. Sci. U.S.A. 2007 104 3793 3798 10.1073/pnas.0611733104 17360432 26 Robinson M.S. Sahlender D.A. Foster S.D. Rapid inactivation of proteins by rapamycin-induced rerouting to mitochondria Dev. Cell 2010 18 324 331 10.1016/j.devcel.2009.12.015 20159602 27 Bear J.E. Loureiro J.J. Libova I. Fassler R. Wehland J. Gertler F.B. Negative regulation of fibroblast motility by Ena/VASP proteins Cell 2000 101 717 728 10.1016/S0092-8674(00)80884-3 10892743 28 Cheeseman L.P. Harry E.F. McAinsh A.D. Prior I.A. Royle S.J. Specific removal of TACC3-ch-TOG-clathrin at metaphase deregulates kinetochore fiber tension J. Cell Sci. 2013 126 2102 2113 10.1242/jcs.124834 23532825 29 Al-Bassam S. Xu M. Wandless T.J. Arnold D.B. Differential trafficking of transport vesicles contributes to the localization of dendritic proteins Cell Rep. 2012 2 89 100 10.1016/j.celrep.2012.05.018 22840400 30 Haruki H. Nishikawa J. Laemmli U.K. The anchor-away technique: rapid, conditional establishment of yeast mutant phenotypes Mol. Cell 2008 31 925 932 10.1016/j.molcel.2008.07.020 18922474 31 Broermann A. Winderlich M. Block H. Frye M. Rossaint J. Zarbock A. Cagna G. Linnepe R. Schulte D. Nottebaum A.F. Vestweber D. Dissociation of VE-PTP from VE-cadherin is required for leukocyte extravasation and for VEGF-induced vascular permeability in vivo J. Exp. Med. 2011 208 2393 2401 10.1084/jem.20110525 22025303 32 Weitzman M. Hahn K.M. Optogenetic approaches to cell migration and beyond Curr. Opin. Cell Biol. 2014 30 112 120 25216352 33 Hughes J. Lamparter T. Mittmann F. Hartmann E. Gärtner W. Wilde A. Börner T. A prokaryotic phytochrome Nature 1997 386 663 10.1038/386663a0 9109482 34 Vierstra R.D. Davis S.J. Bacteriophytochromes: new tools for understanding phytochrome signal transduction Semin. Cell Dev. Biol. 2000 11 511 521 10.1006/scdb.2000.0206 11145881 35 Quail P.H. Phytochrome photosensory signalling networks Nat. Rev. Mol. Cell Biol. 2002 3 85 93 10.1038/nrm728 11836510 36 Kendrick R.E. Kronenberg G.H.M. Photomorphogenesis in Plants 1994 Dordrecht Kluwer Academic Publishers 10.1007/978-94-011-1884-2 37 Kennedy M.J. Hughes R.M. Peteya L.A. Schwartz J.W. Ehlers M.D. Tucker C.L. Rapid blue-light-mediated induction of protein interactions in living cells Nat. Methods 2010 7 973 975 10.1038/nmeth.1524 21037589 38 Liu H. Yu X. Li K. Klejnot J. Yang H. Lisiero D. Lin C. Photoexcited CRY2 interacts with CIB1 to regulate transcription and floral initiation in Arabidopsis Science 2008 322 1535 1539 10.1126/science.1163927 18988809 39 Idevall-Hagren O. Dickson E.J. Hille B. Toomre D.K. De Camilli P. Optogenetic control of phosphoinositide metabolism Proc. Natl. Acad. Sci. U.S.A. 2012 109 E2316 E2323 10.1073/pnas.1211305109 22847441 40 Guglielmi G. Barry J.D. Huber W. DeRenzis S. An optogenetic method to modulate cell contractility during tissue morphogenesis Dev. Cell 2015 35 646 660 10.1016/j.devcel.2015.10.020 26777292 41 Lee S. Park H. Kyung T. Kim N.Y. Kim S. Kim J. Heo W.D. Reversible protein inactivation by optogenetic trapping in cells Nat. Methods 2014 11 633 636 10.1038/nmeth.2940 24793453 42 Nguyen M.K. Kim C.Y. Kim J.M. Park B.O. Lee S. Park H. Heo W.D. Optogenetic oligomerization of Rab GTPases regulates intracellular membrane trafficking Nat. Chem. Biol. 2016 12 431 436 10.1038/nchembio.2064 27065232 43 Bugaj L.J. Choksi A.T. Mesuda C.K. Kane R.S. Schaffer D.V. Optogenetic protein clustering and signaling activation in mammalian cells Nat. Methods 2013 10 249 252 10.1038/nmeth.2360 23377377 44 Rosenfeldt G. Viana R.M. Mootz H.D. von Arnim A.G. Batschauer A. Chemically induced and light-independent cryptochrome photoreceptor activation Mol. Plant 2008 1 4 14 10.1093/mp/ssm002 20031911 45 Wend S. Wagner H.J. Muller K. Zurbriggen M.D. Weber W. Radziwill G. Optogenetic control of protein kinase activity in mammalian cells ACS Synth. Biol. 2014 3 280 285 10.1021/sb400090s 24090449 46 Bugaj L.J. Spelke D.P. Mesuda C.K. Varedi M. Kane R.S. Schaffer D.V. Regulation of endogenous transmembrane receptors through optogenetic Cry2 clustering Nat. Commun. 2015 6 6898 10.1038/ncomms7898 25902152 47 Hahn K.M. Kuhlman B. Hold me tightly LOV Nat. Methods 2010 7 597 595 10.1038/nmeth0810-595 20676079 48 Wu Y.I. Frey D. Lungu O.I. Jaehrig A. Schlichting I. Kuhlman B. Hahn K.M. A genetically encoded photoactivatable Rac controls the motility of living cells Nature 2009 461 104 108 10.1038/nature08241 19693014 49 Wu Y.I. Wang X. He L. Montell D. Hahn K.M. Spatiotemporal control of small GTPases with light using the LOV domain Methods Enzymol. 2011 497 393 407 21601095 50 Wang X. He L. Wu Y.I. Hahn K.M. Montell D.J. Light-mediated activation reveals a key role for Rac in collective guidance of cell movement in vivo Nat. Cell Biol. 2010 12 591 597 10.1038/ncb2061 20473296 51 Yao X. Rosen M.K. Gardner K.H. Estimation of the available free energy in a LOV2-J alpha photoswitch Nat. Chem. Biol. 2008 4 491 497 10.1038/nchembio.99 18604202 52 Strickland D. Yao X. Gawlak G. Rosen M.K. Gardner K.H. Sosnick T.R. Rationally improving LOV domain-based photoswitches Nat. Methods 2010 7 623 626 10.1038/nmeth.1473 20562867 53 Dolmans D.E. Fukumura D. Jain R.K. Photodynamic therapy for cancer Nat. Rev. Cancer 2003 3 380 387 10.1038/nrc1071 12724736 54 Grusch M. Schelch K. Riedler R. Reichhart E. Differ C. Berger W. Ingles-Prieto A. Janovjak H. Spatio-temporally precise activation of engineered receptor tyrosine kinases by light EMBO J. 2014 33 1713 1726 10.15252/embj.201387695 24986882 55 Toettcher J.E. Voigt C. Weiner O.D. Lim W.A. The promise of optogenetics in cell biology: interrogating molecular circuits in space and time Nat. Methods 2011 8 35 38 10.1038/nmeth.f.326 21191370 56 Paulus H. Protein splicing and related forms of protein autoprocessing Annu. Rev. Biochem. 2000 69 447 496 10.1146/annurev.biochem.69.1.447 10966466 57 Topilina N.I. Mills K.V. Recent advances in in vivo applications of intein-mediated protein splicing Mobile DNA 2014 5 5 10.1186/1759-8753-5-5 24490831 58 Cheriyan M. Perler F.B. Protein splicing: a versatile tool for drug discovery Adv. Drug Deliv. Rev. 2009 61 899 907 10.1016/j.addr.2009.04.021 19442693 59 Starokadomskyy P.L. Protein splicing Mol. Biol. 2007 41 278 293 10.1134/S0026893307020094 60 Vila-Perello M. Muir T.W. Biological applications of protein splicing Cell 2010 143 191 200 10.1016/j.cell.2010.09.031 20946979 61 Carrington J.C. Dougherty W.G. A viral cleavage site cassette: identification of amino acid sequences required for tobacco etch virus polyprotein processing Proc. Natl. Acad. Sci. U.S.A. 1988 85 3391 3395 10.1073/pnas.85.10.3391 3285343 62 Dougherty W.G. Parks T.D. Cary S.M. Bazan J.F. Fletterickt R. Characterization of the Catalytic residues of the tobacco etch virus 49-kDa proteinase Virology 1989 172 302 310 10.1016/0042-6822(89)90132-3 2475971 63 Kapust R.B. Tözsér J. Copeland T.D. Waugh D.S. The P1′ specificity of tobacco etch virus protease Biochem. Biophys. Res. Commun. 2002 294 949 955 10.1016/S0006-291X(02)00574-0 12074568 64 Kapust R.B. Waugh D.S. Controlled intracellular processing of fusion proteins by TEV protease Protein Expr. Purif. 2000 19 312 318 10.1006/prep.2000.1251 10873547 65 Uhlmann F. Wernic D. Poupart M.A. Koonin E.V. Nasmyth K. Cleavage of cohesin by the CD clan protease separin triggers anaphase in yeast Cell 2000 103 375 386 10.1016/S0092-8674(00)00130-6 11081625 66 Harder B. Schomburg A. Pflanz R. Küstner K.M. Gerlach N. Schuh R. TEV protease-mediated cleavage in Drosophila as a tool to analyze protein functions in living organisms BioTechniques 2008 44 765 772 10.2144/000112884 18476830 67 Pauli A. Althoff F. Oliveira R.A. Heidmann S. Schuldiner O. Lehner C.F. Dickson B.J. Nasmyth K. Cell-type-specific TEV protease cleavage reveals cohesin functions in Drosophila neurons Dev. Cell 2008 14 239 251 10.1016/j.devcel.2007.12.009 18267092 68 Williams D.J. Puhl H.L. Ikeda S.R. Rapid modification of proteins using a rapamycin-inducible tobacco etch virus protease system PLoS One 2009 4 e7474 10.1371/journal.pone.0007474 19830250 69 Fang J. Chen L. Cheng B. Fan J. Engineering soluble tobacco etch virus protease accompanies the loss of stability Protein Expr. Purif. 2013 92 29 35 10.1016/j.pep.2013.08.015 24012464 70 Kapust R.B. Tözsér J. Fox J.D. Anderson D.E. Cherry S. Copeland T.D. Waugh D.S. Tobacco etch virus protease: mechanism of autolysis and rational design of stable mutants with wild-type catalytic proficiency Protein Eng. 2001 14 993 1000 10.1093/protein/14.12.993 11809930 71 Cesaratto F. Lopez-Requena A. Burrone O.R. Petris G. Engineered tobacco etch virus (TEV) protease active in the secretory pathway of mammalian cells J. Biotechnol. 2015 212 159 166 10.1016/j.jbiotec.2015.08.026 26327323 72 Jacobson K. Rajfur Z. Vitriol E. Hahn K. Chromophore-assisted laser inactivation in cell biology Trends Cell Biol. 2008 18 443 450 10.1016/j.tcb.2008.07.001 18706812 73 Jay D.G. Selective destruction of protein function by chromophore-assisted laser inactivation Proc. Natl. Acad. Sci. U.S.A. 1988 85 5454 5458 10.1073/pnas.85.15.5454 3399501 74 Li W. Stuurman N. Ou G. Chromophore-assisted laser inactivation in neural development Neurosci. Bull. 2012 28 333 341 10.1007/s12264-012-1252-4 22833033 75 Beck S. Sakurai T. Eustace B.K. Beste G. Schier R. Rudert F. Jay D.G. Fluorophore-assisted light inactivation: a high-throughput tool for direct target validation of proteins Proteomics 2002 2 247 255 10.1002/1615-9861(200203)2:3<247::AID-PROT247>3.0.CO;2-K 11921440 76 Marek K.W. Davis G.W. Transgenically encoded protein photoinactivation (FlAsH-FALI): Acute inactivation of synaptotagmin I Neuron 2002 36 805 813 10.1016/S0896-6273(02)01068-1 12467585 77 Boulina M.E. Lukyanov K.A. Britanova O.V. Onichtchouk D. Lukyanov S. Chudakov D.M. Chromophore-assisted light inactivation (CALI) using the phototoxic fluorescent protein KillerRed Nat. Protoc. 2006 1 947 953 10.1038/nprot.2006.89 17406328 78 Ou G. Stuurman N. D'Ambrosio M. Vale R.D. Polarized myosin produces unequal-size daughters during asymmetric cell division Science 2010 330 677 680 10.1126/science.1196112 20929735 79 Boulina M.E. Chudakov D.M. Britanova O.V. Yanushevich Y.G. Staroverov D.B. Chepurnykh T.V. Merzlyak E.M. Shkrob M.A. Lukyanov S. Lukyanov K.A. A genetically encoded photosensitizer Nat. Biotechnol. 2006 24 95 99 10.1038/nbt1175 16369538 80 Pletnev S. Gurskaya N.G. Pletneva N.V. Lukyanov K.A. Chudakov D.M. Martynov V.I. Popov V.O. Kovalchuk M.V. Wlodawer A. Dauter Z. Pletnev V. Structural basis for phototoxicity of the genetically encoded photosensitizer KillerRed J. Biol. Chem. 2009 284 32028 32039 10.1074/jbc.M109.054973 19737938 81 Kobayashi J. Shidara H. Morisawa Y. Kawakami M. Tanahashi Y. Hotta K. Oka K. A method for selective ablation of neurons in C. elegans using the phototoxic fluorescent protein, KillerRed Neurosci. Lett. 2013 548 261 264 10.1016/j.neulet.2013.05.053 23748043 82 Yan L. Kanada M. Zhang J. Okazaki S. Terakawa S. Photodynamic treatment of tumor with bacteria expressing killerred PLoS One 2015 10 e0131518 26213989 83 Takemoto K. Matsuda T. Sakai N. Fu D. Noda M. Uchiyama S. Kotera I. Arai Y. Horiuchi M. Fukui K. SuperNova, a monomeric photosensitizing fluorescent protein for chromophore-assisted light inactivation Sci. Rep. 2013 3 2629 10.1038/srep02629 24043132 84 Shu X. Lev-Ram V. Deerinck T.J. Qi Y. Ramko E.B. Davidson M.W. Jin Y. Ellisman M.H. Tsien R.Y. A genetically encoded tag for correlated light and electron microscopy of intact cells, tissues, and organisms PLoS Biol. 2011 9 e1001041 10.1371/journal.pbio.1001041 21483721 85 Sarkisyan K.S. Zlobovskaya O.A. Gorbachev D.A. Bozhanova N.G. Sharonov G.V. Staroverov D.B. Egorov E.S. Ryabova A.V. Solntsev K.M. Mishin A.S. Lukyanov K.A. KillerOrange, a Genetically encoded photosensitizer activated by blue and green light PLos One 2015 10 e0145287 10.1371/journal.pone.0145287 26679300 86 Guo J. Chen H. Puhl H.L. Ikeda S.R. Fluorophore-assisted light inactivation produces both targeted and collateral effects on N-type calcium channel modulation in rat sympathetic neurons J. Physiol. 2006 576 477 492 10.1113/jphysiol.2006.113068 16873413 87 Yu G. Rosenberg J.N. Betenbaugh M.J. Oyler G.A. Pac-Man for biotechnology: co-opting degrons for targeted protein degradation to control and alter cell function Curr. Opin. Biotechnol. 2015 36 199 204 10.1016/j.copbio.2015.08.023 26435348 88 Jackson P.K. Eldridge A.G. Freed E. Furstenthal L. Hsu J.Y. Kaiser B.K. Reimann J.D.R. The lore of the RINGs: Substrate recognition and catalysis by ubiquitin ligases Trends Cell Biol. 2000 10 429 439 10.1016/S0962-8924(00)01834-1 10998601 89 Bachmair A. Finley D. Varshavsky A. In vivo half-life of a protein is a function of its amino-terminal residue Science 1986 234 179 186 10.1126/science.3018930 3018930 90 Varshavsky A. The N-end rule pathway and regulation by proteolysis Protein Sci. 2011 20 1298 1345 10.1002/pro.666 21633985 91 Dohmen R.J. Wu P. Varshavsky A. Heat-inducible degron: a method for constructing temperature-sensitive mutants Science 1994 263 1273 1276 10.1126/science.8122109 8122109 92 Speese S.D. Trotta N. Rodesch C.K. Aravamudan B. Broadie K. The ubiquitin proteasome system acutely regulates presynaptic protein turnover and synaptic efficacy Curr. Biol. 2003 13 899 910 12781128 93 Kanemaki M. Sanchez-Diaz A. Gambus A. Labib K. Functional proteomic identification of DNA replication proteins by induced proteolysis in vivo Nature 2003 423 720 724 10.1038/nature01692 12768207 94 Su X. Bernal J.A. Venkitaraman A.R. Cell-cycle coordination between DNA replication and recombination revealed by a vertebrate N-end rule degron-Rad51 Nat. Struct. Mol. Biol. 2008 15 1049 1058 10.1038/nsmb.1490 18794841 95 Johnston J.A. Johnson E.S. Waller P.R.H. Varshavsky A. Methotrexate inhibits proteolysis of dihydrofolate reductase by the N-end rule pathway J. Biol. Chem. 1995 270 8172 8178 10.1074/jbc.270.14.8172 7713922 96 Lévy F. Johnston J.A. Varshavsky A. Analysis of a conditional degradation signal in yeast and mammalian cells Eur. J. Biochem. 1999 259 244 252 10.1046/j.1432-1327.1999.00024.x 9914499 97 Bence N.F. Sampat R.M. Kopito R.R. Impairment of the ubiquitin-proteasome system by protein aggregation Science 2001 292 1552 1555 10.1126/science.292.5521.1552 11375494 98 Rajagopalan P.T.R. Zhang Z. McCourt L. Dwyer M. Benkovic S.J. Hammes G.G. Interaction of dihydrofolate reductase with methotrexate: ensemble and single-molecule kinetics Proc. Natl. Acad. Sci. U.S.A. 2002 99 13481 13486 10.1073/pnas.172501499 12359872 99 Taxis C. Stier G. Spadaccini R. Knop M. Efficient protein depletion by genetically controlled deprotection of a dormant N-degron Mol. Syst. Biol. 2009 5 267 10.1038/msb.2009.25 19401679 100 Rodríguez-Limas W.A. Tannenbaum V. Tyo K.E.J. Blocking endocytotic mechanisms to improve heterologous protein titers in Saccharomyces cerevisiae Biotechnol. Bioeng. 2015 112 376 385 10.1002/bit.25360 25154809 101 Zhou P. Bogacki R. McReynolds L. Howley P.M. Harnessing the ubiquitination machinery to target the degradation of specific cellular proteins Mol. Cell 2000 6 751 756 10.1016/S1097-2765(00)00074-5 11030355 102 Gray W.M. Pozo J.C. Walker L. Hobbie L. Risseeuw E. Banks T. Crosby W.L. Yang M. Ma H. Estelle M. Identification of an SCF ubiquitin–ligase complex required for auxin response in Arabidopsis thaliana Genes Dev. 1999 13 1678 1691 10.1101/gad.13.13.1678 10398681 103 Ruegger M. Dewey E. Gray W.M. Hobbie L. Turner J. Estelle M. The TIR1 protein of Arabidopsis functions in auxin response and is related to human SKP2 and yeast Grr1p Genes Dev. 1998 12 198 207 10.1101/gad.12.2.198 9436980 104 Nishimura K. Fukagawa T. Takisawa H. Kakimoto T. Kanemaki M. An auxin-based degron system for the rapid depletion of proteins in nonplant cells Nat. Methods 2009 6 917 922 10.1038/nmeth.1401 19915560 105 Zhang L. Ward J.D. Cheng Z. Dernburg A.F. The auxin-inducible degradation (AID) system enables versatile conditional protein depletion in C. elegans Development 2015 142 4374 4384 10.1242/dev.129635 26552885 106 Holland A.J. Fachinetti D. Han J.S. Cleveland D.W. Inducible, reversible system for the rapid and complete degradation of proteins in mammalian cells Proc. Natl. Acad. Sci. U.S.A. 2012 109 E3350 E3357 10.1073/pnas.1216880109 23150568 107 Morawska M. Ulrich H.D. An expanded tool kit for the auxin-inducible degron system in budding yeast Yeast 2013 30 341 351 10.1002/yea.2967 23836714 108 Sakamoto K.M. Kim K.B. Kumagai A. Mercurio F. Crews C.M. Deshaies R.J. Protacs: chimeric molecules that target proteins to the Skp1-Cullin-F box complex for ubiquitination and degradation Proc. Natl. Acad. Sci. U.S.A. 2001 98 8554 8559 10.1073/pnas.141230798 11438690 109 Deshaies R.J. Protein degradation: prime time for PROTACs Nat. Chem. Biol. 2015 11 634 635 10.1038/nchembio.1887 26284668 110 Toure M. Crews C.M. Small-molecule PROTACS: new approaches to protein degradation Angew. Chem. Int. Ed. Engl. 2016 55 1966 1973 10.1002/anie.201507978 26756721 111 Bondeson D.P. Mares A. Smith I.E. Ko E. Campos S. Miah A.H. Mulholland K.E. Routly N. Buckley D.L. Gustafson J.L. Catalytic in vivo protein knockdown by small-molecule PROTACs Nat. Chem. Biol. 2015 11 611 617 10.1038/nchembio.1858 26075522 112 Lu J. Qian Y. Altieri M. Dong H. Wang J. Raina K. Hines J. Winkler J.D. Crew A.P. Coleman K. Crews C.M. Hijacking the E3 ubiquitin ligase cereblon to efficiently target BRD4 Chem. Biol. 2015 22 755 763 10.1016/j.chembiol.2015.05.009 26051217 113 Winter G.E. Buckley D.L. Paulk J. Roberts J.M. Souza A. Dhe-Paganon S. Bradner J.E. Drug development. Phthalimide conjugation as a strategy for in vivo target protein degradation Science 2015 348 1376 1381 10.1126/science.aab1433 25999370 114 Zengerle M. Chan K.H. Ciulli A. Selective small molecule induced degradation of the BET bromodomain protein BRD4 ACS Chem. Biol. 2015 10 1770 1777 10.1021/acschembio.5b00216 26035625 115 Galdeano C. Gadd M.S. Soares P. Scaffidi S. Van Molle I. Birced I. Hewitt S. Dias D.M. Ciulli A. Structure-guided design and optimization of small molecules targeting the protein–protein interaction between the von Hippel–Lindau (VHL) E3 ubiquitin ligase and the hypoxia inducible factor (HIF) alpha subunit with in vitro nanomolar affinities J. Med. Chem. 2014 57 8657 8663 10.1021/jm5011258 25166285 116 Caussinus E. Kanca O. Affolter M. Fluorescent fusion protein knockout mediated by anti-GFP nanobody Nat. Struct. Mol. Biol. 2011 19 117 121 10.1038/nsmb.2180 22157958 117 Dunst S. Kazimiers T. von Zadow F. Jambor H. Sagner A. Brankatschk B. Mahmoud A. Spannl S. Tomancak P. Eaton S. Brankatschk M. Endogenously tagged rab proteins: a resource to study membrane trafficking in Drosophila Dev. Cell 2015 33 351 365 10.1016/j.devcel.2015.03.022 25942626 118 Brankatschk M. Dunst S. Nemetschke L. Eaton S. Delivery of circulating lipoproteins to specific neurons in the Drosophila brain regulates systemic insulin signaling eLife 2014 3 e02862 10.7554/eLife.02862 25275323 119 Shin Y.J. Park S.K. Jung Y.J. Kim Y.N. Kim K.S. Park O.K. Kwon S.H. Jeon S.H. Trinh le A. Fraser S.E. Nanobody-targeted E3-ubiquitin ligase complex degrades nuclear proteins Sci. Rep. 2015 5 14269 10.1038/srep14269 26373678 120 Janse D.M. Crosas B. Finley D. Church G.M. Localization to the proteasome is sufficient for degradation J. Biol. Chem. 2004 279 21415 21420 10.1074/jbc.M402954200 15039430 121 Renicke C. Schuster D. Usherenko S. Essen L.O. Taxis C. A LOV2 domain-based optogenetic tool to control protein degradation and cellular function Chem. Biol. 2013 20 619 626 10.1016/j.chembiol.2013.03.005 23601651 122 Mettetal J.T. Muzzey D. Gomez-Uribe C. van Oudenaarden A. The frequency dependence of osmo-adaptation in Saccharomyces cerevisiae Science 2008 319 482 484 10.1126/science.1151582 18218902 123 Paszek P. Jackson D.A. White M.R. Oscillatory control of signalling molecules Curr. Opin. Genet. Dev. 2010 20 670 676 10.1016/j.gde.2010.08.004 20850963 124 Tsien R.W. Tsien R.Y. Calcium channels, stores, and oscillations Annu. Rev. Cell Biol. 1990 6 715 760 10.1146/annurev.cb.06.110190.003435 2177344 125 Taslimi A. Vrana J.D. Chen D. Borinskaya S. Mayer B.J. Kennedy M.J. Tucker C.L. An optimized optogenetic clustering tool for probing protein interaction and function Nat. Commun. 2014 5 4925 10.1038/ncomms5925 25233328 126 Cluzel P. Surette M. Leibler S. An ultrasensitive bacterial motor revealed by monitoring signaling proteins in single cells Science 2000 287 1652 1655 10.1126/science.287.5458.1652 10698740 127 Diekmann S. Hoischen C. Biomolecular dynamics and binding studies in the living cell Phys. Life Rev. 2014 11 1 30 10.1016/j.plrev.2013.11.011 24486003 128 Zhou X. Herbst-Robinson K.J. Zhang J. Visualizing dynamic activities of signaling enzymes using genetically encodable FRET-based biosensors from designs to applications Methods Enzymol. 2012 504 317 340 10.1016/B978-0-12-391857-4.00016-1 22264542 129 Kawakami K. Abe G. Asada T. Asakawa K. Fukuda R. Ito A. Lal P. Mouri N. Muto A. Suster M.L. zTrap: zebrafish gene trap and enhancer trap database BMC Dev. Biol. 2010 10 105 10.1186/1471-213X-10-105 20950494 130 Nagarkar-Jaiswal S. Lee P.T. Campbell M.E. Chen K. Anguiano-Zarate S. Gutierrez M.C. Busby T. Lin W.W. He Y. Schulze K.L. A library of MiMICs allows tagging of genes and reversible, spatial and temporal knockdown of proteins in Drosophila Elife 2015 4 e05338 131 Pinder J. Salsman J. Dellaire G. Nuclear domain 'knock-in' screen for the evaluation and identification of small molecule enhancers of CRISPR-based genome editing Nucleic Acids Res. 2015 43 9379 9392 10.1093/nar/gkv993 26429972

PMC005xxxxxx/PMC5003694.txt

==== Front Biochem JBiochem. JppbiochemjBJBiochemical Journal0264-60211470-8728Portland Press Ltd. BCJ2016010710.1042/BCJ20160107Review ArticlesReview Article45815118When fast is better: protein folding fundamentals and mechanisms from ultrafast approaches Ultrafast protein foldingV. Muñoz and M. CerminaraMuñoz Victor *†‡1Cerminara Michele *†* National Biotechnology Center, CSIC, Darwin, 3. E-28049 Madrid, Spain† IMDEA Nanosciences Institute, Calle Faraday 9, E-28049 Madrid, Spain‡ School of Engineering, University of California Merced, 5200 N. Lake Road, Merced, CA 95343, U.S.A.1 To whom correspondence should be addressed (email vmunoz@cnb.csic.es).30 8 2016 1 9 2016 473 17 172545 2559 11 2 2016 11 4 2016 18 4 2016 © 2016 The Author(s)2016This is an open access article published by Portland Press Limited on behalf of the Biochemical Society and distributed under the Creative Commons Attribution Licence 4.0 (CC BY).Protein folding research stalled for decades because conventional experiments indicated that proteins fold slowly and in single strokes, whereas theory predicted a complex interplay between dynamics and energetics resulting in myriad microscopic pathways. Ultrafast kinetic methods turned the field upside down by providing the means to probe fundamental aspects of folding, test theoretical predictions and benchmark simulations. Accordingly, experimentalists could measure the timescales for all relevant folding motions, determine the folding speed limit and confirm that folding barriers are entropic bottlenecks. Moreover, a catalogue of proteins that fold extremely fast (microseconds) could be identified. Such fast-folding proteins cross shallow free energy barriers or fold downhill, and thus unfold with minimal co-operativity (gradually). A new generation of thermodynamic methods has exploited this property to map folding landscapes, interaction networks and mechanisms at nearly atomic resolution. In parallel, modern molecular dynamics simulations have finally reached the timescales required to watch fast-folding proteins fold and unfold in silico. All of these findings have buttressed the fundamentals of protein folding predicted by theory, and are now offering the first glimpses at the underlying mechanisms. Fast folding appears to also have functional implications as recent results connect downhill folding with intrinsically disordered proteins, their complex binding modes and ability to moonlight. These connections suggest that the coupling between downhill (un)folding and binding enables such protein domains to operate analogically as conformational rheostats. conformational rheostatsmolecular dynamics simulationsnuclear magnetic resonanceprotein folding energy landscapessingle-molecule spectroscopyultrafast kinetic methods ==== Body BACKGROUND Proteins are true nanomachines in charge of most biological roles in living cells, a feat they accomplish by self-assembling into sophisticated 3D structures that exploit thermal, and on occasion chemical, energy to change shape in response to stimuli. As proposed by Anfinsen [1] in his seminal work, the process by which proteins fold into their functional structures is dictated by the chemical blueprints encoded into their amino acid sequence. This assertion implies that if we understood protein folding mechanisms in depth, we would have at our fingertips the ability to read genomic information in real functional terms, and to design and engineer synthetic biological components à la carte. Such motivations have fuelled the interest in the ‘protein folding problem’ among scientists from a wide variety of disciplines. Moreover, the physical principles that govern protein self-assembly still hold true in vivo, where folding is tightly coupled to many other processes that control the protein life cycle (Figure 1). In fact, the protein cycle shown in Figure 1 emphasizes that understanding the mechanisms of protein folding is an essential requirement to comprehend protein homoeostasis in vivo. It is thus our contention that molecular biologists and biochemists ought to keep abreast of recent exciting developments in protein folding research, even if those developments are increasingly coming from the realm of physics. Figure 1 Protein folding inside the cell A new protein is synthetized at the ribosome as determined by the activation of gene expression. The nascent chain is typically bound to chaperones that keep it unfolded until the process is complete and the chain is released (A). The recently synthesized unfolded protein folds autonomously and reversibly, establishing a dynamic equilibrium between the unfolded ensemble (B) and the native state (C). This conformational equilibrium determines the protein's ability to perform its function, either directly as a monomer or by further assembling into larger macromolecular complexes (D). The half-life of its biological activity is also controlled by the folded–unfolded equilibrium because the proteasome machinery eliminates targeted proteins from their unfolded state (G). Likewise, the transient formation of partially folded conformations can lead to misfolding events (E) that feed the formation of aberrant and/or toxic protein aggregates (e.g. amyloids) (F). The ribosome image is reproduced with permission from Schmeing and Ramakrishnan (2009) Nature 461 1234–1242. All other structures are available from the PDB. In this regard, the development of ultrafast folding approaches was a major turning point in protein folding research. Previously, experimentalists were constrained by the millisecond resolution of stopped-flow methods, which also offered limited structural resolution. The single-domain proteins that could be studied folded slowly (from tens of milliseconds to minutes) and apparently via a single stroke process, which led to the generalization of the two-state folding model [2]. Although apparently simple two-state folding implies that all of the intermediate structures responsible for defining the mechanism are highly unstable and thus inaccessible to experiment. In a parallel front, analytical theory based on condensed matter and polymer physics defined folding reactions as the stochastic search for the native structure on a corrugated hyper-dimensional energy landscape with an overall funnelled shape that acts as driving force (the energy landscape approach) [3]. Such description had three key implications that departed drastically from conventional interpretations of folding: (i) the existence of myriad microscopic folding pathways instead of a unique sequence of structural events [4]; (ii) the definition of a folding speed limit determined by the timescale of the relevant conformational motions of the protein [5]; and (iii) the prediction that folding free energy barriers originate from entropic bottlenecks and are in general shallow, leading to the possibility of downhill (barrierless) folding [6]. At the time, theory could only be tested against computer simulations that used coarse-grained representations of protein structure [7–9] since the best atomistic simulations were still six or seven orders of magnitude too short in their accessible timescales [10]. Coarse-grained simulations supported energy landscape predictions and showed that a two-state mechanism was not required for efficient folding [11]. There was, however, a deep divide between what theory and simulations predicted and what experiments reported. Access to novel ultrafast folding methods drastically changed this state of affairs, permitting to connect experiments, theory and simulations in synergistic ways. In the present review, we discuss the key contributions as well as the methodological developments that made them possible. In so doing, our goal has not been to be fully comprehensive. After all, the material is too extensive and there already are excellent technical reviews that cover many of these topics separately. Our major motivation has been instead to provide the non-expert reader with an account of the most compelling discoveries in this area and of their significance in shaping the solid conceptual framework in protein folding that we enjoy today. We end the review discussing recent exciting results that highlight the functional and technological implications of fast protein folding as molecular mechanism for the implementation of conformational rheostats. METHODS TO INVESTIGATE ULTRAFAST FOLDING Fast folding is defined as any protein conformational change that takes place in less than 1 ms. In this section, we describe different approaches that reach the sub-millisecond timescales required to investigate fast-folding reactions. Ultrafast kinetic techniques Kinetic experiments measure the conformational relaxation of the protein in response to a perturbation, and have time resolution determined by how quickly the perturbation is enacted (Figure 2). The first fast-folding experiment used a laser trigger to rapidly initiate folding of chemically denatured cytochrome c by photodissociating the haem-bound carbon monoxide, which binds preferentially to the unfolded state [12]. Several photochemical triggers have been used for fast-folding research [13–15], but they are usually protein-specific. An alternative is to use laser pulses to change the surrounding solvent. Photochemical stimulation of caged compounds added to the solution, such as o-nitrobenzaldehyde, releases one proton per molecule resulting in abrupt pH decreases (up to ∼2 units) in ∼100 ns [16]. Ultrafast temperature jumps can be induced by heating the surrounding solvent with an infrared laser pulse at a frequency that overlaps with water vibrational modes [17–19]. The laser-induced T-jump technique attains increases of ∼10–15°C in a few nanoseconds (100000-fold faster than stopped flow), and is universal because protein folding reactions are always temperature-dependent [20]. These characteristics have made the laser-induced T-jump technique the most popular ultrafast kinetic method, including implementations that detect fluorescence intensity [17,19], fluorescence spectra [21], infrared absorption [22,23] and Raman scattering [24]. More recently, this technique has been combined with selective isotope editing to monitor fast-folding kinetics with nanosecond resolution at the level of single peptide bonds [25] and individual side chains [26]. Figure 2 Experimental and computational approaches for investigating fast protein folding Ultrafast kinetic perturbation methods (left): these techniques report on the macroscopic (bulk) relaxation decay to the new thermodynamic conditions imposed by fast perturbation. The key is to devise an efficient procedure to quickly alter the folding–unfolding equilibrium by changing temperature, pH, pressure or chemical potential. The relaxation is monitored using spectroscopic techniques that lead to the determination of the relaxation rate and the amplitude of the change in signal, which, when analysed with a suitable kinetic model, provide the changes in population and the microscopic rates of interconversion between species. The fastest of these methods use laser pulses as triggers, resulting in time responses potentially as short as 1 ps (10−12 s). The laser-induced T-jump method represented produces 10–15°C jumps in less than 10 ns. Relaxation dispersion NMR (top right): different nuclei in the protein rotate (precess) about the magnetization axis at slightly different frequencies given by their chemical shifts, resulting in a loss of phase coherence that broadens the overall magnetization signal that is recorded on the transverse plane (x–y plane). In a system in which chemical shifts do not change with time, the signal decays according to the transverse relaxation rate (R2), which for proteins is ∼5–20 s−1. The interconversion between species in shorter times results in further decoherence, and thus in broader signals/faster decays (enhanced relaxation). The RD-NMR experiment measures this effect using specific radiofrequency pulses that flip the magnetization 180° on the x–y plane to induce refocusing because the faster spins, which are now behind, will eventually catch up with the slower ones. The refocusing pulses cannot compensate for changes in chemical shift due to the molecules exchanging conformation during acquisition. Therefore applying trains of refocusing pulses interspersed at fixed times (Carr–Purcell–Meiboom–Gill or CPMG relaxation dispersion [33]) makes it possible to measure conformational exchange processes. Molecular dynamics simulations (bottom right): MD trajectories of proteins and their analysis are performed in five steps: (A) defining a molecular mechanics force field that calculates the potential energy of the protein as a function of the atomic co-ordinates; (B) building a simulation box that contains all of the atoms from the protein (atomic co-ordinates) plus surrounding solvent (water) molecules; (C) numerical integration of Newton's equations of motion defined by the positions, forces and velocities of every atom in the simulation box over time steps of ∼1 fs (10−15 s) using a supercomputer (shown here is Anton, the computer designed by D.E. Shaw Research for ultra-efficient MD simulations); (D and E) analysis and molecular-mechanistic interpretation of the terabytes of data included in the simulated trajectories. Another ultrafast kinetic perturbation method induces pressure jumps that unfold proteins due to the larger total volume occupied by the native protein and its hydration shell [27]. The fastest implementation of this method achieves microsecond resolution using an electrical discharge to rupture a metallic membrane separating the protein sample from a solution held at very high pressure [28,29]. Advances in chemical dilution techniques using rapid mixers in continuous flow operation have also led to resolutions of ∼50 μs for turbulent mixing [30,31] or only 10 μs for set-ups using microfluidics and hydrodynamic focusing [32]. Single-molecule spectroscopy Single-molecule methods can resolve the stochastic conformational fluctuations of the protein in equilibrium conditions, and thus do not need a fast perturbation. Time resolution is simply determined by the speed of data acquisition. Single-molecule methods available to investigate protein folding can be classified into two groups: (i) force spectroscopy techniques in which single protein molecules are unfolded by pulling from its ends and refolded upon releasing the force; (ii) enhanced optical microscopy methods that detect individual proteins while they undergo conformational transitions, typically using fluorescence imaging (single-molecule fluorescence microscopy). Mechanical unfolding can be achieved with a variety of methods, such as atomic force microscopy [34], magnetic tweezers [35] and optical tweezers [36,37]. Each of these techniques has advantages and disadvantages [38]. Data collection is limited by the oscillation frequency and spring constant of the microdevice that exerts mechanical control (e.g. AFM cantilever or the microbeads in an optical tweezers set-up), which constrains most of these techniques to millisecond resolutions. There are, however, a few applications of optical tweezers that are now reaching microsecond resolution [39,40]. Single-molecule fluorescence spectroscopy employs a confocal microscope to illuminate volumes of only ∼1 fl (10−15 litres) to isolate and detect individual protein molecules, whether immobilized on a surface or freely diffusing. The microscope objective also collects fluorescence photons emitted by the molecule, which are detected by highly efficient, picosecond resolution, avalanche photodiodes [41]. Information about conformational transitions is commonly obtained from the efficiency of FRET (Förster resonance energy transfer) between a donor–acceptor fluorescent pair introduced in specific positions of the protein [41,42]. Time resolution is not set by the detector's response, but by photon emission statistics; that is, by the time it takes to collect the bunches of approximately 50 photons required to measure FRET efficiency (i.e. the ratio between number of photons emitted by the acceptor and total number of emitted photons) with reasonable accuracy [43]. The characteristically low photon collection efficiency of these methods (∼1% [44]) and the low illumination intensity required to avoid photochemical damage of the fluorescent dyes has typically limited their resolution to milliseconds [45]. However, implementation of ultra-efficient photoprotection cocktails have raised the resolution up to ∼50 μs [46]. A complementary approach involves developing methods to analyse photons one by one, like the Gopich–Szabo MLA (maximum likelihood analysis) of photon trajectories [47,48]. Recently, the combination of this approach and a simple theoretical model of protein folding has shown promise to extend the time resolution to 10 μs [49]. Probing fast folding at atomic resolution NMR (nuclear magnetic resonance) is particularly attractive for protein folding studies because it provides both atomic-resolution and dynamic information [50]. Transverse RD-NMR (relaxation dispersion NMR) experiments (Figure 2), in which the NMR signal decoherence caused by chemical shift anisotropy is measured as a function of time, are particularly advantageous. RD-NMR can detect exchange with partially folded conformations that are only minimally populated (down to 0.2%, the so-called invisible states) from the broadening of the native state peaks [51]. Moreover, it permits the extraction of the exchange rate, the population of the invisible species and also its chemical shift values [52]. Chemical shifts provide structural information about the invisible state at atomic resolution [53]. In this technique, time resolution is ultimately determined by technical limitations (e.g. how quickly the refocusing pulses can be applied) that make it ideal for millisecond processes [54]. However, recent developments have pushed the RD-NMR resolution down to 200 μs (rates of 5000 s−1) [55]. Atomistic computer simulations MD (molecular dynamics) simulations offer atomistic structural resolution and dynamic information with virtually infinite time resolution. Therefore MD simulations could potentially provide all of the structural and dynamical insights required to understand the underpinnings of folding mechanisms [56] (Figure 2). Here the problem is not time resolution, but how to reach the much longer timescales of protein folding reactions. For decades, the enormous timescale gap that existed between MD simulations and folding experiments impeded testing and refining the simulations with empirical data and also restricted their use in interpreting experimental results. The first all-atom MD simulation in explicit solvent to hit the 1 μs mark was performed by Duan and Kollman [57] on the small villin headpiece subdomain. The simulation did not achieve complete folding, but it captured hydrophobic collapse and helix formation in the unfolded state [57]. In the last decade, major technical advances have increased sampling to the level of resolving multiple folding–unfolding events on several fast-folding proteins (see [58] for a detailed review). Such approaches have been of two kinds. In the first one, sampling is enhanced by running thousands of short (∼1–5 ns) MD simulations using distributed computing strategies [59]. The idea is to capture some of the fastest possible folding trajectories of a fast-folding protein by taking advantage of the exponential distribution of folding times [for a process with folding rate of 1/(1 μs), the probability of seeing a 1 ns folding trajectory is ∼0.1%]. The second approach aims at making the calculations faster, by optimizing the code [60], by optimizing the force field [61] or by designing new computers hardwired for MD calculations [62]. The latter have resulted in the implementation of MD simulations that routinely reach the 1 ms mark [63], which in turn permits the optimization of force fields by thorough testing and comparison with folding experimental data [64]. MAPPING PROTEIN FOLDING MOTIONS A major contribution from ultrafast kinetic methods has been the determination of the timescales for various folding-related conformational motions [65]. Combining seminal contributions from multiple groups, we can now establish a basic roadmap of the timescales that are relevant to the different structural events that take place during folding reactions (Figure 3). Figure 3 Roadmap of folding timescales The chart shows the timescales associated with the various structural events that take place in protein folding reactions and the experimental and computational methods that are used to probe them. smFRET, single-molecule FRET. Hydrophobic collapse The random collapse of the unfolded polypeptide chain to exclude hydrophobic side chains from surrounding solvent is one of the key events that take place during folding. Collapse was theoretically predicted to be much faster than folding [66], which immediately made it a target for fast-folding experiments. Seminal laser T-jump experiments on a pH-denatured small protein that collapsed as a function of temperature rendered timescales of 100 ns [21]. This nanosecond timescale was later confirmed on other denatured proteins using single-molecule spectroscopy [67]. Therefore non-specific hydrophobic collapse appeared to be among the fastest folding-related processes [21]. The rate, however, can be decreased significantly when collapse is initiated from the fully expanded unfolded state by chemical denaturant dilution [68]. Later efforts have focused on determining the role that interactions other than the hydrophobic force exert on the properties of unfolded polypeptides. Experiments on IDPs (intrinsically disordered proteins) are noteworthy because these proteins do not fold, have low hydrophobicity and have high charge density [69]. Interestingly, single-molecule FRET experiments performed on IDPs also showed a more compact state in the absence of chemical denaturants, indicating that backbone hydrogen bonds also play a role in the solvent-induced contraction of unfolded polypeptide chains [70,71]. Loop formation The closure of loops determines the formation of interactions between secondary-structure elements to form supersecondary and tertiary arrangements. The timescale for loop formation was accordingly considered a proxy for the folding speed limit [72], a parameter that is important for determining the magnitude of folding barriers from the experimental rates (see the next section). The inaugural fast-folding experiment involved the determination of the rate of closing a long loop in chemically unfolded cytochrome c, which occurred in tens of microseconds [12]. Using scaling arguments, these results led to an estimate of 1 μs for the formation of shorter protein-like loops [72]. Subsequent experiments on short flexible peptides in aqueous solution have produced timescales for forming protein-like loops that are much shorter, ranging from ∼50 ns for contact formation between the ends of 12–20-residue unstructured peptides [73,74] down to only 10 ns for contact formation on the shortest three- or four-residue turns [74]. Secondary structure Investigating the timescales for secondary-structure elements required combining the fastest kinetic methods with small peptides that were able to form stable secondary structures on their own. α-Helix formation was studied early on using alanine-rich peptides and found to occur in ∼150 ns [18,19]. Subsequent studies used photoswitches to trigger helix formation from chemically distorted non-helical conformations [14,75]. Interestingly, for richer amino acid sequences (more protein-like) helix formation was found to be only slightly slower (a factor of 2–3) [76]. Although the α-helix is a simple and symmetric structure, higher-resolution experiments showed that helix formation occurs via a complex kinetic process in which nucleation, propagation and the splitting and merging of helical segments conspire to produce different timescales for different positions along the molecule [77–80]. Nevertheless, such complexity takes place within a narrow range of timescales: 150–500 ns for the formation of stable α-helices, and up to ∼1.2 μs for the nucleation of a thermodynamically unstable helical turn on cyclized peptides [81]. The similar timescales for non-specific collapse and α-helix formation suggest that both processes occur almost concomitantly during the early folding stages. The β-hairpin is the minimal β-structure and the basic component of antiparallel β-sheets. Moreover, β-hairpin formation includes local structure (the turn) and collapse of the strands, and thus can be considered the simplest example of protein folding (i.e. ‘mini-protein’). The first experiments on β-hairpin formation were performed on the GB1 hairpin, which was naturally stable when isolated from the protein [82,83]. These experiments showed a relaxation time that was almost 25-fold longer than α-helix formation. Statistical mechanical modelling could neatly explain the differences between both types of structures by invoking a mechanism in which β-hairpins form the central turn first and then zip up the two strands [82,83]. These seminal results and conclusions elicited a great deal of interest in the investigation of β-hairpin formation, which became a benchmark for protein folding studies. A vast array of computer simulations on β-hairpin folding ensued [84–93], producing results that tended to favour the opposite mechanism in which the hydrophobic cluster collapsed first followed by zipping down to form the turn last. However, subsequent experiments confirmed the main conclusions from the original GB1 β-hairpin study. For example, it was shown experimentally that β-hairpin folding is accelerated when the loop connecting both strands is shortened [94], and that turn formation plays a key role in driving β-hairpin folding [95,96]. More recently, an exploration of the speed limit for β-hairpin folding has demonstrated that, when the turn is autonomously stable, β-hairpin formation takes only ∼100 ns, approaching the timescales for the fastest helix formation [97]. Topological reorganization The rearrangement of secondary-structure elements to form native tertiary interactions on a randomly collapsed globule should occur intrinsically more slowly than the motions described before because it requires breaking pre-formed interactions before the protein can reconfigure. Frieden and co-workers measured the reconfiguration dynamics of an unfolded protein under different solvent conditions and found that such conformational rearrangements take place in a few microseconds, and involve formation and dissolution of partially folded structures [98]. Additional experiments by Lapidus et al. [99] with newer and faster continuous flow mixers (20 μs mixing time) have shown that, whereas non-specific collapse occurs within the instrument dead-time, formation of native tertiary contacts is the last event requiring times of at least 100 μs. FAST PROTEIN FOLDING Understanding the determinants of protein folding rates Figure 3 provides an entry point to investigate the determinants of the over six orders of magnitude spread in folding rates that is observed in natural single-domain proteins [100], and, as an ancillary issue, to estimate the speed limit for protein folding reactions [101]. For these purposes, we can utilize a simple folding rate expression derived from the energy landscape approach [6]: 1 kfold=1τ0exp(−ΔG†/RT) where ΔG† is the overall free energy barrier to folding and τ0 is the timescale for the conformational motions undergone by the protein when is crossing the barrier top. The pre-exponential term should be relatively protein invariant, and thus the large differences in folding rates must come from the free energy barrier. According to theory, the barrier arises from an abrupt loss in conformational entropy that occurs at early stages and is not compensated by formation of stabilizing interactions until much later in the process [4]. On the other hand, the folding speed limit is determined by the pre-exponential term and is achieved when the free energy barrier vanishes, resulting in downhill (barrierless) folding [6,102]. Figure 3 suggests that the folding speed limit lies between 107 and 105 s−1 (τ0 from 100 ns to 10 μs). Using this range, it was possible to estimate barrier heights from experimental folding rates, and derive its entropic and enthalpic contributions [103]. This empirical analysis confirmed that folding barriers are entropic bottlenecks as predicted by theory. Moreover, folding rates scale with the size of the protein, which results in inverse correlations between experimental folding rates and a fractional exponent of the number of amino acids [100,104,105] (Figure 4, left). Interestingly, the slope of the rate against size correlation rendered an independent estimate of the folding speed limit of 1 μs [100], which is in the middle of the range estimated from Figure 3, and identical with the estimate obtained from loop formation in cytochrome c (see above). Figure 4 The determinants of protein folding rates Left: the direct correlation existing between protein folding times and a fractional exponent of the number of amino acids in the protein (in this case ½). Right: the compensation between the stabilization energy per residue provided by local and non-local interactions. The higher the non-local fraction the lower the folding rate. Folding rates also depend on the properties of the native 3D structure. This factor was first observed in a correlation between experimental folding rates and the contact order, a parameter related to the average sequence separation between residues in close native contact [106]. The relationship between rates and native structure was later rationalized theoretically using statistical mechanical modelling of protein folding [107]. Size also plays an important role in determining protein stability [108] and unfolding rates [109]. In fact, just ten bits of protein-specific information (number of amino acids and structural class) seem to be sufficient to predict folding and unfolding rates (and thus stability) with errors of magnitude equivalent to the typical perturbations induced by just two point mutations [109]. The combined effects of size and structural class can be condensed on to a single parameter that defines the fractional contributions to native stability from local (between residues close in sequence) and non-local (tertiary) interactions (Figure 4, right). The local fraction increases as the protein becomes smaller because the higher surface/volume ratio of smaller globular structures results in fewer tertiary contacts per residue. Searching for microsecond folding proteins New kinetic techniques also opened the opportunity to resolve the folding–unfolding kinetics of proteins that were too fast for the stopped-flow method. This possibility triggered a quest for the fastest possible protein with the goal of reaching the folding speed limit [101,110], and thus the downhill folding scenario [111]. Inspired by the concepts exemplified in Figure 4, efforts focused on either small protein domains or proteins with maximal energetic contributions from local structure. As a consequence, a large number of proteins that fold to completion in just microseconds has been identified over the last few years [65,112] (see the left-hand panel of Figure 4). Most of these proteins are α-helix bundles [110,113–118] or very small antiparallel β-proteins, such as WW domains [119–122], although there also are examples of mid-size domains with mixed α/β structure [123]. Additional efforts have focused on trying to increase the folding rate by increasing the net stabilization from local interactions, including strategies such as weakening the hydrophobic core of λ-repressor [110], engineering WW domains with very stable β-turns [124,125] or looking at larger de novo designed proteins with maximal helical propensity such as α3D [115]. Downhill folding Another exciting implication of fast folding research is the downhill folding scenario. Barrierless folding raised a lot of interest because it was an opportunity to test experimentally a true prediction from energy landscape theory that could not be explained with traditional descriptions [102,126]. Moreover, on a downhill folding protein, the partially structured intermediates on the folding pathway could be populated and thus become accessible to experimental detection [111]. The problem was how to unambiguously diagnose downhill folding without relying exclusively on timescales. Early attempts focused on the observation of strange kinetics (relaxations over-extended in time) that could be due to the transient accumulation of large numbers of intermediates. Strange kinetics was indeed reported from T-jump experiments that induced refolding from the cold-denatured state of several slow-folding proteins, such as yeast phosphoglycerate kinase and a mutant ubiquitin [127], or a cold-shock protein from Escherichia coli [128]. Later, it was appreciated that thermodynamics provided much more reliable diagnostics than kinetics because a downhill folding protein should (un)fold gradually resulting in structurally complex equilibrium unfolding [129]. The combination of multiple spectroscopic probes, calorimetry and theoretical analysis revealed such gradual unfolding on the small helical protein BBL, which was accordingly identified as the first example of global downhill folding [130]. This classification was subsequently confirmed by measuring the thermal unfolding of BBL at atomic resolution using NMR [131]. Subsequent kinetic studies using multi-probe laser T-jump experiments demonstrated that BBL folds extremely fast (folding time of ∼1 μs at the midpoint temperature), as expected [117]. The connection between how broad and heterogeneous is the equilibrium unfolding of a protein and how fast it folds was elegantly demonstrated in follow up experiments that compared BBL with the ∼10-fold slower structural homologue PDD [132]. A more recent test that relies on measuring folding kinetics after T-jumps of different magnitude to the same final temperature has confirmed the one-state downhill folding of BBL in contrast with the also fast, but barrier-crossing, folding of Trpzip-3c [133]. Other groups have pursued the downhill folding regime by increasing the rate of fast-folding proteins via mutation. Gruebele and co-workers engineered the microsecond folder λ6–85 to make it even faster, and thus induce the emergence of the probe-dependent equilibrium unfolding and complex kinetics of the global downhill folding regime [134–136]. Eaton and co-workers took the villin headpiece subdomain, a small helical protein that folded in few microseconds, and sped up folding 6-fold with two designed point mutations [137]. This superfast villin version was the first example of sub-microsecond folding, and was subsequently identified as a downhill folder from multi-probe and kinetic criteria [138]. The most direct observation of the gradual disordering associated with global downhill folding has been obtained with modern single-molecule fluorescence spectroscopy [139]. In principle, single-molecule methods can directly distinguish between a scenario in which each molecule is either fully unfolded or folded (barrier-crossing folding) and a scenario in which individual molecules are partially unfolded (global downhill folding) [46,139,140]. The challenge was how to make the photon count rates much higher than the ultrafast relaxation expected for a downhill folder. In this case, the authors managed to do so combining advanced photoprotection methods [46] and measurements at very low temperature to slow down the BBL folding–unfolding relaxation 100-fold [139]. PROBING ENERGY LANDSCAPES OF PROTEIN FOLDING The multidimensional energy landscapes of protein folding are supposed to be funnelled towards the native 3D structure, but also rugged (many local valleys and peaks) [102,141]. Landscape topography is in fact very important because it determines the specific folding pathways and mechanisms for each protein, but it has proved to be extremely difficult to probe experimentally. For slow two-state folding domains, the only method available is mutational analysis in which the effects that mutations have on both folding and unfolding rates are used to infer structural properties of the folding transition state ensemble (the conformations that make the barrier top in eqn 1) [142]. This method has been widely used, and there has been ample discussion about the mechanistic interpretation of such mutational data [143–146]. More recently, large-scale analysis of the wealth of mutational data available on multiple proteins has revealed an almost universal trend (one-third of the free energy change on folding and two-thirds on unfolding) [147], which indicates that the specific structural and mechanistic information included in these datasets is very limited [147]. On the other hand, the discovery of fast-folding proteins has made it possible to develop powerful alternatives based on a new set of methods for probing the topographic details of folding energy landscapes. Thermodynamic folding barriers from calorimetry DSC (differential scanning calorimetry) is extremely sensitive to the conformational heterogeneity of protein folding reactions [148]. DSC can thus be employed to estimate the shape of the folding free energy surface. Such a method was originally developed to distinguish between all-or-none (two-state) and gradual (downhill) protein folding [149]. It was subsequently demonstrated that the DSC experiment has enough information to detect small deviations from two-state caused by minimal population (<1%) of the barrier top conformations, leading to the possibility of estimating folding free energy barriers from these thermodynamic experiments [150]. DSC has thus become a powerful tool for characterizing folding energy landscapes, especially for proteins that fold fast [123,132,151,152] or with limited co-operativity [153,154]. The method has solid theoretical grounds, but its practical implementation requires fitting the DSC data to a specific model that represents the folding free energy surface [148]. The results are somewhat dependent on how well the model can represent the underlying conformational ensemble. A recent workaround involves analysing the DSC data with various theoretical models that are then ranked with a Bayesian probabilistic approach to obtain model independent estimates of the folding barrier height [155]. Reconstructing folding landscapes from multi-probe unfolding experiments Fast folding proteins exhibit marginal unfolding co-operativity that results in non-concerted structural disassembly [156]. The decoupling between structural elements can be resolved using multiple spectroscopic probes [129]. The folding landscape is then reconstructed interpreting all the spectroscopic data with statistical mechanical models that include the most relevant partially folded conformations. This approach, which was originally developed to identify one-state downhill folding [129,130], is, in principle, extensible to any fast-folding protein. Such extensibility has been demonstrated by various groups that applied it to reconstruct the folding landscape of villin headpiece subdomain [157], PDD [158], the de novo designed αtα [159], and the P22 subdomain [160]. In recent applications, the multi-probe data is combined with kinetic information [157] or enhanced by incorporating residue-specific infrared probes [159,160]. All of these studies have rendered free energy landscapes that are broad and with shallow barriers, but have also revealed differences that highlight distinctive mechanistic features. For example, comparative analysis between structural homologues (BBL compared with PDD [158] and αtα compared with P22 [160]) showed biases in the relative stability of different structural elements on proteins that share the overall fold. Folding interaction networks at atomic resolution The non-concerted unfolding behaviour of fast-folding proteins can be taken one step further to determine the network of interactions that stabilize the native structure [161]. Here NMR is used to measure the chemical shifts of every relevant atom in the protein at different degrees of unfolding, induced, for example, by increasing temperature [131,161]. The experiments produce hundreds of different chemical shift curves that report on the local changes in electronic environment of the corresponding atoms during unfolding. Such atomic unfolding curves are highly heterogeneous, but collectively represent the global unfolding process [131]. Because the similarities between unfolding curves from pairs of residues is related to their degree of structural coupling during unfolding, such information can be used to infer the folding interaction network [161]. This method was first implemented on the downhill folder BBL, which showed an extremely broad distribution of atomic unfolding behaviours [131]. Recently, it has been successfully extended to the fast folder gpW, demonstrating that is not exclusive of global downhill folding [162]. The gpW data were less heterogeneous and revealed a structurally delineated network of couplings between residues scattered throughout the sequence [162]. For gpW, long-timescale MD simulations at various temperatures were performed in parallel to reproduce the entire unfolding process. From the atomistic trajectories, it was possible to compute chemical shift unfolding curves and thus derive a simulated folding interaction network to be compared directly with the experimental one [162]. The possibility to compare experiments and simulations at this high level of detail is key for the interpretation of experimental data in mechanistic terms and for benchmarking and refining simulation methods. Structural analysis of excited states in protein folding RD-NMR (see above) has been widely used to detect minimally populated species associated with protein conformational changes taking place during catalysis, ligand binding or DNA sliding motions, which tend to occur in the sub-millisecond to millisecond timescale [54]. Likewise, NMR has been increasingly used to resolve the structure of folding intermediates on slow three-state folding proteins [51,163]. Some of these intermediates, like for the 71-residue four-helix bundle FF domain [164], form relatively fast (sub-microsecond), indicating a process that crosses a very small barrier. Chemical shift analysis indicated that such fast-forming intermediates have a compact structure with non-native interactions that need to break before the native state appears in a much slower step [165], thus being examples of kinetic traps [6]. When applied to fast-folding proteins, RD-NMR methods could provide detailed structural information of the conformations corresponding to the top of the folding barrier. In contrast with intermediates that accumulate before the rate-limiting step, the barrier top species determine the overall folding rate and hold the keys to the mechanism. The major challenge is the timescale of the process, which needs to be longer than 0.1 ms to be resolvable with these methods. Recently, a full-blown RD-NMR characterization of the ultrafast folder gpW has been achieved at very low temperature (1°C) to decrease folding kinetics down to ∼4000 s−1 [55]. Under these native conditions, the most populated non-native conformations are expected to be those that sit at the top of the shallow barrier that is found at the denaturation midpoint [55]. RD-NMR experiments revealed an exchange process with rate identical with the overall folding rate in which the excited state had a population of ∼10% and a structure with the two native helices in gpW formed and the β-hairpin unfolded [55]. These RD-NMR experiments are arguably the first example of high-resolution structural analysis of the conformations that determine the folding rate and mechanism of a protein. Interestingly, the structural properties of the barrier top derived by RD-NMR were in very close agreement with the experimental analysis of the folding interaction network and the long-timescale MD simulations obtained independently on the same protein [162] (see the previous section). Folding pathways and mechanisms In addition to obtaining structural information, it is also important to measure dynamic events such as folding transitions and microscopic pathways because they reveal the heterogeneity of mechanisms. For a barrier-crossing process, folding transition paths are the conformational excursions that take the protein over the barrier (Figure 5). Transitions occur very rarely, but are extremely fast. In fact, the typical time the molecule spends crossing the barrier is related to the pre-exponential factor of eqn 1. Therefore one would expect folding transition paths to take a few microseconds and be broadly distributed. Resolving individual transition paths requires methods that simultaneously reach single-molecule, sub-microsecond and atomic resolutions. Not only that, but also the observation times need to be sufficiently long to catch these rare events (Figure 5). Figure 5 Transition paths in protein folding The folding free energy landscape of a single-domain protein is often represented with a simple 1D surface with two minima for the native (N) and unfolded states (U), and a more or less pronounced barrier separating them (left). On this surface, individual molecules dwell on either minimum for most of the time since climbing the barrier is a probabilistically rare event. However, when it happens, the transitions across the barrier are fast because they are only limited by the conformational motions at the barrier top. This results in single-molecule trajectories that slowly alternate between U and N with very sharp transitions (right). The average dwell times on U and N are equivalent to the inverse of the folding and unfolding rate constants measured in bulk kinetic experiments respectively, whereas the sharp transitions correspond to the barrier-crossing paths. The first pass at estimating folding transition path times experimentally came once again from fast-folding proteins [110]. As implemented by Gruebele and co-workers, the idea was to look for evidence of an even ‘faster’ minor kinetic phase in T-jump experiments of proteins that were already near the downhill limit [110,112]. Such a process, termed the molecular phase, should correspond to the depopulation of the barrier top in response to the perturbation. The molecular phase has been observed in several fast-folding proteins and their mutants, including the helix bundle λ6–85 [110,166] and the WW domain FiP35 [167]. The timescale for this process (measured at relatively high temperatures of ∼60–70°C) is 1–2 μs, in line with other estimates of the folding speed limit. Eaton and co-workers have attempted to measure transition paths directly with single-molecule fluorescence methods to obtain estimates of the average folding transition path time for several proteins [168,169]. The time-resolution limitation was overcome by slowing down folding dynamics by addition of viscogens and analysing the single-molecule trajectories photon by photon [47]. The average transition path time of two natural proteins, one that folds fast and another one that folds slowly, was found to be very similar (between 2 and 10 μs) and consistent with previous estimates of the folding speed limit [168]. In contrast, the folding transition path time of the de novo designed α3D protein was much longer [169] even though this protein folds very fast [170] and close to the downhill limit [171]. Further analysis and comparison with MD simulations has revealed that the barrier crossing for this protein involves formation of off-register hydrogen bonds between the helices that need to break to proceed towards the native state, which increases internal friction and thus slows down the pre-exponential term [172]. Despite the impressive advances in experimental methods described above, atomistic MD simulations are possibly the only practical approach to resolve folding transitions of individual molecules with the time and structural resolution required to derive mechanistic information. Fast folding has stimulated the development and benchmarking of various approaches based on MD simulations [59,63,138,173–176]. Recently, Shaw et al. [62] used their Anton supercomputer to reach the near-microsecond simulation times required to watch fast-folding proteins fold and unfold multiple times and simulated 12 experimentally studied fast-folding proteins with diverse topologies [177]. The simulations folded most of these proteins into their native structure multiple times and with rates similar to those determined experimentally [177]. A key result was the confirmation that fast-folding proteins cross very small barriers, and that some of them truly fold in the one-state downhill fashion proposed by Muñoz and co-workers [130,139] (Figure 6). In the simulations, collapse and secondary structure occurred together to form a compact form in which a native-like topology was stabilized by a small subset of key long-range native contacts. Detailed analysis of the trajectories indicated that the productive folding pathways (order of structural events) are relatively homogeneous, although their heterogeneity increased for larger proteins, particularly those containing β-structures [177]. Figure 6 Long-timescale MD simulations of fast-folding proteins MD simulations on the proteins α3D (2A3D), WW domain (2F21), engrailed homedomain (HD, 2P6J) and BBL (2WXC). The upper panels show two individual MD trajectories for each protein revealing multiple folding and unfolding events. The lower panels show the 1D free energy surfaces derived from the MD simulations together with the superposition of the experimentally determined native structure (red) and the native structure identified by the simulations (blue). The free energy surfaces highlight that some fast-folding proteins cross very small free energy barriers (α3D and WW) and others fold downhill (HD and BBL), consistently with experimental interpretation. Figure derived with permission from Lindorff-Larsen et al. [177]. The possibility of simulating multiple folding–unfolding transitions in single trajectories offers very exciting possibilities to investigate folding mechanisms in detail. The simulations provide the extreme resolution that experiments could never achieve, but still use approximate force fields to describe protein energetics and dynamics. Increasingly sophisticated experiments, such as those described in the present review, provide the critical benchmarks for further refinement of simulations in a perfect symbiosis. First steps in this direction have been taken by combining experiments and MD simulations to design mutations that speed up folding of WW domains [125], to investigate the folding interaction network and mechanisms of the protein gpW [162], and to investigate the barrier-crossing process of the designed protein α3D [172]. BIOLOGICAL ROLES AND APPLICATIONS OF FAST FOLDING A related area of interest focuses on the structural and functional analysis of IDPs [178], especially after the realization that IDPs amount to a very large fraction of the proteome [179]. These proteins exhibit structural disorder in native conditions and folding coupled to binding via complex mechanisms [180]. Experimental studies have reported IDPs that bind to their partners through either induced fit or conformational selection mechanisms [181]. Some IDPs bind to multiple partners that are structurally diverse [182], a feature that allows them to moonlight [183] or produce sophisticated allosteric effects [184]. It turns out that IDPs and fast-folding proteins, especially one-state downhill folders, are closely interconnected. It has been noticed recently that folding rate, stability and co-operativity are intimately coupled so proteins that fold fast also unfold fast, are marginally stable and are minimally co-operative [156]. In fact, the stability of domains identified as downhill folders seems to be poised towards exhibiting partial disorder under physiological conditions. This trend has been observed by investigating homologous fast-folding domains from meso-, thermo- and hyper-thermophilic organisms in which the denaturation temperature of the domain tracks the living temperature of the organism [185]. So-called IDPs, on the other hand, have significant residual structure, as shown by NMR [186] and single-molecule fluorescence [187] experiments, and form stable native structures under slightly favourable thermodynamic conditions [188,189]. The ability of IDPs to be both partially disordered and poised to fold up with slight thermodynamic input seems to be a simple manifestation of their one-state downhill folding character [185]. Such folding characteristics enable their operation as conformational rheostats, that is molecular devices capable of producing analogical signals in response to binary stimuli such as binding to specific partners [130,156,185] (Figure 7, left). In this light, the functional complexity and multiple binding modes reported on IDPs could be explained as emerging from the coupling between binding and downhill folding. There is mounting evidence that the complex binding modes observed on IDPs involve gradual conformational changes rather than binary transitions. Figure 7 Biological and technological roles of conformational rheostats Left: a conformational rheostat is based on a downhill folding domain near its denaturation midpoint. The partially folded conformational ensemble of this domain can gradually become more or less structured by coupling folding to a signal such as binding to one or various partners. Right: the conformational rheostat concept has been used to implement folding coupled to binding biosensors that exhibit a much broader dynamic response than conventional conformational switches For example, NCBD (nuclear co-activator-binding domain) has been classified as an IDP [188,190] that binds multiple partners by folding into different conformers [188,191,192]. At the same time, NCBD is capable of folding into a three-helix bundle structure in the presence of stabilizing agents, and it does so following a gradual process (one-state downhill) according to the multivariate analysis of multi-probe experimental data and computer simulations [193]. Another interesting case is the PSBDs (peripheral subunit-binding domains) from several multienzymatic complexes, such as the pyruvate and 2-oxoglutarate dehydrogenases [194], which include the first identified examples of one-state downhill folding [130]. In these multienzymes, the catalytic process involves four steps catalysed by three subunits (E1, E2 and E3) that form a dynamic macromolecular complex, one that is fully controlled by the interactions between the PSBD from the E2 subunit and the E1 and E3 subunits [194]. E1 and E3 are structurally very different, yet the small PSBD (<50 residues) is capable of binding to both exclusively, with high affinity (Kd of 0.33 and 0.58 nM) and 1:1 stoichiometry [195]. Moreover, the crystallographic structures of E1 and E3 bound to the PSBD show very superficial binding interfaces and extremely high mobility in the PSBD region (B-factors up to 80 Å2; 1 Å=0.1 nm) [196,197], suggesting that PSBD is largely disordered when bound. Certain DNA-binding proteins, such as homeodomains, are also likely candidates for conformational rheostats. These domains face the enormous challenge of finding a short target sequence within the enormous pool of potential binding sites provided by genomic DNA. To solve this problem, they exhibit specific and non-specific DNA binding [198] that they aptly combine to slide (1D diffusion) and hop (2D diffusion) over DNA resulting in search speed-ups of at least ∼100-fold relative to a standard 3D diffusion-controlled processes [199]. DNA sliding has been studied theoretically [199], computationally [200] and experimentally using single-molecule methods [201] and paramagnetic relaxation enhancement NMR [202]. The molecular mechanism by which DNA-binding proteins implement these two binding modes remains largely unclear, however. But we now know that DNA-binding domains are marginally co-operative fast folders [155], exhibit partial disorder when unbound to DNA [203], and seem to fold via a downhill mechanism [171]. These properties suggest a molecular rheostat in which the conformational motions of a partially unfolded domain are exploited to counterbalance DNA processivity and sliding speed during non-specific binding, and ensure quick locking into the target sequence. These domains also bind non-specifically to DNA in a manner that seems to be DNA-sequence-dependent [198], which further suggests a homing-to-target mechanism mediated by conformational selection. Finally, conformational rheostats also offer very attractive possibilities for technological applications. A first effort in this direction has targeted the design of high-performance biosensors based on gradual conformational changes coupled to proton binding [204]. The authors of that work exploited the natural properties of the BBL domain in terms of folding [130] and proton binding [205] to engineer a pH ionic strength sensor with linear response over five orders of magnitude in analyte concentration, instead of the two orders that are inherent to conformational switches (Figure 7, right). Moreover, these sensors exhibited ultrafast response thanks to the microsecond folding kinetics of BBL and the gradual coupling between folding and binding [204]. FUNDING This work was supported by the Spanish Ministry of Economy and Competiveness [grant numbers CSD2009-00088 and BIO2011-28092] and the European Research Council [grant number ERC-2012-ADG-323059]. Abbreviations DSCdifferential scanning calorimetry IDPintrinsically disordered protein NCBDnuclear co-activator-binding domain PSBDperipheral subunit-binding domain RD-NMRrelaxation dispersion NMR ==== Refs 1 Anfinsen C.B. The formation and stabilization of protein structure Biochem. J. 1972 128 737 749 10.1042/bj1280737 4565129 2 Jackson S.E. How do small single-domain proteins fold? Folding Des 1998 3 R81 R91 10.1016/S1359-0278(98)00033-9 9710577 3 Onuchic J.N. Wolynes P.G. Theory of protein folding Curr. Opin. Struct. Biol. 2004 14 70 75 10.1016/j.sbi.2004.01.009 15102452 4 Plotkin S.S. Onuchic J.N. Understanding protein folding with energy landscape theory part II: quantitative aspects Q. Rev. Biophys. 2002 35 205 286 12599750 5 Chavez L.L. Onuchic J.N. Clementi C. Quantifying the roughness on the free energy landscape: entropic bottlenecks and protein folding rates J. Am. Chem. Soc. 2004 126 8426 8432 10.1021/ja049510+ 15237999 6 Bryngelson J.D. Onuchic J.N. Socci N.D. Wolynes P.G. Funnels, pathways and the energy landscape of protein folding: a synthesis Proteins 1995 21 167 195 10.1002/prot.340210302 7784423 7 Camacho C.J. Thirumalai D. Kinetics and thermodynamics of folding in model proteins Proc. Natl. Acad. Sci. U.S.A. 1993 90 6369 6372 10.1073/pnas.90.13.6369 8327519 8 Betancourt M. Onuchic J.N. Kinetics of protein-like models: the energy landscape factors that determine folding J. Chem. Phys. 1995 103 773 787 10.1063/1.470109 9 Dill K.A. Bromberg S. Yue K. Fiebig K.M. Yee D.P. Thomas P.D. Chan H.S. Principles of protein folding: a perspective from simple exact models Protein Sci. 1995 4 561 602 10.1002/pro.5560040401 7613459 10 Brooks C.L. Simulations of protein folding and unfolding Curr. Opin. Struct. Biol. 1998 8 222 226 10.1016/S0959-440X(98)80043-2 9631297 11 Knott M. Chan H.S. Criteria for downhill protein folding: calorimetry, chevron plot, kinetic relaxation, and single-molecule radius of gyration in chain models with subdued degrees of cooperativity Proteins 2006 65 373 391 10.1002/prot.21066 16909416 12 Jones C.M. Henry E.R. Hu Y. Chan C.K. Luck S.D. Bhuyan A. Roder H. Hofrichter J. Eaton W.A. Fast events in protein-folding initiated by nanosecond laser photolysis Proc. Natl. Acad. Sci. U.S.A. 1993 90 11860 11864 10.1073/pnas.90.24.11860 8265638 13 Pascher T. Chesick J.P. Winkler J.R. Gray H.B. Protein folding triggered by electron transfer Science 1996 271 1558 1560 10.1126/science.271.5255.1558 8599112 14 Bredenbeck J. Helbing J. Kumita J.R. Woolley G.A. Hamm P. α-Helix formation in a photoswitchable peptide tracked from picoseconds to microseconds by time-resolved IR spectroscopy Proc. Natl. Acad. Sci. U.S.A. 2005 102 2379 2384 10.1073/pnas.0406948102 15699340 15 Schrader T.E. Schreier W.J. Cordes T. Koller F.O. Babitzki G. Denschlag R. Renner C. Löweneck M. Dong S.-L. Moroder L. Light-triggered β-hairpin folding and unfolding Proc. Natl. Acad. Sci. U.S.A. 2007 104 15729 15734 10.1073/pnas.0707322104 17893334 16 Causgrove T.P. Dyer R.B. Nonequilibrium protein folding dynamics: laser-induced pH-jump studies of the helix–coil transition Chem. Phys. 2006 323 2 10 10.1016/j.chemphys.2005.08.032 17 Ballew R.M. Sabelko J. Reiner C. Gruebele M. A single-sweep, nanosecond time resolution laser temperature-jump apparatus Rev. Sci. Instrum. 1996 67 3694 3699 10.1063/1.1147137 18 Williams K. Causgrove T.P. Gilmanshin R. Fang k.S. Callender R.H. Woodruff W.H. Dyer R.B. Fast events in protein folding: helix melting and formation in a small peptide Biochemistry 1996 35 691 697 10.1021/bi952217p 8547249 19 Thompson P.A. Eaton W.A. Hofrichter J. Laser temperature jump study of the helixcoil kinetics of an alanine peptide interpreted with a ‘kinetic zipper’ model Biochemistry 1997 36 9200 9210 10.1021/bi9704764 9230053 20 Cooper A. Thermodynamic analysis of biomolecular interactions Curr. Opin. Chem. Biol. 1999 3 557 563 10.1016/S1367-5931(99)00008-3 10508661 21 Sadqi M. Lapidus L.J. Muñoz V. How fast is protein hydrophobic collapse? Proc. Natl. Acad. Sci. U.S.A. 2003 100 12117 12122 10.1073/pnas.2033863100 14530404 22 Dyer R.B. Gai F. Woodruff W.H. Infrared studies of fast events in protein folding Acc. Chem. Res. 1998 31 709 716 10.1021/ar970343a 23 Huang C.Y. Getahun Z. Wang T. DeGrado W.F. Gai F. Time-resolved infrared study of the helix–coil transition using C-13-labeled helical peptides J. Am. Chem. Soc. 2001 123 12111 12112 10.1021/ja016631q 11724630 24 Lednev I.K. Karnoup A.S. Sparrow M.C. Asher S.A. α-Helix peptide folding and unfolding activation barriers: a nanosecond UV resonance Raman study J. Am. Chem. Soc. 1999 121 8074 8086 10.1021/ja991382f 25 Brewer S.H. Song B.B. Raleigh D.P. Dyer R.B. Residue specific resolution of protein folding dynamics using isotope-edited infrared temperature jump spectroscopy Biochemistry 2007 46 3279 3285 10.1021/bi602372y 17305369 26 Nagarajan S. Taskent-Sezgin H. Parul D. Carrico I. Raleigh D.P. Dyer R.B. Differential ordering of the protein backbone and side chains during protein folding revealed by site-specific recombinant infrared probes J. Am. Chem. Soc. 2011 133 20335 20340 10.1021/ja2071362 22039909 27 Royer C.A. Why and how does pressure unfold proteins? Subcell. Biochem. 2015 72 59 71 10.1007/978-94-017-9918-8 26174377 28 Torrent J. Font J. Herberhold H. Marchal S. Ribó M. Ruan K. Winter R. Vilanova M. Lange R. The use of pressure-jump relaxation kinetics to study protein folding landscapes Biochim. Biophys. Acta 2006 1764 489 496 10.1016/j.bbapap.2006.01.002 16481228 29 Dumont C. Emilsson T. Gruebele M. Reaching the protein folding speed limit with large, sub-microsecond pressure jumps Nat. Methods 2009 6 515 519 10.1038/nmeth.1336 19483692 30 Roder H. Maki K. Cheng H. Shastry M.C.R. Rapid mixing methods for exploring the kinetics of protein folding Methods 2004 34 15 27 10.1016/j.ymeth.2004.03.003 15283912 31 Bilsel O. Kayatekin C. Wallace L.A. Matthews C.R. A microchannel solution mixer for studying microsecond protein folding reactions Rev. Sci. Instrum. 2005 76 014302 10.1063/1.1834698 32 Hertzog D.E. Michalet X. Jager M. Kong X. Santiago J.G. Weiss S. Bakajin O. Femtomole mixer for microsecond kinetic studies of protein folding Anal. Chem. 2004 76 7169 7178 10.1021/ac048661s 15595857 33 Neudecker P. Lundstrom P. Kay L.E. Relaxation dispersion NMR spectroscopy as a tool for detailed studies of protein folding Biophys. J. 2009 96 2045 2054 10.1016/j.bpj.2008.12.3907 19289032 34 Fernandez J.M. Li H. Force-clamp spectroscopy monitors the folding trajectory of a single protein Science 2004 303 1674 1678 10.1126/science.1092497 15017000 35 Chen H. Fu H. Zhu X. Cong P. Nakamura F. Yan J. Improved high-force magnetic tweezers for stretching and refolding of proteins ans short DNA Biophys. J. 2011 100 517 523 10.1016/j.bpj.2010.12.3700 21244848 36 Cecconi C. Shank E.A. Bustamante C. Marqusee S. Direct observation of the three-state folding of a single protein molecule Science 2005 309 2057 2060 10.1126/science.1116702 16179479 37 Gebhardt J.C. Bornschlogl T. Rief M. Full distance-resolved folding energy landscape of one single protein molecule Proc. Natl. Acad. Sci. U.S.A. 2010 107 2013 2018 10.1073/pnas.0909854107 20133846 38 Jagannathan B. Marqusee S. Protein folding and unfolding under force Biopolymers 2013 99 860 869 10.1002/bip.22321 23784721 39 Stigler J. Ziegler F. Gieseke A. Gebhardt J.C. Rief M. The complex folding network of single calmodulin molecules Science 2011 334 512 516 10.1126/science.1207598 22034433 40 Ritchie D.B. Woodside M.T. Probing the structural dynamics of proteins and nucleic acids with optical tweezers Curr. Opin. Struct. Biol. 2015 34 43 51 10.1016/j.sbi.2015.06.006 26189090 41 Michalet X. Weiss S. Jäger M. Single-molecule fluorescence studies of protein folding and conformational dynamics Chem. Rev. 2006 106 1785 1813 10.1021/cr0404343 16683755 42 Schuler B. Eaton W.A. Protein folding studied by single-molecule FRET Curr. Opin. Struct. Biol. 2008 18 16 26 10.1016/j.sbi.2007.12.003 18221865 43 Gopich I. Szabo A. Theory of photon statistics in single-molecule Förster resonance energy transfer J. Chem. Phys. 2005 122 14707 10.1063/1.1812746 15638691 44 Nie S. Zare R.N. Optical detection of single molecules Annu. Rev. Biophys. Biomol. Struct. 1997 26 567 596 10.1146/annurev.biophys.26.1.567 9241430 45 Joo C. Balci H. Ishitsuka Y. Buranachai C. Ha T. Advances in single-molecule fluorescence methods for molecular biology Annu. Rev. Biochem. 2008 77 51 76 10.1146/annurev.biochem.77.070606.101543 18412538 46 Campos L.A. Liu J. Wang X. Ramanathan R. English D.S. Muñoz V. A photoprotection strategy for microsecond-resolution single-molecule fluorescence spectroscopy Nat. Methods 2011 8 143 146 10.1038/nmeth.1553 21217750 47 Gopich I.V. Szabo A. Decoding the pattern of photon colors in single-molecule FRET J. Phys. Chem. B 2009 113 10965 10973 10.1021/jp903671p 19588948 48 Gopich I.V. Szabo A. Theory of the energy transfer efficiency and fluorescence lifetime distribution in single-molecule FRET Proc. Natl. Acad. Sci. U.S.A. 2012 109 7747 7752 10.1073/pnas.1205120109 22550169 49 Ramanathan R. Muñoz V. A method for extracting the free energy surface and conformational dynamics of fast-folding proteins from single molecule photon trajectories J. Phys. Chem. B 2015 119 7944 7956 10.1021/acs.jpcb.5b03176 25988351 50 Kay L.E. NMR studies of protein structure and dynamics J. Magn. Reson. 2005 173 193 207 10.1016/j.jmr.2004.11.021 15780912 51 Korzhnev D.M. Kay L.E. Probing invisible, low-populated states of protein molecules by relaxation dispersion NMR spectroscopy: an application to protein folding Acc. Chem. Res. 2008 41 442 451 10.1021/ar700189y 18275162 52 Mittermaier A. Kay L.E. New tools provide new insights in NMR studies of protein dynamics Science 2006 312 224 228 10.1126/science.1124964 16614210 53 Hansen D.F. Vallurupalli P. Lundstrom P. Neudecker P. Kay L.E. Probing chemical shifts of invisible states of proteins with relaxation dispersion NMR spectroscopy: how well can we do? J. Am. Chem. Soc. 2008 130 2667 2675 10.1021/ja078337p 18237174 54 Hansen A.L. Lundström P. Velyvis A. Kay L.E. Quantifying millisecond exchange dynamics in proteins by CPMG relaxation dispersion NMR using side-chain 1 H probes J. Am. Chem. Soc. 2012 134 3178 3189 10.1021/ja210711v 22300166 55 Sanchez-Medina C. Sekhar A. Vallurupalli P. Cerminara M. Muñoz V. Kay L.E. Probing the free energy landscape of the fast-folding gpW protein by relaxation dispersion NMR J. Am. Chem. Soc. 2014 136 7444 7451 10.1021/ja502705y 24805164 56 Eaton W.A. Muñoz V. Impact of atomistic molecular dynamics simulations on understanding how proteins fold: an experimentalist's perspective 2014 Madrid Roche Institute 57 Duan Y. Kollman P.A. Pathways to a protein folding intermediate observed in a 1-microsecond simulation in aqueous solution Science 1998 282 740 744 10.1126/science.282.5389.740 9784131 58 Best R.B. Atomistic molecular simulations of protein folding Curr. Opin. Struct. Biol. 2012 22 52 61 10.1016/j.sbi.2011.12.001 22257762 59 Snow C.D. Nguyen H. Pande V.S. Gruebele M. Absolute comparison of simulated and experimental protein-folding dynamics Nature 2002 420 102 106 10.1038/nature01160 12422224 60 Freddolino P.L. Liu F. Gruebele M. Schulten K. Ten-microsecond molecular dynamics simulation of a fast-folding WW domain Biophys. J. 2008 94 L75 L77 10.1529/biophysj.108.131565 18339748 61 Lindorff-Larsen K. Piana S. Palmo K. Maragakis P. Klepeis J.L. Dror R.O. Shaw D.E. Improved side-chain torsion potentials for the Amber ff99SB protein force field Proteins 2010 78 1950 1958 20408171 62 Shaw D.E. Dror R.O. Salmon J.K. Grossman J.P. Mackenzie K.M. Bank J.A. Young C. Deneroff M.M. Batson B. Bowers K.J. Millisecond-scale molecular dynamics simulations on Anton Proc. Conf. High Perform. Comput. Networking Storage Anal. 2009 article 39 63 Shaw D.E. Maragakis P. Lindorff-Larsen K. Piana S. Dror R.O. Eastwood M.P. Bank J.A. Jumper J.M. Salmon J.K. Shan Y. Atomic-level characterization of the structural dynamics of proteins Science 2010 330 341 346 10.1126/science.1187409 20947758 64 Piana S. Klepeis J.L. Shaw D.E. Assessing the accuracy of physical models used in protein-folding simulations: quantitative evidence from long molecular dynamics simulations Curr. Opin. Struct. Biol. 2014 24 98 105 10.1016/j.sbi.2013.12.006 24463371 65 Muñoz V. Conformational dynamics and ensembles in protein folding Annu. Rev. Biophys. Biomol. Struct. 2007 36 395 412 10.1146/annurev.biophys.36.040306.132608 17291180 66 Pitard E. Orland H. Dynamics of the swelling or collapse of a homopolymer Europhys. Lett. 1998 41 467 472 10.1209/epl/i1998-00175-8 67 Nettels D. Gopich I.V. Hoffmann A. Schuler B. Ultrafast dynamics of protein collapse from single-molecule photon statistics Proc. Natl. Acad. Sci. U.S.A. 2007 104 2655 2660 10.1073/pnas.0611093104 17301233 68 Ziv G. Haran G. Protein folding, protein collapse, and Tanford's transfer model: lessons from single-molecule FRET J. Am. Chem. Soc. 2009 131 2942 2947 10.1021/ja808305u 19239269 69 Dyson H.J. Wright P.E. Intrinsically unstructured proteins and their functions Nat. Rev. Mol. Cell Biol. 2005 6 197 208 10.1038/nrm1589 15738986 70 Teufel D.P. Johnson C.M. Lum J.K. Neuweiler H. Backbone-driven collapse in unfolded protein chains J. Mol. Biol. 2011 409 250 262 10.1016/j.jmb.2011.03.066 21497607 71 Haran G. How, when and why proteins collapse: the relation to folding Curr. Opin. Struct. Biol. 2012 22 14 20 10.1016/j.sbi.2011.10.005 22104965 72 Hagen S.J. Hofrichter J. Szabo A. Eaton W.A. Diffusion-limited contact formation in unfolded cytochrome c : estimating the maximum rate of protein folding Proc. Natl. Acad. Sci. U.S.A. 1996 93 11615 11617 10.1073/pnas.93.21.11615 8876184 73 Lapidus L.J. Eaton W.A. Hofrichter J. Measuring the rate of intramolecular contact formation in polypeptides Proc. Natl. Acad. Sci. U.S.A. 2000 97 7220 7225 10.1073/pnas.97.13.7220 10860987 74 Krieger F. Fierz B. Bieri O. Drewello M. Kiefhaber T. Dynamics of unfolded polypeptide chains as model for the earliest steps in protein folding J. Mol. Biol. 2003 332 265 274 10.1016/S0022-2836(03)00892-1 12946363 75 Ihalainen J.A. Bredenbeck J. Pfister R. Helbing J. Chi L. van Stokkum I.H.M. Woolley G.A. Hamm P. Folding and unfolding of a photoswitchable peptide from picoseconds to microseconds Proc. Natl. Acad. Sci. U.S.A. 2007 104 5383 5388 10.1073/pnas.0607748104 17372213 76 Du D. Bunagan M.R. Gai F. The effect of charge–charge interactions on the kinetics of α-helix formation Biophys. J. 2007 93 4076 4082 10.1529/biophysj.107.108548 17704172 77 Huang C.Y. Getahun Z. Zhu Y. Klemke J.W. DeGrado W.F. Gai F. Helix formation via conformation diffusion search Proc. Natl. Acad. Sci. U.S.A. 2002 99 2788 2793 10.1073/pnas.052700099 11867741 78 Doshi U. Muñoz V. The principles of α-helix formation: explaining complex kinetics with nucleation-elongation theory J. Phys. Chem. B 2004 108 8497 8506 10.1021/jp049896a 79 Fierz B. Reiner A. Kiefhaber T. Local conformational dynamics in α-helices measured by fast triplet transfer Proc. Natl. Acad. Sci. U.S.A. 2009 106 1057 1062 10.1073/pnas.0808581106 19131517 80 Muñoz V. Ramanathan R. Waltzing α-helices Proc. Natl. Acad. Sci. U.S.A. 2009 106 1299 1300 10.1073/pnas.0812577106 19174518 81 Serrano A.L. Tucker M.J. Gai F. Direct assessment of the α-helix nucleation time J. Phys. Chem. B 2011 115 7472 7478 10.1021/jp200628b 21568273 82 Muñoz V. Thompson P.A. Hofrichter J. Eaton W.A. Folding dynamics and mechanism of β-hairpin formation Nature 1997 390 196 199 10.1038/36626 9367160 83 Muñoz V. Henry E.R. Hofrichter J. Eaton W.A. A statistical mechanical model for β-hairpin kinetics Proc. Natl. Acad. Sci. U.S.A. 1998 95 5872 5879 10.1073/pnas.95.11.5872 9600886 84 Pande V.S. Rokhsar D.S. Molecular dynamics simulations of unfolding and refolding of a β-hairpin fragment of Protein G Proc. Natl. Acad. Sci. U.S.A. 1999 96 9062 9067 10.1073/pnas.96.16.9062 10430895 85 Dinner A.R. Lazaridis T. Karplus M. Understanding β-hairpin formation Proc. Natl. Acad. Sci. U.S.A. 1999 96 9068 9073 10.1073/pnas.96.16.9068 10430896 86 Klimov D.K. Thirumalai D. Mechanisms and kinetics of β-hairpin formation Proc. Natl. Acad. Sci. U.S.A. 2000 97 2544 2549 10.1073/pnas.97.6.2544 10716988 87 Zagrovic B. Sorin E.J. Pande V. β-hairpin folding simulations in atomistic detail using an implicit solvent model J. Mol. Biol. 2001 313 151 169 10.1006/jmbi.2001.5033 11601853 88 Zhou B. Berne B.J. Germain R. The free energy landscape for β hairpin folding in explicit water Proc. Natl. Acad. Sci. U.S.A. 2001 98 14931 14936 10.1073/pnas.201543998 11752441 89 Tsai J. Levitt M. Evidence of turn and salt bridge contributions to β hairpin stability: MD simulations of C-terminal fragment form the B1 domain of Protein G Biophys. Chem. 2002 101–102 187 201 10.1016/S0301-4622(02)00198-9 90 Zhou Y. Zhang C. Stell G. Wang J. Temperature dependence of the distribution of the first passage time: results from discontinuous molecular dynamics simulations of an all-atom model of the second β-hairpin fragment of Protein G J. Am. Chem. Soc. 2003 125 6300 6305 10.1021/ja029855x 12785863 91 Bolhuis P.G. Kinetic pathways of β-hairpin (un)folding in explicit solvent Biophys. J. 2005 88 50 61 10.1529/biophysj.104.048744 15516524 92 Yang S. Onuchic J.N. Garcia A.E. Levine H. Folding time predictions from all-atom replica exchange simulations J. Mol. Biol. 2007 372 756 763 10.1016/j.jmb.2007.07.010 17681536 93 Best R.B. Mittal J. Free-energy landscape of the GB1 hairpin in all-atom explicit solvent simulations with different force fields: similarities and differences Proteins 2011 79 1318 1328 10.1002/prot.22972 21322056 94 Dyer R.B. Maness S.J. Peterson E.S. Franzen S. Fesinmeyer R.M. Andersen N.H. The mechanism of β-hairpin formation Biochemistry 2004 43 11560 11566 10.1021/bi049177m 15350142 95 Chen R.P. Huang J.J. Chen H.L. Jan H. Velusamy M. Lee C.T. Fann W. Larsen R.W. Chan S.I. Measuring the refolding of β-sheets with different turn sequences on a nanosecond time scale Proc. Natl. Acad. Sci. U.S.A. 2004 101 7305 7310 10.1073/pnas.0304922101 15123838 96 Du D. Tucker M.J. Gai F. Understanding the mechanism of β-hairpin folding via φ-value analysis Biochemistry 2006 45 2668 2678 10.1021/bi052039s 16489760 97 Davis C.M. Xiao S. Raleigh D.P. Dyer R.B. Raising the speed limit for β-hairpin formation J. Am. Chem. Soc. 2012 134 14476 14482 10.1021/ja3046734 22873643 98 Chattopadhyay K. Elson E.L. Frieden C. The kinetics of conformational fluctuations in an unfolded protein measured by fluorescence methods Proc. Natl. Acad. Sci. U.S.A. 2005 102 2385 2389 10.1073/pnas.0500127102 15701687 99 Lapidus L.J. Yao S. McGarrity K.S. Hertzog D.E. Tubman E. Bakajin O. Protein hydrophobic collapse and early folding steps observed in a microfluidic mixer Biophys. J. 2007 93 218 224 10.1529/biophysj.106.103077 17416618 100 Naganathan A.N. Muñoz V. Scaling of folding times with protein size J. Am. Chem. Soc. 2005 127 480 481 10.1021/ja044449u 15643845 101 Kubelka J. Hofrichter J. Eaton W.A. The protein folding ‘speed limit’ Curr. Opin. Struct. Biol. 2004 14 76 88 10.1016/j.sbi.2004.01.013 15102453 102 Plotkin S.S. Onuchic J.N. Understanding protein folding with energy landscape theory part I: basic concepts Q. Rev. Biophys. 2002 35 111 167 12197302 103 Akmal A. Muñoz V. The nature of the free energy barriers to two-state folding Proteins 2004 57 142 152 10.1002/prot.20172 15326600 104 Li M.S. Klimov D.K. Thirumalai D. Dependence of folding rates on protein length J. Phys. Chem. B 2002 106 8302 8305 10.1021/jp025837q 105 Ivankov D.N. Garbuzynskiy S.O. Alm E. Plaxco K.W. Baker D. Finkelstein A.V. Contact order revisited: influence of protein size on the folding rate Protein Sci. 2003 12 2057 2062 10.1110/ps.0302503 12931003 106 Plaxco K.W. Simons K.T. Baker D. Contact order, transition state placement and the refolding rates of single domain proteins J. Mol. Biol. 1998 277 985 994 10.1006/jmbi.1998.1645 9545386 107 Muñoz V. Eaton W.A. A simple model for calculating the kinetics of protein folding from three-dimensional structures Proc. Natl. Acad. Sci. U.S.A. 1999 96 11311 11316 10.1073/pnas.96.20.11311 10500173 108 De Sancho D. Muñoz V. Protein folding rates and stability: how much is there beyond size? J. Am. Chem. Soc. 2009 131 2074 2075 10.1021/ja808843h 19170596 109 De Sancho D. Muñoz V. Integrated prediction of protein folding and unfolding rates from only size and structural class Phys. Chem. Chem. Phys. 2011 13 17030 17043 10.1039/c1cp20402e 21670826 110 Yang W.Y. Gruebele M. Folding at the speed limit Nature 2003 423 193 197 10.1038/nature01609 12736690 111 Eaton W.A. Commentary: searching for “downhill scenarios” in protein folding Proc. Natl. Acad. Sci. U.S.A. 1999 96 5897 5899 10.1073/pnas.96.11.5897 10339514 112 Gelman H. Gruebele M. Fast protein folding kinetics Q. Rev. Biophys. 2014 47 1469 8994 10.1017/S003358351400002X 113 Mayor U. Johnson C.M. Daggett V. Fersht A.R. Protein folding and unfolding in microseconds to nanoseconds by experiment and simulation Proc. Natl. Acad. Sci. U.S.A. 2000 97 13518 13522 10.1073/pnas.250473497 11087839 114 Kubelka J. Eaton W.A. Hofrichter J. Experimental tests of villin subdomain folding simulations J. Mol. Biol. 2003 329 625 630 10.1016/S0022-2836(03)00519-9 12787664 115 Zhu Y. Alonso D.O.V. Maki K. Huang C.Y. Lahr S.J. Daggett V. Roder H. DeGrado W.F. Gai F. Ultrafast folding of α3 D: a de novo designed three-helix bundle protein Proc. Natl. Acad. Sci. U.S.A. 2003 100 15486 15491 10.1073/pnas.2136623100 14671331 116 Zhu Y. Fu X. Wang T. Tamura A. Takada S. Saven J.G. Gai F. Guiding the search for a protein's maximum rate of folding Chem. Phys. 2004 307 99 109 10.1016/j.chemphys.2004.05.008 117 Li P. Oliva F.Y. Naganathan A.N. Muñoz V. Dynamics of one-state downhill protein folding Proc. Natl. Acad. Sci. U.S.A. 2009 106 103 108 10.1073/pnas.0802986106 19118204 118 Banachewicz W. Johnson C.M. Fersht A.R. Folding of the Pit1 homeodomain near the speed limit Proc. Natl. Acad. Sci. U.S.A. 2011 108 569 573 10.1073/pnas.1017832108 21187427 119 Crane J.C. Koepf E.K. Kelly J.W. Gruebele M. Mapping the transition state of the WW domain β-sheet J. Mol. Biol. 2000 298 283 292 10.1006/jmbi.2000.3665 10764597 120 Ferguson N. Johnson C.M. Macias M. Oschkinat H. Fersht A. Ultrafast folding of WW domains without structured aromatic clusters in the denatured state Proc. Natl. Acad. Sci. U.S.A. 2001 98 13002 13007 10.1073/pnas.221467198 11687613 121 Nguyen H. Jager M. Moretto A. Gruebele M. Kelly J.W. Tuning the free-energy landscape of a WW domain by temperature, mutation, and truncation Proc. Natl. Acad. Sci. U.S.A. 2003 100 3948 3953 10.1073/pnas.0538054100 12651955 122 Nguyen H. Jäger M. Kelly J.W. Gruebele M. Engineering a β-sheet protein toward the folding speed limit J. Phys. Chem. B 2005 109 15182 15186 10.1021/jp052373y 16852923 123 Fung A. Li P. Godoy-Ruiz R. Sanchez-Ruiz J.M. Muñoz V. Expanding the realm of ultrafast protein folding: gpW, a midsize natural single-domain with α+β topology that folds downhill J. Am. Chem. Soc. 2008 130 7489 7495 10.1021/ja801401a 18479088 124 Xu Y. Purkayastha P. Gai F. Nanosecond folding dynamics of a three-stranded β-sheet J. Am. Chem. Soc. 2006 128 15836 15842 10.1021/ja064865+ 17147395 125 Piana S. Sarkar K. Lindorff-Larsen K. Guo M. Gruebele M. Shaw D.E. Computational design and experimental testing of the fastest-folding β-sheet protein J. Mol. Biol. 2011 405 43 48 10.1016/j.jmb.2010.10.023 20974152 126 Cho S.S. Weinkam P. Wolynes P.G. Origin of barriers and barrierless folding in BBL Proc. Natl. Acad. Sci. U.S.A. 2008 105 118 123 10.1073/pnas.0709376104 18172203 127 Sabelko J. Ervin J. Gruebele M. Observation of strange kinetics in protein folding Proc. Natl. Acad. Sci. U.S.A. 1999 96 6031 6036 10.1073/pnas.96.11.6031 10339536 128 Leeson D.T. Gai F. Rodriguez H.M. Gregoret L.M. Dyer R.B. Protein folding and unfolding on a complex energy landscape Proc. Natl. Acad. Sci. U.S.A. 2000 97 2527 2532 10.1073/pnas.040580397 10681466 129 Muñoz V. Thermodynamics and kinetics of downhill protein folding investigated with a simple statistical mechanical model Int. J. Quant. Chem. 2002 90 1522 1528 10.1002/qua.10384 130 Garcia-Mira M.M. Sadqi M. Fischer N. Sanchez-Ruiz J.M. Muñoz V. Experimental identification of downhill protein folding Science 2002 298 2191 2195 10.1126/science.1077809 12481137 131 Sadqi M. Fushman D. Muñoz V. Atom-by-atom analysis of global downhill protein folding Nature 2006 442 317 321 10.1038/nature04859 16799571 132 Naganathan A.N. Li P. Perez-Jimenez R. Sanchez-Ruiz J.M. Muñoz V. Navigating the downhill protein folding regime via structural homologues J. Am. Chem. Soc. 2010 132 11183 11190 10.1021/ja103612q 20698685 133 Lin C.W. Culik R.M. Gai F. Using VIPT-jump to distinguish between different folding mechanisms: application to BBL and a Trpzip J. Am. Chem. Soc. 2013 135 7668 7673 10.1021/ja401473m 23642153 134 Yang W.Y. Gruebele M. Detection-dependent kinetics as a probe of folding landscape microstructure J. Am. Chem. Soc. 2004 126 7758 7759 10.1021/ja0493751 15212506 135 Ma H. Gruebele M. Kinetics are probe-dependent during downhill folding of an engineered λ6–85 protein Proc. Natl. Acad. Sci. U.S.A. 2005 102 2283 2287 10.1073/pnas.0409270102 15699334 136 DeCamp S.J. Naganathan A.N. Waldauer S.A. Bakajin O. Lapidus L.J. Direct observation of downhill folding of λ-repressor in a microfluidic mixer Biophys. J. 2009 97 1772 1777 10.1016/j.bpj.2009.07.003 19751683 137 Kubelka J. Chiu T.K. Davies D.R. Eaton W.A. Hofrichter J. Sub-microsecond protein folding J. Mol. Biol. 2006 359 546 553 10.1016/j.jmb.2006.03.034 16643946 138 Cellmer T. Buscaglia M. Henry E.R. Hofrichter J. Eaton W.A. Making connections between ultrafast protein folding kinetics and molecular dynamics simulations Proc. Natl. Acad. Sci. U.S.A. 2011 108 6103 6108 10.1073/pnas.1019552108 21441105 139 Liu J. Campos L.A. Cerminara M. Wang X. Ramanathan R. English D.S. Muñoz V. Exploring one-state downhill protein folding in single molecules Proc. Natl. Acad. Sci. U.S.A. 2012 109 179 184 10.1073/pnas.1111164109 22184219 140 Campos L.A. Liu J.W. Muñoz V. The importance of being quantitative Proc. Natl. Acad. Sci. U.S.A. 2009 106 E139 E139 10.1073/pnas.0912057107 20018769 141 Onuchic J.N. Luthey-Schulten Z. Wolynes P.G. Theory of protein folding: the energy landscape perspective Annu. Rev. Phys. Chem. 1997 48 545 600 10.1146/annurev.physchem.48.1.545 9348663 142 Fersht A.R. Matouschek A. Serrano L. The Folding of an enzyme. 1. Theory of protein engineering analysis of stability and pathway of protein folding J. Mol. Biol. 1992 224 771 782 10.1016/0022-2836(92)90561-W 1569556 143 Sanchez I.E. Kiefhaber T. Origin of unusual φ-values in protein folding: evidence against specific nucleation sites J. Mol. Biol. 2003 334 1077 1085 10.1016/j.jmb.2003.10.016 14643667 144 Fersht A.R. Sato S. φ-Value analysis and the nature of protein-folding transition states Proc. Natl. Acad. Sci. U.S.A. 2004 101 7976 7981 10.1073/pnas.0402684101 15150406 145 Raleigh D.P. Plaxco K.W. The protein folding transition state: what are φ-values really telling us? Protein Pept. Lett. 2005 12 117 122 10.2174/0929866053005809 15723637 146 De Los Rios M.A. Muralidhara B.K. Wildes D. Sosnick T.R. Marqusee S. Wittung-Stafshede P. Plaxco K.W. Ruczinski I. On the precision of experimentally determined protein folding rates and φ-values Protein Sci. 2006 15 553 563 10.1110/ps.051870506 16501226 147 Naganathan A.N. Muñoz V. Insights into protein folding mechanisms from large scale analysis of mutational effects Proc. Natl. Acad. Sci. U.S.A. 2010 107 8611 8616 10.1073/pnas.1000988107 20418505 148 Ibarra-Molero B. Naganathan A.N. Sanchez-Ruiz J.M. Muñoz V. Modern analysis of protein folding by differential scanning calorimetry Methods Enzymol. 2016 567 281 318 10.1016/bs.mie.2015.08.027 26794359 149 Muñoz V. Sanchez-Ruiz J.M. Exploring protein-folding ensembles: a variable-barrier model for the analysis of equilibrium unfolding experiments Proc. Natl. Acad. Sci. U.S.A. 2004 101 17646 17651 10.1073/pnas.0405829101 15591110 150 Naganathan A.N. Sanchez-Ruiz J.M. Muñoz V. Direct measurement of barrier heights in protein folding J. Am. Chem. Soc. 2005 127 17970 17971 10.1021/ja055996y 16366525 151 Naganathan A.N. Perez-Jimenez R. Sanchez-Ruiz J.M. Muñoz V. Robustness of downhill folding: guidelines for the analysis of equilibrium folding experiments on small proteins Biochemistry 2005 44 7435 7449 10.1021/bi050118y 15895987 152 Godoy-Ruiz R. Henry E.R. Kubelka J. Hofrichter J. Muñoz V. Sanchez-Ruiz J.M. Eaton W.A. Estimating free-energy barrier heights for an ultrafast folding protein from calorimetric and kinetic data J. Phys. Chem. B 2008 112 5938 5949 10.1021/jp0757715 18278894 153 Halskau O. Jr Perez-Jimenez R. Ibarra-Molero B. Underhaug J. Muñoz V. Martinez A. Sanchez-Ruiz J.M. Large-scale modulation of thermodynamic protein folding barriers linked to electrostatics Proc. Natl. Acad. Sci. U.S.A. 2008 105 8625 8630 10.1073/pnas.0709881105 18550823 154 Farber P. Darmawan H. Sprules T. Mittermaier M. Analyzing protein folding cooperativity by differential scanning calorimetry and NMR spectroscopy J. Am. Chem. Soc. 2010 132 6214 6222 10.1021/ja100815a 20377225 155 Naganathan A.N. Perez-Jimenez R. Muñoz V. Sanchez-Ruiz J.M. Estimation of protein folding free energy barriers from calorimetric data by multi-model Bayesian analysis Phys. Chem. Chem. Phys. 2011 13 17064 17076 10.1039/c1cp20156e 21769353 156 Muñoz V. Campos L.A. Sadqi M. Limited cooperativity in protein folding Curr. Opin. Struct. Biol. 2016 36 58 66 10.1016/j.sbi.2015.12.001 26845039 157 Kubelka J. Henry E.R. Cellmer T. Hofrichter J. Eaton W.A. Chemical, physical, and theoretical kinetics of an ultrafast folding protein Proc. Natl. Acad. Sci. U.S.A. 2008 105 18655 18662 10.1073/pnas.0808600105 19033473 158 Naganathan A.N. Muñoz V. Thermodynamics of downhill folding: multi-probe analysis of PDD, a protein that folds over a marginal free energy barrier J. Phys. Chem. B 2014 118 8982 8994 10.1021/jp504261g 24988372 159 Kubelka G.S. Kubelka J. Site-specific thermodynamic stability and unfolding of a de novo designed protein structural motif mapped by 13 C isotopically edited IR spectroscopy J. Am. Chem. Soc. 2014 136 6037 6048 10.1021/ja500918k 24684597 160 Lai J.K. Kubelka G.S. Kubelka J. Sequence, structure, and cooperativity in folding of elementary protein structural motifs Proc. Natl. Acad. Sci. U.S.A. 2015 112 9890 9895 10.1073/pnas.1506309112 26216963 161 Sborgi L. Verma A. Sadqi M. de Alba E. Muñoz V. Protein folding at atomic resolution: analysis of autonomously folding supersecondary structure motifs by nuclear magnetic resonance Methods Mol. Biol. 2013 932 205 218 10.1007/978-1-62703-065-6 22987355 162 Sborgi L. Verma A. Piana S. Lindorff-Larsen K. Cerminara M. Santiveri C.M. Shaw D.E. de Alba E. Muñoz V. Interaction networks in protein folding via atomic-resolution experiments and long-timescale molecular dynamics simulations J. Am. Chem. Soc. 2015 137 6506 6516 10.1021/jacs.5b02324 25924808 163 Korzhnev D.M. Salvatella X. Vendruscolo M. Di Nardo A.A. Davidson A.A. Dobson C.M. Kay L.E. Low-populated folding intermediates of Fyn SH3 characterized by relaxation dispersion NMR Nature 2005 430 586 590 10.1038/nature02655 15282609 164 Korzhnev D.M. Religa T.L. Lundstrom P. Fersht A.R. Kay L.E. The folding pathway of an FF domain: characterization of an on-pathway intermediate state under folding conditions by 15 N, 13 Cα and 13 C-methyl relaxation dispersion and 1 H/2 H-exchange NMR spectroscopy J. Mol. Biol. 2007 372 497 512 10.1016/j.jmb.2007.06.012 17689561 165 Korzhnev D.M. Vernon R.M. Religa T.L. Hansen A.L. Baker D. Fersht A.R. Kay L.E. Nonnative interactions in the FF domain folding pathway from an atomic resolution structure of a sparsely populated intermediate: an NMR relaxation dispersion study J. Am. Chem. Soc. 2011 133 10974 10982 10.1021/ja203686t 21639149 166 Liu F. Gruebele M. Tuning λ6–85 towards downhill folding at its melting temperature J. Mol. Biol. 2007 370 574 584 10.1016/j.jmb.2007.04.036 17532338 167 Liu F. Nakaema M. Gruebele M. The transition state transit time of WW domain folding is controlled by energy landscape roughness J. Chem. Phys. 2009 131 195101 10.1063/1.3262489 19929078 168 Chung H.S. McHale K. Louis J.M. Eaton W.A. Single-molecule fluorescence experiments determine protein folding transition path times Science 2012 335 981 984 10.1126/science.1215768 22363011 169 Chung H.S. Cellmer T. Louis J.M. Eaton W.A. Measuring ultrafast protein folding rates from photon-by-photon analysis of single molecule fluorescence trajectories Chem. Phys. 2013 422 229 237 10.1016/j.chemphys.2012.08.005 24443626 170 Wang T. Zhu Y.J. Gai F. Folding of a three-helix bundle at the folding speed limit J. Phys. Chem. B 2004 108 3694 3697 10.1021/jp049652q 171 Naganathan A.N. Doshi U. Muñoz V. Protein folding kinetics: barrier effects in chemical and thermal denaturation experiments J. Am. Chem. Soc. 2007 129 5673 5682 10.1021/ja0689740 17419630 172 Chung H.S. Piana-Agostinetti S. Shaw D.E. Eaton W.A. Structural origin of slow diffusion in protein folding Science 2015 349 1504 1510 10.1126/science.aab1369 26404828 173 Garcia A.E. Onuchic J.N. Folding a protein in a computer: an atomic description of the folding/unfolding of protein A Proc. Natl. Acad. Sci. U.S.A. 2003 100 13898 13903 10.1073/pnas.2335541100 14623983 174 Ensign D.L. Kasson P.M. Pande V.S. Heterogeneity even at the speed limit of folding: large-scale molecular dynamics study of a fast-folding variant of the villin headpiece J. Mol. Biol. 2007 374 806 816 10.1016/j.jmb.2007.09.069 17950314 175 Best R.B. Hummer G. Diffusion models of protein folding Phys. Chem. Chem. Phys. 2011 13 16902 16911 10.1039/c1cp21541h 21842082 176 Piana S. Lindorff-Larsen K. Shaw D.E. How robust are protein folding simulations with respect to force field parameterization? Biophys. J. 2011 100 L47 L49 10.1016/j.bpj.2011.03.051 21539772 177 Lindorff-Larsen K. Piana S. Dror R.O. Shaw D.E. How fast-folding proteins fold Science 2011 334 517 520 10.1126/science.1208351 22034434 178 Wright P.C. Dyson H.J. Intrinsically disordered proteins in cellular signalling and regulation Nat. Rev. Mol. Cell Biol. 2015 16 18 29 10.1038/nrm3920 25531225 179 Uversky V.N. Introduction to intrinsically disordered proteins Chem. Rev. 2014 114 6557 6560 10.1021/cr500288y 25004990 180 Berlow R.B. Dyson H.J. Wright P.E. Functional advantages of dynamic protein disorder FEBS Lett. 2015 589 2433 2440 10.1016/j.febslet.2015.06.003 26073260 181 Arai M. Sugase K. Dyson H.J. Wright P.E. Conformational propensities of intrinsically disordered proteins influence the mechanism of binding and folding Proc. Natl. Acad. Sci. U.S.A. 2015 112 9614 9619 10.1073/pnas.1512799112 26195786 182 Waters L. Yue B.G. Veverka V. Renshaw P. Bramham J. Matsuda S. Frenkiel T. Kelly G. Muskett F. Carr M. Heery D.M. Structural diversity in p160/CREB-binding protein coactivator complexes J. Biol. Chem. 2006 281 14787 14795 10.1074/jbc.M600237200 16540468 183 Jeffery C.J. Moonlighting proteins: old proteins learning new tricks Trends Genet. 2003 19 415 417 10.1016/S0168-9525(03)00167-7 12902157 184 Ferreon A.C. Ferreon J.C. Wright P.E. Deniz A.A. Modulation of allostery by protein intrinsic disorder Nature 2013 498 390 394 10.1038/nature12294 23783631 185 Naganathan A.N. Doshi U. Fung A. Sadqi M. Muñoz V. Dynamics, energetics, and structure in protein folding Biochemistry 2006 45 8466 8475 10.1021/bi060643c 16834320 186 Dyson H.J. Wright P.E. Unfolded proteins and protein folding studied by NMR Chem. Rev. 2004 104 3607 3622 10.1021/cr030403s 15303830 187 Lee T. Moran-Gutierrez C.R. Deniz A.A. Probing protein disorder and complexity at single-molecule resolution Semin. Cell Dev. Biol. 2015 37 26 34 10.1016/j.semcdb.2014.09.027 25305580 188 Kjaergaard M. Teilum K. Poulsen F.M. Conformational selection in the molten globule state of the nuclear coactivator binding domain of CBP Proc. Natl. Acad. Sci. U.S.A. 2010 107 12535 12540 10.1073/pnas.1001693107 20616042 189 Moosa M.M. Ferreon A.C. Deniz A.A. Forced folding of a disordered protein accesses an alternative folding landscape ChemPhysChem 2015 16 90 94 10.1002/cphc.201402661 25345588 190 Demarest S.J. Martinez-Yamout M. Chung J. Chen H. Xu W. Dyson H.J. Evans R.M. Wright P.E. Mutual synergistic folding in recruitment of CBP/p300 by p160 nuclear receptor coactivators Nature 2002 415 549 553 10.1038/415549a 11823864 191 Qin B.Y. Liu C. Srinath H. Lam S.S. Correia J.J. Derynck R. Lin K. Crystal structure of IRF-3 in complex with CBP Structure 2005 13 1269 1277 10.1016/j.str.2005.06.011 16154084 192 Ferreon J.C. Lee C.W. Arai M. Martinez-Yamout M.A. Dyson H.J. Wright P.E. Cooperative regulation of p53 by modulation of ternary complex formation with CBP/p300 and HDM2 Proc. Nat. Acad. Sci. U.S.A. 2009 106 6591 6596 10.1073/pnas.0811023106 193 Naganathan A.N. Orozco M. The native ensemble and folding of a protein molten-globule: functional consequence of downhill folding J. Am. Chem. Soc. 2011 133 12154 12161 10.1021/ja204053n 21732676 194 Perham R.N. Swinging arms and swingin domains in multifunctional enzymes: catalytic machines for multistep reactions Annu. Rev. Biochem. 2000 69 961 1004 10.1146/annurev.biochem.69.1.961 10966480 195 Lessard I.A. Perham R.N. Interaction of component enzymes with the peripheral subunit-binding domain of the pyruvate dehydrogenase multienzyme complex of Bacillus stearothermophilus : stoichiometry and specificity in self-assembly Biochem. J. 1995 306 727 733 10.1042/bj3060727 7702567 196 Mande S.S. Sarfaty S. Allen M.D. Perham R.N. Hol W.G. Protein–protein interactions in the pyruvate dehydrogenase multienzyme complex: dihydrolipoamide dehydrogenase complexed with the binding domain of dihydrolipoamide acetyltransferase Structure 1996 4 277 286 10.1016/S0969-2126(96)00032-9 8805537 197 Frank R.A. Pratap J.V. Pei X.Y. Perham R.N. Luisi B.F. The molecular origins of specificity in the assembly of a multienzyme complex Structure 2005 13 1119 1130 10.1016/j.str.2005.04.021 16084384 198 Ades S.E. Sauer R.T. Differential DNA-binding specificity of the engrailed homeodomain: the role of residue 50 Biochemistry 1994 33 9187 9194 10.1021/bi00197a022 8049221 199 Florescu A.M. Joyeux M. Comparison of kinetic and dynamical models of DNA–protein interaction and facilitated diffusion J. Phys. Chem. A 2010 114 9662 9672 10.1021/jp101151a 20394450 200 Givaty O. Levy Y. Protein sliding along DNA: dynamics and structural characterization J. Mol. Biol. 2009 385 1087 1097 10.1016/j.jmb.2008.11.016 19059266 201 Gorman J. Greene E.C. Visualizing one-dimensional diffusion of proteins along DNA Nat. Struct. Mol. Biol. 2008 15 768 774 10.1038/nsmb.1441 18679428 202 Iwahara J. Clore G.M. Detecting transient intermediates in macromolecular binding by paramagnetic NMR Nature 2006 440 1227 1230 10.1038/nature04673 16642002 203 Dyson H.J. Roles of intrinsic disorder in protein–nucleic acid interactions Mol. Biosyst. 2012 8 97 104 10.1039/C1MB05258F 21874205 204 Cerminara M. Desai T.M. Sadqi M. Muñoz V. Downhill protein folding modules as scaffolds for broad-range ultrafast biosensors J. Am. Chem. Soc. 2012 134 8010 8013 10.1021/ja301092z 22554075 205 Cerminara M. Campos L.A. Ramanathan R. Muñoz V. Slow proton transfer coupled to unfolding explains the puzzling results of single-molecule experiments on BBL, a paradigmatic downhill folding protein PLoS One 2013 8 e78044 10.1371/journal.pone.0078044 24205082

PMC005xxxxxx/PMC5003698.txt

==== Front Biochem JBiochem. JppbiochemjBJBiochemical Journal0264-60211470-8728Portland Press Ltd. BCJ2016055710.1042/BCJ20160557Research ArticlesResearch Article105147Phos-tag analysis of Rab10 phosphorylation by LRRK2: a powerful assay for assessing kinase function and inhibitors New method to assess LRRK2 Rab phosphorylationG. Ito and othersIto Genta *13Katsemonova Kristina *Tonelli Francesca *Lis Pawel *Baptista Marco A.S. †Shpiro Natalia *Duddy Graham ‡2Wilson Steve §Ho Philip Wing-Lok ║Ho Shu-Leong ║Reith Alastair D. ¶4Alessi Dario R. *1* MRC Protein Phosphorylation and Ubiquitylation Unit, School of Life Sciences, University of Dundee, Dundee DD1 5EH, U.K.† The Michael J. Fox Foundation for Parkinson's Research, Grand Central Station, P.O. Box 4777, New York, NY 10163, U.S.A.‡ Molecular Discovery Research, GlaxoSmithKline Pharmaceuticals R&D, New Frontiers Science Park, Harlow, Essex CM19 5AD, U.K.§ RD Platform Technology & Science, GlaxoSmithKline, U.K.║ Division of Neurology, Department of Medicine, University of Hong Kong, Hong Kong¶ Neurodegeneration Discovery Performance Unit, RD Neurosciences, GlaxoSmithKline Pharmaceuticals R&D, Stevenage, U.K.1 Correspondence may be addressed to either of these authors (email genta-ito@umin.ac.jp or d.r.alessi@dundee.ac.uk).2 Present address: The Wellcome Trust Sanger Institute, Hinxton, Cambridge, U.K. 3 Present address: Laboratory of Neuropathology and Neuroscience, Graduate School of Pharmaceutical Sciences, University of Tokyo, 7-3-1 Hongo Bunkyoku, Tokyo 113-0033, Japan. 4 Requests for LRRK2 G2019SGSK mice or LRRK2 D2017AGSK mice should be directed to alastair.d.reith@gsk.com 30 8 2016 1 9 2016 473 17 172671 2685 8 6 2016 1 7 2016 5 7 2016 © 2016 The Author(s)2016This is an open access article published by Portland Press Limited on behalf of the Biochemical Society and distributed under the Creative Commons Attribution Licence 4.0 (CC BY).Autosomal dominant mutations that activate the leucine-rich repeat kinase 2 (LRRK2) cause inherited Parkinson's disease. Recent work has revealed that LRRK2 directly phosphorylates a conserved threonine/serine residue in the effector-binding switch-II motif of a number of Rab GTPase proteins, including Rab10. Here we describe a facile and robust method to assess phosphorylation of endogenous Rab10 in mouse embryonic fibroblasts (MEFs), lung and spleen-derived B-cells, based on the ability of the Phos-tag reagent to retard the electrophoretic mobility of LRRK2-phosphorylated Rab10. We exploit this assay to show that phosphorylation of Rab10 is ablated in kinase-inactive LRRK2[D2017A] knockin MEFs and mouse lung, demonstrating that LRRK2 is the major Rab10 kinase in these cells/tissue. We also establish that the Phos-tag assay can be deployed to monitor the impact that activating LRRK2 pathogenic (G2019S and R1441G) knockin mutations have on stimulating Rab10 phosphorylation. We show that upon addition of LRRK2 inhibitors, Rab10 is dephosphorylated within 1–2 min, markedly more rapidly than the Ser935 and Ser1292 biomarker sites that require 40–80 min. Furthermore, we find that phosphorylation of Rab10 is suppressed in LRRK2[S910A+S935A] knockin MEFs indicating that phosphorylation of Ser910 and Ser935 and potentially 14-3-3 binding play a role in facilitating the phosphorylation of Rab10 by LRRK2 in vivo. The Rab Phos-tag assay has the potential to significantly aid with evaluating the effect that inhibitors, mutations and other factors have on the LRRK2 signalling pathway. Parkinson's diseaseprotein kinasesRab GTPasesignal transduction ==== Body INTRODUCTION Our knowledge of the origins of Parkinson's disease has been transformed by the identification of genes whose mutation in humans leads to Mendelian inherited disease [1,2]. One of these genes encodes the leucine-rich repeat kinase 2 (LRRK2) protein kinase where autosomal dominant mutations account for ∼1% of sporadic Parkinson's disease [3,4]. The most common LRRK2 mutation converts Gly2019 into a serine within the kinase domain magnesium ion-binding motif [5]. This mutation enhances in vitro protein kinase activity ∼3-fold [6,7], indicating that abnormal increase in the kinase activity of LRRK2 is involved in the pathogenesis of Parkinson's disease, suggesting that LRRK2 kinase inhibitors have therapeutic benefit for the treatment of Parkinson's disease. LRRK2 is a large enzyme (2527 residues), consisting of leucine-rich repeats (residues 1010–1287), a GTPase domain (residues 1335–1504), a COR [C-terminal of ROC (Ras of complex GTPase domain)] domain (residues 1517–1843), a serine/threonine protein kinase domain (residues 1875–2132) and a WD40 repeat (residues 2231–2276) [8]. Three well-characterized pathogenic mutations occur within the GTPase domain (R1441C, R1441G and R1441H) [9,10] and one within the COR domain (Y1699C) [11]. Unlike the G2019S mutation, the R1441G/H/C and Y1699C mutations do not directly enhance LRRK2 in vitro kinase activity [12]. We recently reported that members of the Rab GTPase family, including Rab8A and Rab10 were direct physiological substrates for LRRK2 [13]. The LRRK2 phosphorylation site (Thr72 for Rab8A and Thr73 for Rab10) is conserved in ∼50 different Rab proteins [13], and lies within the effector-binding switch-II motif [14–16]. LRRK2 phosphorylation of Rab8A and Rab10 proteins is inhibitory as it suppresses binding to the Rab GDP-dissociation inhibitor (GDI) factors that are required for membrane delivery and recycling [13]. Furthermore, LRRK2 phosphorylation also inhibits binding of Rab8A to Rabin-8, its guanine-nucleotide-exchange factor (GEF) activator [13]. Other work has also linked Rab GTPases with Parkinson's disease. For example, Rab7L1 (also known as Rab29) is one of five genes that is mutated with Parkinson's disease patients that have the PARK16 mutation [17,18]. Depletion of Rab7L1 reportedly induced loss of dopaminergic neurons, similar to that observed with LRRK2-[G2019S] expression [19]. Furthermore, genetic analysis has recently revealed that loss of function mutations in the poorly studied Rab39B protein is responsible for a rare form of X-linked Parkinson's disease [20,21]. Moreover, overexpression of Rab8a, Rab1 and Rab3a protein attenuated α-synuclein-induced cytotoxicity in cellular and animal models of Parkinson's disease [22,23]. Finally, another protein kinase mutated in Parkinson's disease termed PINK1, indirectly controls the phosphorylation of a small group of Rabs including Rab8A at a site distinct from that used by LRRK2 (Ser111 on Rab8A) [24]. Taken together these results strongly suggest a functional interplay between Rab GTPases and known Parkinson's disease factors. In 2004, an agent (1,3-bis[bis(pyridin-2-ylmethyl) amino]propan-2-olato dizinc(II) complex) commonly referred to as ‘Phos-tag’ was described that binds to phosphate ions with much higher affinity (Kd ∼ 25 nM for phenyl phosphate) than other anions [25]. The Phos-tag reagent was subsequently shown to interact with high affinity with proteins as well as peptides phosphorylated on serine, threonine and tyrosine residues [26]. A modified version of the Phos-tag reagent termed ‘Phos-tag acrylamide’ (N-(5-(2-acryloylaminoethylcarbamoyl)pyridin-2-ylmethyl)-N,N′,N′-tris(pyridin-2-yl-methyl)-1,3-diaminopropan-2-ol) was developed that when polymerized into SDS/polyacrylamide gels retarded electrophoretic mobility of phosphorylated proteins, resulting in substantial mobility shifts [27]. The Phos-tag approach is particularly suited for analysing phosphorylation of relatively small proteins such as Rab protein that are phosphorylated at a single site. We previously observed that in a human embryonic kidney (HEK)-293 cell overexpression system, LRRK2 phosphorylation of haemagglutinin (HA)–Rab8A and HA–Rab10 resulted in an electrophoretic mobility shift of the phosphorylated Rab protein [13]. We also observed that pathogenic LRRK2 mutations tested, including the R1441G, Y1699C and G2019S, stimulated phosphorylation of Rab protein to a greater extent than wild-type (WT) LRRK2 [13]. An important goal is to develop robust methods to rapidly assess LRRK2 phosphorylation of endogenous Rab proteins in samples where sample material may be limiting. In the present study we develop a straightforward procedure based on the Phos-tag approach to quantitatively assess phosphorylation of endogenous Rab10 in mouse embryonic fibroblasts (MEFs), lung tissue as well as spleen-derived B-cells. We demonstrate that ablation of LRRK2 catalytic activity in a novel kinase-inactive LRRK2[D2017A] knockin mouse model blocks Rab10 phosphorylation in MEFs as well as lung, demonstrating that LRRK2 is indeed the major Rab10 kinase in these cells and tissue. We establish that the Phos-tag assay can be used to monitor the impact of LRRK2 inhibitors, as well as pathogenic knockin mutations (G2019S and R1441G) on Rab10 phosphorylation. There is also significant interest in studying the roles that LRRK2 Ser910 and Ser935 phosphorylation play, as phosphorylation of these residues promotes 14-3-3 binding and LRRK2 inhibitors induce their dephosphorylation [12,28]. To address whether Ser910 and Ser935 play a role in regulating Rab10 phosphorylation in vivo, we generated LRRK2[S910A+S935A] knockin MEFs and found that this mutation significantly inhibits Rab10 phosphorylation. The Rab10 Phos-tag assay will aid assessment of the impact that inhibitors, mutations and other factors have on the LRRK2 signalling pathway. MATERIALS AND METHODS Reagents GSK2578215A was obtained from GlaxoSmithKline [29]. HG-10-102-01 was custom synthesized as described previously [30]. MLi-2 was obtained from Merck and also synthesized as described in [31a]. Phos-tag acrylamide was synthesized as described in [31b]. Phos-tag acrylamide was stored at 5 mM aqueous solution (3.43 mg of compound in 1 ml of solution) at 4°C in black tubes that block out light as Phos-tag acrylamide is light-sensitive. HPLC analysis of stock Phos-tag acrylamide was undertaken every 4–5 weeks to ensure stock had not started to polymerize. All recombinant proteins, DNA constructs and antibodies generated for the present study can be requested via our reagents website (https://mrcppureagents.dundee.ac.uk/). General methods DNA procedures were undertaken using standard protocols. DNA constructs were purified from E. coli DH5α using a Maxi Prep kit (Qiagen). DNA sequence of the DNA constructs used in the present study was performed by our Sequencing Service (http://www.dnaseq.co.uk). Antibodies Anti-Rab10 antibody was from Cell Signaling Technology (#8127) and used at 1:1000 dilution. Rabbit monoclonal antibodies for total LRRK2 (UDD3) and pS935-LRRK2 (UDD2) were purified at the University of Dundee and used at 1:10000 and 1:2000 dilutions respectively. Rabbit monoclonal antibody detecting phospho-Ser1292 LRRK2 was from Abcam (ab203181) and used at a final concentration of 1 μg/ml. Anti-glyceraldehyde-3-phosphate dehydrogenase (GAPDH) antibody was from Santa Cruz Biotechnology (sc-32233) and used at 1:5000 dilution. Sheep polyclonal antibody for phospho-Thr73 Rab10 (S873D) was described previously [13] and used at final concentration of 1 μg/ml in the presence of 10 μg/ml non-phosphorylated peptide. Horseradish peroxidase-conjugated anti-mouse (#31450), -rabbit (#31460), -rat (#31470) and -sheep IgG secondary antibodies (#31480) were from Thermo Fisher Scientific. Plasmids The following constructs were used for protein production: 6His-SUMO-Rab10 WT (DU51062), 6His-SUMO-Rab8A WT (DU47363). The following constructs were used for overexpression in cells: HA–Rab10 WT/T73A (DU44250/DU51006), FLAG–LRRK2 R1441G (DU13077). The following constructs were used for generation of Rab10 knockout (KO) A549 cells: Rab10 KO N-terminal antisense guide and Cas9 D10A (DU52110) and Rab10 KO N-terminal sense guide (DU52100). Full datasheets for each plasmid are available from https://mrcppureagents.dundee.ac.uk/. Mice All animal studies were ethically reviewed and carried out in accordance with Animals (Scientific Procedures) Act 1986, the GSK Policy on the Care, Welfare and Treatment of Animals, regulations set by the University of Dundee and the U.K. Home Office. Animal studies and breeding were approved by the University of Dundee ethical committee and performed under a U.K. Home Office project licence and maintained under specific pathogen-free conditions at the University of Dundee. Animals (unless otherwise stated) were multiply housed at an ambient temperature (20–24°C) and humidity (45–55%) maintained on a 12 h light/12 h dark cycle, with free access to food (SDS RM No. 3 autoclavable) and water. The LRRK2[G2019S]GSK knockin mice, the LRRK2[A2016T] knockin mice and the LRRK2[R1441G] knockin mice were described previously [13,32]. The LRRK2 KO mice were generated and provided by Dr Huaibin Cai (National Institutes of Health, Bethesda, MD, U.S.A.) and have been described previously [33]. For experiments shown in Figures 5(B) and 7, littermate matched WT and LRRK2 knockin mice (3–6 months of age) were injected subcutaneously either with vehicle [40% (w/v) (2-hydroxypropyl)-β-cyclodextrin (Sigma–Aldrich)] or MLi-2 dissolved in vehicle at the indicated dose and killed by cervical dislocation 1 h after treatment. Lung was rapidly isolated and snap frozen in liquid nitrogen. No specific randomization method or blinding was applied to experiments. Generation of LRRK2[D2017A] knockin mice The LRRK2[D2017A] knockin mouse line was generated by a targeting strategy devised to introduce the point mutation D2017A into exon 41 of the LRRK2 gene by homologous recombination in mouse embryonic stem (ES) cells. 5′ and 3′ homology arms (approximately 4.8 and 3.8 kb respectively) flanking exon 41 were generated using Phusion High-Fidelity DNA Polymerase (New England Biolabs) on a C57BL/6J genomic DNA template. Similarly, a 739 bp fragment carrying exon 41 lying between these two homology arms was isolated and subjected to site-directed mutagenesis with the QuikChangeII site-directed mutagenesis kit (Stratagene) to introduce the appropriate point mutation (A to C mutation at bp 102 of exon 41). The 5′ and 3′ homology arms and the mutated exon 41 fragments were subcloned into a parental targeting vector to achieve the positioning of the loxP and FRT sites and PGKneo cassette. Gene targeting was performed in de novo generated hybrid C57BL/6J;129Ola-derived ES cells. The targeting construct was linearized and electroporated into ES cells according to standard methods. ES cells correctly targeted at the 3′ end was identified by Southern blot analysis of EcoRV digested genomic DNA using a PCR-derived external probe. Correct gene targeting at the 5′ end and presence of the point mutation was confirmed by sequencing of a ∼6 kb PCR product. The latter was generated by high-fidelity PCR of ES cell clone-derived genomic DNA using primers spanning the 5′ homology arm. Correctly targeted ES cell clones were injected into BALB/c blastocysts and implanted into foster mothers according to standard procedures. Male chimaeras resulting from the D2017A-targeted ES cells were bred with C57BL/6J female mice expressing CRE recombinase from the ROSA26 locus to facilitate removal of the loxP flanked PGKneo cassette in vivo, and germline transmission of the targeted allele was confirmed by PCR. Germline mice were back-crossed once to C57BL/6J mice, and confirmed to be >98% C57BL/6J by single nucleotide polymorphism (SNP) analysis. The line was subsequently maintained by breeding with C57BL/6J, and crossing mice heterozygous for the point mutation generated homozygous mice at the expected Mendelian ratio. Separate colonies of WT and homozygous animals were subsequently maintained to minimize breeding wastage. Standard genotyping which distinguishes WT from point mutation knockin alleles was used throughout. Genotyping of mice was performed by PCR using genomic DNA isolated from ear biopsies. Generation of LRRK2[S910A+S935A] knockin mice The constitutive LRRK2[S910A+S935A] knockin mouse line was produced by implementing a targeting strategy based on NCBI transcript NM_025730.3, to introduce two point mutations S910A and S935A into exon 21 of the LRRK2 gene by homologous recombination in mouse ES cells (TaconicArtemis). To start with, the S910A and S935A mutations have been introduced into exon 21 by site-directed mutagenesis with the QuikChangeII site-directed mutagenesis kit (Stratagene) (S910A: TCA to GCC and S935A: TCG to GCG of exon 21). The positive selection marker PuroR has been flanked by FRT sites and inserted into intron 21. 5′ and 3′ homology arms (approximately 4.1 and 6 kb respectively) flanking exon 21 were generated using Phusion High-Fidelity DNA Polymerase (New England Biolabs) on a C57BL/6J genomic DNA template. The 5′ and 3′ homology arms comprising mutated exon 21 were subcloned into a parental targeting vector to achieve the positioning of the loxP and FRT sites and PGKneo cassette. For this purpose, the targeting vector was generated using bacterial artificial chromosome (BAC) clones from the C57BL/6J RPCIB-731 BAC library which then were transfected into the TaconicArtemis C57BL/6N Tac ES cell line. Homologous recombinant clones were selected using positive (PuroR) and negative (thymidine kinase–Tk) selection. The constitutive knockin allele comprising desired mutations was obtained after Flp-mediated removal of the selection marker. The targeting construct was linearized and electroporated into ES cells according to standard methods. Successful gene targeting of ES cells at the 5′ and 3′ ends was confirmed by sequencing of a ∼6 kb PCR product. Properly targeted ES cell clones were then subjected to the diploid injection into BALB/c blastocysts and implanted into foster mothers according to standard procedures. Male chimaeras resulting from the LRRK2[S910A+S935A]-targeted ES cells were bred with C57BL/6J female mice expressing Cre recombinase from the ROSA26 locus to facilitate removal of the loxP flanked PGKneo cassette in vivo, and germline transmission was identified by the presence of black, strain C57BL/6, offspring (G1) and PCR. Genotyping of mice For LRRK2[D2017A] knockin mice, primers 5′-CCGAG-CCAAAAACTAAGCTC-3′ and 5′-CCATCTTGGGTACTT-GACC-3′ were used to detect the WT and knockin alleles (WT, 400 bp; knockin, 550 bp; heteroduplex formation). For LRRK2[S910A+S935A] knockin mice, primers 5′-GTG-CTTGAAGTTTGATCATAATGC-3′ and 5′-GCATATAGCA-TGTAGTGTCATCTCC-3′ were used to detect the WT and knockin alleles (WT, 326 bp; knockin, 401 bp; heteroduplex formation). The PCR programme consisted of 5 min at 95°C, then 35 cycles of 30 s at 95°C, 30 s at 60°C and 30 s at 72°C, and then 5 min at 72°C. DNA sequencing was used to confirm the knockin mutation and performed by DNA Sequencing & Services (MRC-PPU; http://www.dnaseq.co.uk/) using Applied Biosystems Big-Dye version 3.1 chemistry on Applied Biosystems model 3730 automated capillary DNA sequencer. Generation and culture of MEFs Littermate matched WT and homozygous LRRK2[S910A+S935A] or homozygous LRRK2[R1441G] knockin MEFs were isolated from mouse embryos at embryonic day (E)12.5 resulting from crosses between heterozygous LRRK2[S910A+S935A]/WT or LRRK2[R1441G]/WT mice using a previously described protocol [34]. LRRK2[S910A+S935A] cells were genotyped as described above and LRRK2[R1441G] cells were genotyped as described previously [32]. Homozygous LRRK2[S910A+S935A] knockin as well as the WT cells generated from the same littermate were spontaneously immortalized by prolonged passaging in parallel for at least 20 passages before being used for Phos-tag experiments. Genotype of these cells was also confirmed by immunoblot analysis with anti-phospho-Ser910 and -Ser935 antibodies (Figure 8A). Homozygous LRRK2[R1441G] knockin MEFs used for the experiment shown in Figure 6(A) were used on passage 5. Littermate matched WT and homozygous LRRK2[D2017A] knockin MEFs were isolated by Dr Francisco Inesta-Vaquera (University of Dundee) from mouse embryos at E12.5 resulting from crosses between heterozygous LRRK2[D2017A]/WT mice using a previously described protocol [34]. Cells were genotyped as described above for mice, and WT and homozygous LRRK2[D2017A] knockin cells generated from the same littermate were selected for subsequent experiments. Cells were continuously passaged in parallel for at least 20 passages before being used for Phos-tag experiments. An identical approach was used to generate littermate WT and LRRK2[S910A+S935A] and LRRK2[R1441G] knockin MEFs. Littermate matched WT and homozygous LRRK2[G2019S]GSK MEFs, littermate matched WT and homozygous LRRK2[A2016T] MEFs and littermate matched WT and homozygous LRRK2 KO MEFs were isolated as described previously and used at over passage 20 [13,35]. All MEFs were cultured in Dulbecco's modified Eagle's medium (DMEM) containing 10% FBS, 2 mM L-glutamine, 100 units/ml penicillin, 100 μg/ml streptomycin, non-essential amino acids (Life Technologies) and 1 mM sodium pyruvate (Life Technologies). All knockin and KO cell lines were verified by allelic sequencing. Mouse tissue lysate preparation Frozen mouse tissues were quickly defrosted in the ice-cold lysis buffer containing 50 mM Tris/HCl, pH 7.5, 1% (v/v) Triton X-100, 1 mM EGTA, 1 mM sodium orthovanadate, 50 mM NaF, 0.1% (v/v) 2-mercaptoethanol, 10 mM 2-glycerophosphate, 5 mM sodium pyrophosphate, 0.1 μg/ml mycrocystin-LR (Enzo Life Sciences), 270 mM sucrose and Complete EDTA-free protease inhibitor cocktail (Roche) and homogenized using a POLYTRON homogenizer (KINEMATICA) on ice (5 s homogenization, 10 s interval and 5 s homogenization). Lysates were centrifuged at 20800 g for 30 min at 4°C and supernatants were used for Bradford assay and immunoblot analysis. Generation of Rab10 KO A549 cells A549 cells at ∼80% confluency were co-transfected in a six-well plate with DU52110 and DU52100 plasmids using Lipofectamine LTX (Life Technologies) according to the manufacturer's instructions, with a final amount of 9 μl of Lipofectamine LTX and 2.5 μg of DNA per well. The cells were then incubated for 24 h in DMEM supplemented with 10% FBS, 2 mM L-glutamine, 100 units/ml penicillin and 100 μg/ml streptomycin. The medium was then replaced with fresh medium supplemented with 2 μg/ml puromycin. After 24 h of puromycin selection the medium was replaced again with fresh medium without puromycin and the cells were left to recover for 48 h before performing single-cell sorting. Cell sorting was performed using influx cell sorter (Becton Dickinson). Single cells were placed in individual wells of a 96-well plate containing DMEM supplemented with 10% FBS, 2 mM L-glutamine, 100 units/ml penicillin, 100 μg/ml streptomycin and 100 μg/ml Normocin (InvivoGen). After reaching ∼80% confluency individual clones were transferred into six-well plates. After reaching ∼80% confluency the clones were screened by Western blotting for the presence of Rab10. Selected clones lacking expression of Rab10 were sequenced to confirm the KO. Genomic DNA was isolated using a GenElute Mammalian Genomic DNA Miniprep Kit (Sigma–Aldrich). The PCR was performed using PfuUltra High-Fidelity DNA Polymerase (Agilent Technologies) with primers 5′-TTCCTCAAAGCTGTTCGTAGGTCG-3′ and 5′-TCCTCCCACAGGTCTTACCTATGG-3′ to amplify the region targeted for KO, followed by incubation with Taq polymerase (New England Biolabs) to add 3′ A overhangs. The PCR products were then cloned into pSC-A-amp/kan vector using StrataClone PCR Cloning Kit (Agilent Technologies). For each cloning reaction 20 positive bacterial colonies were selected and the plasmids were isolated using QIAprep Spin Miniprep Kit (Qiagen). The inserts in each individual clone were sequenced using M13 primers (DNA sequencing facility of Division of Signal Transduction Therapy at the University of Dundee). This procedure allowed us to confirm that there were no WT alleles of the Rab10 gene present in the genome of selected clone thus confirming a successful KO. Cell culture, transfection, treatments and lysis HEK-293 and A549 cells were maintained in DMEM containing 10% (v/v) FBS, 2 mM L-glutamine, 100 units/ml penicillin and 100 μg/ml streptomycin at 37°C in a humidified atmosphere with 5% CO2. HEK-293 cells were seeded into six-well plates at 3×105 cells/well, and after 24 h culture cells were transfected with Lipofectamine 2000 (Life Technologies) using 0.5 μg of the Rab10 plasmid, 2 μg of the LRRK2 plasmid and 6 μl of Lipofectamine 2000 according to the manufacturer's protocol. Cells were lysed 24 h after transfection. Inhibitors were dissolved in DMSO. An equivalent volume of DMSO was added to negative control samples. Following treatment, cells were washed with TBS (20 mM Tris/HCl, pH 7.5, and 150 mM NaCl) on ice and lysed in an ice-cold lysis buffer containing 50 mM Tris/HCl, pH 7.5, 1% (v/v) Triton X-100, 1 mM EGTA, 1 mM sodium orthovanadate, 50 mM NaF, 0.1% (v/v) 2-mercaptoethanol, 10 mM 2-glycerophosphate, 5 mM sodium pyrophosphate, 0.1 μg/ml mycrocystin-LR (Enzo Life Sciences), 270 mM sucrose and Complete EDTA-free protease inhibitor cocktail (Roche). Lysates were centrifuged at 20800 g for 15 min at 4°C and supernatants were used for Bradford assay (Thermo Scientific) and immunoblot analysis. Phos-tag SDS/PAGE and immunoblot analysis Cell/tissue lysates were mixed with 4× SDS/PAGE sample buffer [250 mM Tris/HCl, pH 6.8, 8% (w/v) SDS, 40% (v/v) glycerol, 0.02% (w/v) Bromophenol Blue and 4% (v/v) 2-mercaptoethanol] and heated at 95°C for 5 min. For normal SDS/PAGE, 10–20 μg samples were loaded on to NuPAGE Bis-Tris 4–12% gels (Life Technologies) and electrophoresed at 150 V. For Phos-tag SDS/PAGE, samples were supplemented with 10 mM MnCl2 before loading gels. Phos-tag SDS/PAGE was carried out essentially as described previously [27]. Gels for Phos-tag SDS/PAGE consisted of a stacking gel [4% (w/v) acrylamide, 125 mM Tris/HCl, pH 6.8, 0.1% (w/v) SDS, 0.2% (v/v) N,N,N′,N′_tetramethylethylenediamine (TEMED) and 0.08% (w/v) ammonium persulfate (APS)] and a separating gel [12% (w/v) acrylamide, 375 mM Tris/HCl, pH 8.8, 0.1% (w/v) SDS, 75 μM Phos-tag acrylamide, 150 μM MnCl2, 0.1% (v/v) TEMED and 0.05% (w/v) APS]. The gel mixture was degassed for 10 min before adding TEMED and APS. After centrifugation at 20800 g for 1 min, 10–30 μg samples were loaded and electrophoresed at 70 V for the stacking part and at 150 V for the separating part with the running buffer [25 mM Tris/HCl, 192 mM glycine and 0.1% (w/v) SDS]. For Coomassie Blue staining, gels were stained with Colloidal Coomassie Blue Staining Kit (Life Technologies) according to the manufacturer's instructions. For immunoblot analysis, gels were washed for 10 min in the transfer buffer [48 mM Tris/HCl, 39 mM glycine and 20% (v/v) methanol] containing 10 mM EDTA and 0.05% (w/v) SDS three times, followed by one wash in the transfer buffer containing 0.05% SDS for 10 min. Proteins were electrophoretically transferred onto nitrocellulose membranes (Amersham Protran 0.45 μm NC; GE Healthcare) at 100 V for 180 min on ice in the transfer buffer without SDS/EDTA. Transferred membranes were blocked with 5% (w/v) non-fat dry milk (NFDM) dissolved in TBS-T [20 mM Tris/HCl, pH 7.5, 150 mM NaCl and 0.1% (v/v) Tween 20] at room temperature for 30 min. Membranes were then incubated with primary antibodies diluted in 5% NFDM and skim milk powder in TBS-T overnight at 4°C. After washing membranes in TBS-T, membranes were incubated with horseradish peroxidase-labelled secondary antibodies diluted in 5% NFDM and skim milk powder in TBS-T at room temperature for 1 h. After washing membranes in TBS-T, protein bands were detected by exposing films [Medical Film (Konica Minolta) for normal immunoblot and Amersham Hyperfilm ECL (GE Healthcare) for Phos-tag immunoblot] to the membranes using an ECL solution [Amersham ECL Western Blotting Detection Reagents (GE Healthcare) for normal immunoblot and SuperSignal West Dura Extended Duration (Thermo Fisher Scientific) for Phos-tag immunoblot]. Purification of Rab proteins Rab10 The coding sequence for human Rab10 (accession number: NM_016131.4) was cloned into pET15b so that the protein was N-terminally tagged with 6His-SUMO (clone number DU51062). BL21(DE3) cells were co-transformed with pET15b-6His-SUMO-Rab10 and a plasmid encoding the chaperone GroEL/S, and clones were allowed to grow in the presence of 100 μg/ml carbenicillin and 20 μg/ml chloramphenicol. Transformed bacteria were grown overnight and used to inoculate 6 litres of LB containing 50 μg/ml carbenicillin and 20 μg/ml chloramphenicol. After growing cells to a D600 of 0.4 at 37°C, temperature was lowered to 16°C and cells were grown until reaching a D600 of 0.6. Expression of Rab10 was induced by adding 125 μM IPTG for 14–18 h at 16°C with agitation at 200 rotations/min. Cells were collected by sedimentation and resuspended in ice-cold 50 mM Tris/HCl, pH 7.5, 250 mM NaCl, 0.2% Triton X-100, 5 mM MgCl2, 10 μM GDP, 1 mM tris-(2-carboxyethyl)phosphine (TCEP), 1 μM Pefabloc and 0.1% leupeptin. The suspension was sonicated and insoluble material was removed by centrifugation (20 min at 40000 g). The supernatant was supplemented with 10% glycerol, 20 mM imidazole, 50 μM ATP and 1 ml Ni-agarose and incubated on a roller mixer for 1 h at 4°C. Contaminants were removed with five washes (5×12 vol.) of 50 mM Tris/HCl, pH 7.5, 250 mM NaCl, 10% glycerol, 25 mM imidazole, 5 mM MgCl2, 0.2% Triton X-100, 0.03% Brij-35, 10 μM GDP, 50 μM ATP and 1 mM TCEP. Rab10 was removed from the resin by cleaving the His-SUMO tag using 1 mg of a catalytic domain of SUMO-specific protease His-SENP1 (amino acids 415–643) for 16 h at 4°C and collected in four resin volumes. The protein was diluted 10-fold into 50 mM HEPES/NaOH, pH 7.5, 25 mM NaCl, 10% glycerol, 5 mM MgCl2, 0.03% Brij-35, 10 μM GDP, 50 μM ATP and 1 mM TCEP and purified further over a 1 ml heparin HiTrap HP column (GE Healthcare), which was developed with a total 18 ml gradient of NaCl (25–1200 mM). Rab10 was eluted in two peaks, of which the earlier peak, eluting at approximately 200 mM NaCl, contained 90% pure Rab10. The yield is very low at only 50 μg/l expression. Rab8A The coding sequence for human Rab8A (accession number: NM_005370.4) was cloned into pET15b (clone number DU47363) and purified as described previously [13]. Assessment of kinase activity of endogenous LRRK2 In Figure 2(B), the kinase activity of endogenous LRRK2 immunoprecipitated from littermate WT and kinase-inactive LRRK2[D2017A] knockin MEFs was assessed in an in vitro kinase reaction as previously described [35]. Briefly, endogenous LRRK2 was immunoprecipitated from lysates (5 mg of protein) using 10 μg of anti-LRRK2 antibody UDD3 coupled to Protein A–Sepharose beads. A control was also included when UDD3 was replaced by pre-immune IgG. Peptide kinase assays were set up with immunoprecipitated LRRK2 in 50 mM Tris/HCl (pH 7.5), 0.1 mM EGTA, 10 mM MgCl2 and 0.1 mM [γ-32P]ATP (∼300–500 c.p.m./pmol, PerkinElmer) in the presence of 20 μM Nictide peptide substrate (RLGWWRFYTLRRARQGNTKQR) in the presence of either 1 μM MLi-2 or the equivalent volume of DMSO. After incubation for 20 min at 30°C with shaking, reactions were terminated by applying the reaction mixture on to P81 phosphocellulose papers and immersing in 50 mM orthophosphoric acid. After extensive washing, reaction products were quantified by Cerenkov counting. For experiments performed in Figure 7(B), the endogenous LRRK2 was immunoprecipitated from littermate WT and LRRK2[S910A+S935A] knockin MEFs as described above. Kinase assays were carried out using purified Rab8A protein as a substrate as described previously [13]. Assessment of phosphorylation at Ser1292 of endogenous LRRK2 Endogenous LRRK2 was immunoprecipitated as described above from lysates (3.5 mg of protein). Immunoprecipitated LRRK2 was washed twice with the lysis buffer containing 0.5 M NaCl and eluted from the beads with 30 μl of 2× NuPAGE lithium dodecyl sulfate (LDS) Sample Buffer (Thermo Fisher Scientific). Eluted samples at 5 and 15 μl were loaded for detecting total LRRK2 and phospho-Ser1292 LRRK2 respectively. For detecting phospho-Ser1292 LRRK2 VeriBlot secondary antibody (Abcam, ab131366) was used instead of normal anti-rabbit IgG secondary antibody. In vitro phosphorylation of Rab10 by LRRK2 Purified Rab10 (6.5 μg per 25 μl reaction) was phosphorylated using full-length LRRK2[G2019S] (0.8 μg) in a buffer containing 50 mM Tris/HCl, pH 7.5, 0.1 mM EGTA, 10 mM MgCl2 and 1 mM ATP, in the absence or presence of the LRRK2 inhibitor MLi-2 (1 μM final concentration). A reaction where no LRRK2 was added was also included as a negative control. Assays were carried out in Dispo-Biodialysers of 1 kDa molecular mass cut-off (Sigma–Aldrich) put in 0.5 litre of the same buffer to allow for ADP exchange for the indicated times at room temperature. Kinase reactions were terminated by addition of sample buffer containing 2-mercaptoethanol. Isolation of B-cells from mouse spleen Mouse B-cells were isolated from spleen using the MACSTM B-Cell Isolation Kit (Miltenyi Biotec, catalogue number 130-090-862) according to manufacturer's instructions. After isolation, B-cells were cultured in RPMI 1640 medium supplemented with 10% heat-inactivated FBS, 2 mM L-glutamine, 50 units/ml penicillin, 50 μg/ml streptomycin, sodium pyruvate and non-essential amino acids (Life Technologies) for 90 min before being treated with the LRRK2 inhibitor MLi-2 (50 nM final concentration) for 60 min. RESULTS Validation of the Phos-tag approach to assess LRRK2-mediated phosphorylation of Rab10 We first explored the effect of phosphorylation of recombinant bacterial expressed Rab10 with LRRK2[G2019S] on the electrophoretic mobility of Rab10 on Phos-tag-containing polyacrylamide gels. LRRK2 phosphorylation induced a time-dependent retardation in the migration of phosphorylated Rab10, an effect that was prevented by inclusion of the MLi-2 LRRK2 kinase inhibitor in the kinase reaction [36] (Figure 1A). Immunoblot analysis with a phospho-specific antibody confirmed that the slower migrating Rab10 species that appears following LRRK2 phosphorylation is indeed Rab10 phosphorylated at Thr73 (Figure 1A). We also studied LRRK2-mediated phosphorylation of HA–Rab10 following its co-expression with pathogenic LRRK2[R1441G] in HEK-293 cells. Under these conditions ∼70% of Rab10 was phosphorylated and the phosphorylation-induced mobility was blocked by mutation of the LRRK2 phosphorylation site (Thr73 to alanine) or by treatment of cells with MLi-2 LRRK2 inhibitor (Figure 1B). Figure 1 Phos-tag analysis of LRRK2 mediated Rab10 phosphorylation (A) Time course of LRRK2-mediated phosphorylation of recombinant Rab10, in the absence or presence of the LRRK2 inhibitor MLi-2. Rab10 phosphorylation was analysed by a Phos-tag assay using an anti-total Rab10 antibody or a phospho-specific antibody. A Coomassie Blue-stained Phos-tag gel is also shown (top panel). Control immunoblots (Rab10 total and LRRK2) were done on normal gels using the indicated antibodies (bottom panels). (B) HEK-293 cells were transfected with FLAG–LRRK2 R1441G and HA–Rab10 WT or T73A mutant and treated with or without 100 nM MLi-2 for 1 h. Phosphorylation of overexpressed Rab10 was analysed by a Phos-tag assay (top panel). Equal levels of expression of HA–Rab10 and FLAG–LRRK2 R1441G were confirmed by immunoblotting on normal gels using an anti-HA (second panel from the top) and anti-LRRK2 (third panel from the top) antibodies respectively. Equal loading was shown by immunoblotting with an anti-GAPDH antibody (bottom panel). Bands corresponding to phosphorylated and non-phosphorylated Rab10 were marked with open (○) and closed (●) circles respectively. Similar results were obtained in at least two separate experiments. Use of the Phos-tag approach to assess LRRK2-mediated phosphorylation of endogenous Rab10 in MEFs We tested whether the Phos-tag approach could be used to assess LRRK2 phosphorylation of endogenous Rab10 in MEFs. The Rab10 antibody used for these studies was selective as it detected endogenous Rab10 in WT, but not in clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated 9 (Cas9) Rab10 knockout A549 cells (Supplementary Figure S1). Phos-tag analysis of Rab10 derived from WT MEFs revealed that the bulk of Rab10 was in the unphosphorylated form; nevertheless, a significant minor phosphorylated Rab10 species was observed (Figure 2A). Treatment of WT MEFs with structurally diverse LRRK2 inhibitors (GSK2578215A, HG-10-102-01 and MLi-2) prevented Rab10 phosphorylation as judged by loss of the phosphorylated slower migrating Rab10 species (Figure 2A). Figure 2 Phosphorylation of endogenous Rab10 in MEFs analysed by Phos-tag assay (A) WT MEFs were treated with 0.1% (v/v) DMSO (−), 1 μM GSK2578215A (GSK), 3 μM HG-10-102-01 (HG) or 10 nM MLi-2 for 1 h in duplicate. Cell lysates were prepared and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. (B) LRRK2 immunoprecipitated from littermate WT or kinase-inactive LRRK2[D2017A] knockin MEFs was assessed for phosphorylation of Nictide peptide substrate in the absence or presence of MLi-2 (1 μM). IgG controls were also included where LRRK2 immunoprecipitating antibody was replaced by pre-immune IgG. Western blots below show the levels of immunoprecipitated LRRK2 used for the kinase assays and LRRK2 levels in pre-immune lysates. Results are means ± S.D. (n=3). (C) Littermate WT and kinase-inactive LRRK2[D2017A] knockin MEFs were treated with or without 1 μM GSK2578215A for 1 h. Cell lysates were prepared and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. (D) As in (C) except littermate WT and LRRK2 KO MEFs were used. (E) Littermate WT and MLi-2-resistant LRRK2[A2016T] knockin MEFs were treated with the indicated concentrations of MLi-2 for 1 h in duplicate. Cell lysates were prepared and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. Bands corresponding to phosphorylated and non-phosphorylated Rab10 were marked with open (○) and closed (●) circles respectively. Similar results were obtained in at least two separate experiments. We next analysed Rab10 phosphorylation in MEFs derived from a novel kinase-inactive LRRK2[D2017A] knockin mouse model, described here for the first time. LRRK2 isolated from the LRRK2[D2017A] knockin MEFs is expressed at slightly elevated levels compared with that in littermate WT cells (Figure 2B). Kinase activity measurements following LRRK2 immunoprecipitation confirmed that LRRK2 in the LRRK2[D2017A] knockin cells is devoid of kinase activity (Figure 2B). Phos-tag analysis revealed that Rab10 phosphorylation was strikingly absent from LRRK2[D2017A] knockin MEFs (Figure 2C). We also observed that Rab10 was also not phosphorylated in LRRK2 knockout MEFs (Figure 2D). Phos-tag analysis permitted detection of MLi-2 inhibition of Rab10 phosphorylation in WT and previously described MLi-2 inhibitor resistant LRRK2[A2016T] knockin MEFs [13] (Figure 2E). Doses of 3–10 nM MLi-2 suppressed Rab10 phosphorylation in WT MEFs, but concentrations of ≥100 nM were required to equivalently reduce phosphorylation in LRRK2[A2016T] knockin cells (Figure 2E). MLi-2 induced dephosphorylation of the LRRK2 Ser935 biomarker site [28], paralleled Rab dephosphorylation in the WT as well as the inhibitor-resistant LRRK2[A2016T] knockin MEFs (Figure 2E). Two other structurally diverse GSK2578215A [29], HG-10-102-01 [30] LRRK2 inhibitors induced a dose-dependent inhibition of Rab10 phosphorylation in WT MEFs (Figures 3A and 3B), with suppression of Rab10 phosphorylation coinciding with loss of LRRK2 Ser935 phosphorylation. LRRK2[G2019S]GSK knockin MEFs were treated with the LRRK2 inhibitors (Figures 3C–3E), showing inhibition of Rab10 phosphorylation and loss of LRRK2 Ser935 phosphorylation at a similar dose to that required in WT MEFs (Figures 2E, 3A and 3B). Figure 3 Dose-dependent inhibition of Rab10 phosphorylation in MEFs analysed by Phos-tag assay WT MEFs were treated with the indicated concentrations of (A) GSK2578215A or (B) HG-10-102-01 for 1 h in duplicate. Cell lysates were prepared and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. (C and D) As in (A) and (B) except LRRK2[G2019S]GSK knockin MEFs were used. (E) As in (C) and (D) except MLi-2 was used. Bands corresponding to phosphorylated and non-phosphorylated Rab10 were marked with open (○) and closed (●) circles respectively. Similar results were obtained in at least two separate experiments. LRRK2 inhibitors induce more rapid dephosphorylation of Rab10 than kinase biomarker residues (Ser935 and Ser1292) We next compared the rate at which Rab10 and LRRK2 Ser935 are dephosphorylated following treatment of WT MEFs with structurally diverse kinase inhibitors. This revealed that Rab10 was rapidly dephosphorylated within 1–2 min following treatment with 1 μM GSK2578215A (Figure 4A) or 3 μM HG-10-102-01 (Figure 4B) and 5–10 min with 10 nM MLi-2 (Figure 4C). In contrast, dephosphorylation of LRRK2 Ser935 was markedly slower requiring 40–80 min (Figures 4A–4C). Previous work revealed that the autophosphorylation of LRRK2 at Ser1292 can also be deployed as a read out for LRRK2 kinase activity and that phosphorylation of this residue is enhanced by pathogenic mutations including G2019S [37]. To investigate the rate at which Ser1292 is dephosphorylated, we treated LRRK2[G2019S] knockin MEFs (in which Ser1292 is more readily detected than in WT MEFs) with 1 μM GSK2578215A for various time points. Ser1292 phosphorylation was analysed employing a Ser1292 phospho-specific antibody following immunoprecipitation of LRRK2. These studies revealed that dephosphorylation of Ser1292 occurred on a longer time course more similar to that of Ser935 requiring 80–160 min to attain maximal dephosphorylation. As observed in WT MEFs (Figure 4A), GSK2578215A induced rapid dephosphorylation of Rab10 within 1–2 min in LRRK2[G2019S] knockin MEFs (Figure 4D). Figure 4 Time-course experiments to compare phosphorylation of endogenous LRRK2 Ser935 and endogenous Rab10 in MEFs analysed by Phos-tag assay WT MEFs were treated with (A) 1 μM GSK2578215A, (B) 3 μM HG-10-102-01 or (C) 10 nM MLi-2. Cell lysates were prepared at the indicated time points and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. (D) LRRK2[G2019S]GSK knockin MEFs were treated with 1 μM GSK2578215A. Cell lysates were prepared at the indicated time points and Rab10 phosphorylation was analysed by a Phos-tag assay. Control immunoblots were done on normal gels with the indicated antibodies. Endogenous LRRK2 was also immunoprecipitated from cell lysates and blotted with the anti-pS1292 or total LRRK2 antibody (top panel). Bands corresponding to phosphorylated and non-phosphorylated Rab10 were marked with open (○) and closed (●) circles respectively. Similar results were obtained in at least two separate experiments. Use of the Phos-tag approach to assess LRRK2-mediated phosphorylation of endogenous Rab10 in mouse lung and spleen derived B-cells We next analysed Rab10 phosphorylation in littermate WT and kinase-inactive LRRK2[D2017A] knockin mouse lung tissue. This revealed that phosphorylation of Rab10 was readily observed in WT but not in the kinase-inactive LRRK2[D2017A] knockin lung (Figure 5A). Injection of WT mice with doses of 1–3 mg/kg MLi-2 blocked Rab10 phosphorylation, whereas doses of ≥10 mg/kg MLi-2 were needed to induce equivalent blockade in LRRK2[A2016T] inhibitor-resistant lung (Figure 5B). MLi-2 induced dephosphorylation of LRRK2 Ser935 paralleled Rab dephosphorylation, with significantly higher doses of MLi-2 required to induce equivalent Ser935 and Rab10 dephosphorylation in LRRK2[A2016T] lung compared with WT (Figure 5B). Phosphorylated Rab10 was also detected in splenic B-cells derived from WT mice, which was lost following incubation of B-cells with MLi-2 in RPMI 1640 medium for 60 min prior to cell lysis (Figure 5C). Figure 5 Phosphorylation of endogenous Rab10 in mouse lung and spleen-derived B-cells analysed by Phos-tag assay (A) Lung tissues were collected from two littermate WT and two kinase-inactive LRRK2[D2017A] knockin mice. Tissue lysates were prepared and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. (B) Littermate WT and MLi-2-resistant LRRK2[A2016T] mice were subcutaneously injected with the indicated doses of MLi-2 and treated for 1 h (two mice for each dose). Lung tissues were collected and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. (C) B-cells were isolated from eight WT mouse spleens and treated with or without 50 nM MLi-2 for 1 h (four replicates for each condition). Cell lysates were prepared and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. Bands corresponding to phosphorylated and non-phosphorylated Rab10 were marked with open (○) and closed (●) circles respectively. Similar results were obtained in at least two separate experiments. Use of the Phos-tag approach to assess the impact of LRRK2 pathogenic mutations We next employed the Phos-tag approach to assess the impact of homozygous LRRK2[R1441G] (Figure 6A) and LRRK2[G2019S]GSK (Figure 6B) knockin mutations on LRRK2 Rab10 phosphorylation in MEFs. Compared with WT controls, the LRRK2[R1441G] knockin enhanced Rab10 phosphorylation ∼3–4-fold (Figure 6A) and the G2019S mutation enhanced phosphorylation ∼2-fold (Figure 6B). In both R1441G and G2019S knockin MEFs, LRRK2 inhibitors suppressed Rab10 phosphorylation (Figure 6). Consistent with a previous report [12], the R1441G knockin mutation markedly inhibited basal levels of LRRK2 Ser935 phosphorylation (Figure 6B). Figure 6 Phosphorylation of endogenous Rab10 in MEFs harbouring pathogenic mutations analysed by Phos-tag assay (A) Littermate WT and pathogenic LRRK2[R1441G] knockin MEFs were treated with or without 1 μM GSK2578215A for 1 h. Cell lysates were prepared and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. (B) As in (A) except littermate WT and pathogenic LRRK2[G2019S]GSK knockin MEFs were used. Bands corresponding to phosphorylated and non-phosphorylated Rab10 were marked with open (○) and closed (●) circles respectively. Similar results were obtained in at least two separate experiments. We next analysed Rab10 phosphorylation in littermate WT and LRRK2[R1441G] knockin mouse lung tissue. This revealed that phosphorylation of Rab10 was markedly elevated in LRRK2[R1441G] knockin lung compared with WT (Figure 7A). Injection of 3 mg/kg MLi-2 for 60 min blocked Rab10 phosphorylation in both WT and LRRK2[R1441G] knockin mouse lung (Figure 7B). Figure 7 Phosphorylation of endogenous Rab10 in mouse lung harbouring pathogenic mutations analysed by Phos-tag assay Littermate WT and pathogenic LRRK2[R1441G] knockin mice were subcutaneously injected with vehicle only or MLi-2 at 3 mg/kg and treated for 1 h (three mice for vehicle control and five mice for MLi-2). Lung tissues were collected and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. (A) Side-by-side comparison of WT and LRRK2[R1441G] knockin lungs on a same gel. (B) Side-by-side comparison of lungs injected with vehicle or MLi-2 on a same gel. Bands corresponding to phosphorylated and non-phosphorylated Rab10 were marked with open (○) and closed (●) circles respectively. Similar results were obtained in at least two separate experiments. Use of the Phos-tag approach to assess the impact of S910A/S935A mutations There is significant interest in understanding the roles that LRRK2 phosphorylation at LRRK2 Ser910 and Ser935 residues plays in controlling LRRK2 activity, as these phosphorylations regulate interaction of LRRK2 with 14-3-3 proteins and are also sensitive to LRRK2 inhibitors [12,28]. To better understand the role of Ser910 and Ser935 phosphorylations, we generated homozygous LRRK2[S910A+S935A] knockin MEFs. Immunoblot analysis confirmed that the LRRK2[S910A+S935A] mutant kinase was expressed at the same level as LRRK2 derived from littermate WT cells (Figure 8A). Moreover, following immunoprecipitation, the LRRK2[S910A+S935A] mutant was capable of phosphorylating recombinant Rab8A in vitro to a similar extent as the WT LRRK2 (Figure 8B). Rab8A rather than Rab10 was used for these experiments as rates of phosphorylation of Rab8A by immunoprecipitated endogenous LRRK2 was much higher and could be more robustly quantified than with Rab10. Strikingly, we observed that endogenous Rab10 phosphorylation was markedly reduced in the LRRK2[S910A+S935A] knockin MEFs compared with littermate-derived WT cells (Figure 8C). Figure 8 Phosphorylation of endogenous Rab10 in MEFs harbouring S910A/S935A knockin mutation analysed by Phos-tag assay (A) Phosphorylation of LRRK2 at Ser910 was analysed by immunoprecipitating LRRK2 from cell lysates of WT, LRRK2[S910A+S935A] knockin and LRRK2 KO MEFs and immunoblotting with an anti-pSer910 antibody (top panel). Phosphorylation of LRRK2 at Ser935 was analysed by immunoblotting of the cell lysates with an anti-pSer935 antibody (second panel from the top). Equal expression of LRRK2 in WT and knockin MEFs lysates was confirmed by immunoblotting with an anti-total LRRK2 antibody (third panel from the top). (B) Endogenous LRRK2 proteins were immunoprecipitated with a monoclonal anti-total LRRK2 antibody from WT, LRRK2[S910A+S935A] knockin and LRRK2 KO MEFs. Purified LRRK2 proteins were assessed for phosphorylation of Rab8A protein in the absence or presence of GSK2578215A. Immunoprecipitates were then subjected to electrophoresis on a polyacrylamide gel, autoradiography and immunoblot analysis with the indicated antibodies. After autoradiography, the bands corresponding to Rab8A were cut out to measure the radioactivity by scintillation counting. Results are means ± S.D. (n=3). (C) Littermate WT and LRRK2[S910A+S935A] knockin MEFs were treated with or without 1 μM GSK2578215A for 1 h. Cell lysates were prepared and Rab10 phosphorylation was analysed by a Phos-tag assay (top panel). Control immunoblots were done on normal gels with the indicated antibodies. Bands corresponding to phosphorylated and non-phosphorylated Rab10 were marked with open (○) and closed (●) circles respectively. Similar results were obtained in at least two separate experiments. DISCUSSION Here we show that the Rab10 Phos-tag assay can readily be used to assess LRRK2-mediated phosphorylation of endogenous Rab10 in MEFs, mouse lung, mouse spleen-derived B-cells. We expect that the Rab10 Phos-tag assay will work in other cell lines in which LRRK2 and Rab10 are well expressed. The Rab10 Phos-tag assay is straightforward, necessitating only SDS/polyacrylamide gel and immunoblotting apparatus and moderate amounts of cell extracts (10–45 μg of protein). Moreover, the two key reagents required for the assay, namely the anti-Rab10 monoclonal antibody and Phos-tag acrylamide are both commercially available. To reduce assay costs, we undertook chemical synthesis of the Phos-tag acrylamide reagent. It should also be noted that Phos-tag reagent requires Mn2+ ions in order to interact with phosphate groups [25–27]. We have also found that the Phos-tag acrylamide reagent can undergo polymerization following long-term storage which results in reduced separation of dephosphorylated and LRRK2-phosphorylated Rab10. The optimal conditions we have found to store Phos-tag acrylamide is 5 mM in aqueous solution at 4°C in black tubes that block out light as the reagent is light-sensitive. We would also recommend that purity of Phos-tag acrylamide be assessed by HPLC analysis periodically. We would also recommend that if samples to be analysed and/or the SDS sample buffer contain EDTA, an excess of MnCl2 over EDTA is added to the sample prior to loading the samples on to the Phos-tag gel. A single researcher could readily analyse a few dozen of samples per day using the Rab10 Phos-tag assay. The finding that diverse LRRK2 inhibitors, kinase-inactivating LRRK2[D2017A] knockin mutation as well as LRRK2 knockout, ablate all detectable phosphorylation of Rab10, strongly suggests that LRRK2 is the major kinase that phosphorylates Rab10 at least in MEFs and lung tissue that we have analysed. The finding that LRRK2[A2016T] inhibitor-resistant knockin increases the dose of LRRK2 inhibitor required to reduced Rab10 phosphorylation in both MEFs and mouse lung provides a fundamental demonstration that LRRK2 is the major kinase controlling Rab10 phosphorylation in MEFs. Another advantage of the Phos-tag method is that it allows assessment of stoichiometry of phosphorylation. In MEFs and lung tissue that we have analysed, the data indicate that only a small fraction of Rab10 is phosphorylated at steady state. This probably accounts for why it was challenging to identify phosphorylated species of Rab10 by mass spectrometry, as such a low proportion of the Rab protein is phosphorylated by LRRK2 in vivo. However, the low basal levels of LRRK2-phosphorylated Rab10 may make it easier to monitor the impact of activating LRRK2 pathogenic mutations have on enhancing Rab10 phosphorylation (Figure 6 and 7). We have also examined total brain and kidney tissue extracts to see whether we could detect LRRK2-mediated phosphorylation of Rab10, but failed to observe significant Rab10 phosphorylation using the described Phos-tag assays under conditions where LRRK2 phosphorylation of Rab10 in lung and spleen was observed. Further work is warranted to develop methodology to assess LRRK2 phosphorylation of Rab10 in brain and kidney. In the future it will be interesting to explore whether it is possible to observe LRRK2-dependent phosphorylation of Rab proteins using the Phos-tag approach in human derived cells such as fibroblast, peripheral blood mononuclear cells (PBMCs) or other blood cells, as well as bodily fluids such as in cerebrospinal fluid. It will also be important to explore whether elevated Rab protein phosphorylation can be observed in Parkinson's disease patients who are carriers of LRRK2 mutations and whether a subgroup of Parkinson's disease patients with idiopathic disease also display elevated Rab phosphorylation. For the benefit of future clinical trials of LRRK2 inhibitors, it would be desirable to determine whether target engagement of LRRK2 inhibitors could be demonstrated by monitoring the effect these compounds have on Rab protein phosphorylation in human blood cells. It will also be intriguing to investigate whether the Rab Phos-tag assay can be used to detect LRRK2-phosphorylated Rab proteins in human urinary exosomes that contain LRRK2 [38]. Recent studies have reported elevated phosphorylation of LRRK2 at its Ser1292 autophosphorylation site [37] in urine exosomes and concluded that this can predict Parkinsonian phenotypes in G2019S LRRK2 subjects [39]. There has been a lot of interest in studying the roles of the LRRK2 Ser910 and Ser935 phosphorylation sites, as these mediate 14-3-3 binding and become dephosphorylated when cells are exposed to LRRK2 inhibitors [12,28]. Most of the data suggest that Ser910 and Ser935 are likely to be phosphorylated by kinases distinct to LRRK2 [12,28]. Although several candidates for the LRRK2 Ser910 and Ser935 kinase(s) have been proposed [40–42], further studies are required to pinpoint these kinase(s) and characterize how inhibition of LRRK2 leads to dephosphorylation of these residues. Consistent with the notion that an LRRK2-distinct kinase phosphorylates Ser935, we observe that Ser935 is still phosphorylated in the LRRK2[D2017A] kinase-inactive MEFs (Figure 2C) and lungs (Figure 5A). However, following MLi-2 administration, in contrast with wild type situation where Ser935 becomes dephosphorylated, in the LRRK2[D2017A] knockin MEFs, Ser935 is not dephosphorylated (Figure 2C). This is consistent with a model in which the LRRK2 Ser935 kinase is uncoupled from LRRK2 in the LRRK2[D2017A] knockin MEFs. The finding that treatment of cells with LRRK2 inhibitors induces more rapid dephosphorylation of Rab10 (1–2 min with GSK2578215A and HG-10-102-01) than Ser935 (40–80 min, Figures 4A–4C), is consistent with the regulation of Rab10 being directly mediated by LRRK2, whereas phosphorylation of Ser935 is indirectly controlled. The rapid dephosphorylation Rab10 that is observed following suppression of LRRK2 kinase activity may indicate that the phosphatase that acts on Rab10 is highly active and/or the Thr73 residue is exposed and accessible to the phosphatase. In contrast, dephosphorylation of the Ser1292 autophosphorylation site of LRRK2 was significantly slower than Rab10, necessitating 40–80 min (Figure 4D). This slower dephosphorylation might result if the phosphatase that targets Ser1292 had low activity and/or access of phosphorylated Ser1292 to the protein phosphatase was hindered. The finding that the LRRK2[S910A+S935A] knockin mutation suppresses phosphorylation of Rab10 in MEFs, provides evidence for a functional role of Ser910 and Ser935 phosphorylation in enabling LRRK2 to optimally phosphorylate Rab GTPases. More work is needed to unravel this mechanism. One possibility is that this is mediated through localization of LRRK2. Previous work in a HEK-293 cell overexpression system suggested that the LRRK2[S910A+S935A] mutant was assembled into inclusion-like bodies very different from the WT LRRK2 that was diffusely localized throughout the cytosol. As functional Rab proteins are largely localized on membranes, perhaps LRRK2 Ser910 and Ser935 phosphorylation and 14-3-3 binding facilitate recruitment of LRRK2 on to membranes where it can phosphorylate Rab proteins. We thank Matthew J. Fell and John A. Morrow (Merck Research Laboratories, Early Discovery Neuroscience, Boston, MA, U.S.A.) for providing MLi-2 used in experiments shown in Figures 1(B), 5(B) and 5(C), Axel Knebel for purifying Rab10, Francisco de Asis Inesta Vaquera for generating LRRK2[D2017A] knockin MEFs, and express gratitude to for the excellent technical support of the MRC-Protein Phosphorylation and Ubiquitylation Unit (PPU) DNA Sequencing Service (co-ordinated by Nicholas Helps), the MRC-PPU tissue culture team (co-ordinated by Laura Fin), the Division of Signal Transduction Therapy (DSTT) antibody purification teams (co-ordinated by Hilary McLauchlan and James Hastie). We thank Dr Suzanne Pfeffer (Department of Biochemistry, Stanford School of Medicine, Stanford, CA, U.S.A.), Dr Martin Steger and Dr Matthias Mann (Department of Proteomics and Signal Transduction, Max Planck Institute of Biochemistry, Martinsried, Germany) for helpful discussions. AUTHOR CONTRIBUTION Genta Ito designed and executed the majority of the experiments and analysed data as well as playing a major role in the preparation of the paper; Kristina Katsemonova undertook the initial characterization of the LRRK2[S910A+S935A] knockin mice that are described in the present paper for the first time and generated LRRK2[S910A+S935A] knockin MEFs as well as being involved in the design and execution of experiments and analysis of data; Francesca Tonelli designed the experiments, designed and executed experiments and analysed data; Pawel Lis generated the CRISPR/Cas9 Rab10 knockout A549 cells that were required to validate the anti-Rab10 antibody; Marco Baptista helped with the design of the experiments, suggested performing Rab10 Phos-tag analysis in mouse lung tissue, provided LRRK2[A2016T] knockin mice and helped with the analysis of the data; Natalia Shpiro contributed essential reagents by synthesizing the MLi-2 and Phos-tag reagents utilized in the present study; Graham Duddy, Steve Wilson and Alastair Reith generated the novel LRRK2[D2017A] knockin mice that are described in the present paper for the first time and also provided LRRK2[G2019S] knockin mice and helped with the design and execution of experiments and analysis of the data; Philip Wing-Lok Ho and Shu-Leong Ho generated the LRRK2[R1441G] knockin mouse line, performed MLi-2 injections and generated the LRRK2[R1441G] knockin MEFs and helped with the design of the experiments; Dario Alessi conceived the project, helped with experimental design and analysis and interpretation of data and wrote the paper. CONFLICT OF INTEREST S.W. and A.D.R. are employees of GlaxoSmithKline, a global healthcare company that may conceivably benefit financially through this publication. The other authors of this paper declare no conflict of interest. FUNDING This work was supported by the Michael J. Fox Foundation for Parkinson's research [grant number 357811350 R60 (to D.R.A.)]; the Medical Research Council [grant number MC_UU_12016/2 (to D.R.A.)]; the pharmaceutical companies supporting the Division of Signal Transduction Therapy Unit (AstraZeneca, Boehringer-Ingelheim, GlaxoSmithKline, Merck KGaA, Janssen Pharmaceutica and Pfizer) (to D.R.A.); and the Health and Medical Research Fund, Food and Health Bureau, Hong Kong [grant number #1120976 (to S.L.H.)]. Abbreviations APSammonium persulfate BACbacterial artificial chromosome Cas9CRISPR-associated 9 CORC-terminal of ROC (Ras of complex GTPase domain) CRISPRclustered regularly interspaced short palindromic repeats DMEMDulbecco's modified Eagle's medium Eembryonic day ESembryonic stem GAPDHglyceraldehyde-3-phosphate dehydrogenase HAhaemagglutinin HEKhuman embryonic kidney KOknockout LRRK2leucine-rich repeat kinase 2 MEFmouse embryonic fibroblast NFDMnon-fat dry milk TCEPtris-(2-carboxyethyl)phosphine TEMEDN,N,N′,N′_tetramethylethylenediamine WTwild-type ==== Refs 1 Bras J. Guerreiro R. Hardy J. SnapShot: genetics of Parkinson's disease Cell 2015 160 570 570 10.1016/j.cell.2015.01.019 25635463 2 Singleton A.B. Farrer M.J. Bonifati V. The genetics of Parkinson's disease: progress and therapeutic implications Mov. Disord. 2013 28 14 23 10.1002/mds.25249 23389780 3 Zimprich A. Biskup S. Leitner P. Lichtner P. Farrer M. Lincoln S. Kachergus J. Hulihan M. Uitti R.J. Calne D.B. Mutations in LRRK2 cause autosomal-dominant parkinsonism with pleomorphic pathology Neuron 2004 44 601 607 10.1016/j.neuron.2004.11.005 15541309 4 Paisan-Ruiz C. Jain S. Evans E.W. Gilks W.P. Simon J. van der Brug M. Lopez de Munain A. Aparicio S. Gil A.M. Khan N. Cloning of the gene containing mutations that cause PARK8-linked Parkinson's disease Neuron 2004 44 595 600 10.1016/j.neuron.2004.10.023 15541308 5 Thaler A. Ash E. Gan-Or Z. Orr-Urtreger A. Giladi N. The LRRK2 G2019S mutation as the cause of Parkinson's disease in Ashkenazi Jews J. Neural. Transm. (Vienna) 2009 116 1473 1482 10.1007/s00702-009-0303-0 19756366 6 West A.B. Moore D.J. Biskup S. Bugayenko A. Smith W.W. Ross C.A. Dawson V.L. Dawson T.M. Parkinson's disease-associated mutations in leucine-rich repeat kinase 2 augment kinase activity Proc. Natl. Acad. Sci. U.S.A. 2005 102 16842 16847 10.1073/pnas.0507360102 16269541 7 Jaleel M. Nichols R.J. Deak M. Campbell D.G. Gillardon F. Knebel A. Alessi D.R. LRRK2 phosphorylates moesin at threonine-558: characterization of how Parkinson's disease mutants affect kinase activity Biochem. J. 2007 405 307 317 10.1042/BJ20070209 17447891 8 Mata I.F. Wedemeyer W.J. Farrer M.J. Taylor J.P. Gallo K.A. LRRK2 in Parkinson's disease: protein domains and functional insights Trends Neurosci 2006 29 286 293 10.1016/j.tins.2006.03.006 16616379 9 Li Y. Liu W. Oo T.F. Wang L. Tang Y. Jackson-Lewis V. Zhou C. Geghman K. Bogdanov M. Przedborski S. Mutant LRRK2(R1441G) BAC transgenic mice recapitulate cardinal features of Parkinson's disease Nat. Neurosci. 2009 12 826 828 10.1038/nn.2349 19503083 10 Tong Y. Pisani A. Martella G. Karouani M. Yamaguchi H. Pothos E.N. Shen J. R1441C mutation in LRRK2 impairs dopaminergic neurotransmission in mice Proc. Natl. Acad. Sci. U.S.A. 2009 106 14622 14627 10.1073/pnas.0906334106 19667187 11 Daniels V. Vancraenenbroeck R. Law B.M. Greggio E. Lobbestael E. Gao F. De Maeyer M. Cookson M.R. Harvey K. Baekelandt V. Taymans J.M. Insight into the mode of action of the LRRK2 Y1699C pathogenic mutant J. Neurochem. 2011 116 304 315 10.1111/j.1471-4159.2010.07105.x 21073465 12 Nichols J. Dzamko N. Morrice N.A. Campbell D.G. Deak M. Ordureau A. Macartney T. Tong Y. Shen J. Prescott A. Alessi D.R. 14-3-3 binding to LRRK2 is disrupted by multiple Parkinson's disease associated mutations and regulates cytoplasmic localisation Biochem. J. 2010 430 393 404 10.1042/BJ20100483 20642453 13 Steger M. Tonelli F. Ito G. Davies P. Trost M. Vetter M. Wachter S. Lorentzen E. Duddy G. Wilson S. Phosphoproteomics reveals that Parkinson's disease kinase LRRK2 regulates a subset of Rab GTPases eLife 2016 5 e12813 10.7554/eLife.12813 26824392 14 Pereira-Leal J.B. Seabra M.C. The mammalian Rab family of small GTPases: definition of family and subfamily sequence motifs suggests a mechanism for functional specificity in the Ras superfamily J. Mol. Biol. 2000 301 1077 1087 10.1006/jmbi.2000.4010 10966806 15 Cherfils J. Zeghouf M. Regulation of small GTPases by GEFs, GAPs, and GDIs Physiol. Rev. 2013 93 269 309 10.1152/physrev.00003.2012 23303910 16 Pfeffer S.R. Rab GTPases: specifying and deciphering organelle identity and function Trends Cell Biol. 2001 11 487 491 10.1016/S0962-8924(01)02147-X 11719054 17 Tucci A. Nalls M.A. Houlden H. Revesz T. Singleton A.B. Wood N.W. Hardy J. Paisan-Ruiz C. Genetic variability at the PARK16 locus Eur. J. Hum. Genet. 2010 18 1356 1359 10.1038/ejhg.2010.125 20683486 18 Pihlstrom L. Rengmark A. Bjornara K.A. Dizdar N. Fardell C. Forsgren L. Holmberg B. Larsen J.P. Linder J. Nissbrandt H. Fine mapping and resequencing of the PARK16 locus in Parkinson's disease J. Hum. Genet. 2015 60 357 362 10.1038/jhg.2015.34 25855069 19 MacLeod D.A. Rhinn H. Kuwahara T. Zolin A. Di Paolo G. McCabe B.D. Marder K.S. Honig L.S. Clark L.N. Small S.A. Abeliovich A. RAB7L1 interacts with LRRK2 to modify intraneuronal protein sorting and Parkinson's disease risk Neuron 2013 77 425 439 10.1016/j.neuron.2012.11.033 23395371 20 Wilson G.R. Sim J.C. McLean C. Giannandrea M. Galea C.A. Riseley J.R. Stephenson S.E. Fitzpatrick E. Haas S.A. Pope K. Mutations in RAB39B cause X-linked intellectual disability and early-onset Parkinson disease with alpha-synuclein pathology Am. J. Hum. Genet. 2014 95 729 735 10.1016/j.ajhg.2014.10.015 25434005 21 Mata I.F. Jang Y. Kim C.H. Hanna D.S. Dorschner M.O. Samii A. Agarwal P. Roberts J.W. Klepitskaya O. Shprecher D.R. The RAB39B p.G192R mutation causes X-linked dominant Parkinson's disease Mol. Neurodegener. 2015 10 50 10.1186/s13024-015-0045-4 26399558 22 Cooper A.A. Gitler A.D. Cashikar A. Haynes C.M. Hill K.J. Bhullar B. Liu K. Xu K. Strathearn K.E. Liu F. Alpha-synuclein blocks ER-Golgi traffic and Rab1 rescues neuron loss in Parkinson's models Science 2006 313 324 328 10.1126/science.1129462 16794039 23 Gitler A.D. Bevis B.J. Shorter J. Strathearn K.E. Hamamichi S. Su L.J. Caldwell K.A. Caldwell G.A. Rochet J.C. McCaffery J.M. The Parkinson's disease protein alpha-synuclein disrupts cellular Rab homeostasis Proc. Natl. Acad. Sci. U.S.A. 2008 105 145 150 10.1073/pnas.0710685105 18162536 24 Lai Y.C. Kondapalli C. Lehneck R. Procter J.B. Dill B.D. Woodroof H.I. Gourlay R. Peggie M. Macartney T.J. Corti O. Phosphoproteomic screening identifies Rab GTPases as novel downstream targets of PINK1 EMBO J. 2015 34 2840 2861 10.15252/embj.201591593 26471730 25 Kinoshita E. Takahashi M. Takeda H. Shiro M. Koike T. Recognition of phosphate monoester dianion by an alkoxide-bridged dinuclear zinc(II) complex Dalton Trans. 2004 1189 1193 10.1039/b400269e 15252659 26 Kinoshita E. Yamada A. Takeda H. Kinoshita-Kikuta E. Koike T. Novel immobilized zinc(II) affinity chromatography for phosphopeptides and phosphorylated proteins J. Sep. Sci. 2005 28 155 162 10.1002/jssc.200401833 15754823 27 Kinoshita E. Kinoshita-Kikuta E. Takiyama K. Koike T. Phosphate-binding tag, a new tool to visualize phosphorylated proteins Mol. Cell. Proteomics 2006 5 749 757 10.1074/mcp.T500024-MCP200 16340016 28 Dzamko N. Deak M. Hentati F. Reith A.D. Prescott A.R. Alessi D.R. Nichols R.J. Inhibition of LRRK2 kinase activity leads to dephosphorylation of Ser(910)/Ser(935), disruption of 14-3-3 binding and altered cytoplasmic localization Biochem. J. 2010 430 405 413 10.1042/BJ20100784 20659021 29 Reith A.D. Bamborough P. Jandu K. Andreotti D. Mensah L. Dossang P. Choi H.G. Deng X. Zhang J. Alessi D.R. Gray N.S. GSK2578215A; a potent and highly selective 2-arylmethyloxy-5-substitutent-N-arylbenzamide LRRK2 kinase inhibitor Bioorg. Med. Chem. Lett. 2012 22 5625 5629 10.1016/j.bmcl.2012.06.104 22863203 30 Choi H.G. Zhang J. Deng X. Hatcher J.M. Patricelli M.P. Zhao Z. Alessi D.R. Gray N.S. Brain penetrant LRRK2 inhibitor ACS Med. Chem. Lett. 2012 3 658 662 10.1021/ml300123a 23066449 31a Miller M. Basu K. Demong D. Scott J. Li W. Stamford A. Poirier M. Tempest P. Compounds inhibiting leucine-rich repeat kinase enzyme activity Int. Pat. 2014 WO2014134774 31b Koike T. Kawasaki A. Kobashi T. Polyacrylamide gel for electrophoresis, polyacrylamide gel electrophoresis method using the same, method of producing the same, and acrylamide compound Int. Pat. 2007 WO2007015312A1 32 Liu H.F. Lu S. Ho P.W. Tse H.M. Pang S.Y. Kung M.H. Ho J.W. Ramsden D.B. Zhou Z.J. Ho S.L. LRRK2 R1441G mice are more liable to dopamine depletion and locomotor inactivity Ann. Clin. Transl. Neurol. 2014 1 199 208 10.1002/acn3.45 25356398 33 Parisiadou L. Xie C. Cho H.J. Lin X. Gu X.L. Long C.X. Lobbestael E. Baekelandt V. Taymans J.M. Sun L. Cai H. Phosphorylation of ezrin/radixin/moesin proteins by LRRK2 promotes the rearrangement of actin cytoskeleton in neuronal morphogenesis J. Neurosci. 2009 29 13971 13980 10.1523/JNEUROSCI.3799-09.2009 19890007 34 Wiggin G.R. Soloaga A. Foster J.M. Murray-Tait V. Cohen P. Arthur J.S. MSK1 and MSK2 are required for the mitogen- and stress-induced phosphorylation of CREB and ATF1 in fibroblasts Mol. Cell. Biol. 2002 22 2871 2881 10.1128/MCB.22.8.2871-2881.2002 11909979 35 Davies P. Hinkle K.M. Sukar N.N. Sepulveda B. Mesias R. Serrano G. Alessi D.R. Beach T.G. Benson D.L. White C.L. Comprehensive characterization and optimization of anti-LRRK2 (leucine-rich repeat kinase 2) monoclonal antibodies Biochem. J. 2013 453 101 113 10.1042/BJ20121742 23560750 36 Fell M.J. Mirescu C. Basu K. Cheewatrakoolpong B. DeMong D.E. Ellis J.M. Hyde L.A. Lin Y. Markgraf C.G. Mei H. MLi-2, a potent, selective, and centrally active compound for exploring the therapeutic potential and safety of LRRK2 kinase inhibition J. Pharmacol. Exp. Ther. 2015 355 397 409 10.1124/jpet.115.227587 26407721 37 Sheng Z. Zhang S. Bustos D. Kleinheinz T. Le Pichon C.E. Dominguez S.L. Solanoy H.O. Drummond J. Zhang X. Ding X. Ser1292 autophosphorylation is an indicator of LRRK2 kinase activity and contributes to the cellular effects of PD mutations Sci. Transl. Med. 2012 4 164ra161 10.1126/scitranslmed.3004485 23241745 38 Fraser K.B. Moehle M.S. Daher J.P. Webber P.J. Williams J.Y. Stewart C.A. Yacoubian T.A. Cowell R.M. Dokland T. Ye T. LRRK2 secretion in exosomes is regulated by 14-3-3 Hum. Mol. Genet. 2013 22 4988 5000 10.1093/hmg/ddt346 23886663 39 Fraser K.B. Moehle M.S. Alcalay R.N. West A.B. LRRK2 Cohort Consortium Urinary LRRK2 phosphorylation predicts parkinsonian phenotypes in G2019S LRRK2 carriers Neurology 2016 86 994 999 10.1212/WNL.0000000000002436 26865512 40 Chia R. Haddock S. Beilina A. Rudenko I.N. Mamais A. Kaganovich A. Li Y. Kumaran R. Nalls M.A. Cookson M.R. Phosphorylation of LRRK2 by casein kinase 1alpha regulates trans-Golgi clustering via differential interaction with ARHGEF7 Nat. Commun. 2014 5 5827 10.1038/ncomms6827 25500533 41 Dzamko N. Inesta-Vaquera F. Zhang J. Xie C. Cai H. Arthur S. Tan L. Choi H. Gray N. Cohen P. The IkappaB kinase family phosphorylates the Parkinson's disease kinase LRRK2 at Ser935 and Ser910 during Toll-like receptor signaling PLos One 2012 7 e39132 10.1371/journal.pone.0039132 22723946 42 Li X. Wang Q.J. Pan N. Lee S. Zhao Y. Chait B.T. Yue Z. Phosphorylation-dependent 14-3-3 binding to LRRK2 is impaired by common mutations of familial Parkinson's disease PLos One 2011 6 e17153 10.1371/journal.pone.0017153 21390248

PMC005xxxxxx/PMC5003744.txt

==== Front 00455031920Brain ResBrain Res.Brain research0006-89931872-62402699641210.1016/j.brainres.2016.02.037nihpa770807ArticlePrion-like domains as epigenetic regulators, scaffolds for subcellular organization, and drivers of neurodegenerative disease March Zachary M. abKing Oliver D. oliver.king@umassmed.educ*Shorter James jshorter@mail.med.upenn.eduab**a Department of Biochemistry and Biophysics, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA 19104, United States of Americab Biochemistry and Molecular Biophysics Graduate Group, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA 19104, United States of Americac Department of Cell and Developmental Biology, University of Massachusetts Medical School, Worcester, MA 01655, United States of America** Co-corresponding author at: Department of Biochemistry and Biophysics, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA 19104, United States of America* Co-corresponding author at: Department of Cell and Developmental Biology, University of Massachusetts Medical School, Worcester, MA 01655, United States of America8 4 2016 18 3 2016 15 9 2016 15 9 2017 1647 9 18 This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).Key challenges faced by all cells include how to spatiotemporally organize complex biochemistry and how to respond to environmental fluctuations. The budding yeast Saccharomyces cerevisiae harnesses alternative protein folding mediated by yeast prion domains (PrDs) for rapid evolution of new traits in response to environmental stress. Increasingly, it is appreciated that low complexity domains similar in amino acid composition to yeast PrDs (prion-like domains; PrLDs) found in metazoa have a prominent role in subcellular cytoplasmic organization, especially in relation to RNA homeostasis. In this review, we highlight recent advances in our understanding of the role of prions in enabling rapid adaptation to environmental stress in yeast. We also present the complete list of human proteins with PrLDs and discuss the prevalence of the PrLD in nucleic-acid binding proteins that are often connected to neurodegenerative disease, including: ataxin 1, ataxin 2, FUS, TDP-43, TAF15, EWSR1, hnRNPA1, and hnRNPA2. Recent paradigm-shifting advances establish that PrLDs undergo phase transitions to liquid states, which contribute to the structure and biophysics of diverse membraneless organelles. This structural functionality of PrLDs, however, simultaneously increases their propensity for deleterious protein-misfolding events that drive neurodegenerative disease. We suggest that even these PrLD-misfolding events are not irreversible and can be mitigated by natural or engineered protein disaggregases, which could have important therapeutic applications. RNA-binding proteinsPrion-like domainsPrionALSDisaggregasePhase transition ==== Body 1. Prions as epigenetic regulators in yeast Alternative protein folding underpins prion-based phenomena (Lindquist, 1997; Prusiner, 1998; Shorter, 2010). Prions are infectious proteins that can adopt many functionally distinct conformations, at least one of which is self-replicating (Prusiner, 1998; Shorter, 2010). Whereas the mammalian prion protein (PrP) forms prions that cause devastating neurodegenerative diseases (Aguzzi and Lakkaraju, 2016; Colby and Prusiner, 2011; Collinge and Clarke, 2007; Ma and Wang, 2014), prions in yeast are often benign and can even confer selective advantages (Liebman and Chernoff, 2012; Lindquist, 1997; Newby and Lindquist, 2013; Shorter and Lindquist, 2005; Suzuki et al., 2012; True and Lindquist, 2000). Amyloid fibril formation is a unifying feature of both mammalian and fungal prion phenomena, and indeed underpins several age-related neurodegenerative diseases including Alzheimer’s disease and Parkinson’s disease (Colby and Prusiner, 2011; Cushman et al., 2010; Guo and Lee, 2014; Shorter and Lindquist, 2005). Thus, understanding key determinants of this basic biophysical process as it relates to yeast prions will help illuminate new therapeutic strategies for human disease. The yeast prion states, [PSI+] and [URE3], are embodied by self-replicating conformers of their protein determinants Sup35 and Ure2, respectively (Cox, 1965; Tuite and Serio, 2010; Tuite et al., 2015; Wickner, 1994). Sup35 is a translation termination factor in yeast; the yeast prion state [PSI+] is associated with a nonsense suppression phenotype (Tuite et al., 2015). Prion-switching behavior of Sup35 can be readily assessed in yeast harboring a premature stop codon in their ADE1 gene: [psi~] cells (which lack Sup35 prions) are red on rich media and require adenine whereas [PSI+] cells appear white on rich media and can grow on media lacking adenine (Tuite et al., 2015). Yeast prion formation is typically mediated by modular and transferable prion domains (PrDs), which are low complexity sequences enriched in polar, uncharged amino acids such as glutamine, asparagine, tyrosine, and serine as well as glycine (Alberti et al., 2009; An et al., 2016; Edskes et al., 1999; Li and Lindquist, 2000; Michelitsch and Weissman, 2000; Santoso et al., 2000; Sondheimer and Lindquist, 2000). An exception is found in Mod5, which forms beneficial [MOD+] prions without a canonical PrD, and instead harbors an amyloid core-forming region enriched in hydrophobic residues (Suzuki and Tanaka, 2013). Importantly, deletion of the PrD from Sup35 and Ure2 does not affect normal protein function (Coschigano and Magasanik, 1991; Ter-Avanesyan et al., 1993). Conversely, PrDs can be appended to model proteins to generate novel engineered prions (Alberti et al., 2009; Li and Lindquist, 2000; Osherovich and Weissman, 2001; Tyedmers et al., 2010). Furthermore, the sequence of the PrD can be scrambled and still encode prion behavior (Ross et al., 2004, 2005). Thus, it is the amino acid composition and not the exact primary sequence of a PrD, which is critical for prionogenesis (Ross et al., 2004, 2005). The modular, transferable, and randomizable nature of PrDs has proven enormously instructive in the quest to discover novel prions. A hidden Markov model, based on the amino acid composition of PrDs of Sup35, Ure2, and Rnq1, was developed to query the S. cerevisiae genome for novel prions (Alberti et al., 2009), yielding ~200 candidates (cPrDs). To characterize these, chimeras consisting of cPrDs fused to the C-terminal domain of Sup35 were expressed in yeast and cPrDs were assessed based on their ability to mimic the [PSI+] phenotype (e.g. red-white colony color switching) (Alberti et al., 2009). Further refinement of cPrDs by biochemical and yeast phenotypic analysis led to the discovery of Mot3 prions (Alberti et al., 2009; Holmes et al., 2013). Importantly, not all cPrDs conferred prion behavior (Alberti et al., 2009), and subsequent algorithms have been used as filters to improve prediction accuracy and even to design PrDs capable of bona fide prionogenesis (Paul et al., 2015; Toombs et al., 2010, 2012). [MOT3+] is a prion state formed by self-replicating con-formers of its protein determinant, Mot3 (Alberti et al., 2009; Halfmann et al., 2012). Mot3 is a transcription factor that modulates a variety of processes including mating, carbon metabolism, and stress response by repressing anaerobic genes such as DAN1 during aerobic growth (Grishin et al., 1998). Perhaps most importantly, [MOT3+] governs the acquisition of multicellular growth phenotypes in yeast through transcriptional regulation of FLO11 (Holmes et al., 2013). [MOT3+] enables acquisition of facultative multicellular growth phenotypes including invasive growth on poor nitrogen sources, complex colony morphology as a starvation response, and flocculation in liquid media (Holmes et al., 2013). Furthermore, similar results were found in wild strains harboring [MOT3+] (Halfmann et al., 2012). These findings illustrate the potential of prions to facilitate rapid adaptation to environmental cues (Lindquist, 1997; Shorter and Lindquist, 2005). Yeast prions can confer selective advantages in various circumstances, but can also be neutral or detrimental in other settings (Du et al., 2015; Halfmann et al., 2012; Holmes et al., 2013; Newby and Lindquist, 2013; Shorter and Lindquist, 2005; Suzuki et al., 2012; Wickner et al., 2011). The beneficial phenotypes conferred by yeast prions are often observed under stress conditions, which has led to the suggestion that yeast prions constitute bet-hedging devices, which can reveal potentially adaptive genetic diversity in fluctuating environments (Du et al., 2015; Garcia and Jarosz, 2014; Halfmann et al., 2010; Halfmann and Lindquist, 2010; Masel and Bergman, 2003; Newby and Lindquist, 2013; Tyedmers et al., 2008). This process is facilitated by the conformational range of PrDs, which can access multiple, distinct cross-β structures or strains (Shorter, 2010). Moreover, protein folding is exquisitely sensitive to environment, allowing even subtle changes to favor one conformation, and therefore one function, over another (Halfmann et al., 2010, 2012; Halfmann and Lindquist, 2010; Holmes et al., 2013). In the case of Sup35 prions, [PSI+]-mediated stop codon read-through allows expression of cryptic genetic variation that accumulates in 3′ untranslated regions at many genetic loci (Baudin-Baillieu et al., 2014; Halfmann et al., 2012; Namy et al., 2008; True and Lindquist, 2000; True et al., 2004). Indeed, Sup35 prions act as evolutionary capacitors that release cryptic genetic variation under stress to facilitate the rapid evolution of adaptive traits (Masel and Siegal, 2009; Shorter and Lindquist, 2005). The adaptive significance of yeast prions, particularly Sup35 prions, has been contested (McGlinchey et al., 2011; Nakayashiki et al., 2005; Wickner et al., 2011, 2015). There has been a lack of evidence that Sup35 and Ure2 prions arise in wild yeast (although Rnq1 prions were readily found) (Chernoff et al., 2000; Resende et al., 2003), leading to speculation that prions were merely ‘diseases’ or artifacts of laboratory cultivation (Wickner et al., 2011, 2015). However, a survey of 690 wild yeast strains from diverse biological niches found 10 strains harboring Sup35 prions, 43 harboring Rnq1 prions, and 6 harboring Mot3 prions (Halfmann et al., 2012). Prions conferred a range of phenotypes that increased fitness of these yeast strains under a wide variety of stresses, and the prion phenotypes could become genetically fixed, thus fulfilling key predictions for bet-hedging prions (Halfmann et al., 2012). However, several of the advantageous [PSI+]-dependent phenotypes of wild yeast strains were not replicated in another study (Wickner et al., 2015). Nonetheless, the preponderance of evidence suggests that yeast prion proteins undergo environmentally-sensitive, alternative folding to effect epigenetic changes that increase fitness in response to fluctuating environments (Garcia and Jarosz, 2014; Halfmann et al., 2012; Newby and Lindquist, 2013; Suzuki et al., 2012). 2. Prion-like domains in humans Do human proteins contain domains similar in amino acid composition to yeast PrDs? We applied an updated PrD detection algorithm, PLAAC (for Prion-Like Amino Acid Composition, with default core length of 60 and alpha set to 0.5) (Lancaster et al., 2014), to the human genome (Ensembl GRCh38.p5, release 83) and uncovered 240 genes out of ~20,000 protein-coding genes (~1.2%) harboring a domain compositionally similar to annotated yeast PrDs, termed a prion-like domain (PrLD) (Couthouis et al., 2011; Cushman et al., 2010; Kim et al., 2013; King et al., 2012; Lancaster et al., 2014; Li et al., 2013). The complete list of human proteins with a PrLD, including the location of each PrLD, is presented in Table S1. Remarkably, 72/240 (30%) of these proteins are annotated with the gene ontology (GO) molecular function “RNA binding” and 79/240 (~33%) with the GO molecular function “DNA binding” (Table 1). These are among the nine terms from the generic GO slim (an abridged versions of the full gene ontology) that are significantly enriched for human PrLD-containing proteins (Fisher’s exact test, with Holm’s adjusted p<0.05); the others are molecular functions “transcription factor activity, protein binding,” “nucleic acid binding transcription factor activity,” and “transcription factor binding”; biological processes “chromosome organization”, “mRNA processing,” and cellular components “nucleoplasm” and ‘nucleolus’ (Table 1); 174/240 (~73%) of human PrLD-containing proteins were annotated in at least one of these categories (Fig. 1). There are also three GO Slim categories in which PrLD-containing proteins are significantly underrepresented: molecular functions “signal transducer activity” and cellular components “plasma membrane” and “mitochondrion” (Table 1). Our findings suggest that RNA-binding proteins (RBPs) and DNA-binding proteins that reside primarily in the nucleus are significantly overrepresented among the collection of PrLD-containing human proteins (King et al., 2012; Li et al., 2013). Thus, PrLDs feature prominently at the critical functional interfaces between nucleic acid and protein. Several human RBPs with PrLDs including ataxin 1, ataxin 2, TDP-43, FUS, TAF15, EWSR1, hnRNPA1, and hnRNPA2 (Table S1) feature prominently in the pathology and genetics of a number of fatal neurodegenerative diseases, including amyotrophic lateral sclerosis (ALS), frontotemporal dementia (FTD), and spinocereballar ataxias (Couthouis et al., 2011, 2012, 2014; Cushman et al., 2010; Elden et al., 2010; Kim et al., 2013; Kwiatkowski et al., 2009; Neumann et al., 2006; Orr and Zoghbi, 2007; Orr, 2012; Vance et al., 2009; Zoghbi and Orr, 2009). For example, TDP-43 mislocalizes from the nucleus to cytoplasmic inclusions in degenerating neurons in ALS and FTD, and is an intrinsically aggregation prone protein (Johnson et al., 2009; Ling et al., 2013; Neumann et al., 2006). The PrLD of TDP-43 confers this intrinsically aggregation-prone property (Johnson et al., 2009). Almost all ALS- and FTD-linked TDP-43 mutations lie in the PrLD, and several of these mutations can promote deleterious TDP-43 misfolding and enhance proteotoxicity in diverse model systems (Barmada et al., 2010; Guo et al., 2011; Johnson et al., 2009; Kabashi et al., 2010; Li et al., 2010; Lim et al., 2016; Ling et al., 2013; Ritson et al., 2010; Sreedharan et al., 2008; Zhang et al., 2009). Likewise, multisystem proteinopathy (MSP) can be caused by missense mutations in the PrLD of hnRNPA1 or hnRNPA2 (Kim et al., 2013; Shorter and Taylor, 2013). These mutations alter a gatekeeper aspartate residue and introduce a potent steric zipper motif into the PrLD, which accelerates formation of self-templating hnRNPA1 and hnRNPA2 fibrils (Kim et al., 2013; Shorter and Taylor, 2013). Furthermore, polyglutamine expansions in the PrLD of ataxin 1 cause spinocereballar ataxia 1, and promote ataxin 1 aggregation (Banfi et al., 1994; Cummings et al., 1998). These three striking examples and many others (Chesi et al., 2013; Couthouis et al., 2011, 2012, 2014; Hackman et al., 2013; Klar et al., 2013; Mori et al., 2013; Patel et al., 2015; Vieira et al., 2014) suggest that human proteins with PrLDs are prone to deleterious misfolding events that underpin neurodegenerative disease. Thus, special attention to these proteins is urgently warranted. We suggest that human proteins bearing a PrLD (Table 1) should be scrutinized as potential etiological agents of various degenerative diseases, which might be revealed via gene sequencing and histopathological examination of protein localization (King et al., 2012). Do human proteins with PrLDs form bona fide prions? Currently, there is no evidence that any human protein with a PrLD can form a prion like PrP, which can naturally transmit devastating neurodegenerative disease between individuals (Colby and Prusiner, 2011; Collinge, 1999; Prusiner, 1998). Nonetheless, PrLDs do enable proteins to spontaneously form self-templating fibrils in isolation (Kim et al., 2013). Intriguingly, ataxin 1 bearing a polyQ expansion within the PrLD can form oligomeric structures that induce local spread of ataxin 1 pathology in transgenic mice (Lasagna-Reeves et al., 2015). Moreover, TDP-43 and TDP-43 fragments containing the PrLD (193–414) can form fibrils that elicit TDP-43 aggregation in cell culture (Furukawa et al., 2011). Furthermore, detergent-insoluble fractions from ALS brains contain TDP-43 fibrils and induce TDP-43 aggregation in cell culture (Nonaka et al., 2013). Thus, TDP-43 may access a prion-like conformation, which may even be transmitted across axon terminals (Feiler et al., 2015). Indeed, phosphorylated TDP-43 pathology in ALS has been interpreted to spread in a sequential manner with highly discernible stages that might indicate involvement of axonal pathways (Brettschneider et al., 2013, 2014; Ludolph and Brettschneider, 2015). Prion-like conformers have been proposed to underlie this spreading phenomenon in ALS and other disorders (Cushman et al., 2010; Grad et al., 2015; King et al., 2012; Li et al., 2013; Maniecka and Polymenidou, 2015; Polymenidou and Cleveland, 2011, 2012; Ravits and La Spada, 2009). Although intriguing, compelling proof of formation of prions will require their de novo construction from purely synthetic protein and an ability to infect wild-type mice with a neurodegenerative disease, as has been achieved with PrP and α–synuclein (Luk et al., 2012; Wang et al., 2010, 2011a, 2011b). Even then, only PrP has been shown to form prions that can spread disease naturally between individuals in a population (Colby and Prusiner, 2011; Collinge, 1999; Prusiner, 1998). Evidence is currently lacking that α–synuclein conformers can be infectious in this way. These various connections with neurodegenerative disease have led to a negative view of PrLDs in human proteins (Cushman et al., 2010; Gitler and Shorter, 2011; King et al., 2012; Li et al., 2013). However, PrLDs are found in 240 human proteins, and so presumably may serve some beneficial or essential function. Indeed, many genes that encode proteins containing PrLDs are essential in mammals (Kraemer et al., 2010; Sephton et al., 2010; Wang et al., 2015). Furthermore, unlike the PrDs of Sup35 and Ure2, the PrLDs of several human proteins, including TDP-43, hnRNPA1, and hnRNPA2 play a critical role in protein function (Li et al., 2013). For example, the PrLD of TDP-43 is not required for RNA- or DNA-binding activity, but is critical for alternative splicing of some mRNAs and for protein-protein interactions with other hnRNPs, including hnRNPA1, hnRNPA2, and FUS, as well components of the Dicer and Drosha complexes (Ayala et al., 2005; Buratti et al., 2005; D’Ambrogio et al., 2009; Kawahara and Mieda-Sato, 2012; Kim et al., 2010). Likewise, the PrLDs of hnRNPA1 and hnRNPA2 make important contributions to the splicing activity of these proteins (Mayeda et al., 1994). Recent reports suggest that PrLDs may have an important role in critical phase transition events that provide organizational scaffolds for various membraneless organelles, including RNP granules and nuclear subcompartments (Brangwynne et al., 2015; Courchaine and Neugebauer, 2015; Guo and Shorter, 2015; Hennig et al., 2015; Kawaguchi and Hirose, 2015; Li et al., 2013; Ramaswami et al., 2013). 3. Structure and function of membraneless cellular compartments In addition to classical membrane-delimited organelles, the eukaryotic cell is also organized by membraneless organelles. These include nucleoli, Cajal bodies, gems, paraspeckles, and PML bodies in the nucleus and processing (P) bodies, stress granules, and P granules in the cytoplasm (Zhu and Brangwynne, 2015). Yet, many questions remain about the physical basis by which these compartments form and function. One hypothesis that has gained significant attention is that these membraneless organelles form through phase separation (Brangwynne et al., 2009; Brangwynne, 2013; Hyman et al., 2014; Li et al., 2012). However, precisely what phase architecture these membraneless organelles adopt has remained the subject of intense scrutiny. Early evidence that membraneless organelles may be liquid-like came from the study of P granules. In the Caenorhabditis elegans embryo, polarization along the anterior-posterior axis leads to accumulation of P granules in the embryo posterior to mark germ cells (Seydoux and Braun, 2006; Strome and Lehmann, 2007). Three-dimensional tracking of fluorescently labeled P granule components revealed that upon symmetry breaking, P granules form spontaneously in the embryo posterior in the vicinity of polarity proteins. Thus, P granule formation is driven by local decreases in the saturating concentration of P granule components in the embryo posterior and an increased flux of P granule components into the embryo posterior (Brangwynne et al., 2009). Furthermore, P granules exhibit classic liquid properties such as fusion, dripping and wetting (Brangwynne et al., 2009, 2015). Together, these data strongly argued for liquid-liquid phase separation as a mechanism for subcellular cytoplasmic architecture. A small molecule screen for compounds to promote neuronal progenitor cell differentiation into mature neurons led to the serendipitous discovery that biotinylated 5-aryl-isoxazole-3-carboxyamide (b-isox) selectively precipitates known protein components of ribonucleoprotein (RNP) granules, including TDP-43 and FUS (Kato et al., 2012). In vitro, recombinant full-length FUS phase transitions to a hydrogel-like state, but very high protein concentrations are required (Kato et al., 2012; Patel et al., 2015). Furthermore, it was observed that hydrogels formed by the PrLD of FUS were capable of retaining soluble FUS PrLD, suggesting that hydro-gel assembly is mediated by homotypic interactions in the FUS PrLD (Kato et al., 2012). Electron microscopy of FUS PrLD hydrogels revealed a composition of amyloid-like fibrils (Kato et al., 2012). X-ray diffraction analysis showed prominent reflections at 4.6–4.7 Å and 10 Å typical of cross-β structure common to amyloid fibrils, again strongly suggesting an amyloid-like structural basis to hydrogel architecture (Kato et al., 2012). These hydrogels and fibrils were readily dissolved upon exposure to SDS or mild (37 °C) heating, suggesting that they were distinct and more dynamic than the SDS-resistant amyloid fibrils formed by yeast prions (Kato et al., 2012; Kryndushkin et al., 2003; Serio et al., 2000; Shorter and Lindquist, 2006). This finding was consistent with previous work demonstrating that purified FUS spontaneously assembles into SDS-soluble fibrils (Sun et al., 2011). However, whether the hydrogels formed in vitro were reflective of RNP granules found in cells remained uncertain (Weber and Brangwynne, 2012). Multiple reports have suggested that RNA granules exist as predominantly liquid compartments and not gel-like compartments in cells (Burke et al., 2015; Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015; Zhang et al., 2015). For example, in cells and in vitro, FUS forms dynamic assemblies that are rapidly recruited to sites of DNA damage, and display physical characteristics of liquid droplets as predicted by classical physics of polymeric phase transitions, including fast internal dynamics, spherical morphology, and a propensity for two droplets to readily fuse when in close contact with one another (Altmeyer et al., 2015; Burke et al., 2015; Lin et al., 2015; Murakami et al., 2015; Patel et al., 2015). Importantly, the PrLD mediates the phase transition as deletion of the PrLD abrogates droplet assembly. Detailed study of RBPs hnRNPA1, Lsm4, Tia1, and Pub1 revealed similar behavior (Burke et al., 2015; Lin et al., 2015; Molliex et al., 2015). Droplet formation was influenced by the presence of molecular crowding agents, ionic strength, and presence of RNA or other polyanions, such as poly-ADP ribose (Burke et al., 2015; Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015; Zhang et al., 2015). The identity of the bound RNA tunes the biophysical properties of the RNP granule (Zhang et al., 2015). In the filamentous fungus Ashbya gossypii, the RBP Whi3 possesses a PrLD that enables assembly into liquid droplets to organize cyclin transcripts (CLN3) at sites of nuclear division and formin transcripts (BNI1) at polarity sites where new branch sites are located (Zhang et al., 2015). Microrheology studies revealed that Whi3 droplets bound to BNI1 are less viscous than CLN3-bound droplets (Zhang et al., 2015). Additionally, BNI1 droplets fuse with one another faster than CLN3 droplets (Zhang et al., 2015). This suggests that client RNA identity is critical to tuning the biophysical properties of RNP granules, and that different physical properties of RNP granules may be optimized for specific cellular functions (Guo and Shorter, 2015; Zhang et al., 2015). This latter point raises a provocative parallel with yeast prions, where conformational diversity gives rise to distinct prion strains with unique phenotypes (King and Diaz-Avalos, 2004; Roberts et al., 2009; Shorter, 2010; Tanaka et al., 2004). It is possible that bound RNAs may define ‘strains’ of RNP granules by tuning their biophysical properties. The case that biophysical properties of RNP granules reflect their functional role is perhaps most strikingly made in S. cerevisiae. Specifically, whereas P bodies are constitutively active sites involved in mRNA processing and degradation and exist as liquid droplets in yeast, stress granules are inactive storage sites for proteins and RNA that form rapidly upon onset of stress, and are solid, gel-like aggregates (Balagopal and Parker, 2009; Kroschwald et al., 2015). Yeast rely on the protein disaggregase and hexameric AAA+ ATPase, Hsp104 (DeSantis and Shorter, 2012), to maintain the fluidity of P bodies (Kroschwald et al., 2015). Metazoa lack an Hsp104 homolog (Erives and Fassler, 2015; Shorter, 2008), and curiously in mammalian cells P bodies and stress granules exist in more liquid-like states (Kroschwald et al., 2015). Thus, the powerful disaggregase activity of Hsp104 may enable yeast cells to readily exploit solid, gel, and liquid states in RNP granules (Kroschwald et al., 2015). What structure(s) do the PrLDs of RNA-binding proteins adopt in liquid droplets? Solution NMR study of the FUS PrLD in monodisperse solution and condensed into droplets revealed that the PrLD retains disordered character in droplets (Burke et al., 2015). This finding suggests a model in which these liquid droplets maintain rapid internal dynamics while being held together by transient intermolecular contacts between adjacent PrLDs (Burke et al., 2015). In an attempt to further address this question, a mass spectrometry-based chemical footprinting method has been employed in which N-acetylimidazole (NAI) is used to acetylate serine, tyrosine, lysine, threonine, arginine, and asparagine side chains in proteins (Xiang et al., 2015). Using two model proteins, recombinant glutathione-S transferase (GST) and poly-ADP-ribose polymerase (PARP) isolated from HEK 293T cell nuclei, it was shown that solvent-accessible side chains (as assessed by available crystallographic data for these two proteins) are more readily acetylated (Xiang et al., 2015). Thus, the acetylation pattern or “footprint” for a given protein can be used as a conformational proxy (Xiang et al., 2015). This technique was then deployed to show that the PrLD of hnRNPA2 in droplets or polymerized into hydrogels in vitro, or hnRNPA2 isolated from nuclei adopt similar conformations, as assessed by their NAI footprints (Xiang et al., 2015). This congruence might suggest that cross-β fibrillization underpins both phase transitions to liquid droplets and hydrogel formation (Xiang et al., 2015). Curiously, however, hnRNPA2 PrLD fused to maltose binding protein (MBP) to maintain the hnRNPA2 PrLD in a soluble, monomeric state also displayed a footprint qualitatively similar to that obtained for hydrogels, liquid droplets, and fibrils (Xiang et al., 2015). Therefore, it is difficult to interpret precisely how the hnRNPA2 chemical footprints relate to hnRNPA2 structure. One possibility is that the observed NAI footprint in the monomeric state may be due to contamination by small amounts of fibrils. This possibility is supported by the fact that the intensity of the footprint progressively increased with time after MBP cleavage, and suggests that the NAI footprinting method detects hnRNPA2 fibril abundance. Alternatively, a fraction of hnRNPA2 PrLDs might exhibit cross-β structure even in the context of monomeric, soluble hnRNPA2. Although the hnRNPA2 PrLD is predicted to be intrinsically disordered in the soluble, monomeric state (Kim et al., 2013), circular dichroism studies suggest that this domain may adopt β-sheet-rich structures in solution (Landsberg et al., 2006). Another possibility is that similar regions within the hnRNPA2 PrLD may be invariably solvent accessible (or inaccessible) in distinct structures for monomeric forms in solution, in liquid phases, and in cross-β fibrils, and consequently the NAI footprint does not resolve between them. We suggest that further structural studies using complementary techniques in addition to NAI foot-printing are needed to further resolve the structure of the hnRNPA2 PrLD in various soluble, liquid, and fibrillar states. Parker and colleagues observed that stress granules contain stable subcompartments that can be isolated in cell lysates, thus suggesting that stress granules are not simply liquids (Jain et al., 2016). Instead, they propose that stress granules contain stable, gel-like cores surrounded by a dynamic liquid shell (Jain et al., 2016). They demonstrate that the cores of stress granules isolated from yeast are larger than their counterparts from mammalian cells, thus reconciling their data with that of Alberti and colleagues (Kroschwald et al., 2015). The existence of a stable, gel-like core within stress granules may even provide an explanation for the observation by McKnight and colleagues that nuclear hnRNPA2 has a similar chemical footprint to recombinant hnRNPA2 polymerized into hydrogels (Xiang et al., 2015). However, it remains unclear whether cross-β polymerization is at the root of the stable stress granule core (Jain et al., 2016; Xiang et al., 2015). Thus, the structure of PrLDs within RNP granules is likely to remain the subject of intense focus in future studies. Moreover, other membraneless organelles, such as the nucleolus, contain distinct subcompartments (Boisvert et al., 2007). It will be important to determine whether these are due to separated gel and liquid phases, or immiscible liquid phases with different viscosities. 4. Membraneless organelles and prion-like domains: a mechanistic link between normal physiology and neurodegenerative disease What is the connection between the functional role of PrLDs in beneficial phase transitions in the formation of membraneless organelles and the deleterious misfolding events that these domains undergo in neurodegenerative disease? Recent work in vitro demonstrates that liquid droplets composed of FUS and hnRNPA1 harboring disease-linked mutations in their PrLDs (FUSG156E and hnRNPA1D262V) mature to a solid, hydrogel-like state more rapidly than droplets formed by wild-type protein (Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015). Moreover, hydrogel-like forms of mutant FUS have been specifically associated with neurodegenerative phenotypes in a C. elegans model of FUS proteinopathy (Murakami et al., 2015). Thus, a direct link emerges between the biophysical propensity of these proteins to adopt more solid-like structures and neurodegeneration. Interestingly, the majority of ALS-causing mutations in FUS are found not in the PrLD but in the C-terminal region (Da Cruz and Cleveland, 2011; Kwiatkowski et al., 2009; Vance et al., 2009), where they disrupt a proline-tyrosine (PY) nuclear localization signal (Lee et al., 2006; Suel et al., 2008; Zhang and Chook, 2012). Mutations in the FUS PY-NLS lead to persistent cytoplasmic FUS mislocalization, which correlates with ALS severity (Dormann et al., 2010), but curiously does not directly increase the biophysical propensity of FUS to aggregate (Sun et al., 2011). However, the biophysics of membraneless organelle assembly shed new light on this observation. Membraneless organelles such as nucleoli and P granules in Caenorhabditis elegans have the property that local concentration of granular components drive granule droplet condensation (Brangwynne et al., 2015; Weber and Brangwynne, 2015; Zhu and Brangwynne, 2015). Thus, a model emerges where impaired nuclear import of FUS leads to persistent cytoplasmic FUS droplets that mature to more intractable solid aggregates, in accordance with findings in vitro (Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015). Therefore, strategies to boost FUS nuclear import or maintain FUS droplet fluidity should represent a significant therapeutic opportunity for ALS and FTD. 5. Clearance mechanisms for solid protein aggregates In S. cerevisiae, the protein-remodeling factor Hsp104 regulates the formation, elimination, and propagation of beneficial yeast prions (Sweeny and Shorter, 2008). Hsp104 severs yeast prions to ensure their dissemination to daughter cells upon division (Satpute-Krishnan et al., 2007; Shorter and Lindquist, 2004, 2006, 2008; Sweeny et al., 2015; Sweeny and Shorter, 2015), and is required to clear solid stress granules in yeast and to maintain the fluid integrity of yeast P bodies (Cherkasov et al., 2013; Kroschwald et al., 2015). Indeed, it is hypothesized that this robust disaggregase machinery coevolved with solid stress granules as a way for yeast to cope with their vulnerability to environmental fluctuation (Kroschwald et al., 2015). However, metazoa lack a clear Hsp104 homolog (Erives and Fassler, 2015), and fibrils formed from human RBPs with PrLDs represent an intractable substrate for wild-type Hsp104 (Jackrel et al., 2014). Recently, however, engineered forms of Hsp104 have been generated that potently suppress the aggregation and toxicity of various disease-linked RBPs with PrLDs, including TDP-43, FUS, and TAF15 (Jackrel et al., 2014, 2015; Jackrel and Shorter, 2014a; Sweeny and Shorter, 2015; Torrente et al., 2016). Moreover, these potentiated Hsp104 variants dissolve preformed TDP-43, FUS, and TAF15 fibrils in vitro (Jackrel et al., 2014; Jackrel and Shorter, 2014a). Thus, Hsp104 could represent a disruptive technology to enhance metazoan proteostasis to counter RBP misfolding that causes neurodegenerative disease such as ALS and FTD (Jackrel and Shorter, 2014b, 2015). In metazoan cells and yeast, the Hsp110, Hsp70, and Hsp40 protein-disaggregase machinery (Nillegoda and Bukau, 2015; Shorter, 2011; Torrente and Shorter, 2013), contributes to the clearance of stress granules (Cherkasov et al., 2013; Kroschwald et al., 2015; Walters et al., 2015). Hsp70 and Hsp40 chaperones often get sequestered and inactivated by misfolded protein aggregates (Auluck et al., 2002; Derkatch and Liebman, 2013; Yu et al., 2014). Thus, enhancement or engineering of this disaggregase machinery might also open potential therapeutic avenues for ALS, FTD, and a variety of other neurodegenerative disorders. Deletion of several autophagy-related genes and Cdc48 adaptor proteins gives rise to constitutive stress granules in S. cerevisiae (Buchan et al., 2013). Cdc48, or valosin-containing protein (VCP) in humans, is another hexameric AAAþ ATPase that can promote autophagy (Ju et al., 2009; Krick et al., 2010; Meyer et al., 2012), and mutations in VCP cause familial forms of ALS and MSP (Johnson et al., 2010). These diseases are characterized by the formation of cytoplasmic protein aggregates containing RBP components of stress granules (Li et al., 2013). Deletion of the autophagy gene ATG7 or siRNA-mediated knockdown of VCP leads to impaired ability to clear stress granules in mammalian cells, and disease-causing mutations in VCP cause accumulation of constitutive stress granules that contain TDP-43 (Buchan et al., 2013). However, it remains unclear whether the sole role of Cdc48/VCP is to target stress granule components for autophagic degradation or whether Cdc48/VCP may have also have an active role in disaggregation of stress granules and reactivation of their components (Buchan et al., 2013). Indeed, an exciting possibility is that Cdc48/VCP may represent a triage center for stress granule components, effecting the reactivation of salvageable components and the degradation of others. 6. Concluding remarks Here, we have reviewed recent advances in our understanding of prion-like phenomena and architecture across biology, from epigenetic regulation in the simple model organism S. cerevisiae to complex mechanisms of eukaryotic subcellular organization. We suggest that PrDs and PrLDs may have been biologically conserved for their wide-ranging biological utility. The ability of prions to rapidly switch between distinct conformational and functional states confers selective advantages for yeast in the face of environmental stress (Halfmann et al., 2012; Holmes et al., 2013; Suzuki et al., 2012; True and Lindquist, 2000; True et al., 2004). In fungi and metazoa, PrLDs now have a clear functional role in mediating the reversible coalescence of RNP granules. However, in humans the persistence and maturation of these RNP granules via complex mechanisms leads to pathological protein accumulation and neurodegenerative disease (Fig. 2) (Guo and Shorter, 2015; Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015; Xiang et al., 2015). We suggest that PrLDs may be general scaffolds for membraneless subcellular organization. However, this activity places PrLDs at risk to accessing deleterious misfolding trajectories that cause neurodegenerative disease (Li et al., 2013). PrLDs have a very distinctive amino acid composition, but this role in subcellular compartmentalization via phase transitions and simultaneous risk of protein misfolding may extend to other intrinsically unfolded, low complexity domains with different amino acid composition. Regardless, we suggest that deleterious misfolding events can be reversed by select protein disaggregases, which could have important therapeutic applications (Jackrel and Shorter, 2015; Shorter, 2008; Torrente and Shorter, 2013). Supplementary Material 1 We thank Lin Guo, Alice Ford, and Korrie Mack for comments on the manuscript and Mary Leonard for artwork. J.S. was supported by NIH Grant R01GM099836, a Muscular Dystrophy Association Research Award (MDA277268), the ALS association (15-IIP-214), the Life Extension Foundation, the Packard Center for ALS Research at Johns Hopkins University, and Target ALS. Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.brainres.2016.02.037. Fig. 1 Associations between human PrLD-containing proteins and the Gene Ontology categories enriched for them. Columns correspond to the nine GO Slim categories enriched for human PrLD-containing proteins (Table 1), and rows correspond to the PrLD-containing proteins; the cell indexed by a given row and column is colored blue if the corresponding protein is annotated as belonging to the corresponding category, and gray otherwise. The rows and columns are hierarchically clustered based on correlation of GO Slim annotations. The PLAAC COREscore is also indicated in the far right column using a red color gradient with ranging from black (score 0) to saturated red (score 64), as indicated by the color bar. Rows and columns are ordered by a correlation-based clustering. The 66 of 240 PrLD-containing proteins not associated with any of these categories are not shown (the full list of human proteins with PrLDs is presented in Table S1). Fig. 2 Phase transitions of prion-like domains. RNA-binding proteins (RNA recognition motif depicted by purple circles) can reversibly coalesce into dynamic liquid droplets through transient interactions in their prion-like domains (PrLDs depicted as purple lines). Droplet persistence over time, or mutations in PrLDs that introduce potent steric zippers, can drive further condensation of aged droplets into a less dynamic form that can give rise to solid fibrillar protein aggregates that accrue in neurodegenerative disease. Table 1 Gene Ontology categories in which human PrLD-containing proteins are significantly over- or under-represented. Fisher’s exact test was used to test for independence between the set of human proteins with PrLDs and GO Slim gene annotations for 129 categories (14 with fewer than 5 or more than 5000 annotated genes were excluded); the categories with p-value<0.05 after adjusting for multiple hypothesis testing with Holm’s method are shown. Columns give the GO ID, branch of the ontology, GO term, number of genes with this annotation overall (n.total) and among PrLD containing-proteins (n.PrLD), odds ratio, raw p-value, and Holm’s-adjusted p-value. Only the longest protein-coding transcript for each HUGO gene name was included in the analyses to avoid redundancies, and 906 genes with no GO Slim annotations (including 4 with PrLDs) were excluded from the gene universe, as were those encoding proteins of fewer than 60 amino acids, leaving 18,672 genes in the universe, 236 of them with PrLDs. GO.id GO.branch GO.term n.total n.PrLD Odds.ratio p.raw p.adjusted Over-represented   GO:0005654 CC Nucleoplasm 2879 116 5.48 2.2e–34 2.9e–32   GO:0003723 MF RNA binding 1545 72 5.05 2.5e–23 3.2e–21   GO:0000988 MF Transcription factor activity, protein binding 516 37 6.97 8.8e–18 1.1e–15   GO:0003677 MF DNA binding 2320 79 3.64 2.3e–17 2.9e–15   GO:0006397 BP mRNA processing 430 31 6.83 4.4e–15 5.4e–13   GO:0051276 BP Chromosome organization 971 44 4.33 1.1e–13 1.4e–11   GO:0001071 MF Nucleic acid binding transcription factor activity 1077 39 3.32 2.5e–09 3.1e–07   GO:0008134 MF Transcription factor binding 432 16 3.15 1.3e–04 1.6e–02   GO:0005730 CC Nucleolus 810 23 2.42 2.8e–04 3.4e–02 Under-represented   GO:0005886 CC Plasma membrane 4539 23 0.33 1.6e–08 1.9e–06   GO:0004871 MF Signal transducer activity 1566 4 0.19 2.5e–05 3.0e–03   GO:0005739 CC Mitochondrion 1319 4 0.22 2.8e–04 3.4e–02 ==== Refs REFERENCES Aguzzi A Lakkaraju AK Cell biology of prions and prionoids: a status report Trends Cell Biol 2016 26 40 51 26455408 Alberti S Halfmann R King O Kapila A Lindquist S A systematic survey identifies prions and illuminates sequence features of prionogenic proteins Cell 2009 137 146 158 19345193 Altmeyer M Neelsen KJ Teloni F Pozdnyakova I Pellegrino S Grofte M Rask MB Streicher W Jungmichel S Nielsen ML Lukas J Liquid demixing of intrinsically disordered proteins is seeded by poly(ADP-ribose) Nat. Commun 2015 6 8088 26286827 An L Fitzpatrick D Harrison PM Emergence and evolution of yeast prion and prion-like proteins BMC Evol. Biol 2016 16 24 26809710 Auluck PK Chan HY Trojanowski JQ Lee VM Bonini NM Chaperone suppression of alpha-synuclein toxicity in a Drosophila model for Parkinson’s disease Science 2002 295 865 868 11823645 Ayala YM Pantano S D’Ambrogio A Buratti E Brindisi A Marchetti C Romano M Baralle FE Human, Drosophila, and C.elegans TDP43: nucleic acid binding properties and splicing regulatory function J. Mol. Biol 2005 348 575 588 15826655 Balagopal V Parker R Polysomes, P bodies and stress granules: states and fates of eukaryotic mRNAs Curr. Opin. Cell Biol 2009 21 403 408 19394210 Banfi S Servadio A Chung MY Kwiatkowski TJ Jr McCall AE Duvick LA Shen Y Roth EJ Orr HT Zoghbi HY Identification and characterization of the gene causing type 1 spinocerebellar ataxia Nat. Genet 1994 7 513 520 7951322 Barmada SJ Skibinski G Korb E Rao EJ Wu JY Finkbeiner S Cytoplasmic mislocalization of TDP-43 is toxic to neurons and enhanced by a mutation associated with familial amyotrophic lateral sclerosis J. Neurosci 2010 30 639 649 20071528 Baudin-Baillieu A Legendre R Kuchly C Hatin I Demais S Mestdagh C Gautheret D Namy O Genome-wide translational changes induced by the prion [PSI+] Cell Rep 2014 8 439 448 25043188 Boisvert FM van Koningsbruggen S Navascues J Lamond AI The multifunctional nucleolus Nat. Rev. Mol. Cell Biol 2007 8 574 585 17519961 Brangwynne CP Phase transitions and size scaling of membrane-less organelles J. Cell Biol 2013 203 875 881 24368804 Brangwynne CP Eckmann CR Courson DS Rybarska A Hoege C Gharakhani J Julicher F Hyman AA Germline P granules are liquid droplets that localize by controlled dissolution/condensation Science 2009 324 1729 1732 19460965 Brangwynne Clifford P Tompa P Pappu Rohit V Polymer physics of intracellular phase transitions Nat. Phys 2015 11 899 904 Brettschneider J Arai K Del Tredici K Toledo JB Robinson JL Lee EB Kuwabara S Shibuya K Irwin DJ Fang L Van Deerlin VM Elman L McCluskey L Ludolph AC Lee VM Braak H Trojanowski JQ TDP-43 pathology and neuronal loss in amyotrophic lateral sclerosis spinal cord Acta Neuropathol 2014 128 423 437 Brettschneider J Del Tredici K Toledo JB Robinson JL Irwin DJ Grossman M Suh E Van Deerlin VM Wood EM Baek Y Kwong L Lee EB Elman L McCluskey L Fang L Feldengut S Ludolph AC Lee VM Braak H Trojanowski JQ Stages of pTDP-43 pathology in amyotrophic lateral sclerosis Ann. Neurol 2013 74 20 38 23686809 Buchan JR Kolaitis RM Taylor JP Parker R Eukaryotic stress granules are cleared by autophagy and Cdc48/VCP function Cell 2013 153 1461 1474 23791177 Buratti E Brindisi A Giombi M Tisminetzky S Ayala YM Baralle FE TDP-43 binds heterogeneous nuclear ribonucleoprotein A/B through its C-terminal tail: an important region for the inhibition of cystic fibrosis transmembrane conductance regulator exon 9 splicing J Biol. Chem 2005 280 37572 37584 16157593 Burke KA Janke AM Rhine CL Fawzi NL Residue-by-Residue View of In Vitro FUS Granules that Bind the C-Terminal Domain of RNA Polymerase II Mol. Cell 2015 60 231 241 26455390 Cherkasov V Hofmann S Druffel-Augustin S Mogk A Tyedmers J Stoecklin G Bukau B Coordination of translational control and protein homeostasis during severe heat stress Curr. Biol 2013 23 2452 2462 24291094 Chernoff YO Galkin AP Lewitin E Chernova TA Newnam GP Belenkiy SM Evolutionary conservation of prion-forming abilities of the yeast Sup35 protein Mol. Microbiol 2000 35 865 876 10692163 Chesi A Staahl BT Jovicic A Couthouis J Fasolino M Raphael AR Yamazaki T Elias L Polak M Kelly C Williams KL Fifita JA Maragakis NJ Nicholson GA King OD Reed R Crabtree GR Blair IP Glass JD Gitler AD Exome sequencing to identify de novo mutations in sporadic ALS trios Nat. Neurosci 2013 16 851 855 23708140 Colby DW Prusiner SB Prions Cold Spring Harb. Perspect. Biol 2011 3 a006833 21421910 Collinge J Variant Creutzfeldt-Jakob disease Lancet 1999 354 317 323 10440324 Collinge J Clarke AR A general model of prion strains and their pathogenicity Science 2007 318 930 936 17991853 Coschigano PW Magasanik B The URE2 gene product of Saccharomyces cerevisiae plays an important role in the cellular response to the nitrogen source and has homology to glutathione s-transferases Mol. Cell Biol 1991 11 822 832 1990286 Courchaine E Neugebauer KM Paraspeckles: paragons of functional aggregation J Cell Biol 2015 210 527 528 26283795 Couthouis J Raphael AR Daneshjou R Gitler AD Targeted exon capture and sequencing in sporadic amyotrophic lateral sclerosis PLoS Genet 2014 10 e1004704 25299611 Couthouis J Hart MP Erion R King OD Diaz Z Nakaya T Ibrahim F Kim HJ Mojsilovic-Petrovic J Panossian S Kim CE Frackelton EC Solski JA Williams KL Clay-Falcone D Elman L McCluskey L Greene R Hakonarson H Kalb RG Lee VM Trojanowski JQ Nicholson GA Blair IP Bonini NM Van Deerlin VM Mourelatos Z Shorter J Gitler AD Evaluating the role of the FUS/ TLS-related gene EWSR1 in amyotrophic lateral sclerosis Hum. Mol. Genet 2012 21 2899 2911 22454397 Couthouis J Hart MP Shorter J DeJesus-Hernandez M Erion R Oristano R Liu AX Ramos D Jethava N Hosangadi D Epstein J Chiang A Diaz Z Nakaya T Ibrahim F Kim HJ Solski JA Williams KL Mojsilovic-Petrovic J Ingre C Boylan K Graff-Radford NR Dickson DW Clay-Falcone D Elman L McCluskey L Greene R Kalb RG Lee VM Trojanowski JQ Ludolph A Robberecht W Andersen PM Nicholson GA Blair IP King OD Bonini NM Van Deerlin V Rademakers R Mourelatos Z Gitler AD A yeast functional screen predicts new candidate ALS disease genes Proc. Natl. Acad. Sci. USa 2011 108 20881 20890 22065782 Cox BS [PSI], a cytoplasmic suppressor of super-suppression in yeast Heredity 1965 20 505 521 Cummings CJ Mancini MA Antalffy B DeFranco DB Orr HT Zoghbi HY Chaperone suppression of aggregation and altered subcellular proteasome localization imply protein misfolding in SCA1 Nat. Genet 1998 19 148 154 9620770 Cushman M Johnson BS King OD Gitler AD Shorter J Prion-like disorders: blurring the divide between trans-missibility and infectivity J. Cell Sci 2010 123 1191 1201 20356930 D’Ambrogio A Buratti E Stuani C Guarnaccia C Romano M Ayala YM Baralle FE Functional mapping of the interaction between TDP-43 and hnRNP A2 in vivo Nucleic Acids Res 2009 37 4116 4126 19429692 Da Cruz S Cleveland DW Understanding the role of TDP-43 and FUS/TLS in ALS and beyond Curr. Opin. Neurobiol 2011 21 904 919 21813273 Derkatch IL Liebman SW The story of stolen chaperones: how overexpression of Q/N proteins cures yeast prions Prion 2013 7 294 300 23924684 DeSantis ME Shorter J The elusive middle domain of Hsp104 and ClpB: location and function Biochim. Biophys. Acta 2012 1823 29 39 21843558 Dormann D Rodde R Edbauer D Bentmann E Fischer I Hruscha A Than ME Mackenzie IR Capell A Schmid B Neumann M Haass C ALS-associated fused in sarcoma (FUS) mutations disrupt Transportin-mediated nuclear import EMBO J 2010 29 2841 2857 20606625 Du Z Zhang Y Li L The yeast prion [SWI(+)] abolishes multicellular growth by triggering conformational changes of multiple regulators required for flocculin gene expression Cell Rep 2015 13 2865 2878 26711350 Edskes HK Gray VT Wickner RB The [URE3] prion is an aggregated form of Ure2p that can be cured by overexpression of Ure2p fragments Proc. Natl. Acad. Sci. USA 1999 96 1498 1503 9990052 Elden AC Kim HJ Hart MP Chen-Plotkin AS Johnson BS Fang X Armakola M Geser F Greene R Lu MM Padmanabhan A Clay-Falcone D McCluskey L Elman L Juhr D Gruber PJ Rub U Auburger G Trojanowski JQ Lee VM Van Deerlin VM Bonini NM Gitler AD Ataxin-2 intermediate-length polyglutamine expansions are associated with increased risk for ALS Nature 2010 466 1069 1075 20740007 Erives AJ Fassler JS Metabolic and chaperone gene loss marks the origin of animals: evidence for Hsp104 and Hsp78 chaperones sharing mitochondrial enzymes as clients PLoS One 2015 10 e0117192 25710177 Feiler MS Strobel B Freischmidt A Helferich AM Kappel J Brewer BM Li D Thal DR Walther P Ludolph AC Danzer KM Weishaupt JH TDP-43 is intercellularly transmitted across axon terminals J. Cell Biol 2015 211 897 911 26598621 Furukawa Y Kaneko K Watanabe S Yamanaka K Nukina N A seeding reaction recapitulates intracellular formation of Sarkosyl-insoluble transactivation response element (TAR) DNA-binding protein-43 inclusions J. Biol. Chem 2011 286 18664 18672 21454603 Garcia DM Jarosz DF Rebels with a cause: molecular features and physiological consequences of yeast prions FEMS Yeast Res 2014 14 136 147 25667942 Gitler AD Shorter J RNA-binding proteins with prion-like domains in ALS and FTLD-U Prion 2011 5 179 187 21847013 Grad LI Fernando SM Cashman NR From molecule to molecule and cell to cell: prion-like mechanisms in amyotrophic lateral sclerosis Neurobiol. Dis 2015 77 257 265 25701498 Grishin AV Rothenberg M Downs MA Blumer KJ Mot3, a Zn finger transcription factor that modulates gene expression and attenuates mating pheromone signaling in Saccharomyces cerevisiae Genetics 1998 149 879 892 9611199 Guo JL Lee VM Cell-to-cell transmission of pathogenic proteins in neurodegenerative diseases Nat. Med 2014 20 130 138 24504409 Guo L Shorter J It’s raining liquids: RNA tunes viscoelasticity and dynamics of membraneless organelles Mol. Cell 2015 60 189 192 26474062 Guo W Chen Y Zhou X Kar A Ray P Chen X Rao EJ Yang M Ye H Zhu L Liu J Xu M Yang Y Wang C Zhang D Bigio EH Mesulam M Shen Y Xu Q Fushimi K Wu JY An ALS-associated mutation affecting TDP-43 enhances protein aggregation, fibril formation and neuro-toxicity Nat. Struct. Mol. Biol 2011 18 822 830 21666678 Hackman P Sarparanta J Lehtinen S Vihola A Evila A Jonson PH Luque H Kere J Screen M Chinnery PF Ahlberg G Edstrom L Udd B Welander distal myopathy is caused by a mutation in the RNA-binding protein TIA1 Ann. Neurol 2013 73 500 509 23401021 Halfmann R Lindquist S Epigenetics in the extreme: prions and the inheritance of environmentally acquired traits Science 2010 330 629 632 21030648 Halfmann R Alberti S Lindquist S Prions, protein homeostasis, and phenotypic diversity Trends Cell Biol 2010 20 125 133 20071174 Halfmann R Jarosz DF Jones SK Chang A Lancaster AK Lindquist S Prions are a common mechanism for phenotypic inheritance in wild yeasts Nature 2012 482 363 368 22337056 Hennig S Kong G Mannen T Sadowska A Kobelke S Blythe A Knott GJ Iyer KS Ho D Newcombe EA Hosoki K Goshima N Kawaguchi T Hatters D Trinkle-Mulcahy L Hirose T Bond CS Fox AH Prion-like domains in RNA binding proteins are essential for building subnuclear paraspeckles J. Cell Biol 2015 210 529 539 26283796 Holmes DL Lancaster AK Lindquist S Halfmann R Heritable remodeling of yeast multicellularity by an environmentally responsive prion Cell 2013 153 153 165 23540696 Hyman AA Weber CA Julicher F Liquid-liquid phase separation in biology Annu. Rev. Cell Dev. Biol 2014 30 39 58 25288112 Jackrel ME Shorter J Potentiated Hsp104 variants suppress toxicity of diverse neurodegenerative disease-linked proteins Dis. Model. Mech 2014a 7 1175 1184 25062688 Jackrel ME Shorter J Reversing deleterious protein aggregation with re-engineered protein disaggregases Cell Cycle 2014b 13 1379 1383 24694655 Jackrel ME Shorter J Engineering enhanced protein disaggregases for neurodegenerative disease Prion 2015 9 90 109 25738979 Jackrel ME Yee K Tariq A Chen AI Shorter J Disparate mutations confer therapeutic gain of Hsp104 function ACS Chem. Biol 2015 10 2672 2679 26441009 Jackrel ME DeSantis ME Martinez BA Castellano LM Stewart RM Caldwell KA Caldwell GA Shorter J Potentiated Hsp104 variants antagonize diverse proteotoxic misfolding events Cell 2014 156 170 182 24439375 Jain S Wheeler JR Walters RW Agrawal A Barsic A Parker R ATPase-modulated stress granules contain a diverse proteome and substructure Cell 2016 164 487 498 26777405 Johnson BS Snead D Lee JJ McCaffery JM Shorter J Gitler AD TDP-43 is intrinsically aggregation-prone, and amyotrophic lateral sclerosis-linked mutations accelerate aggregation and increase toxicity J. Biol. Chem 2009 284 20329 20339 19465477 Johnson JO Mandrioli J Benatar M Abramzon Y Van Deerlin VM Trojanowski JQ Gibbs JR Brunetti M Gronka S Wuu J Ding J McCluskey L Martinez-Lage M Falcone D Hernandez DG Arepalli S Chong S Schymick JC Rothstein J Landi F Wang YD Calvo A Mora G Sabatelli M Monsurro MR Battistini S Salvi F Spataro R Sola P Borghero G Consortium I Galassi G Scholz SW Taylor JP Restagno G Chio A Traynor BJ Exome sequencing reveals VCP mutations as a cause of familial ALS Neuron 2010 68 857 864 21145000 Ju JS Fuentealba RA Miller SE Jackson E Piwnica-Worms D Baloh RH Weihl CC Valosin-containing protein (VCP) is required for autophagy and is disrupted in VCP disease J. Cell Biol 2009 187 875 888 20008565 Kabashi E Lin L Tradewell ML Dion PA Bercier V Bourgouin P Rochefort D Bel Hadj S Durham HD Vande Velde C Rouleau GA Drapeau P Gain and loss of function of ALS-related mutations of TARDBP (TDP-43) cause motor deficits in vivo Hum. Mol. Genet 2010 19 671 683 19959528 Kato M Han TW Xie S Shi K Du X Wu LC Mirzaei H Goldsmith EJ Longgood J Pei J Grishin NV Frantz DE Schneider JW Chen S Li L Sawaya MR Eisenberg D Tycko R McKnight SL Cell-free formation of RNA granules: low complexity sequence domains form dynamic fibers within hydrogels Cell 2012 149 753 767 22579281 Kawaguchi T Hirose T Chromatin remodeling complexes in the assembly of long noncoding RNA-dependent nuclear bodies Nucleus 2015 1 6 25644654 Kawahara Y Mieda-Sato A TDP-43 promotes microRNA biogenesis as a component of the Drosha and Dicer complexes Proc. Natl. Acad. Sci. USA 2012 109 3347 3352 22323604 Kim HJ Kim NC Wang YD Scarborough EA Moore J Diaz Z MacLea KS Freibaum B Li S Molliex A Kanagaraj AP Carter R Boylan KB Wojtas AM Rademakers R Pinkus JL Greenberg SA Trojanowski JQ Traynor BJ Smith BN Topp S Gkazi AS Miller J Shaw CE Kottlors M Kirschner J Pestronk A Li YR Ford AF Gitler AD Benatar M King OD Kimonis VE Ross ED Weihl CC Shorter J Taylor JP Mutations in prion-like domains in hnRNPA2B1 and hnRNPA1 cause multisystem proteinopathy and ALS Nature 2013 495 467 473 23455423 Kim SH Shanware NP Bowler MJ Tibbetts RS Amyotrophic lateral sclerosis-associated proteins TDP-43 and FUS/TLS function in a common biochemical complex to co-regulate HDAC6 mRNA J. Biol. Chem 2010 285 34097 34105 20720006 King CY Diaz-Avalos R Protein-only transmission of three yeast prion strains Nature 2004 428 319 323 15029195 King OD Gitler AD Shorter J The tip of the iceberg: RNA-binding proteins with prion-like domains in neurodegenerative disease Brain Res 2012 1462 61 80 22445064 Klar J Sobol M Melberg A Mabert K Ameur A Johansson AC Feuk L Entesarian M Orlen H Casar-Borota O Dahl N Welander distal myopathy caused by an ancient founder mutation in TIA1 associated with perturbed splicing Hum. Mutat 2013 34 572 577 23348830 Kraemer BC Schuck T Wheeler JM Robinson LC Trojanowski JQ Lee VM Schellenberg GD Loss of murine TDP-43 disrupts motor function and plays an essential role in embryogenesis Acta Neuropathol 2010 119 409 419 20198480 Krick R Bremer S Welter E Schlotterhose P Muehe Y Eskelinen EL Thumm M Cdc48/p97 and Shp1/p47 regulate autophagosome biogenesis in concert with ubiquitin-like Atg8 J. Cell Biol 2010 190 965 973 20855502 Kroschwald S Maharana S Mateju D Malinovska L Nuske E Poser I Richter D Alberti S Promiscuous interactions and protein disaggregases determine the material state of stress-inducible RNP granules Elife 2015 4 e06807 26238190 Kryndushkin DS Alexandrov IM Ter-Avanesyan MD Kushnirov VV Yeast [PSI+] prion aggregates are formed by small Sup35 polymers fragmented by Hsp104 J. Biol. Chem 2003 278 49636 49643 14507919 Kwiatkowski TJ Jr Bosco DA Leclerc AL Tamrazian E Vanderburg CR Russ C Davis A Gilchrist J Kasarskis EJ Munsat T Valdmanis P Rouleau GA Hosler BA Cortelli P de Jong PJ Yoshinaga Y Haines JL Pericak-Vance MA Yan J Ticozzi N Siddique T McKenna-Yasek D Sapp PC Horvitz HR Landers JE Brown RH Jr Mutations in the FUS/TLS gene on chromosome 16 cause familial amyotrophic lateral sclerosis Science 2009 323 1205 1208 19251627 Lancaster AK Nutter-Upham A Lindquist S King OD PLAAC: a web and command-line application to identify proteins with prion-like amino acid composition Bioinformatics 2014 30 2501 2502 24825614 Landsberg MJ Moran-Jones K Smith R Molecular recognition of an RNA trafficking element by heterogeneous nuclear ribonucleoprotein A2 Biochemistry 2006 45 3943 3951 16548521 Lasagna-Reeves CA Rousseaux MW Guerrero-Munoz MJ Vilanova-Velez L Park J See L Jafar-Nejad P Richman R Orr HT Kayed R Zoghbi HY Ataxin-1 oligomers induce local spread of pathology and decreasing them by passive immunization slows Spinocerebellar ataxia type 1 phenotypes Elife 2015 4 e10891 http://dx.doi.org/10.1016/j.jmb.2015.11.016 26673892 Lee BJ Cansizoglu AE Suel KE Louis TH Zhang Z Chook YM Rules for nuclear localization sequence recognition by karyopherin beta 2 Cell 2006 126 543 558 16901787 Li L Lindquist S Creating a protein-based element of inheritance Science 2000 287 661 664 10650001 Li P Banjade S Cheng HC Kim S Chen B Guo L Llaguno M Hollingsworth JV King DS Banani SF Russo PS Jiang QX Nixon BT Rosen MK Phase transitions in the assembly of multivalent signalling proteins Nature 2012 483 336 340 22398450 Li Y Ray P Rao EJ Shi C Guo W Chen X Woodruff EA 3rd Fushimi K Wu JY A Drosophila model for TDP-43 proteinopathy Proc. Natl. Acad. Sci. USA 2010 107 3169 3174 20133767 Li YR King OD Shorter J Gitler AD Stress granules as crucibles of ALS pathogenesis J. Cell Biol 2013 201 361 372 23629963 Liebman SW Chernoff YO Prions in yeast Genetics 2012 191 1041 1072 22879407 Lim L Wei Y Lu Y Song J ALS-causing mutations significantly perturb the self-assembly and interaction with nucleic acid of the intrinsically disordered prion-like domain of TDP-43 PLoS Biol 2016 14 e1002338 26735904 Lin Y Protter DS Rosen MK Parker R Formation and maturation of phase-separated liquid droplets by RNA-binding proteins Mol. Cell 2015 60 208 219 26412307 Lindquist S Mad cows meet psi-chotic yeast: the expansion of the prion hypothesis Cell 1997 89 495 498 9160741 Ling SC Polymenidou M Cleveland DW Converging mechanisms in ALS and FTD: disrupted RNA and protein homeostasis Neuron 2013 79 416 438 23931993 Ludolph AC Brettschneider J TDP-43 in amyotrophic lateral sclerosis - is it a prion disease? Eur. J. Neurol 2015 22 753 761 25846565 Luk KC Kehm V Carroll J Zhang B O’Brien P Trojanowski JQ Lee VM Pathological alpha-synuclein transmission initiates Parkinson-like neurodegeneration in nontransgenic mice Science 2012 338 949 953 23161999 Ma J Wang F Prion disease and the ‘protein-only hypothesis’ Essays Biochem 2014 56 181 191 25131595 Maniecka Z Polymenidou M From nucleation to widespread propagation: a prion-like concept for ALS Virus Res 2015 207 94 105 25656065 Masel J Bergman A The evolution of the evolvability properties of the yeast prion [PSI+] Evolution 2003 57 1498 1512 12940355 Masel J Siegal ML Robustness: mechanisms and consequences Trends Genet 2009 25 395 403 19717203 Mayeda A Munroe SH Caceres JF Krainer AR Function of conserved domains of hnRNP A1 and other hnRNP A/B proteins EMBO J 1994 13 5483 5495 7957114 McGlinchey RP Kryndushkin D Wickner RB Suicidal [PSI+] is a lethal yeast prion Proc. Natl. Acad. Sci. USA 2011 108 5337 5341 21402947 Meyer H Bug M Bremer S Emerging functions of the VCP/p97 AAA-ATPase in the ubiquitin system Nat. Cell Biol 2012 14 117 123 22298039 Michelitsch MD Weissman JS A census of glutamine/ asparagine-rich regions: implications for their conserved function and the prediction of novel prions Proc. Natl. Acad. Sci. USA 2000 97 11910 11915 11050225 Molliex A Temirov J Lee J Coughlin M Kanagaraj AP Kim HJ Mittag T Taylor JP Phase separation by low complexity domains promotes stress granule assembly and drives pathological fibrillization Cell 2015 163 123 133 26406374 Mori K Lammich S Mackenzie IR Forne I Zilow S Kretzschmar H Edbauer D Janssens J Kleinberger G Cruts M Herms J Neumann M Van Broeckhoven C Arzberger T Haass C hnRNP A3 binds to GGGGCC repeats and is a constituent of p62-positive/TDP43-negative inclusions in the hippocampus of patients with C9orf72 mutations Acta Neuropathol 2013 125 413 423 23381195 Murakami T Qamar S Lin JQ Schierle GS Rees E Miyashita A Costa AR Dodd RB Chan FT Michel CH Kronenberg-Versteeg D Li Y Yang SP Wakutani Y Meadows W Ferry RR Dong L Tartaglia GG Favrin G Lin WL Dickson DW Zhen M Ron D Schmitt-Ulms G Fraser PE Shneider NA Holt C Vendruscolo M Kaminski CF St George-Hyslop P ALS/FTD mutation-induced phase transition of FUS liquid droplets and reversible hydrogels into irreversible hydrogels impairs rnp granule function Neuron 2015 88 678 690 26526393 Nakayashiki T Kurtzman CP Edskes HK Wickner RB Yeast prions [URE3] and [PSI+] are diseases Proc. Natl. Acad. Sci. USA 2005 102 10575 10580 16024723 Namy O Galopier A Martini C Matsufuji S Fabret C Rousset JP Epigenetic control of polyamines by the prion [PSI+] Nat. Cell Biol 2008 10 1069 1075 19160487 Neumann M Sampathu DM Kwong LK Truax AC Micsenyi MC Chou TT Bruce J Schuck T Grossman M Clark CM McCluskey LF Miller BL Masliah E Mackenzie IR Feldman H Feiden W Kretzschmar HA Trojanowski JQ Lee VM Ubiquitinated TDP-43 in frontotemporal lobar degeneration and amyotrophic lateral sclerosis Science 2006 314 130 133 17023659 Newby GA Lindquist S Blessings in disguise: biological benefits of prion-like mechanisms Trends Cell Biol 2013 23 251 259 23485338 Nillegoda NB Bukau B Metazoan Hsp70-based protein disaggregases: emergence and mechanisms Front. Mol. Biosci 2015 2 57 26501065 Nonaka T Masuda-Suzukake M Arai T Hasegawa Y Akatsu H Obi T Yoshida M Murayama S Mann DM Akiyama H Hasegawa M Prion-like properties of pathological TDP-43 aggregates from diseased brains Cell Rep 2013 4 124 134 23831027 Orr HT Cell biology of spinocerebellar ataxia J. Cell Biol 2012 197 167 177 22508507 Orr HT Zoghbi HY Trinucleotide repeat disorders Annu. Rev. Neurosci 2007 30 575 621 17417937 Osherovich LZ Weissman JS Multiple Gln/Asn-rich prion domains confer susceptibility to induction of the yeast [PSI+] prion Cell 2001 106 183 194 11511346 Patel A Lee Hyun O Jawerth L Maharana S Jahnel M Hein Marco Y Stoynov S Mahamid J Saha S Franzmann Titus M Pozniakovski A Poser I Maghelli N Royer Loic A Weigert M Myers Eugene W Grill S Drechsel D Hyman Anthony A Alberti S A liquid-to-solid phase transition of the ALS protein FUS accelerated by disease mutation Cell 2015 162 1066 1077 26317470 Paul KR Hendrich CG Waechter A Harman MR Ross ED Generating new prions by targeted mutation or segment duplication Proc. Natl. Acad. Sci. USA 2015 112 8584 8589 26100899 Polymenidou M Cleveland DW The seeds of neurodegeneration: prion-like spreading in ALS Cell 2011 147 498 508 22036560 Polymenidou M Cleveland DW Prion-like spread of protein aggregates in neurodegeneration J. Exp. Med 2012 209 889 893 22566400 Prusiner SB Prions. Proc. Natl. Acad. Sci. USA 1998 95 13363 13383 9811807 Ramaswami M Taylor JP Parker R Altered ribostasis: RNA-protein granules in degenerative disorders Cell 2013 154 727 736 23953108 Ravits JM La Spada AR ALS motor phenotype heterogeneity, focality, and spread: deconstructing motor neuron degeneration Neurology 2009 73 805 811 19738176 Resende CG Outeiro TF Sands L Lindquist S Tuite MF Prion protein gene polymorphisms in Saccharomyces cerevisiae Mol. Microbiol 2003 49 1005 1017 12890024 Ritson GP Custer SK Freibaum BD Guinto JB Geffel D Moore J Tang W Winton MJ Neumann M Trojanowski JQ Lee VM Forman MS Taylor JP TDP-43 mediates degeneration in a novel Drosophila model of disease caused by mutations in VCP/p97 J. Neurosci 2010 30 7729 7739 20519548 Roberts BE Duennwald ML Wang H Chung C Lopreiato NP Sweeny EA Knight MN Shorter J A synergistic small-molecule combination directly eradicates diverse prion strain structures Nat. Chem. Biol 2009 5 936 946 19915541 Ross ED Baxa U Wickner RB Scrambled prion domains form prions and amyloid Mol. Cell Biol 2004 24 7206 7213 15282319 Ross ED Edskes HK Terry MJ Wickner RB Primary sequence independence for prion formation Proc. Natl. Acad. Sci. USA 2005 102 12825 12830 16123127 Santoso A Chien P Osherovich LZ Weissman JS Molecular basis of a yeast prion species barrier Cell 2000 100 277 288 10660050 Satpute-Krishnan P Langseth SX Serio TR Hsp104-dependent remodeling of prion complexes mediates protein-only inheritance PLoS Biol 2007 5 e24 17253904 Sephton CF Good SK Atkin S Dewey CM Mayer P 3rd Herz J Yu G TDP-43 is a developmentally regulated protein essential for early embryonic development J. Biol. Chem 2010 285 6826 6834 20040602 Serio TR Cashikar AG Kowal AS Sawicki GJ Moslehi JJ Serpell L Arnsdorf MF Lindquist SL Nucleated conformational conversion and the replication of conformational information by a prion determinant Science 2000 289 1317 1321 10958771 Seydoux G Braun RE Pathway to totipotency: lessons from germ cells Cell 2006 127 891 904 17129777 Shorter J Hsp104: a weapon to combat diverse neurodegenerative disorders Neurosignals 2008 16 63 74 18097161 Shorter J Emergence and natural selection of drug-resistant prions Mol. Biosyst 2010 6 1115 1130 20422111 Shorter J The mammalian disaggregase machinery: Hsp110 synergizes with Hsp70 and Hsp40 to catalyze protein disaggregation and reactivation in a cell-free system PLoS One 2011 6 e26319 22022600 Shorter J Lindquist S Hsp104 catalyzes formation and elimination of self-replicating Sup35 prion conformers Science 2004 304 1793 1797 15155912 Shorter J Lindquist S Prions as adaptive conduits of memory and inheritance Nat. Rev. Genet 2005 6 435 450 15931169 Shorter J Lindquist S Destruction or potentiation of different prions catalyzed by similar Hsp104 remodeling activities Mol. Cell 2006 23 425 438 16885031 Shorter J Lindquist S Hsp104, Hsp70 and Hsp40 interplay regulates formation, growth and elimination of Sup35 prions EMBO J 2008 27 2712 2724 18833196 Shorter J Taylor JP Disease mutations in the prion-like domains of hnRNPA1 and hnRNPA2/B1 introduce potent steric zippers that drive excess RNP granule assembly Rare Dis 2013 1 e25200 25002999 Sondheimer N Lindquist S Rnq1: an epigenetic modifier of protein function in yeast Mol. Cell 2000 5 163 172 10678178 Sreedharan J Blair IP Tripathi VB Hu X Vance C Rogelj B Ackerley S Durnall JC Williams KL Buratti E Baralle F de Belleroche J Mitchell JD Leigh PN Al-Chalabi A Miller CC Nicholson G Shaw CE TDP-43 mutations in familial and sporadic amyotrophic lateral sclerosis Science 2008 319 1668 1672 18309045 Strome S Lehmann R Germ versus soma decisions: lessons from flies and worms Science 2007 316 392 393 17446385 Suel KE Gu H Chook YM Modular organization and combinatorial energetics of proline-tyrosine nuclear localization signals PLoS Biol 2008 6 e137 18532879 Sun Z Diaz Z Fang X Hart MP Chesi A Shorter J Gitler AD Molecular determinants and genetic modifiers of aggregation and toxicity for the ALS disease protein FUS/TLS PLoS Biol 2011 9 e1000614 21541367 Suzuki G Tanaka M Expanding the yeast prion world: active prion conversion of non-glutamine/asparagine-rich Mod5 for cell survival Prion 2013 7 109 113 23117914 Suzuki G Shimazu N Tanaka M A yeast prion, Mod5, promotes acquired drug resistance and cell survival under environmental stress Science 2012 336 355 359 22517861 Sweeny EA Shorter J Prion proteostasis: Hsp104 meets its supporting cast Prion 2008 2 135 140 19242125 Sweeny EA Shorter J Mechanistic and Structural Insights into the Prion-Disaggregase Activity of Hsp104 J. Mol. Biol 2015 http://dxdoi.org/10.1016/j.jmb.2015.11.016 Sweeny EA Jackrel ME Go MS Sochor MA Razzo BM DeSantis ME Gupta K Shorter J The Hsp104 N-terminal domain enables disaggregase plasticity and potentiation Mol. Cell 2015 57 836 849 25620563 Tanaka M Chien P Naber N Cooke R Weissman JS Conformational variations in an infectious protein determine prion strain differences Nature 2004 428 323 328 15029196 Ter-Avanesyan MD Kushnirov VV Dagkesamanskaya AR Didichenko SA Chernoff YO Inge-Vechtomov SG Smirnov VN Deletion analysis of the SUP35 gene of the yeast Saccharomyces cerevisiae reveals two non-overlapping functional regions in the encoded protein Mol. Microbiol 1993 7 683 692 8469113 Toombs JA McCarty BR Ross ED Compositional determinants of prion formation in yeast Mol. Cell Biol 2010 30 319 332 19884345 Toombs JA Petri M Paul KR Kan GY Ben-Hur A Ross ED De novo design of synthetic prion domains Proc. Natl. Acad. Sci. USA 2012 109 6519 6524 22474356 Torrente MP Shorter J The metazoan protein disaggregase and amyloid depolymerase system: Hsp110, Hsp70, Hsp40, and small heat shock proteins Prion 2013 7 457 463 24401655 Torrente MP Chuang E Noll MM Jackrel ME Go MS Shorter J Mechanistic Insights Into Hsp104 Potentiation J Biol. Chem 2016 291 5101 5115 26747608 True HL Lindquist SL A yeast prion provides a mechanism for genetic variation and phenotypic diversity Nature 2000 407 477 483 11028992 True HL Berlin I Lindquist SL Epigenetic regulation of translation reveals hidden genetic variation to produce complex traits Nature 2004 431 184 187 15311209 Tuite MF Serio TR The prion hypothesis: from biological anomaly to basic regulatory mechanism Nat. Rev. Mol. Cell Biol 2010 11 823 833 21081963 Tuite MF Staniforth GL Cox BS [PSI(+)] turns 50 Prion 2015 9 318 332 26645632 Tyedmers J Madariaga ML Lindquist S Prion switching in response to environmental stress PLoS Biol 2008 6 e294 19067491 Tyedmers J Treusch S Dong J McCaffery JM Bevis B Lindquist S Prion induction involves an ancient system for the sequestration of aggregated proteins and heritable changes in prion fragmentation Proc. Natl. Acad. Sci. USA 2010 107 8633 8638 20421488 Vance C Rogelj B Hortobagyi T De Vos KJ Nishimura AL Sreedharan J Hu X Smith B Ruddy D Wright P Ganesalingam J Williams KL Tripathi V Al-Saraj S Al-Chalabi A Leigh PN Blair IP Nicholson G de Belleroche J Gallo JM Miller CC Shaw CE Mutations in FUS, an RNA processing protein, cause familial amyotrophic lateral sclerosis type 6 Science 2009 323 1208 1211 19251628 Vieira NM Naslavsky MS Licinio L Kok F Schlesinger D Vainzof M Sanchez N Kitajima JP Gal L Cavacana N Serafini PR Chuartzman S Vasquez C Mimbacas A Nigro V Pavanello RC Schuldiner M Kunkel LM Zatz M A defect in the RNA-processing protein HNRPDL causes limb-girdle muscular dystrophy 1G (LGMD1G) Hum. Mol. Genet 2014 23 4103 4110 24647604 Walters RW Muhlrad D Garcia J Parker R Differential effects of Ydj1 and Sis1 on Hsp70-mediated clearance of stress granules in Saccharomyces cerevisiae RNA 2015 21 1660 1671 26199455 Wang F Wang X Ma J Conversion of bacterially expressed recombinant prion protein Methods 2011a 53 208 213 21176786 Wang F Wang X Yuan CG Ma J Generating a prion with bacterially expressed recombinant prion protein Science 2010 327 1132 1135 20110469 Wang F Zhang Z Wang X Li J Zha L Yuan CG Weissmann C Ma J Genetic informational RNA is not required for recombinant prion infectivity J. Virol 2011b 86 1874 1876 22090130 Wang T Birsoy K Hughes NW Krupczak KM Post Y Wei JJ Lander ES Sabatini DM Identification and characterization of essential genes in the human genome Science 2015 350 1096 1101 26472758 Weber SC Brangwynne CP Getting RNA and protein in phase Cell 2012 149 1188 1191 22682242 Weber SC Brangwynne CP Inverse size scaling of the nucleolus by a concentration-dependent phase transition Curr. Biol 2015 25 641 646 25702583 Wickner RB [URE3] as an altered URE2 protein: evidence for a prion analog in Saccharomyces cerevisiae Science 1994 264 566 569 7909170 Wickner RB Edskes HK Bateman D Kelly AC Gorkovskiy A The yeast prions [PSI+] and [URE3] are molecular degenerative diseases Prion 2011 5 258 262 22052353 Wickner RB Shewmaker FP Bateman DA Edskes HK Gorkovskiy A Dayani Y Bezsonov EE Yeast prions: structure, biology, and prion-handling systems Microbiol. Mol. Biol. Rev 2015 79 1 17 25631286 Xiang S Kato M Wu LC Lin Y Ding M Zhang Y Yu Y McKnight SL The LC domain of hnRNPA2 adopts similar conformations in hydrogel polymers, liquid-like droplets, and nuclei Cell 2015 163 829 839 26544936 Yu A Shibata Y Shah B Calamini B Lo DC Morimoto RI Protein aggregation can inhibit clathrin-mediated endocytosis by chaperone competition Proc. Natl. Acad. Sci. USA 2014 111 E1481 E1490 24706768 Zhang H Elbaum-Garfinkle S Langdon EM Taylor N Occhipinti P Bridges AA Brangwynne CP Gladfelter AS RNA Controls PolyQ Protein Phase Transitions Mol. Cell 2015 60 220 230 26474065 Zhang YJ Xu YF Cook C Gendron TF Roettges P Link CD Lin WL Tong J Castanedes-Casey M Ash P Gass J Rangachari HT Buratti E Baralle F Golde TE Dickson DW Petrucelli L Aberrant cleavage of TDP-43 enhances aggregation and cellular toxicity Proc. Natl. Acad. Sci. USA 2009 106 7607 7612 19383787 Zhang ZC Chook YM Structural and energetic basis of ALS-causing mutations in the atypical proline-tyrosine nuclear localization signal of the Fused in Sarcoma protein (FUS) Proc. Natl. Acad. Sci. USA 2012 109 12017 12021 22778397 Zhu L Brangwynne CP Nuclear bodies: the emerging biophysics of nucleoplasmic phases Curr. Opin. Cell Biol 2015 34 23 30 25942753 Zoghbi HY Orr HT Pathogenic mechanisms of a polyglutamine-mediated neurodegenerative disease, spino-cerebellar ataxia type 1 J. Biol. Chem 2009 284 7425 7429 18957430

PMC005xxxxxx/PMC5003778.txt

==== Front 10131695735616Stem Cell ResStem Cell ResStem cell research1873-50611876-77532656193810.1016/j.scr.2015.10.015nihpa810460ArticleDiscovery of survival factor for primitive chronic myeloid leukemia cells using induced pluripotent stem cells Suknuntha Kran aIshii Yuki bTao Lihong cHu Kejin cMcIntosh Brian E. dYang David aSwanson Scott dStewart Ron dWang Jean Y.J. bThomson James defSlukvin Igor ac*a Department of Pathology and Laboratory Medicine, University of Wisconsin, Madison, WI 53792, United Statesb Department of Medicine, Moores Cancer Center, School of Medicine, University of California, San Diego, La Jolla, CA 92093-0820, United Statesc Wisconsin National Primate Research Center, University of Wisconsin, Madison, WI 53715, United Statesd Morgridge Institute for Research, Madison, WI 53707, United Statese Department of Cell and Regenerative Biology, School of Medicine and Public Health, University of Wisconsin, Madison, WI 53707, United Statesf Department of Molecular, Cellular & Developmental Biology, University of California, Santa Barbara, CA 93106, United States* Corresponding author at: National Primate Research Center, University of Wisconsin, 1220 Capitol Court, Madison, WI 53715, United States. islukvin@wisc.edu (I. Slukvin).19 8 2016 31 10 2015 11 2015 29 8 2016 15 3 678 693 This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).A definitive cure for chronic myeloid leukemia (CML) requires identifying novel therapeutic targets to eradicate leukemia stem cells (LSCs). However, the rarity of LSCs within the primitive hematopoietic cell compartment remains a major limiting factor for their study in humans. Here we show that primitive hematopoietic cells with typical LSC features, including adhesion defect, increased long-term survival and proliferation, and innate resistance to tyrosine kinase inhibitor (TKI) imatinib, can be generated de novo from reprogrammed primary CML cells. Using CML iPSC-derived primitive leukemia cells, we discovered olfactomedin 4 (OLFM4) as a novel factor that contributes to survival and growth of somatic lin−CD34+ cells from bone marrow of patients with CML in chronic phase, but not primitive hematopoietic cells from normal bone marrow. Overall, this study shows the feasibility and advantages of using reprogramming technology to develop strategies for targeting primitive leukemia cells. ==== Body 1. Introduction CML is a myeloproliferative disorder characterized by unregulated growth of predominantly myeloid cells, and their subsequent accumulation in the bone marrow and peripheral blood. CML originates in hematopoietic stem cells (HSCs) with t(9;22)(q34;q11.2) translocation, which causes the constitutive expression of the BCR-ABL kinase driving the expansion of leukemic progeny (Holtz et al., 2002; Holyoake et al., 2001; Ramaraj et al., 2004). Ex vivo cultures of CML-derived cell lines and primary CML cells, ectopic expression of BCR-ABL in CD34+ cells and mouse models have provided important insights into CML pathogenesis and led to the development of targeted therapy for this neoplastic disease with BCR-ABL tyrosine kinase inhibitor (TKI), imatinib (Druker et al., 2006; Druker et al., 2001). Despite these achievements, eradication of CML remains challenging. Although the majority of patients treated with imatinib achieve a complete cytogenetic response, discontinuation of imatinib treatment is commonly associated with relapse (Mahon et al., 2010). Multiple lines of evidence suggest that the major cause of disease persistence is innate resistance of leukemia stem cells (LSCs) to TKIs (Corbin et al., 2011; Graham et al., 2002; Holyoake et al., 2001). Thus, studies of primitive leukemia cells are essential for better understanding leukemia pathogenesis and developing curative therapies for CML. Due to the limited number of BCR-ABL+ cells within the most primitive hematopoietic cell compartments (Holyoake et al., 1999; Holyoake et al., 2001; Vargaftig et al., 2012), establishing technologies for de novo generation of LSC-like cells would provide a significant benefit to the CML field. Reprogramming human somatic cells to pluripotency allows for the generation of induced pluripotent stem cells (iPSCs) that behave similarly to embryonic stem cells (ESCs), i.e., they are capable of self-renewal, large-scale expansion, and differentiation toward derivatives of all three germ layers, including blood (Choi et al., 2009b; Park et al., 2008; Takahashi et al., 2007; Yu et al., 2009). Because iPSCs capture the entire genome of diseased cells, they are already being used in modeling human genetic diseases (Grskovic et al., 2011). Recently, we and other groups successfully generated iPSCs from primary CML cells and showed that CML-iPSCs capture the genetic alterations present in leukemia cells, and possess the ability to produce differentiated leukemia cells (Bedel et al., 2013; Hu et al., 2011; Kumano et al., 2012). Here, we tested the hypothesis that reprogramming CML cells to pluripotency and then differentiating them back into blood cells can be used as a novel approach to produce an unlimited number of primitive hematopoietic cells with LSC properties and identify novel primitive leukemia cell survival factors and drug targets. We validated this hypothesis by demonstrating the successful application of the iPSC-based platform to discover OLFM4 as a novel primitive leukemia cell survival factor in patients in the chronic phase of CML. This finding provides a basis for development of novel approaches for treating CML by targeting OLFM4 or OLFM4-mediated signaling pathways in primitive leukemia cells. 2. Results 2.1. Generation of LSC-like cells from CML-iPSCs Recently we generated transgene-free iPSCs from the bone marrow mononuclear cells of a patient with a newly diagnosed CML in the chronic phase (CML15 iPSCs and CML17 iPSCs) and showed that these iPSCs capture the entire genome of neoplastic cells, including the unique 4-way translocation between chromosomes 1, 9, 22, and 11 that was present in the patient bone marrow (BM) (Hu et al., 2011). Sequencing analysis revealed that the BCR-ABL translocation in these CML-iPSCs expresses the p210 oncoprotein with a typical b3a2 rearrangement and lack of mutations in the kinase domain (Supplementary Fig. S1a and b). These findings were consistent with the observed sensitivity of parental bone marrow progenitors to imatinib (Supplementary Fig. S1c). CML LSCs have been identified within the most primitive hematopoietic compartments as cells with long-term culture initiating cell (LTC-IC) or in vivo repopulating activities. (Corbin et al., 2011; Li et al., 2012; Petzer et al., 1996; Sloma et al., 2010; Udomsakdi et al., 1992b) Similar to normal HSCs, CML LSCs express typical markers of primitive hematopoietic cells including, CD34 and CD90, and are negative for hematopoietic lineage markers (lin−) and CD45RA (Udomsakdi et al., 1992b). They also have high aldehyde dehydrogenase (ALDH) activity and are able to efflux Rhodamine-123 (Udomsakdi et al., 1992a). Since we previously found that cells with such features can be generated from karyotypically normal iPSCs by coculture with OP9 (Choi et al., 2009a; Choi et al., 2009b; Vodyanik et al., 2005), we used this system to induce hematopoietic differentiation from CML iPSCs. In OP9 coculture, CML iPSCs and control bone marrow-derived iPSCs (BM1K and BM9) formed CD34+CD43+ hematopoietic progenitors, including CD235a−CD41a−CD45+ cells that are highly enriched in myeloid progenitors (Hu et al., 2011; Vodyanik et al., 2006) (Fig. 1a and b). As shown in Fig. 1b, CD235a−CD41a−CD45+ cells obtained at day 9 of CML iPSC differentiation had the phenotypic features of CML LSCs including the expression of the primitive hematopoietic cell markers CD34, CD90 and CD117, and an absence of CD38 and CD45RA and line-age markers (lin−). Similar to somatic CML LSCs, day 9 CML iPSC-derived CD45+ cells had ALDH activity and were able to efflux Rhodamine-123 (Fig. 1c-d). After expansion for 2 days, CD45+ cells retained expression of CD34 and CD90 and CD117 stem cell markers and remained lin− and CD45RA−. However, up to 25% of the population acquired CD38 expression (Fig. 1b, day 11) and the levels of ALDHhigh and Rholow cells within CD45+ population decreased (Fig. 1c and d). CML iPSC-derived lin−CD34+CD90+CD117+CD45+38+/− (hereafter referred to as induced CD34+; iCD34+) cells acquired expression of myeloid lineage-specific markers and lost CD34 expression (Fig. 1b, day 15), i.e. became lin+CD34− (hereafter referred to as induced CD34− cells; iCD34−) following further culture with hematopoietic cytokines. Thus, we concluded that cells with the LSC phenotype could be generated de novo from reprogrammed CML cells (Fig. 1e). Although CML LSCs share many properties with HSCs, they have three distinct characteristics: the increased proliferation and long-term survival (Corbin et al., 2011; Holyoake et al., 1999); the ability to grow in vitro without added cytokines (Jiang et al., 1999); and an adhesion defect (Bhatia et al., 2001; Verfaillie et al., 1992). Therefore, next we evaluated whether de novo generated CML iCD34+ cells possess similar properties. By analyzing the growth of iCD34+ cells in serum-free medium with growth factors, we found an enhanced expansion of iCD34+CD38− cells from CML iPSCs relative to similar cells from normal BM iPSCs. Treatment with imatinib decreased expansion of CML iCD34+CD38− cells to the level observed in normal BM iCD34+CD38− cultures (Fig. 2a). CML iCD34+CD38− cells also generated significantly more and larger myeloid colonies as compared to normal BM iCD34+CD38− cells (Fig. 2b). Although we did not see differences in the number of colonies between CML and BM cells within CD38− compartment, CML iCD34+CD38− cells consistently generated larger colonies. Using the LTC-IC assay, we found that CML iCD34+ cells produced a much higher number of LTC-IC-derived CFCs than BM iCD34+ cells, indicating an increase in long-term survival (Fig. 2c). In serum-free cultures, CML iCD34+ cells were able to withstand growth factor deprivation in both CD38+ and CD38− compartments, which was reversed by treatment with imatinib (Fig. 2d). CML iCD34+ cells also showed reduced adhesion to fibronectin (Fig. 2ei), which was partially restored by imatinib treatment (Fig. 2eii). Overall, these findings provide strong evidence that iCD34+ cells derived from CML iPSCs behave similarly to their somatic LSC counterparts (Fig. 2f). 2.2. Induced LSC-like cells are resistant to imatinib The dependence of CML cells on BCR-ABL signaling enables the suppression of the disease by TKIs. However, resistances of CML LSCs to imatinib preclude a complete cure for CML (Corbin et al., 2011; Graham et al., 2002; Holyoake et al., 2001). Analysis of BCR-ABL expression by qPCR and Western blot demonstrated that BCR-ABL mRNA and protein, including phosphorylated protein (p-BCR-ABL) were present in undifferentiated CML iPSCs and their iCD34+ and iCD34− progeny (Fig. 3a and b). Interestingly, expression of BCR-ABL mRNA was greater in undifferentiated CML iPSCs when compared to parental bone marrow CD34+ cells. However, in iCD34+ cells, the level of BCR-ABL expression downregulated to the level observed in parental CD34+ cells. Evaluation of the phosphorylation status of the BCR-ABL-specific substrate CRK-like protein (CRKL) revealed the presence of a phosphorylated form of CRKL (p-CRKL) in CML, but not in the control BM iCD34+ and iCD34− cells (Fig. 3c), thus providing evidence that BCR-ABL is active in CML iPSC-derived hematopoietic progeny. Studies in CML patients have shown that imatinib inhibits BCR-ABL kinase and cell proliferation in primitive hematopoietic compartments without affecting survival of LSCs (Copland et al., 2006; Corbin et al., 2011; Graham et al., 2002; Holtz et al., 2002; Schemionek et al., 2010). To find out whether de novo generated CML iCD34+ cells respond to imatinib in a similar fashion, we evaluated the effect of imatinib on these cells in vitro. After treatment with imatinib for 4 h, p-CRKL dramatically decreased in both primitive iCD34+ and in more mature iCD34− blood cells (Fig. 3c and d). This indicates that imatinib efficiently inhibits the majority of BCR-ABL kinase activity in de novo generated CML cells, independent of the stage of maturation. The inhibition of BCR-ABL kinase in iCD34− cells by imatinib treatment was associated with increased apoptosis as determined by using Annexin V staining and caspase 3/7 fluorogenic substrate (Fig. 3e and fii). In contrast, imatinib failed to induce apoptosis in CML iCD34+ cells (Fig. 3fi), despite the abrogation of p-CRKL signaling in these cells (Fig. 3c and d). The increased resistance to imatinib within the more primitive hematopoietic compartment was also evident from a significant increase of inhibitory concentration 50% (IC50) for CML iCD34+ cells as compared to CML iCD34− cells (Fig. 3gi). Both, iCD34+ and iCD34− cells, generated from normal BM iPSCs, were not affected by imatinib treatment (Fig. 3fi, fii and gii). To confirm the maturation stage-dependent sensitivity to imatinib, we analyzed the distribution of apoptotic cells within different compartments and generations following expansion of CFSE-labeled CML iCD34+ cells. Imatinib treatment of CML iCD34+ cell cultures was associated with a significant increase of slowly dividing CD34+ cells (generations 2–4) in bulk cultures (Supplementary Fig. S2a and b) and retention of CD34+ expression by rapidly dividing cells (generations 5–7; Supplementary Fig. S2c) as compared to non-treated cells. Annexin V staining of CFSE-labeled cell cultures revealed that the most primitive iCD34bright cells were resistant to imatinib-induced apoptosis, while a substantial increase in apoptosis was observed in iCD34− cells and in more mature and proliferative iCD34dim cells the majority of which were CD38+ (Supplementary Fig. S2d–f). These findings imply that differentiated CML iCD34− cells became sensitive to imatinib and could not survive in culture, while the most primitive CML iCD34+CD38− cells did not undergo apoptosis upon imatinib treatment and were selected to dominate in the culture. Taken together, these results indicate that CML iCD34+ cells reproduce many aspects of drug resistance observed in somatic primitive hematopoietic cells from CML patients in the chronic phase (Fig. 3h). 2.3. Identication of olfactomedin 4 as a novel survival factor in LSC-like cells To find out whether CML iPSC-derived primitive hematopoietic cells can be used to discover novel LSC survival factors and drug targets, we compared the molecular profile of CML and normal BM iCD34+ cells that were either treated or not treated with imatinib for 16 h in expansion cultures with cytokines. Similar to the findings with somatic CML CD34+ cells (Leng et al. 2013; Li and Dewey 2011), untreated CML iCD34+ cells showed significant differences compared to normal BM iCD34+ cells in the expression of genes regulating chemotaxis, adhesion, cytokine production, proliferation, programmed cell death, regulation of phosphorylation, and fatty acid metabolism (Supplementary Fig. S3a). These differences included the upregulation of genes associated with cancer development, such as BCL2, CDK6, PRKCQ, MYCN, CDKNA2A, GFI1B, TP53RK, and RASGRP3, and the downregulation of adhesion molecules ICAM1, ICAM3, ITGB2 and ITGB7 (Supplementary Fig. S3b and Supplementary Table S1). We next evaluated the effect of imatinib on the transcriptome of iCD34+ cells by performing principal component analysis (PCA) to define the change in transcriptional distance following drug treatment. As shown in Fig. 4a, after treatment with 5 μM imatinib, CML iCD34+ cells moved away from non-treated CML iCD34+ cells toward non-treated control BM iCD34+ cells in the dimension defined by principal component PC1 and PC2, i.e. became more similar to control BM iCD34+ cells, thereby suggesting the important role of BCR-ABL signaling in establishing the unique transcriptional signature of neoplastic cells in CML. To find candidate genes associated with imatinib resistance, we initially selected genes that were significantly upregulated or downregulated by imatinib in CML iCD34+ cells. From the list of upregulated genes, we chose 137 that were induced ≥2 fold by imatinib treatment (Fig. 4b). Then, we selected genes that were upregulated in CML iCD34+ cells, but suppressed in BM iCD34+ cells, i.e. excluded genes that were upregulated or not affected by imatinib in BM iCD34+ cells (Fig. 4b and c and Supplementary Table S2). A similar algorithm was applied for down-regulated genes. From the list of significantly dowregulated genes, we chose those that were suppressed ≥2 fold by imatinib in CML iCD34+ (16 genes). We then selected genes that were suppressed in CML iCD34+ cells, but upregulated in BM iCD34+ cells (Fig. 4b and c and Supplementary Table S2). Overall, based on this selection algorithm we identified 127 upregulated and 11 downregulated genes following imatinib treatment. Upregulation of selected genes (BCL2A1, ALOX15 and SMAD7) by imatinib in CML iCD34+ cells was confirmed by qPCR analysis (Supplementary Figure S3c). Interestingly, the top-ranking up-regulated gene identified by our algorithm was OLFM4 (olfactomedin 4), which has been reported to have anti-apoptotic activity in malignancy (Liu et al., 2012; Oh et al., 2011; Zhang et al., 2004). According to AceView analysis (Thierry-Mieg and Thierry-Mieg, 2006), transcription of human OLFM4 produces 5 different mRNAs (OLFM4a–OLFM4e) encoding four alternatively spliced variants and one unspliced variant. These mRNAs encode four complete proteins, including OLFM4a and OLFM4d secreted isoforms and OLFM4b and OLFM4c cytoplasmic isoforms. OLFM4e isoform is reconstructed from partial mRNA and is not fully characterized (http://www.ncbi.nlm.nih.gov/IEB/Research/Acembly/av.cgi?db=human&c=Gene&l=OLFM4). Using isoform specific primers, we demonstrated the presence of four out of five OLFM4 mRNA isoforms (OLFM4a–OLFM4d) in CML iCD34+ cells and the entire set of five mRNA isoforms (OLFM4a–OLFM4e) in CML iCD34− cells (Supplementary Fig. S4a). Consistent with RNAseq data, qPCR analysis revealed the augmentation of OLFM4 expression by imatinib in CML iCD34+ cells, including all four OLFM4a–OLFM4d mRNA isoforms (Fig. 4d). To evaluate the potential role of OLFM4 in survival of CML iCD34+ cells, we tested the effect of OLFM4 knocking-down on iCD34+ cell apoptosis using OLFM4-targeted small interference RNA (siOLFM4) composed of four OLFM4-specific siRNAs. Transfection of CML iPSC-derived hematopoietic cells with this set of siRNAs enables 60%–85% reduction in expression of all five OLFM4 mRNA isoforms (Supplementary Fig. S4b). As shown in Fig. 4e, imatinib or siOLFM4 had no effect on CML iCD34+ cell apoptosis. However, siOLFM4 treatment in the presence of imatinib induced significant increase of apoptosis in CML, but not normal bone marrow iCD34+ cells. In addition, OLFM4 knockdown selectively inhibited expansion of CML iCD34+ cells in vitro (Fig. 4f). These findings imply that OLFM4 is involved in regulation of imatinib resistance and survival of de novo generated primitive CML cells (Fig. 4g). 2.4. OLFM4 is essential for somatic primitive leukemia cell survival in vitro and in vivo engraftment To find out whether the findings obtained using iCD34+ cells can be translated to somatic cells, we isolated lin−CD34+ cells (hereafter referred to as sCD34+) form parental bone marrow (patient 1; P1) and bone marrow from several other CML patients in the chronic phase. Using RT-PCR, we detected OLFM4 mRNAs in sCD34+ cells isolated from all CML patients. However, in contrast to iCD34+ cells generated de novo from CML iPSCs, sCD34+ cells from CML patients expressed a restricted range of OLFM4 isoforms (Fig. 5a and b). Cytoplasmic OLFM4b isoform was found in sCD34+ cells from all CML patients, while none of the patients showed expression of isoform OLFM4e. The expression of cytoplasmic OLFM4c isoform, and secreted OLFM4a and OLFM4d iso-forms varied between patients, but none of the patients expressed the entire range of isoforms. In contrast, sCD34− cells expressed all range of isoforms, although isoform e was not detected in some patients (Fig. 5a and b). None of the OLFM4 isoforms were detectable in normal bone marrow sCD34+ cells by PCR (Fig. 5a). To examine whether OLFM4 is expressed in primitive CML cells in situ, we performed immunofluorescent staining of BM biopsies from normal individuals and patients in the chronic phase of CML. As shown in Fig.5c, OLFM4 was detected in CD34+ cells from CML patients, but not in normal donors. In contrast, CD34− cells showed OLFM4 protein expression in both normal and CML biopsies. To confirm the production of OLFM4 by the most primitive CML cells, we evaluated OLFM4 secretion by sCD34+ (lin−CD34+) cells isolated from P1 and P6 following the short-term in vitro cultures. Using ELISA, we detected OLFM4 protein in sCD34+ cultures from CML patients, while the OLFM4 levels in control BM cultures were not exceeded background level (Fig. 5d). Since previous studies have demonstrated the induction of OLFM4 expression by G-CSF in myeloid progenitors (Chin et al., 2008; Zhang et al., 2002), we tested whether G-CSF enhances OLFM4 production in sCD34+ cells from CML patients. We found that G-CSF stimulation indeed enhances OLFM4 production by sCD34+ cells and that this effect is potentiated by imatinib (Fig. 5d). To confirm the specificity of G-CSF effect, we cultured sCD34+ cells from patient P6 with G-SCF and G-SCF blocking monoclonal antibody. The level of OLFM4 production in presence of antibody was reduced from 205 pg/ml to 47.5 pg/ml. To assess whether OLFM4 provides a survival advantage for CML sCD34+ cells, we performed CFC assay after a 24 h treatment with imatinib, siOLFM4, or both. Consistent with prior observations in iCD34+ cells, CML cells in clonogenic cultures displayed a significant upregulation of OLFM4 expression when pretreated with imatinib (Fig. 6a and Supplementary Fig. S5). The OLFM4a mRNA isoform was the most abundant in these cultures (Fig. 6a), while other isoforms were expressed at much lower levels (Supplementary Fig. S5). Knockdown of OLFM4 in CML sCD34+ cells using siRNA potentiated the inhibitory effect of imatinib on CFCs, which was reversed by pretreatment of cells with OLFM4 protein (Fig. 6b). In contrast, no significant effect was observed following similar treatments of sCD34+ cells from BM of three normal donors (Fig. 6b). To determine whether OLFM4 provides a survival advantage for cells more primitive than CFCs and whether it can affect the response of these cells to imatinib, we performed an LTC-IC assay using CML sCD34+ cells treated with siOLFM4, imatinib or both together. Consistent with previous studies (Bellodi et al., 2009; Corbin et al., 2011), imatinib had little effect on somatic LTC-ICs. In contrast, treatment of cells with siOLFM4 significantly reduced OLFM4 production in sCD34+ cultures (Fig. 6c) and the number of colonies in LTC-IC cultures (Fig. 6d and Supplementary Table S3). Addition of imatinib to siOLFM4-treated cultures had no significant impact on LTC-IC numbers (Fig. 6d). Both imatinib and siOLFM4 treatment did not affect substantially the proportion of BCR-ABL positive colonies from LTC-IC cultures (Supplementary Table S4). The observed effect of OLFM4 knockdown on the LTC-ICs was different from the one we observed with less primitive leukemia cells, CFCs. While OLFM4 downregulation alone had minimal effect on CFCs, it significantly augmented the inhibitory effect of imatinib on these progenitors. In contrast, OLFM4 knockdown alone was able to abrogate LTC-ICs and no synergistic action was observed between OLFM4 knockdown and imatinib treatment, thereby suggesting that OLFM4-mediated survival of the most primitive CML cells may not depend on BCR-ABL tyrosine kinase-mediated signaling. Although our studies indicate that knockdown of OLFM4 in sCD34+ cells affects primitive CML cell survival in vitro, it remains unclear whether this effect is mediated through cell-intrinsic pathways or through paracrine signaling. Following culture, lin−CD34+ cells undergo differentiation into lin+CD34− cells, that highly express OLFM4. Since siRNAs are transferred to cells following cellular division and the duration of RNA silencing in rapidly dividing cells approaches 1 week (McIntosh et al. 2014), siOLFM4 knockdown most likely reduces OLFM4 production in primitive sCD34+ cells as well as in CD34− cells differentiated from sCD34+ cells. Thus, OLFM4 secreted by differentiated cells (neoplastic and non-neoplastic) may provide survival cues to primitive CML cells. The ability to engraft in immunocompromised mice is used as important criterion to identify LSC function. However, hematopoietic cells derived from human pluripotent stem cells have markedly impaired engraftment capacity (Slukvin, 2013; Vo and Daley, 2015). Although previous mouse studies have found that forced retroviral expression of BCR-ABL in hematopoietic progenitors generated from pluripotent embryonic stem cells endowed them with engraftment potential (Perlingeiro et al., 2001), we failed to detect meaningful engraftment following transplantation of CML iPSC-derived iCD34+ cells (not shown). This could be related to a more “physiological” level of BCR-ABL expression in de novo generated human primitive leukemia cells that could be insufficient to enable their engraftment. Therefore, to assess the role of OLFM4 in vivo, we performed transplantation of somatic lin−CD34+ cells from CML patients in chronic phase using NOD,B6.SCID IL2Rγ−/− KitW41/W41 (NBSGW mice possessing a defective Kit allele; c-kit, W41) (McIntosh et al., 2014). Mice models with a defective Kit allele support engraftment of human cells without irradiation, and are superior to irradiated mice for transplantation of human leukemia cells (Cosgun et al., 2014; McIntosh et al., 2014). Using NBSGW mice we were able to detect human CD45+ cells in bone marrow following transplantation as low as 10,000 lin−CD34+ cells from patients in chronic phase of CML. As shown in Fig. 7a and b, most of the NBSGW mice transplanted with scrambled siRNA-transfected CML sCD34+ cells demonstrated the presence of human CD45+ cells in bone marrow. In contrast, only two out eight mice transplanted with siOLFM4-transfected sCD34+ cells showed detectable human CD45+ cells at much lower levels compared to corresponding mice in scrambled siRNA group. RT-PCR analysis confirmed that engrafted human CD45+ cells expressed BCR-ABL (Fig. 7c). Thus, these studies provide the evidence that OLFM4 is important for the engraftment of primitive leukemia cells in vivo, although it has yet to be determined whether OLFM4 knockdown affects bone marrow homing or in vivo survival of CML lin−CD34+ cells. 3. Discussion Multiple studies have already demonstrated the validity of the iPSC model for studying pathogenesis of monogenic human diseases and drug screening (reviewed in (Rais et al., 2013)). Although the potential use of iPSC technology in studying neoplasia has been suggested (Ramos-Mejia et al., 2012; Slukvin, 2009; Ye et al., 2009), only a few groups have reported the generation of iPSCs from malignant cells. The first successful reprogramming of cancer cells was achieved by Hochedlinger, who generated mouse pluripotent stem cells by transplanting nuclei from melanoma into oocytes (Hochedlinger et al., 2004). After the discovery of pluripotency factors capable of reprogramming blood cells, iPSC lines have been generated from a patient with a myeloproliferative disorder bearing the JAK2-V617F mutation (Ye et al., 2009), CML cell lines (Carette et al., 2010), primary bone marrow cells of chronic phase CML patients (Bedel et al., 2013; Hu et al., 2011; Kumano et al., 2012), juvenile myelomonocytic leukemia (Gandre-Babbe et al., 2013), and mouse MLL-AF9 leukemia cells (Liu et al., 2013). These studies demonstrated that iPSCs generated from neo-plastic cells carry a disease-specific genetic mutation and can be used to generate blood cells affected by that particular mutation. Here, we advanced the iPSC model to study primitive leukemia cells and proved the utility of this model for discovering drug targets by identifying the novel CML LSC survival factor OLFM4. The iPSC technology offers several major benefits for studying CML. Current protocols for obtaining primitive CML cells are based on selection of lin−CD34+ cells from patient bone marrow samples. However, this approach does not reliably isolate LSCs because normal primitive hematopoietic cells usually outnumber BCR-ABL+ cells within the lin− CD34+ compartment by several folds (Petzer et al., 1996). The capacity of iPSCs to capture CML genetic aberrations, self-renew indefinitely and differentiate into blood cells allows for the production of pure populations of patient-specific BCR-ABL+ LSC-like cells in unlimited numbers for functional studies and drug screening. As shown in present studies, the use of pure population of induced lin−CD34+CD90+CD117+-CD45+CD38+/− BCR-ABL+ cells enabled identification of genes affected by imatinib treatment and eventually led to discovery of novel survival factor for primitive leukemia cells OLFM4. iPSCs can be genetically modified to generate CML iPSC lines in which specific gene expression can be conditionally manipulated, thereby facilitating the identification of genes regulating human LSCs’ survival and expansion. iPSCs obtained from the patients before and after acquisition of drug resistance or progression to blast crisis would be a valuable tool for studying the molecular mechanisms of CML progression and drug resistance. Understanding the impact that clonal evolution and diversity has on CML progression requires the selection and expansion of individual leukemia clones; and this represents the major technical challenge. Because iPSCs are of a single cell origin, multiple iPSC lines can be generated from the same patient to capture the diversity of genetic alterations within leukemia cell populations and address the role of non-BCR-ABL-associated mutations in disease development. Another important finding of this study is the discovery of OLFM4 as a novel survival factor for primitive CML cells. The human OLFM4 (also called GW112 and hGC-1) gene encodes a glycoprotein with a multimer structure (Liu et al., 2006). Transcription for this gene produces four spliced variants and one unspliced OLFM4a–OLFM4e mRNA isoforms encoding secreted and cytoplasmic proteins (http://www.ncbi.nlm.nih.gov/IEB/Research/Acembly/av.cgi?db=human&c=Gene&l=OLFM4). OLFM4 plays an important role in a variety of cellular functions including cell adhesion, cell cycle and apoptosis (Tomarev and Nakaya, 2009). In the human intestine, OLFM4 was identified as a robust marker of LGR5+ stem cells and a subset of cancer cells (van der Flier et al., 2009). OLFM4 is involved in cell growth and apoptosis in human malignancies (Liu et al., 2010; Oh et al., 2011; Park et al., 2012; Zhang et al., 2004) and is considered to be an inducible resistance factor to apoptotic stimuli (Koshida et al., 2007). OLFM4 interacts with GRIM-19 (Zhang et al., 2004), which is a component of the respiratory complex I of mitochondria with an anti-apoptotic role in prostate cancer cells (Huang et al., 2004). However, the effects of OLFM4 on apoptosis appear to be a cell-type dependent. In contrast to prostate cancer cells, overexpression of OLFM4 in HL-60 leukemia cells induces their differentiation and apoptosis (Liu et al., 2010). Although OLFM4 was initially identified in myeloblasts and later in neutrophils (Clemmensen et al., 2012; Zhang et al., 2002), OLFM4's biological function in normal and neoplastic hematopoiesis remains largely unknown. It has been shown that OLFM4 expression is upregulated in a subset of patients with acute myeloid leukemia and primary myelo-fibrosis (Hasselbalch et al., 2014; Liu et al., 2010). Here, we demonstrated the expression of OLFM4 protein in lin−CD34+ and CD34− bone marrow cells from CML patients in chronic phase and revealed the distinct pattern of OLFM4 isoform expression in these cells at the mRNA level. CML lin−CD34+ cells expressed predominantly cytoplasmic OLFM4 mRNAs. In contrast, CML CD34− differentiated cells displayed the broad spectrum of cytoplasmic and secreted OLFM4a–OLFM4d isoforms. Using OLFM4 knockdown studies, we discovered the critical role OLFM4 in the survival and drug resistance of primitive CML cells and demonstrated differences in OLFM4-mediated regulation of cell survival in CFCs and more primitive LTC-ICs (Fig. 7d). While OLFM4 had a minimal role in CML CFC maintenance in normal conditions, it was required for protection of CML CFCs from imatinib-mediated inhibition. In contrast, maintenance of CML LTC-ICs was dependent on OLFM4 and independent of BCRABL kinase activity. Consistent with in vitro LTC-IC studies, OLFM4 knockdown also abrogated the engraftment of BCR-ABL+ cells in NBSGW mice, thereby providing another critical evidence for the essential role of OLFM4 in homeostasis of CML cells in vivo. The mechanisms responsible for the distinct role of OLFM4 in the survival of leukemia cells at different stages of maturation remain to be investigated. It is possible that cell-context specific effects may be related to differences in the expression and/or regulation of the OLFM4 isoforms in a stage-specific fashion. It is also important to investigate whether OLFM4 mediates cell survival in a self-autonomous or non-autonomous manner. Given that lin−CD34+ cells acquire a CD34− differentiated phenotype which is associated with a high level of OLFM4 expression in culture, it is possible that secretion of OLFM4 by differentiated cells generates a milieu supportive for the growth of undifferentiated CML cells in vitro. Similarly, we observed that differentiated CD34− cells next to CD34+ cells in bone marrow biopsy samples of CML patients express high levels of OLFM4. In addition, OLFM4 is produced by normal non-leukemic CD34− cells. Thereby, OLFM4 secreted by differentiated leukemic or nonleukemic cells may confer a selective growth advantage to the more primitive CML cells and LSCs in vivo. Overall, our studies indicate that OLFM4-mediated signaling supports the survival of primitive leukemia cells. Thus, analysis of OLFM4 and its isoforms in peripheral blood could be exploited as predictive markers for treatment regimens, while targeting OLFM4 or OLFM4-mediated signaling could be explored as an option to eradicate primitive CML cells. Potentially, OLFM4 could be targeted using anti-OLFM4 antibodies. An antibody approach has already been described to modulate the activity of another protein within the OLFM family, OLFM3. As shown by Miljkovic-Licina (Miljkovic-Licina et al., 2012), administration of anti-OLFM3 antibodies efficiently blocked proangiogenic signaling mediated by this protein in vivo. Finally, our studies provide a strong background and novel model for the further exploration of the mechanisms of OLFM4 action in primitive CML cells. 4. Material and methods 4.1. iPSCs maintenance and hematopoietic differentiation In this study, we used previously described transgene-free BM9, CML15 and CML17 iPSCs produced by reprogramming of normal bone marrow mononuclear cells and bone marrow mononuclear cells from patient with newly diagnosed CML in the chronic phase (Hu et al., 2011). BM1K iPSC was generated from the normal control using the same approach (Hu et al., 2011). Undifferentiated iPSCs were maintained in cocultures with mouse embryonic fibroblasts (MEFs). Hematopoietic differentiation was induced by transferring the iPSCs to overgrown OP9 feeders as we have previously described in detail (Choi et al., 2009a; Vodyanik et al., 2005). VEGF (100 ng ml−1) was added to some cultures at initiation of differentiation to enhance blood production. CD43+ cells were collected on day 9 of differentiation using MACS and cultured in α-MEM supplemented with 10% FBS, 50 μg ml−1 ascorbic acid, 100 μM monothioglycerol (complete serum supplemented medium (CSSM)) and 200 ng ml−1 GM-CSF to selectively expand myeloid progenitors (Choi et al., 2009a). After two days of expansion with GM-CSF CD43+ cells enriched in myeloid progenitors were cultured for an additional four days in the same media supplemented with 10 ng ml−1 IL-3, 100 ng ml−1 IL-6, 100 ng ml−1 Flt3L, 100 ng ml−1 SCF, and 200 ng ml−1 GM-CSF (all from Peprotech). 4.2. Isolation of lin−CD34+ cells from iPSC cultures CD43+ hematopoietic cells were collected from differentiated iPSC cultures using MACS and labeled with CD235a/CD41a FITC, CD45 APC and CD38 PE (BD Pharmingen). Lin-CD45+CD38+ and lin-CD45+CD38-subpopulations were obtained by fluorescence-activated cell sorting using FACSAria (BD) (Vodyanik et al., 2006). 4.3. Collection of bone marrow specimens from CML patients in the chronic phase, and controls, and purification of lin-CD34+ cells Bone marrow mononuclear cells from CML patients in the chronic phase were purchased commercially (AllCells or Applied StemCells), or obtained from the patients at the University of Wisconsin Hospital and Clinics (Madison, WI) with approval from the University of Wisconsin Institutional Review Board. Donors had previously signed an Institutional Review Board-approved consent. Patients P1, P2, P5–P7 were newly diagnosed with CML. All studied (P1–P7) patients were sensitive to imatinib. Bone marrow cells from healthy donors were obtained from Cincinnati Children's Hospital Medical Center (CCHMC). Mononuclear cells were labeled with the lineage-specific markers CD2, CD3, CD14, CD15, CD16, CD19, CD20, CD24, CD41a, CD56, CD66b, and Glycophorin A (FITC-conjugated antibodies), CD34 APC (BD Pharmingen) and DAPI to exclude dead cells. Live lin−CD34+ cells were isolated using FACSAria (BD). In some experiments, mononuclear cells were thawed and cultured overnight followed by labeling and isolation of lin−CD34+ cells as described above. 4.4. Hematopoietic colony-forming assay Hematopoietic clonogenic assays were performed using serum-containing StemMACS semisolid clonogenic medium (Miltenyi Biotec, CA). Cells were incubated with either DMSO or 5 μM imatinib in serum-free medium (SFM) composed of IMDM, 10% BIT (Stem Cell Technologies), 2-mercaptoethanol, and EXCYTE (Millipore) and supplemented with a low concentration of growth factors (1 ng ml−1 of each SCF, IL3, IL6, Flt3L and GM-CSF). After 24 h of cultures, cell collected and plated in clonogenic medium without imatinib. 4.5. Rhodamine 123 exclusion aldehyde dehydrogenase activity assays Rho exclusion assay was performed as previously described (Vodyanik et al., 2005). Aldehyde dehydrogenase (ALDH) staining was performed with the Aldefluor kit (Stem Cell Technologies) per manufacturer instructions. 4.6. Apoptosis assay CML iCD34+ cells were cultured in CSSM containing 10 ng ml−1 IL3, 100 ng ml−1 IL6, 200 ng ml−1 GM-CSF, 100 ng ml−1 SCF, and 100 ng ml−1 Flt3L with or without 5 μM imatinib for 24 h before analysis of apoptosis. When indicated, cells were transfected with either OLFM4 or scrambled (control) siRNA. Cells were stained with Annexin-V–PE and 7-AAD using the Annexin V: PE Apoptosis Detection Kit (BD Bioscience) according to the manufacturer's protocol and analyzed by flow cytometry. In some experiments green Caspase 3/7 detection reagent (Life Technologies) was used in combination with Annexin-V-PE staining to detect apoptotic cells. 4.7. Western blotting Cells were cultured in serum-free medium without growth factor in the presence or the absence of 5 μM imatinib for 4 h prior to harvesting. Lysates were prepared in RIPA buffer containing 1% Nonidet P-40 and 1% sodium deoxycholate supplemented with 1 mM phenylmethylsulfonyl fluoride, protease inhibitors cocktail and 1 mM sodium vanadate. Proteins were separated by SDS-PAGE, transferred to nitrocellulose membrane and immunoblotted for phospho-CRKL (py207), phospho-Abl (py245) and phospho-BCR-ABL (pY245) antibodies (Cell Signaling Technology). Signals were detected with HRP-conjugated secondary antibody using the ECL kit (Amersham). CML cell line K562 was included as a positive control. p-CRKL expression levels were determined by densitometry using ImageJ software (NIH). 4.8. Cell proliferation assay Total lin−CD34+, lin−CD34+CD38− or lin−CD34+CD38+ were plated in triplicate in 96-well plate at 103 cells/well. Cells were then cultured in SFM with or without 5 μM imatinib. When specified, the 300 ng ml−1 of OLFM4 (Acro Biosystem) or the following growth factors were added: 10 ng ml−1 IL3, 100 ng ml−1 SCF, 100 ng ml−1 Flt3L, 100 ng ml−1 IL6, and 200 ng ml−1 GM-CSF. Viable cell yields were determined by counting trypan blue excluding viable cells using a hemocytometer. 4.9. Long-term culture initiating cell assay (LTC-IC) Sorted iPSC-derived lin−CD34+CD45+ cells were plated in a six-well plate at 104 cells/well containing 5–7 day-old cultures of murine 1 × 105 M2-10B4 and OP9 stromal cells (1:1 ratio mix) in a LTC-IC medium consisting of SFM supplemented with 10 μM hydrocortisone, 50 ng ml−1 SCF, 5 ng ml−1 IL3, and 50 ng ml−1 IL6. Cultures were maintained at 37 °C in a humidified atmosphere with 5% CO2, and fed at weekly intervals. After five weeks, cells were harvested and analyzed for CFC potential as described above. LTC-IC assay for somatic lin− CD34+CD45+ cells was performed using M2-10B4 cells exactly as described in the Stem Cell Technologies protocol (http://www.stemcell.com/en/Technical-Resources/db5a9/28412_ltc_ic-H.aspx). When indicated, cells were pretreated for one week with 5 μM imatinib or DMSO (control), in SFM supplemented with low concentration of growth factors (1 ng ml−1 of each SCF, IL3, IL6, Flt3L and GM-CSF), and transferred to LTC-IC cultures for an additional five weeks of culture. In some experiments, cells were transfected with either OLFM4 or negative control siRNA and then pretreated with imatinib or DMSO as described above. 4.10. IC50 assay The lin−CD34+ and lin+CD34− cells were plated at 103 cells/well in a 96-well plate in CSSM containing 10 ng ml−1 IL3, 100 ng/ml IL6, 200 ng ml−1 GM-CSF, 100 ng ml−1 SCF, and 100 ng ml−1 Flt3L with 0–100 μM imatinib. After 24 h of culture, a viable cell count was performed using trypan blue. The IC50 was determined as the drug concentration where cell death was 50% of maximum in the upper plateau (Sebaugh, 2011). Data from three assays performed in triplicate were used for statistical analysis and graph plots for IC50 determinations. Relative IC50 were determined by fitting an exponential dose–response curve to the cell proliferation data by using GraphPad Prism software (GraphPad, San Diego, CA). 4.11. CFSE tracking of cell division Differentiated cells were labeled with 1 μM carboxy-fluorescein diacetate succinimidyl diester (CFSE; Molecular Probes, Eugene, OR) as previously described (Copland et al., 2006; Holtz et al., 2002). These cells were then incubated overnight in CSSM medium supplemented with growth factors to allow excess unbound dye to leak out of the cells. Cells cultured in the presence of 10 μg/ml mitomycin C (Sigma Aldrich) were used to establish the CFSEmax (undivided cell population). The next day, CFSEbright cells were sorted by FACArias to exclude non-labeled and CFSEdim populations. These cells were then cultured for four days in CSSM supplemented with growth factors with or without 5 μM imatinib. At the end of the culture period, cells were stained with CD34-APC and 7AAD for flow cytometry analysis. The percentage of cells in each generation was determined using FlowJo software (Tree Star, Ashland, OR), with the position of the parent generation set on the basis of the fluorescence profile of undivided cells. 4.12. Adhesion assay The lin−CD34+CD45+ cells were incubated in SFM supplemented with 1 ng/ml each of SCF, IL-3, IL-6, Flt3L and GM-CSF, with and without 5 μM imatinib for 24 h. Cells were then washed and resuspended in SFM and plated onto either fibronectin- or BSA-coated wells of 96-well plates at 103 cells/well. After 2 h, nonadherent and adherent fractions were separated as described previously (Bhatia et al., 2001; Holtz et al., 2002). Both fractions were plated in serum-containing StemMACS HSC-CFU medium (Miltenyi Biotec, CA), and the percentage of CFCs in adherent fraction was calculated. 4.13. Immunofluorescence microscopy Heat-induced antigen-retrieved paraffin sections of bone marrow biopsy from normal subjects and CML patients (N = 2 each) were stained with mouse anti-human CD34 (Ventana), and rabbit anti-human OLFM4 (Abcam), or matched IgG isotype primary antibodies. After staining with secondary antibodies, anti-mouse IgG Alexa Fluor 555 and anti-rabbit IgG Alexa Fluor 488 (Life Technology), and DAPI, pictures were acquired by Nikon Eclipse Ti-E configured with an A1R confocal system (Nikon Instruments Inc. Melville, NY) and Nikon Elements (NIS — element C) imaging software (Nikon Instruments Inc. Melville, NY). 4.14. Detection of OLFM4 by ELISA The lin−CD34+ cells from patients P1 and P6 (2000 cells per well) were cultured in 120 μl serum-free medium with 1 ng ml−1 SCF, IL3, IL6 and Flt3L with or without 10 ng ml−1 G-CSF, 5 μM imatinib, or together for 24 h. For G-CSF neutralization, cells treated with cultures treated with 0.03 μg ml−1 mouse human G-CSF neutralizing antibody (clone #3316 R&D Systems). OLFM4 protein in cultures was quantified using a sandwich ELISA kit (USCN Life Science) according to manufacturer instruction. 4.15. siRNA transfection OLFM4 siRNA composed of four OLFM4-specific siRNAs (siOLFM4) or control scrambled siRNA (AllStars Neg. siRNA AF 488) were obtained from Qiagen. iPSCs-derived lin−CD34+CD45+ and somatic CML bone marrow lin−CD34+ cells were transfected with a total 100 nM of either OLFM4 or scrambled siRNA using the HiPerfect transfection reagent according to the manufacturer's protocol (all from Qiagen). The transfection efficiency was 50–60% as evaluated by using control ALLStars Neg. siRNA AF488. As determined by qPCR performed 24 h after transfection, the silencing efficiency was 60–85% (Supplementary Fig. S4b). 4.16. Gene expression analysis by real-time PCR (qPCR) RNA was isolated from the cell subpopulations using the PureLink RNA mini kit (Life Technologies). cDNA synthesis was carried out using Advantage RT-for-PCR kit (Clontech). Quantitative real-time PCR analysis was performed using the PlatinumSYBR Green qPCR SuperMix-UDG kit (Life Technologies) and primers listed in Supplementary Table S5. The reactions were run on a Mastercyclerep realplex thermal cycler (Eppendorf) and expression levels were calculated by minimal cycle threshold values (Ct) normalized to the reference expression of GAPDH in each sample (Pfaffl, 2001). When specified, K562 was used as a reference. All qPCR products were analyzed on 1.2% agarose gels to confirm the specificity of detection. 4.17. RNA-Seq analysis Day 11 iCD34+ cells were isolated and cultured CSSM supplemented with 200 ng ml−1 of GM-CSF with DMSO or 5 μm of imatinib. After 16 h, total RNA from cells was isolated with PureLink RNA mini kit (Life Technologies) and treated with DNaseI TURBO DNase™ kit (Ambion). Total RNA was quantitated using the Life Technologies Qubit fluorometer (Q32857) and the Agilent Bioanalyzer 2100. Samples were then prepared for sequencing using the Illumina TruSeq RNA Sample Preparation Kit v2 (RS-122-2001), according to the manufacturer's protocol. Final sample libraries were quantitated with the Life Technologies Qubit fluorometer and sequenced on the Illumina HiSeq 2500 (SY-401-1003-PRE). Base-calling and demultiplexing were done with the Illumina Genome Analyzer Casava Software, version 1.8.2. Following quality assessment and filtering for adapter molecules and other sequencing artifacts, the remaining sequencing reads were aligned to 19,084 RefSeq genes extracted from the Illumina iGenomes annotation, selecting only “NM_” designated genes. Bowtie v 0.12.9 was used, allowing two mismatches in a 28 bp seed, and excluding reads with more than 200 alignments (Langmead et al., 2009). RSEM v 1.2.3 was used to estimate isoform or gene relative expression levels in units of “transcripts per million” (tpm) (Li and Dewey, 2011; Li et al., 2010). To determine differentially expressed genes, RNAseq output data were analyzed using EBseq(v.1.1.6) http://www.biostat.wisc.edu/~kendzior/EBSEQ/ (Leng et al., 2013). Genes with a posterior probability equal to 1.000 were considered differentially expressed. Genes with tpm < 10 across all studied samples were excluded from analysis. Remaining genes demonstrating significant differences in expression between the studied groups were assigned to biological process categories using the DAVID bioinformatics program (Huang da et al., 2008). To visualize the gene-expression levels, a heat-map was composed using MultiExperiment Viewer v4.2 (http://www.tm4.org). PCA was performed with TPM (transcripts per million) normalized data using “stats” package for R programming language, with “scale” option disabled. The data was preliminary restricted to genes that were called differentially expressed between CML-control and CML-Imatinib with Benjamini-Hochberg false discovery rate below 0.01. 4.18. Animal transplantation Engraftment of sCD34+ cells was evaluated using NBSGW mice possessing a defective Kit allele; (c-kit, W41) (McIntosh et al., 2014). Bone marrow lin−CD34+ cells from the CML patients in chronic phase were equally divided and transduced with scramble siRNA or siOLFM4. 5000 or 10,000 lin−CD34+ cells were injected into the retroorbital vein of NBSGW mice. Animals were analyzed at 12–16 weeks after transplantation by flow cytometry and presence of BCR-ABL transcripts was confirmed by RT-PCR. To evaluate BCR-ABL expression in engrafted cells, bone marrow samples from transplanted animals were cultured in SFM with human SCF, IL3, IL6 and Flt3L (each at 100 ng ml−1) to expand human cells. After 6–7 days of culture, human CD45+ cells were isolated by FACS and analyzed by for BCR-ABL expression by RT-PCR. 4.19. Statistical analysis Data obtained from multiple experiments were reported as the mean ± SEM. Significance levels were determined by one-tailed Student-t test analysis. Supplementary Material Figures Table 1 Table 2 Acknowledgments We thank Dr. Toru Nakano for providing OP9 cells, Stem Cell Technologies for providing M2-10B4 cells, Mitchell Probasco for cell sorting, Oleg Moskvin for help with PCA analysis and Matt Raymond, Gene Ueneshi, Derek Theisen and Nicole Denison for editorial assistance. We also thank the Cell Processing and Manipulation Core in the Translational Cores, and Physicians and Nurses at UWCCC and CCHMC for obtaining and processing bone marrow samples and the Translational Research Trials Office for providing the regulatory and administrative support for this endeavor. This work was supported by funds from the National Institute of Health (U01HL099773, P01 GM081629, P51 OD011106, and P30 CA014520), and The Charlotte Geyer Foundation. K.S. is supported by funds from the Department of Pharmacology, Faculty of Science, Mahidol University, Bangkok, Thailand. Author contribution KS designed, conducted and analyzed experiments, interpreted experimental data, made figures and contributed to paper writing; YI conducted western blood experiments; LT performed PCR analysis; KH generated CML iPSCs; BM generated NBSGW mice and assisted with transplantation studies; DY provided patient samples and relevant clinical data; RS and SS performed bioinformatics analysis of RNAseq data; JW planned and contributed to BCR-ABL and pCRKL analysis; JT contributed to concept development and directed molecular profiling studies; IS developed concept, led and supervised all aspects of the studies, analyzed and interpreted data, and wrote the paper. Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.scr.2015.10.015. Disclosure of conflicts of interest Authors declare no conflicts of interest. Additional information Accession codes: The RNAseq data has been deposited in Gene Expression Omnibus under accession number GSE60065. Fig. 1 Generation of lin−CD34+CD43+CD45+ primitive hematopoietic cells from CML-iPSCs. (a) Schematic diagram of hematopoietic differentiation from iPSCs. Phenotype and its designation are shown under and on the top of cells, respectively. Prefix i indicates iPSC-derived. FSGm36 = Flt3L, SCF, GM-CSF, IL3, IL6. (b) Phenotypes of CD45+ cells obtained from CML-iPSCs after differentiation on OP9 (day 9) and following their expansion/differentiation with cytokines in stroma-free cultures (as indicated in A). Plots show isotype control (open) and specific antibody (shaded) histograms. Representative results from three to four independent experiments are shown. (c) and (d) Rhodamine (Rho) efflux and ALDH activity in isolated CML iCD34+ cells. Representative dot plots show Rho efflux and ALDH activity in CD34+CD38+ and CD34+CD38− populations (c). Graph shows quantification of Rho efflux (di) and ALDH assay (dii). The values are mean ± SEM of % of Rholow cells (% of Rholowverapamil− – Rholowverapamil+) and ALDH+ cells (% of ALDH+DEAB− – ALDH+DEAB+) from three experiments respectively. DEAB = diethylaminobenzaldehyde, ALDH = aldehyde dehydrogenase. (e) Summary of findings. Fig. 2 Stem/progenitor cell properties of CML and control BM iPSC-derived iCD34+ cells. (a) Expansion of CML and BM iCD34+ cells in serum-free medium (SFM) with growth factors (GF) in presence of DMSO (control) or 5 μM imatinib. Results are mean ± SEM from six independent experiments (three from each BM1K and BM9 iCD34+, and three from each CML15 and CML17 iCD34+). * indicates significant differences between cell counts in BM and CML iCD34+ cultures without imatinib (p < 0.05). Ψ indicates significant differences between imatinib treated and untreated CML iCD34+ cells (p < 0.05). (b) Colony formation from BM and CML iCD34+ cells. Graph shows mean ± SEM from 3 independent experiments. * p < 0.01. Right panels show typical GM colony from BM and CML iCD34+ cells. Both, CML iCD34+CD38− and iCD34+CD38+ cells grew much larger colonies in CFC medium as compare to control BM iCD34+ cells. (c) LTC-IC assay of CML and BM iCD34+ cells. Numbers next to the plot show the total number of CFCs detected in LTC-IC cultures of iCD34+ cells. Results are mean ± SEM of two biological replicates performed in triplicates. Representative colony and Wright stained cytospin from the colony formed by CML iCD34+ cells after 5 weeks of LTC-IC culture are shown below the chart. Scale bar = 100 μm. (d) Expansion of CML and BM iCD34+ cells in serum-free medium (SFM) without growth factors (GF) in presence of DMSO (control) or 5 μM imatinib. Results are mean ± SEM from six independent experiments (three from each BM1K and BM9 iCD34+, and three from each CML15 and CML17 iCD34+). * indicates significant differences between cell counts in BM and CML iCD34+ cultures without imatinib (p < 0.05). Ψ indicates significant differences between imatinib treated and untreated CML iCD34+ cells (p < 0.05). (e) Adhesion of BM and CML iCD34+ cells to fibronectin or control bovine serum albumin (BSA)-adsorbed plates (ei) and the effect of imatinib on adhesion of iCD34+ cells to fibronectin (eii). * p ≤ 0.02. Results are mean ± SEM from three independent experiments. (f) Summary of findings. Fig. 3 Effects of imatinib on CML iPSC-derived iCD34+ cells. (a) Western blot shows expression of BCR-ABL, c-ABL and p-BCR-ABL protein in normal BM iPSCs and CML iPSCs and their hematopoietic derivatives. GAPDH and K562 were used as a loading control and a positive control respectively. (b) qPCR quantification of BCR-ABL mRNA expression in CML iPSCs, their derivatives and paternal sCD34+ bone marrow cells. The expression levels were calculated relatively to K562 as calibrator. Results are mean ± SEM from three experiments in duplicate. (c) Western blot shows phospho-CRKL (p-CRKL) expression in iCD34+ and iCD34− cells following 4 h treatment with 5 μM imatinib or DMSO. GAPDH was used as a loading control. (d) Quantitative analysis of p-CRKL expression by densitometry. Results are calculated as p-CRKL/GAPDH ratio. Graph shows % of inhibition relative to DMSO control. Results are the mean ± SEM from three independent experiments. (e) Dot plot shows staining of CML iCD34− cells with Annexin V and caspase 3/7 detection agent after treatment with imatinib or DMSO. (f) Apoptosis in BM and CML iCD34+ (fi) and iCD34− (fii) cells treated and non-treated with 5 μM imatinib for 24 h. Apoptosis was evaluated by Annexin V staining. Results are the mean ± SEM from three independent experiments. K562 was used as an IM-sensitive control. * p < 0.05. (g) 50% Inhibition concentration (IC50) assay from CML (gi) and bone marrow (gii) iCD34+ and iCD34− cells is shown as relative response (% of cell death relative to cell death in the upper plateau) versus log concentration of imatinib. * indicates a significant IC50 shift (p < 0.05). Results are the mean ± SEM from three independent experiments performed in triplicate. (h) Summary of findings. Fig. 4 Gene expression analysis reveals set of imatinib-induced genes in CML iPSC-derived iCD34+ cells. (a) Principal component (PC) analysis of global gene expression in iCD34+ cells treated (IM+) and non-treated (IM−) with imatinib. After treatment with imatinib the position of CML iCD34+ cells shifted toward normal non-treated BM iCD34+ cells. (b) Selection of candidate genes associated with imatinib resistance. Upper left Venn diagram shows genes significantly upregulated by imatinib in CML iCD34+(≥2 fold induction). Upper right Venn diagram shows genes that were significantly suppressed by imatinib (≥2 fold reduction). Genes with tpm < 10 across all studied samples were excluded from analysis. (c) Heat maps showing expression of genes selectively induced or suppressed by imatininb in CML iCD34+. The top differentially expressed genes are arranged based on ratio of expression in imatinib-treated CML/imatinib-treated BM iCD34+ cells. (d) qPCR analysis of the effect of imatinib treatment on expression of OLFM4 mRNA isoforms in CML and BM iCD34+ cells. The results are mean ± SEM of three experiments in duplicate. The expression levels were calculated relative to untreated control. (e) OLFM4 knockdown with siRNA induced apoptosis in imatinib treated CML iCD34+ cells. Results are mean ± SEM of three independent experiments. The statistical differences were only observed between CML iCD34+ samples treated with imatinib alone and imatinib plus siOLFM4; * p < 0.05 (f) Effect of OLFM4 knockdown on BM or CML iCD34+ cell proliferation in CSSM medium supplemented with growth factors. Graph displays the ratio of cell numbers in cultures treated with siOLFM4 relative to cell numbers in corresponding cultures treated with scrambled siRNA. Results are mean ± SEM of three independent experiments. * indicates significant (p < 0.05) decrease in the expansion cell ratios (siOLFM/scramble siRNA cultures) in CML iCD34+ cultures when compared to normal BM controls. (g) Summary of findings. Fig. 5 Expression of OLFM4 in somatic CD34+ cells from CML patients in chronic phase. (a) Expression of OLFM4 mRNA isoforms in freshly isolated lin−CD34+ (sCD34+) and sCD34− cells from CML patients and normal controls, as determined by RT-PCR. iCD34− cells were used as a positive control (Pos). L is DNA ladder. (b) qPCR analysis of the expression of OLFM4 mRNA isoforms in sCD34+ and sCD34− cells from P1 and P6 CML patients. * p < 0.05 (c) Confocal microscopy showing the expression of OLFM4 in CD34+ cells (arrow) in the bone marrow (BM) biopsy specimen from normal donor and CML patient (scale bar = 20 μm). (d) Quantitation of OLFM4 production in sCD34+ cultures by ELISA. The lin−CD34+ cells from patients P1 and P6 were cultured in serum-free medium with 1 ng ml−1 SCF, IL3, IL6 and Flt3L with or without 10 ng ml−1 G-CSF, 5 μM imatinib, or together for 24 h. Fig. 6 Effects of OLFM4 on somatic lin−CD34+ CML cells. (a) The effect of imatinib on expression of OLFM4a mRNA isoform in hematopoietic colonies from patients P1 and P6. sCD34+ cells were pretreated for 24 h with imatinib and transferred to clonogenic cultures. Hematopoietic colonies were collected seven days after transfection and analyzed by qPCR. Results are mean ± SEM of two experiments performed in triplicates. * p < 0.01 (b) Knockdown of OLFM4 potentiates the inhibitory effect of imatinib on CML sCD34+ CFCs (G + GM). After transfection with OLFM4 siRNA (siOLFM4) or scramble (negative control) siRNA, CML or normal bone marrow sCD34+ were cultured with or without 5 μM imatinib in presence or absence of OLFM4 protein (300 ng ml−1) for 24 h in serum-free medium with low concentrations of growth factors and transferred to clonogenic medium. Results are the mean ± SEM of three to five independent experiments. Significant differences (* p < 0.05) were observed only in CML sCD34+ cultures, but not in normal bone marrow sCD34+ controls from three patients (NBM1–NBM3). (c) Quantification of OLFM4 protein in the media from 5-week LTC-IC cultures by ELISA. Results are mean ± SEM of three independent experiments. Significant differences (* p < 0.01) were observed in cultures of CML sCD34+ transfected with scramble siRNA and siOLFM4. (d) Knockdown of OLFM4 with siRNA dramatically reduced CFC output in LTCIC cultures of CML patients but not of normal bone marrow controls (NBM). Results are mean ± SEM of three experiments. * p < 0.05. NBM = normal bone marrow. Fig. 7 Effect of OLFM4 knockdown with siRNA on in vivo survival of CML sCD34+. (a) Representative dot plots show human hCD45 vs. mouse mCD45 labeling in freshly isolated bone marrow samples at 12–14 weeks after transplantation. The samples of sCD34+ cells from each CML patient were divided equally, treated with siOLFM4 or scramble siRNA and then injected into retroorbital vein of NBSGW mice. Corresponding mice pairs injected with cells from patient 1 (P1) and 6 (P6) are shown. (b) Frequency of human CD45+ cells in bone marrow of NBSGW mice transplanted with scrambled and OLFM4 siRNA. Each point shows individual mouse. & indicates a mice pair injected with 5000 lin−CD34+ (sCD34+) cells. All remaining pairs of mice were injected with 10,000 sCD34+ cells. The differences between studied groups are significant (* p = 0.0265). (c) Detection of BCR-ABL expression by RT-PCR in human CD45+ cells isolated from bone marrow of six mice injected with scrambled siRNA. Pos = positive control (mRNA isolated from CML iCD34− cells), m = mouse, h = human. (d) The hypothetical model of OLFM4 action on primitive CML cells. OLFM4 produced by CD34+ and CD34− cells supports survival and expansion of the most primitive leukemia cells. Since imatinib enhances the production of OLFM4, it can potentially affect the survival of primitive leukemia cells through modulation of OLFM4-mediated signaling. ==== Refs References Bedel A Pasquet JM Lippert E Taillepierre M Lagarde V Dabernat S Dubus P Charaf L Beliveau F de Verneuil H Variable behavior of iPSCs derived from CML patients for response to TKI and hematopoietic differentiation. PLoS One 2013 8 e71596 24058405 Bellodi C Lidonnici MR Hamilton A Helgason GV Soliera AR Ronchetti M Galavotti S Young KW Selmi T Yacobi R Targeting autophagy potentiates tyrosine kinase inhibitor-induced cell death in Philadelphia chromosome-positive cells, including primary CML stem cells. J. Clin. Invest 2009 119 1109 1123 19363292 Bhatia R Munthe HA Forman SJ Abnormal growth factor modulation of beta1-integrin-mediated adhesion in chronic myelogenous leukaemia haematopoietic progenitors. Br. J. Haematol 2001 115 845 853 11843818 Carette JE Pruszak J Varadarajan M Blomen VA Gokhale S Camargo FD Wernig M Jaenisch R Brummelkamp TR Generation of iPSCs from cultured human malignant cells. Blood 2010 115 4039 4042 20233975 Chin KL Aerbajinai W Zhu J Drew L Chen L Liu W Rodgers GP The regulation of OLFM4 expression in myeloid precursor cells relies on NF-kappaB transcription factor. Br. J. Haematol 2008 143 421 432 18764868 Choi KD Vodyanik MA Slukvin II Generation of mature human myelomonocytic cells through expansion and differentiation of pluripotent stem cell-derived lin-CD34+CD43+CD45+ progenitors. J. Clin. Invest 2009a 119 2818 2829 19726877 Choi KD Yu J Smuga-Otto K Salvagiotto G Rehrauer W Vodyanik M Thomson J Slukvin I Hematopoietic and endothelial differentiation of human induced pluripotent stem cells. Stem Cells 2009b 27 559 567 19259936 Clemmensen SN Bohr CT Rorvig S Glenthoj A Mora-Jensen H Cramer EP Jacobsen LC Larsen MT Cowland JB Tanassi JT Olfactomedin 4 defines a subset of human neutrophils. J. Leukoc. Biol 2012 91 495 500 22187488 Copland M Hamilton A Elrick LJ Baird JW Allan EK Jordanides N Barow M Mountford JC Holyoake TL Dasatinib (BMS-354825) targets an earlier progenitor population than imatinib in primary CML but does not eliminate the quiescent fraction. Blood 2006 107 4532 4539 16469872 Corbin AS Agarwal A Loriaux M Cortes J Deininger MW Druker BJ Human chronic myeloid leukemia stem cells are insensitive to imatinib despite inhibition of BCR-ABL activity. J. Clin. Invest 2011 121 396 409 21157039 Cosgun KN Rahmig S Mende N Reinke S Hauber I Schafer C Petzold A Weisbach H Heidkamp G Purbojo A Kit Regulates HSC Engraftment across the Human-Mouse Species Barrier. Cell Stem Cell 2014 Druker BJ Talpaz M Resta DJ Peng B Buchdunger E Ford JM Lydon NB Kantarjian H Capdeville R Ohno-Jones S Efficacy and safety of a specific inhibitor of the BCR-ABL tyrosine kinase in chronic myeloid leukemia. N. Engl. J. Med 2001 344 1031 1037 11287972 Druker BJ Guilhot F O'Brien SG Gathmann I Kantarjian H Gattermann N Deininger MW Silver RT Goldman JM Stone RM Five-year follow-up of patients receiving imatinib for chronic myeloid leukemia. N. Engl. J. Med 2006 355 2408 2417 17151364 Gandre-Babbe S Paluru P Aribeana C Chou ST Bresolin S Lu L Sullivan SK Tasian SK Weng J Favre H Patient-derived induced pluripotent stem cells recapitulate hematopoietic abnormalities of juvenile myelomonocytic leukemia. Blood 2013 121 4925 4929 23620576 Graham SM Jorgensen HG Allan E Pearson C Alcorn MJ Richmond L Holyoake TL Primitive, quiescent, Philadelphia-positive stem cells from patients with chronic myeloid leukemia are insensitive to STI571 in vitro. Blood 2002 99 319 325 11756187 Grskovic M Javaherian A Strulovici B Daley GQ Induced pluripotent stem cells–opportunities for disease modelling and drug discovery. Nat. Rev. Drug Discov 2011 10 915 929 22076509 Hasselbalch HC Skov V Stauffer Larsen T Thomassen M Hasselbalch Riley C Jensen MK Bjerrum OW Kruse TA Transcriptional profiling of whole blood identifies a unique 5-gene signature for myelofibrosis and imminent myelofibrosis transformation. PLoS One 2014 9 e85567 24454890 Hochedlinger K Blelloch R Brennan C Yamada Y Kim M Chin L Jaenisch R Reprogramming of a melanoma genome by nuclear transplantation. Genes Dev 2004 18 1875 1885 15289459 Holtz MS Slovak ML Zhang F Sawyers CL Forman SJ Bhatia R Imatinib mesylate (STI571) inhibits growth of primitive malignant progenitors in chronic myelogenous leukemia through reversal of abnormally increased proliferation. Blood 2002 99 3792 3800 11986238 Holyoake T Jiang X Eaves C Eaves A Isolation of a highly quiescent subpopulation of primitive leukemic cells in chronic myeloid leukemia. Blood 1999 94 2056 2064 10477735 Holyoake TL Jiang X Jorgensen HG Graham S Alcorn MJ Laird C Eaves AC Eaves CJ Primitive quiescent leukemic cells from patients with chronic myeloid leukemia spontaneously initiate factor-independent growth in vitro in association with up-regulation of expression of interleukin-3. Blood 2001 97 720 728 11157490 Hu K Yu J Suknuntha K Tian S Montgomery K Choi KD Stewart R Thomson JA Slukvin II Efficient generation of transgene-free induced pluripotent stem cells from normal and neoplastic bone marrow and cord blood mononuclear cells. Blood 2011 117 e109 e119 21296996 Huang da W Sherman BT Stephens R Baseler MW Lane HC Lempicki RA DAVID gene ID conversion tool. Bioinformation 2008 2 428 430 18841237 Huang G Lu H Hao A Ng DC Ponniah S Guo K Lufei C Zeng Q Cao X GRIM-19, a cell death regulatory protein, is essential for assembly and function of mitochondrial complex I. Mol. Cell. Biol 2004 24 8447 8456 15367666 Jiang X Lopez A Holyoake T Eaves A Eaves C Autocrine production and action of IL-3 and granulocyte colony-stimulating factor in chronic myeloid leukemia. Proc. Natl. Acad. Sci. U. S. A 1999 96 12804 12809 10536003 Koshida S Kobayashi D Moriai R Tsuji N Watanabe N Specific overexpression of OLFM4(GW112/HGC-1) mRNA in colon, breast and lung cancer tissues detected using quantitative analysis. Cancer Sci 2007 98 315 320 17270020 Kumano K Arai S Hosoi M Taoka K Takayama N Otsu M Nagae G Ueda K Nakazaki K Kamikubo Y Generation of induced pluripotent stem cells from primary chronic myelogenous leukemia patient samples. Blood 2012 119 6234 6242 22592606 Langmead B Trapnell C Pop M Salzberg SL Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol 2009 10 R25 19261174 Leng N Dawson JA Thomson JA Ruotti V Rissman AI Smits BM Haag JD Gould MN Stewart RM Kendziorski C EBSeq: an empirical Bayes hierarchical model for inference in RNA-seq experiments. Bioinformatics 2013 29 1035 1043 23428641 Li B Dewey CN RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinforma 2011 12 323 Li G Chan TM Leung KS Lee KH A cluster refinement algorithm for motif discovery. IEEE/ACM Trans. Comput. Biol. Bioinforma 2010 7 654 668 Li L Wang L Wang Z Ho Y McDonald T Holyoake TL Chen W Bhatia R Activation of p53 by SIRT1 inhibition enhances elimination of CML leukemia stem cells in combination with imatinib. Cancer Cell 2012 21 266 281 22340598 Liu W Chen L Zhu J Rodgers GP The glycoprotein hGC-1 binds to cadherin and lectins. Exp. Cell Res 2006 312 1785 1797 16566923 Liu W Lee HW Liu Y Wang R Rodgers GP Olfactomedin 4 is a novel target gene of retinoic acids and 5-aza-2′-deoxycytidine involved in human myeloid leukemia cell growth, differentiation, and apoptosis. Blood 2010 116 4938 4947 20724538 Liu RH Yang MH Xiang H Bao LM Yang HA Yue LW Jiang X Ang N Wu LY Huang Y Depletion of OLFM4 gene inhibits cell growth and increases sensitization to hydrogen peroxide and tumor necrosis factor-alpha induced-apoptosis in gastric cancer cells. J. Biomed. Sci 2012 19 38 22471589 Liu Y Cheng H Gao S Zou Z He F Hu L Hou D Lu X Li Y Zhang H Reprogramming of MLL-AF9 leukemia cells into pluripotent stem cells. Leukemia 2013 Mahon FX Rea D Guilhot J Guilhot F Huguet F Nicolini F Legros L Charbonnier A Guerci A Varet B Discontinuation of imatinib in patients with chronic myeloid leukaemia who have maintained complete molecular remission for at least 2 years: the prospective, multicentre Stop Imatinib (STIM) trial. Lancet Oncol 2010 11 1029 1035 20965785 McIntosh BE Brown ME Duffin BE Maufort JP Vereide DT Slukvin II Thomson JA Non-irradiated NOD, B6.SCID Il2rg−/− KitW41/W41 (NBSGW) Mice Support Multi-lineage Engraftment of Human Hematopoietic Cells. Stem Cell Rep 2014 (accepted for publication) Miljkovic-Licina M Hammel P Garrido-Urbani S Lee BP Meguenani M Chaabane C Bochaton-Piallat ML Imhof BA Targeting olfactomedin-like 3 inhibits tumor growth by impairing angiogenesis and pericyte coverage. Mol. Cancer Ther 2012 11 2588 2599 23002094 Oh HK Tan AL Das K Ooi CH Deng NT Tan IB Beillard E Lee J Ramnarayanan K Rha SY Genomic loss of miR-486 regulates tumor progression and the OLFM4 antiapoptotic factor in gastric cancer. Clin. Cancer Res 2011 17 2657 2667 21415212 Park IH Zhao R West JA Yabuuchi A Huo H Ince TA Lerou PH Lensch MW Daley GQ Reprogramming of human somatic cells to pluripotency with defined factors. Nature 2008 451 141 146 (Epub 2007 Dec 2023) 18157115 Park KS Kim KK Piao ZH Kim MK Lee HJ Kim YC Lee KS Lee JH Kim KE Olfactomedin 4 suppresses tumor growth and metastasis of mouse melanoma cells through downregulation of integrin and MMP genes. Mol. Cells 2012 34 555 561 23161172 Perlingeiro RC Kyba M Daley GQ Clonal analysis of differentiating embryonic stem cells reveals a hematopoietic progenitor with primitive erythroid and adult lymphoid-myeloid potential. Development 2001 128 4597 4604 11714684 Petzer AL Eaves CJ Lansdorp PM Ponchio L Barnett MJ Eaves AC Characterization of primitive subpopulations of normal and leukemic cells present in the blood of patients with newly diagnosed as well as established chronic myeloid leukemia. Blood 1996 88 2162 2171 8822936 Pfaffl MW A new mathematical model for relative quantification in real-time RT-PCR. Nucleic Acids Res 2001 29 e45 11328886 Rais Y Zviran A Geula S Gafni O Chomsky E Viukov S Mansour AA Caspi I Krupalnik V Zerbib M Deterministic direct reprogramming of somatic cells to pluripotency. Nature 2013 502 65 70 24048479 Ramaraj P Singh H Niu N Chu S Holtz M Yee JK Bhatia R Effect of mutational inactivation of tyrosine kinase activity on BCR/ABL-induced abnormalities in cell growth and adhesion in human hematopoietic progenitors. Cancer Res 2004 64 5322 5331 15289338 Ramos-Mejia V Montes R Bueno C Ayllon V Real PJ Rodriguez R Menendez P Residual expression of the reprogramming factors prevents differentiation of iPSC generated from human fibroblasts and cord blood CD34+ progenitors. PLoS One 2012 7 e35824 22545141 Schemionek M Elling C Steidl U Baumer N Hamilton A Spieker T Gothert JR Stehling M Wagers A Huettner CS BCR-ABL enhances differentiation of long-term repopulating hematopoietic stem cells. Blood 2010 115 3185 3195 20053753 Sebaugh JL Guidelines for accurate EC50/IC50 estimation. Pharm. Stat 2011 10 128 134 22328315 Sloma I Jiang X Eaves AC Eaves CJ Insights into the stem cells of chronic myeloid leukemia. Leukemia 2010 24 1823 1833 20861912 Slukvin II Neoplastic blood cells become pluripotent. Blood 2009 114 5409 5410 20035040 Slukvin II Hematopoietic specification from human pluripotent stem cells: current advances and challenges toward de novo generation of hematopoietic stem cells. Blood 2013 122 4035 4046 24124087 Takahashi K Tanabe K Ohnuki M Narita M Ichisaka T Tomoda K Yamanaka S Induction of pluripotent stem cells from adult human fibroblasts by defined factors. Cell 2007 131 861 872 18035408 Thierry-Mieg D Thierry-Mieg J AceView: a comprehensive cDNA-supported gene and transcripts annotation. Genome Biol 2006 7 Suppl. 1 S12 11 14 16925834 Tomarev SI Nakaya N Olfactomedin domain-containing proteins: possible mechanisms of action and functions in normal development and pathology. Mol. Neurobiol 2009 40 122 138 19554483 Udomsakdi C Eaves CJ Lansdorp PM Eaves AC Phenotypic heterogeneity of primitive leukemic hematopoietic cells in patients with chronic myeloid leukemia. Blood 1992a 80 2522 2530 1384787 Udomsakdi C Lansdorp PM Hogge DE Reid DS Eaves AC Eaves CJ Characterization of primitive hematopoietic cells in normal human peripheral blood. Blood 1992b 80 2513 2521 1384786 van der Flier LG Haegebarth A Stange DE van de Wetering M Clevers H OLFM4 is a robust marker for stem cells in human intestine and marks a subset of colorectal cancer cells. Gastroenterology 2009 137 15 17 19450592 Vargaftig J Taussig DC Griessinger E Anjos-Afonso F Lister TA Cavenagh J Oakervee H Gribben J Bonnet D Frequency of leukemic initiating cells does not depend on the xenotransplantation model used. Leukemia 2012 26 858 860 21926966 Verfaillie CM McCarthy JB McGlave PB Mechanisms underlying abnormal trafficking of malignant progenitors in chronic myelogenous leukemia. Decreased adhesion to stroma and fibronectin but increased adhesion to the basement membrane components laminin and collagen type IV. J. Clin. Invest 1992 90 1232 1241 1383271 Vo LT Daley GQ De novo generation of HSCs from somatic and pluripotent stem cell sources. Blood 2015 125 2641 2648 25762177 Vodyanik MA Bork JA Thomson JA Slukvin II Human embryonic stem cell-derived CD34+ cells: efficient production in the coculture with OP9 stromal cells and analysis of lymphohematopoietic potential. Blood 2005 105 617 626 15374881 Vodyanik MA Thomson JA Slukvin II Leukosialin (CD43) defines hematopoietic progenitors in human embryonic stem cell differentiation cultures. Blood 2006 108 2095 2105 16757688 Ye Z Zhan H Mali P Dowey S Williams DM Jang YY Dang CV Spivak JL Moliterno AR Cheng L Human-induced pluripotent stem cells from blood cells of healthy donors and patients with acquired blood disorders. Blood 2009 114 5473 5480 19797525 Yu J Hu K Smuga-Otto K Tian S Stewart R Slukvin II Thomson JA Human induced pluripotent stem cells free of vector and transgene sequences. Science 2009 324 797 801 19325077 Zhang J Liu WL Tang DC Chen L Wang M Pack SD Zhuang Z Rodgers GP Identification and characterization of a novel member of olfactomedin-related protein family, hGC-1, expressed during myeloid lineage development. Gene 2002 283 83 93 11867215 Zhang X Huang Q Yang Z Li Y Li CY GW112, a novel antiapoptotic protein that promotes tumor growth. Cancer Res 2004 64 2474 2481 15059901

PMC005xxxxxx/PMC5003783.txt

==== Front 3 Biotech3 Biotech3 Biotech2190-5738Springer Berlin Heidelberg Berlin/Heidelberg 50110.1007/s13205-016-0501-zOriginal ArticleImmobilized lipase catalyzing glucose stearate synthesis and their surfactant properties analysis Sebatini A. Maria 1Jain Manisha 1Radha P. 1Kiruthika S. 2Tamilarasan Krishnamurthi tamilbio@gmail.com 21 Department of Biotechnology, School of Bioengineering, SRM University, Chennai, Tamilnadu 603203 India 2 Department of Chemical Engineering, School of Bioengineering, SRM University, Chennai, Tamilnadu 603203 India 29 8 2016 29 8 2016 12 2016 6 2 18429 6 2016 16 8 2016 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.Sugar fatty acid esters are practical importance and have a variety of applications that include surfactants and as an emulsifying agent. In this study, we report glucose stearate synthesis using lipase-Fe3O4 nanoparticles catalyst. The influence of various reaction factors, such as silica gel concentration, molar ratio of sugar/acid, reaction temperature and speed of agitation on esterification by immobilized enzyme was analyzed. The glucose stearate esterification degree of 87.2 % was obtained under the optimized condition: 1:2 molar ratio of glucose/stearic acid, 2 % (w/v) of silica gel at 120 rpm and 40 °C. Glucose esters were characterized according to their surfactant activity like emulsification index, oil displacement activity and antimicrobial activity. The results indicated glucose stearate acts as biosurfactant, with emulsification index of 66 % in mustard oil and oil displacement activity of 19.64 cm2. Keywords Glucose stearateSilica gelEmulsification indexAntimicrobial activityissue-copyright-statement© King Abdulaziz City for Science and Technology 2016 ==== Body Introduction Biosurfactants are surface active substances, consisting both hydrophobic (lipids) and hydrophilic (sugar) portion on its molecules (Mulligan 2005). The main function of surfactants is to reduce surface and interface tensions between hydrophobic substances (oil, hydrocarbons and sterols) and hydrophilic water molecules (Desai and Banat 1997). In recent years, surfactants play an important role in medical, food, agricultural, cosmetic and bioremediation process industries (Amézcua et al. 2007; Dickinson 2009; Joshi et al. 2008; Nguyen et al. 2010; Rodrigues et al. 2006; Seydlová and Svobodová 2008). Many carbohydrate esters synthesis by lipase enzyme, for instance, are used as antibacterial agents in food industry. The ester may be produced from renewable source and inexpensive substances under mild reaction conditions, which minimize side reactions compared to the chemical process (Nair et al. 2012). The environmental concern about chemical surfactants is promoting the exploration on the use of biosurfactant. In the past few years, there is a huge demand for biosurfactants over chemical surfactant because of its low toxicity to environment and biodegradability (Chamouleau et al. 2001; Nitschke and Csta 2007; Park et al. 2004; Tsavas et al. 2002). The application of chemically synthesized sugar esters is limited, because they are produced at high temperatures in toxic solvents and expensive purification is required. Even though the application of this surfactant is diversified, it is being limited due to problems associated with its preparation (Maja et al. 2008; Hill and Rhode 1999). The nonionic biosurfactants of sugar ester are obtained from microbial and enzymatic process using renewable and inexpensive substrate (Atanu et al. 2008; Kshirsagar and Singhal 2007; Sabeder et al. 2006; Ye et al. 2010). Enzymatic process is quite favorable for sugar ester synthesis due to high specificity, high efficiency and lesser downstream process (Li et al. 2010; Plou et al. 2002; Sengupta et al. 2010). Many of the reports suggested that removal of water molecule during the esterification in reaction medium via molecular sieves, azeotropic distillation and pervaporation methods (Seydlová and Svobodová 2008). However, the industrial application of the biocatalysts is limited because of its high cost and difficult processing (Yong et al. 2008). To overcome the above limitations enzyme immobilization on supports materials was used (Ming et al. 2012; Yan et al. 2014; Hamidah et al. 2009). Immobilizations help for better dispersion of enzyme in reaction medium and also reduce enzyme contamination and facilitate easy separation of products (Badgujar et al. 2014). In recent years, nanoparticles have been used as support material for lipase immobilization. Among the nanostructures, magnetic nanoparticles are low in toxicity and are easy to separate from reaction medium by applying a magnetic field (Jiang et al. 2009; Liu et al. 2011; Sohrabi et al. 2014). Recently, glucose esters synthesis by immobilized Candida antarctica lipase catalysis has been reported (Jiang et al. 2009). In the present work, we report lipase covalently immobilized on functionalized magnetic nanoparticles. We have investigated enzymatic synthesis of glucose ester using glucose and stearic acid as substrate. The effect of various reaction conditions for glucose ester synthesis was studied. We have also characterized the glucose esters by their surfactant and antimicrobial properties. Materials and methods Chemicals and microorganism Microorganisms were obtained from MTCC, Institute of Microbial Technology (IMTECH), Chandigarh, India. All the media components used were of analytical grade, and were purchased from Hi Media Laboratories Pvt Ltd, (Mumbai, India). The solvents and stearic acid were purchased from Merck specialties Pvt Ltd, Mumbai, India. Lipase enzyme (Rhizobium oryzae) was purchased from sigma chemical, USA. Preparation of immobilized lipase enzyme Magnetic nanoparticles (MNP) were prepared by co-precipitation method (Chia et al. 2012; Pan et al. 2009; Manisha et al. 2015). 400 mg of nanoparticles were dispersed in 0.4 g chitosan containing 20 mL of 1 % acetic acid solution. 1 N sodium hydroxide was slowly added to the reaction mixture to precipitate the chitosan coated MNP. 30 mg of lipase was added to 0.25 % N-(3-dimethylaminoproyl)-N-ethylcarbodiimide (EDC) containing 4 mL of 50 mM phosphate buffer solution (with 200 mM NaCl) and the solution was incubated at 25 °C for 1 h with shaking. Then, 30 mg of N-hydroxysuccinimide (NHS) was added to the solution and the incubation was continued for another 1 h. 50 mg of the chitosan coated MNP was added to this solution and incubated for further 4 h. The activated carboxyl groups of enzyme combined with amino group of nanoparticle by simple carbodiimide reaction to synthesized complex immobilized lipase (Manisha et al. 2016). Experimental setup Glucose esters were synthesized by esterification reactions. The ester synthesis was carried out in glass vials by adding various molar ratio of glucose and stearic acid with silica gel in 25 mL of isooctane. The reaction mixture was stirred at 120 rpm for 10 min, later on reaction was initiated by adding 100 mg of immobilized lipase enzyme then kept at room temperature for 24 h. At the end of the reaction, the immobilized enzyme and silica gel were removed by filtration using filter paper with a pore size of 60 mm. Afterward, the sample was analysed to determine amount of ester content in the reaction mixture. Optimization of esterification condition of glucose ester with immobilized lipase The experiments were conducted under 1:1–1:4 molar ratio of glucose to stearic acid, reaction temperature (20–60 °C) and speed of agitation (0–240 rpm) with 0–6 % (w/v) silica gel content. The sugar ester yield was analyzed using the sample withdrawn from the reaction mixtures. At the end of the esterification reaction, immobilized enzyme and unconverted substrate were separated by filtration and centrifugation methods. The filtrate containing the ester product was quantified by titration method. Figure 1 shows the scheme for lipase catalyzed synthesis of glucose stearate ester.Fig. 1 Lipase-catalyzed synthesis of glucose stearate ester Quantification of glucose ester The glucose ester content was quantified by calculating the residual fatty acid amount in the reaction mixture, which was determined by the titration method (Leitgeb and Knez 1990). Briefly, 2 mL sample from the reaction mixture were titrated against 0.001 N standardized sodium hydroxide solution using phenolphthalein indicator. The yield was calculated using the formula: 1 Yield%=100-X1X2]]×100 X1sample=V1×NW1 X2(control)=V2×NW2 where V 1 volume of sodium hydroxide used for sample, V 2 volume of sodium hydroxide used for control, N normality of sodium hydroxide, W 1 weight of sample, W 2 weight of control. All the measurements were performed in triplicate and the results represent the standard deviation. Emulsification index and oil displacement activity of glucose ester The emulsifying capacity of glucose stearate was analyzed by emulsification index (Leitgeb and Knez 1990; Nair et al. 2012). The emulsification index of ester sample was determined by adding oil and sugar ester (1:2 ratio) was homogenized using a vortex for 5 min. The emulsions were left to settle for 48 h, to measured height of the emulsion layer at 2 min, 24 and 48 h intervals. The emulsification index was calculated as the ratio of measured height of the emulsion layer to the total height of mixture and multiplying it by 100 (Nair et al. 2012). The oil displacement assay is a convenient method for surfactant activity studied (Morikawa et al. 1993). 30 mL of distilled water was added to a petri plate. 50 µL of oil was then added to the water surface followed by 10 µL of glucose ester (1 mg/mL) dropped on the center of oil surface. The occurrence of clear zone is an indication of the surfactant activity of ester. The activity of ester was directly proportional to oil displacement area in petri plate. Antimicrobial activity of glucose ester The antimicrobial activity of glucose esters was tested against the Bacillus subtilis, Bacillus cereus, Bacillus megaterium and E. coli (Manuel et al. 2005; Vagi et al. 2005). The activity of esters was tested in nutrient liquid media for the growth of microorganism. The stock solution of sugar ester (5 mg/mL) was prepared to add in the liquid media. 100 µL of overnight culture microorganism was used as inoculate for test medium and incubation for 24 h at 37 °C. Antimicrobial activity was tested by measuring the turbidity of growth using UV–vis spectrophotometer at 600 nm. The growth of micro-organisms in the medium containing defined concentration of sugar fatty acid ester were compared with those obtained in a medium without sugar fatty acid ester (control) (Maja et al. 2008). The percentage of inhibition of microorganism in test medium was calculated compared to growth of control medium. Results and discussion Effect of silica gel content In esterification reaction, water content in the reaction mixture affects the lipase activity and favors the equilibrium state (Chortyk 1996). Therefore, the effect of water content in reaction medium for enzymatic synthesis of glucose ester was measured. In this work esterification reaction were carried out with various concentrations (0–6 % w/v) of silica gel adsorbent was studied. Silica gel not only dried the reaction medium but also shifted the equilibrium to the synthesis of glucose ester by adsorbing the water molecule (Jiugao et al. 2008). As shown in Fig. 2, the concentration of silica gel increased to 2 %, the esterification increased up to 75 %, but on further increase in silica gel concentration, the esterification decreased significantly. It might be due to adsorption of the minimal water necessary for the enzyme activity in the reaction. From the result maximum conversion (85.4 %) of glucose stearate was obtained, when 2 % silica gel was used.Fig. 2 Effect of silica gel concentration on glucose stearate synthesis Effect of temperature and speed of agitation Temperature is an important parameter for any enzymatic reaction that increases the molecular collision and improves the substrate solubility in reaction media (Badgujar et al. 2013). The effect of reaction temperature (20–60 °C) on esterification was tested. As shown in Fig. 3a, the lower reaction temperature resulted in poor conversion because of the increase in viscosity of medium that led to higher mass transfer resistance. When reaction temperature reached higher than 40 °C, the esterification decreased slowly, because of the equilibrium of the reaction and the loss of enzyme activity. From the results the highest esterification 86 % was obtained at 40 °C.Fig. 3 a Effect of temperature on glucose stearate synthesis, b effect of speed of agitation on glucose stearate synthesis The effect of external mass transfer limit was analyzed by varying the speed of agitation, which is imperative for the enzymatic reaction. This is a micro aqueous solid–liquid system in which reactants are in liquid phase while enzyme is in solid phase (Badgujar et al. 2013). The effect of speed of agitation on esterification reaction was examined in solvent medium (Fig. 3b). The conversion of glucose stearate increased with increasing speed of agitation from 60–120 rpm and then decreased slowly. High esterification (88 %) was obtained at 120 rpm after 24 h incubation. These results shows that mass transfer effect was observed in between the 60 to 120 rpm; as soon as mass transfer barrier is reached then there is no influence of the mass transfer diffusion on the reaction rate and after 220 rpm there was no significant increase in the activity (Yadav and Pawar 2012). Effect of substrate molar ratio The effect of stearic acid concentration on glucose stearate synthesis was examined at constant glucose concentration. In a set of experiments, glucose was kept constant at 1 mM and the quantity of stearic acid was varied as 1:1, 1:2, 1:3 and 1:4 mM. The esterification rate increased up to 1:2 ratio, after that the esterification rate decreased (Fig. 4). The highest conversion of 87 % was obtained at the glucose/stearic acid molar ratios of 1:2. Lower the molar ratio of stearic acid, probability to get low active sites then difficult to catalyze reaction. At higher molar ratio of stearic acid, lower solubility in organic solvent destroys the balance of esterification, and the reaction is hindered. Similar type of glucose ester was synthesized by an immobilized lipase with 0.8 g molecular sieves/2 mL acetone at 40 °C for 48 h, and the 38 % acid conversion was obtained (Jiugao et al. 2008). In addition, a higher stearic acid concentration may change the catalytic environment and the active site of immobilized enzyme (He et al. 2010).Fig. 4 Effect of molar ratio of sugar/fatty acid on glucose stearate synthesis Reusability of the immobilized lipase Reusability of the immobilized enzyme is relatively important for its industrial application. To investigate the reusability, immobilized enzyme was first washed with ethanol and then with deionized water after one reaction cycle and reintroduced into a new esterification reaction. The effect of repeated use of immobilized enzyme on esterification of glucose is shown in Fig. 5. The results were observed that the esterification was still retained (60 %) after the three reuses. After using three times, the conversion marginally decreased and it was due to the denaturation of the enzyme on MNP support during separation from reaction mixture.Fig. 5 Reusability of the immobilized lipase for glucose stearate synthesis Glucose stearate ester production The reaction mixture consists of 2 mM glucose and 4 mM stearic acid (1:2 molar ratio of sugar and acid) dissolved in 50 mL of isooctane with 500 mg of immobilized lipase. To improve conversion, 100 mg silica gel was added to adsorb water generated from the reaction mixture during esterification. The reaction mixture was incubated at 40 °C, 120 rpm for 48 h. At the end of the esterification reaction, reaction mixture was separated by filtration and centrifugation methods. The filtrate contains ester product; it was concentrated by rotary vacuum evaporator. A schematic diagram of the overall process of glucose ester synthesis in batch process is shown in Fig. 6.Fig. 6 Schematic diagrams of glucose stearate synthesis and purification process Glucose stearate structural analysis The Fourier transform-IR spectrum chart of purified glucose stearate is shown in Fig. 7. The observed band characteristic of C–O (1113 cm−1) indicated that glucose stearate contained some sugars. The C=O stretching is calculated from 1702 cm−1 and combination band of OCH and COH (1463, 1422 cm−1) and C–H (2918, 2850 cm−1) can be observed. Similar results were obtained by other authors (Jiugao et al. 2008) when determining the chemical structure of 6-O-glucose stearate synthesized by Candida lipase.Fig. 7 Fourier-transform infrared spectra of glucose stearate Oil displacement assay Oil displacement assay is a more sensitive analysis for surface active compounds (Mulligan 2005). Oil displacement activity of ester attributed to the formation of critical micelle concentration which helps to reduce the surface tension between two interfaces. In this study oil displacement activity was measured by the area of clear zone formed on the oil–water surface (Fig. 8). 19.64 m2 clear zone was formed after adding 1 mg/mL of glucose ester on the surface of oil drop. The reduction in the surface tension was observed to formation of clear zone at the surface. The zone area formation is directly proportional to surfactant concentration.Fig. 8 Oil displacement activity of glucose stearate. a Before adding glucose stearate and b after adding glucose stearate Emulsification index Emulsion index was studied at 2 min, 24 and 48 h time intervals with 1:2 (glucose ester: oil sample) ratio mixture. The comparison emulsification results of different oil are shown in Fig. 9. The highest emulsification indexes of 71 and 61 % were obtained from mustard oil and neem oil, respectively, at 24 h. Emulsification indexes of olive oil and castor oil was calculated to be 57.2 and 57.1 %, respectively. Higher emulsification index represents higher stability. Sugar ester achieved highest stability (EI 66 %) in mustard oil at 48 h (Nitschke and Csta 2007).Fig. 9 Emulsification index of glucose stearate Antimicrobial activity of sugar ester Sugar esters are valuable compound in food industry as antibacterial agents because it is biodegradable and nontoxic. Most of the previous studies on antimicrobial properties of commercial sugar esters were tested against different microorganisms (Plou et al. 2002; Hathcox and Beuchat 1996; Tsuchido et al. 1993). Our enzymatically synthesized glucose stearate antimicrobial properties were studied against various bacterial species as shown in Fig. 10. Among the bacteria tested, Bacillus subtilis, Bacillus cereus and Bacillus megaterium was inhibited 31, 26 and 42 %, respectively, compare to original growth. In addition, glucose ester showed strong inhibition (52 %) of E. coli growth at 1 mg/mL concentration. Similarly, other reported that sucrose monolaurate strongly inhibits the growth of E. coli in concentration of 1 mg/ml (Kato and Arima 1971). But pervious report show that about 10 % higher inhibition against Bacillus cereus was obtained after 24 h of growth, when enzymatically synthesized sugar ester.Fig. 10 Growth inhibition activity of glucose stearate Conclusion We prepared lipase immobilized nanoparticles and employed it as a biocatalyst for sugar ester synthesis. The optimal reaction conditions that achieved highest esterification rate (87.2 %) were as follows: 1:2 molar ratio (glucose: stearic acid) with 2 % (w/v) silica gel at 40 °C, and 120 rpm. Furthermore, the glucose ester showed better surfactant properties with 67 % of emulsification index (EI) and 19.64 cm2 oil displacement activities. Glucose ester showed highest antibacterial activity against E. coli with 50 % of inhibition. Authors are acknowledge and thanks the Management of SRM University and Director (E&T) for their support to carry out this research work and also thank the Department of Biotechnology and Department of Chemical Engineering for providing necessary facilities. Compliance with ethical standards Conflict of interest The authors declared that there is no conflict of interest on publication of this article. ==== Refs References Amézcua VC Poggi VHM Esparza GF Ríos LE Rodríguez VR Effect of culture conditions on fatty acids composition of a biosurfactant produced by Candida ingens and changes of surface tension of culture media Bioresour Technol 2007 98 237 240 10.1016/j.biortech.2005.11.025 16413180 Atanu B Shogren RL Gordon S Salch J Willett JL Charles MB Rapid and environmentally friendly preparation of starch esters Carbohydr Polym 2008 74 137 141 10.1016/j.carbpol.2008.01.013 Badgujar KC Bhanage BM Application of lipase immobilized on the biocompatible ternary blend polymer matrix for synthesis of citronellyl acetate in non-aqueous media: kinetic modelling study Enzyme Microb Technol 2014 57 16 25 10.1016/j.enzmictec.2014.01.006 24629263 Badgujar KC Dhake KP Bhanage BM Immobilization of Candida cylindracea lipase on polylactic acid, polyvinyl alcohol and chitosan based ternary blend film: characterization, activity, stability and its application for N -acylation reactions Process Biochem 2013 48 1335 1347 10.1016/j.procbio.2013.06.009 Chamouleau F Coulon D Girardin M Ghoul M Influence of water activity and water content on sugar esters lipase-catalyzed synthesis in organic media J Mol Catal B Enzym 2001 11 949 954 10.1016/S1381-1177(00)00166-1 Chia HK Yung CL Chieh MJC Jiann HC Cheng C Chwen JS Optimum conditions for lipase immobilization on chitosan-coated Fe3 O4 nanoparticles Carbohydr Polym 2012 87 2538 2545 10.1016/j.carbpol.2011.11.026 Chortyk OT Synthesis and characterization of insecticidal sucrose esters J Agric Food Chem 1996 44 1551 1557 10.1021/jf950615t Desai JD Banat IM Microbial production of surfactants and their commercial potential Mol Biol Rev 1997 61 47 64 Dickinson E Hydrocolloids as emulsifiers and emulsion stabilizers Food Hydroc 2009 23 1473 1482 10.1016/j.foodhyd.2008.08.005 Hamidah B Raizatul AGAR Noorullhamezon MdN Suzaini B Hamidah S Enzymatic synthesis of palm-based ascorbyl esters J Mol Catal B Enzym 2009 58 153 157 10.1016/j.molcatb.2008.12.012 Hathcox AK Beuchat LR Inhibitory effects of sucrose fatty acid esters, alone and in combination with ethylenediamine tetraacetic acid and other organic acids, on viability of Escherichia coli O157:H7 Food Microbiol 1996 13 213 225 10.1006/fmic.1996.0027 He WS Jia CS Ma Y Yang YB Zhang XM Feng B Yue L Lipasecatalyzed synthesis of phytostanyl esters in non-aqueous media J Mol Catal B: Enzym 2010 67 60 65 10.1016/j.molcatb.2010.07.006 Hill K Rhode O Sugar-based surfactants for consumer products and technical applications Fett Lipid 1999 101 25 33 10.1002/(SICI)1521-4133(19991)101:1<25::AID-LIPI25>3.0.CO;2-N Jiang Y Guo C Xia H Mahmood I Liu C Liu H Magnetic nanoparticles supported ionic liquids for lipase immobilization: enzyme activity in catalyzing esterification J Mol Catal B Enzym 2009 58 103 109 10.1016/j.molcatb.2008.12.001 Jiugao YU Jianshe Z Ang Z Xiaofei M Study of glucose ester synthesis by immobilized lipase from Candida sp Catal Commun 2008 9 1369 1374 10.1016/j.catcom.2007.11.036 Joshi S Bharucha C Desai AJ Production of biosurfactant and antifungal compound by fermented food isolate Bacillus subtilis 20B Bioresour Technol 2008 99 4603 4608 10.1016/j.biortech.2007.07.030 17855083 Kato A Arima K Inhibitory effect of sucrose ester of lauric acid on the growth of Escherichia coli Biochem Biophys Res Commun 1971 42 596 601 10.1016/0006-291X(71)90529-8 5543940 Kshirsagar AC Singhal RS Optimization of starch oleate derivatives from native corn and hydrolyzed corn starch by response surface methodology Carbohydr Polym 2007 69 455 461 10.1016/j.carbpol.2007.01.007 Leitgeb M Knez Z The influence of water on the synthesis of n -butyl oleate by immobilized mucor miehei lipase J Am Oil Chem Soc 1990 67 775 778 10.1007/BF02540490 Li R Jia CS Yue L Zhang XM Xia QY Zhao SL Feng B Zhong F Chen WJ Lipase-catalyzed synthesis of conjugated linoleyl β-sitosterol and its cholesterol-lowering properties in mice J Agric Food Chem 2010 58 1898 1902 10.1021/jf902943y 20055411 Liu Y Jia S Wu Q Ran J Zhang W Wu S Studies of Fe3 O4 -chitosan nanoparticles prepared by co-precipitation under the magnetic field for lipase immobilization Catal Commun 2011 12 717 720 10.1016/j.catcom.2010.12.032 Maja H Sasa S Zeljko K Enzymatic synthesis of sugar fatty acid esters in organic solvent and in supercritical carbon dioxide and their antimicrobial activity J of Supercrit Fluids 2008 45 338 345 10.1016/j.supflu.2008.01.002 Manisha J Mariya S Sharmila G Muthukumaran C Baskar G Tamilarasan K Fabrication of a chitosan-coated magnetic nanobiocatalyst for starch hydrolysis Chem Eng Technol 2015 38 1444 1451 10.1002/ceat.201400493 Manisha J Mariya S Radha S Kiruthika S Muthukumaran C Tamilarasan K Synthesis, characterization and kinetic analysis of chitosan coated magnetic nanobiocatalyst and its application on glucose oleate ester synthesis J Mol Catal B Enzym 2016 128 1 9 10.1016/j.molcatb.2016.02.006 Manuel F Juan S Francisco JP Nieves LC Dolores RD Morten C Jose LCP Antonio B Synthesis of sugar esters in solvent mixtures by lipases from Thermomyces lanuginosus and Candida antarctica B, and their antimicrobial properties Enzyme Microb Technol 2005 36 391 398 10.1016/j.enzmictec.2004.02.009 Ming MZ Ling D Yong L Ping MG Qian CD Wen LL Yu QF Feng HH Immobilization of Candida rugosa lipase on magnetic poly (allyl glycidylether-co-ethylene glycol dimethacrylate) polymer microsphere for synthesis of phytosterol esters of unsaturated fatty acids J Mol Catal B Enzym 2012 74 16 23 10.1016/j.molcatb.2011.08.008 Morikawa M Daido H Takao T Murata S Shimonishi Y Imanaka T A new lipopeptide biosurfactant produced by Arthrobacter sp. strain MIS38 J Bacteriol 1993 175 6459 6466 10.1128/jb.175.20.6459-6466.1993 8407822 Mulligan CN Enviromental applications for biosurfactants Environ Pollut 2005 133 183 198 10.1016/j.envpol.2004.06.009 15519450 Nair NAS RJosé CSS Soraya OS Sueli R Luciana RBG Ligia RR José AT Enzymatic synthesis of sugar esters and their potential as surface-active stabilizers of coconut milk emulsions Food Hydroc 2012 27 324 331 10.1016/j.foodhyd.2011.10.009 Nguyen TTL Edelen A Neighbors B Sabatini DA Biocompatible lecithin-based microemulsions with rhamnolipid and sophorolipid biosurfactants: formulation and potential applications J Colloid Interface Sci 2010 348 498 504 10.1016/j.jcis.2010.04.053 20471022 Nitschke M Csta SGVAO Biosurfactants in food industry Trends Food Sci Tech 2007 18 252 259 10.1016/j.tifs.2007.01.002 Pan C Hu B Li W Sun Y Ye Zeng X Novel and efficient method for immobilization and stabilization of β-d-galactosidase by covalent attachment onto magnetic Fe3 O4 –chitosan nanoparticles J Mol Catal B Enzym 2009 61 208 215 10.1016/j.molcatb.2009.07.003 Park D Haam S Ahn I Lee TG Kim G Kim W Enzymatic esterification of B-methylglucoside with acrylic/methacrylic acid in organic solvents J Biotechnol 2004 107 151 160 10.1016/j.jbiotec.2003.09.004 14711498 Plou FJ Cruces MA Ferrer M Fuentes G Pastor E Bernabe M Enzymatic acylation of di- and trisaccharides with fatty acids: choosing the appropriate enzyme, support and solvent J Biotechnol 2002 96 55 66 10.1016/S0168-1656(02)00037-8 12142143 Rodrigues L Banat IM Teixeira J Oliveira R Biosurfactants: potential applications in medicine J Antimicrob Chemoth 2006 57 609 618 10.1093/jac/dkl024 Sabeder S Habulin H Knez Z Lipase-catalyzed synthesis of fatty acid fructose esters J Food Eng 2006 77 880 886 10.1016/j.jfoodeng.2005.08.016 Sengupta A Pal M SilRoy S Ghosh M Comparative study on sterol ester synthesis using Thermomyces lanuginosus lipase in stirred tank and packed-bed bioreactors J Am Oil Chem Soc 2010 87 1019 1025 10.1007/s11746-010-1587-9 Seydlová G Svobodová I Review of surfactin chemical properties and the potential biomedical applications Cent Eur J Med 2008 3 123 133 Sohrabi N Rasouli N Torkzadeh M Enhanced stability and catalytic activity of immobilized α-amylase on modified Fe3 O4 nanoparticles Chem Eng J 2014 240 426 433 10.1016/j.cej.2013.11.059 Tsavas P Polydorou S Faflia I Voutsas EC Tassios D Flores MV Naraghi K Halling PJ Chamouleau F Ghoul M Engasser JM Ferrer M Plou FJ Solubility of glucose in mixtures containing 2-methyl-2-butanol, dimethyl sulfoxide, acids, esters, and water J Chem Eng Data 2002 47 807 810 10.1021/je0102457 Tsuchido T Yokosula N Takano M Isolation and characteristics of a Bacillus subtilis mutant tolerant to the lytic action of sucrose esters of long-chain fatty-acids J Ferment Bioeng 1993 75 191 195 10.1016/0922-338X(93)90114-N Vagi E Simandi B Suhajda A Hethelyi E Essential oil composition and antimicrobial activity of Origanum majorana L. extracts obtained with ethyl alcohol and supercritical carbon dioxide Food Res Int 2005 38 51 57 10.1016/j.foodres.2004.07.006 Yadav GD Pawar SV Synergism between microwave irradiation and enzyme catalysis in transesterification of ethyl-3-phenylpropanoate with n -butanol Bioresour Technol 2012 109 1 6 10.1016/j.biortech.2012.01.030 22305539 Yan W Jiaying X Jia S Wenlong W Chungu X A kinetic study of starch palmitate synthesis by immobilized lipase-catalyzed esterification in solvent free system J Mol Catal B Enzym 2014 101 73 79 10.1016/j.molcatb.2014.01.003 Ye R Pyo SH Hayes DG Lipase-catalyzed synthesis of saccharide-fatty acid esters using suspensions of saccharide crystals in solvent-free media J Am Oil Chem Soc 2010 87 281 293 10.1007/s11746-009-1504-2 Yong Y Bai YX Li YF Lin L Cui YJ Xi CG Characterization of Candida rugosa lipase immobilized onto magnetic microspheres with hydrophilicity Process Biochem 2008 43 1179 1185 10.1016/j.procbio.2008.05.019

PMC005xxxxxx/PMC5003784.txt

==== Front Protein CellProtein CellProtein & Cell1674-800X1674-8018Higher Education Press Beijing 28510.1007/s13238-016-0285-2Research ArticleNovel matrine derivative MD-1 attenuates hepatic fibrosis by inhibiting EGFR activation of hepatic stellate cells Feng Yi orchidfy@hotmail.com Ying Hai-yan Qu Ying Cai Xiao-bo Xu Ming-yi Lu Lun-gen Lungenlu1965@163.com Department of Gastroenterology, Shanghai General Hospital, Nanjing Medical University, Shanghai, 200080 China 24 6 2016 24 6 2016 9 2016 7 9 662 672 27 3 2016 24 5 2016 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.Matrine (MT), the effective component of Sophora flavescens Ait, has been shown to have anti-inflammation, immune-suppressive, anti-tumor, and anti-hepatic fibrosis activities. However, the pharmacological effects of MT still need to be strengthened due to its relatively low efficacy and short half-life. In the present study, we report a more effective thio derivative of MT, MD-1, and its inhibitory effects on the activation of hepatic stellate cells (HSCs) in both cell culture and animal models. Cytological experiments showed that MD-1 can inhibit the proliferation of HSC-T6 cells with a half-maximal inhibitory concentration (IC50) of 62 μmol/L. In addition, MD-1 more strongly inhibits the migration of HSC-T6 cells compared to MT and can more effectively induce G0/G1 arrest and apoptosis. Investigating the biological mechanisms underlying anti-hepatic fibrosis in the presence of MD-1, we found that MD-1 can bind the epidermal growth factor receptor (EGFR) on the surface of HSC-T6 cells, which can further inhibit the phosphorylation of EGFR and its downstream protein kinase B (Akt), resulting in decreased expression of cyclin D1 and eventual inhibition of the activation of HSC-T6 cells. Furthermore, in rats with dimethylnitrosamine (DMN)-induced hepatic fibrosis, MD-1 slowed the development and progression of hepatic fibrosis, protecting hepatic parenchymal cells and improving hepatic functions. Therefore, MD-1 is a potential drug for anti-hepatic fibrosis. Electronic supplementary material The online version of this article (doi:10.1007/s13238-016-0285-2) contains supplementary material, which is available to authorized users. Keywords matrine derivativehepatic stellate cellhepatic fibrosisepidermal growth factor receptorsignal transduction pathwayissue-copyright-statement© HEP and Springer 2016 ==== Body INTRODUCTION Hepatic fibrosis is a pathological process comprising atypical hyperplasia (dysplasia) of intrahepatic connective tissue and excessive accumulation of extracellular matrix (ECM) that can ultimately develop into hepatic cirrhosis. The main pathogenic factors that induce hepatic fibrosis include chronic hepatitis virus (i.e., HBV and HCV) and parasite infection (e.g., schistosomes), alcoholic liver disease, fatty liver disease, cholestasis, drug-induced liver injury, and autoimmune hepatitis. The persistent presence of these factors causes the production and activation of hepatic fibrosis-initiating cells, especially the activation of hepatic stellate cells (HSCs). Other cells that can be activated are fibroblasts, liver sinusoidal endothelial cells, and Kupffer cells surrounding the portal vein; mesenchymal stem cells derived from bone marrow; blood fibrocytes; and hepatocytes or biliary epithelial cells that have undergone epithelial-mesenchymal transition (EMT) (Fausther et al., 2013; Iwaisako et al., 2014b; Pinzani, 2015; Koo et al., 2016). These cells can proliferate and transdifferentiate into myofibroblasts (MFs) and then synthesize and secret a large amount of ECM, causing hepatic fibrosis and cirrhosis (Fagone et al., 2015). Studies of the mechanisms underlying the development of hepatic fibrosis indicate that many factors associated with liver injuries can induce HSC activation (Seki and Brenner, 2015). Hepatic parenchymal cell injury can be caused by a variety of pathological factors and induce the synthesis and release of a variety of cytokines, including transforming growth factor β (TGF-β), epidermal growth factor (EGF), platelet-derived growth factor (PDGF), endothelin (ET), fibroblast growth factor (FGF), connective tissue growth factor (CTGF), and leptin, which can further activate the TGF-β/Smad, epidermal growth factor receptor (EGFR), and Ras/ERK signaling pathways to promote HSC proliferation and activity (Cuevas et al., 2011; Handy et al., 2011; Shimada et al., 2011; Liu et al., 2013). After activation, HSCs can transdifferentiate into MFs, which express α-smooth muscle actin (α-SMA), secrete ECM components (including type I and type II collagens and proteoglycan) that cause ECM accumulation, and promote the development of hepatic fibrosis. Therefore, inhibition of hepatic fibrosis through HSC inactivation is a major direction in the prevention and treatment of hepatic cirrhosis. Two major treatments are used against hepatic fibrosis: those that eliminate pathogenic factors that cause hepatic fibrosis (e.g., antiviral therapy and glucocorticoid treatment of autoimmune or alcoholic liver diseases) and those that target ECM accumulation by decreasing its synthesis and facilitating its degradation. Clinical treatment measures for anti-hepatic fibrosis are currently limited. Therefore, it is crucial to explore new types of anti-hepatic fibrosis drugs. Matrine (MT), the effective component of Sophora flavescens Ait, has been shown to have anti-inflammation, immune-suppressive, anti-tumor, and anti-hepatic fibrosis activities. MT also has protective effects against acute liver injury in mice and rats (Zhang et al., 2011; Liu et al., 2014). However, the clinical pharmacological effects of this drug still need to be strengthened due to its relatively low efficacy and short half-life. In the present study, we structurally modified MT and obtained a novel thio derivative called MD-1 (C30N4H40SO5F) through thiosulfate and side chain Michael addition to prepare maleate for anti-hepatic fibrosis experiments. We examined the effects of MD-1 on HSC activation and inhibition of hepatic fibrosis in rats. In addition, the possible mechanism of action of this novel MT derivative was investigated to determine if it is a potential clinical treatment option for hepatic fibrosis. RESULTS Inhibition of HSC-T6 cell proliferation by novel thio derivatives of MT A series of novel thio derivatives of MT, including MD-1, MD-2, and MD-3, were obtained from structural modifications to MT (Fig. 1A, also see Supplementary Information for details). First, we determined the effects of MT and its derivatives on the proliferation of HSC-T6 cells at a concentration of 100 µmol/L. MT treatment decreased the survival rate to 62.9% ± 10.0% (n = 3) compared to control (114.2% ± 8.5%, n = 3; P < 0.001, Fig. 1B). Among the three derivatives of MT, MD-1 had the strongest inhibitory effect on HSC-T6 cells, decreasing the cell survival rate to 24.2% ± 5.0%, which is significantly lower than that of MT (n = 3; P = 0.004). For MD-2 the cell survival rate was 27.4% ± 1.6% (n = 3; P < 0.01 compared to MT). The inhibitory effect of MD-3 on HSC-T6 cells was equivalent to that of MT (47.2% ± 8.1%, n = 3; P = 0.10). We also examined the effects of different concentrations of MD-1 on the proliferation of HSC-T6 cells. The survival rates of HSC-T6 cells decreased as the MD-1 concentration increased. The IC50 values of MD-1 and MT on HSC-T6 cells were 62 µmol/L and 128 µmol/L, respectively (Fig. 1C).Figure 1 Novel thio derivatives of MT inhibit the proliferation of HSC-T6 cells. (A) The chemical structure of MD-1, MD-2, and MD-3. (B) The inhibitory effects of MD-1, MD-2, and MD-3 on the proliferation of HSC-T6 cells (MD-1 group: 24.2% ± 5.0%; MD-2 group: 27.4% ± 1.6%; MD-3 group: 47.2% ± 8.1%; MT group: 62.9% ± 10.0%; control group: 114.2% ± 8.5%; n = 3 in each group). (C) A concentration-dependent plot of the effect of MD-1 and MT on the proliferation of HSC-T6 cells. IC50 of MD-1 and MT was 62 µmol/L and 128 µmol/L, respectively Inhibitory effects of MD-1 on HSC-T6 cell motility The inhibitory effects of MD-1 and MT on the migration of HSC-T6 cells were observed using the Transwell assay. To exclude the influence of inhibited cell proliferation, the experimental concentrations of MD-1 and MT were set at their corresponding IC50 values to compare their effects on cell migration at the same cell survival rate (50%). After 24 h of treatment, the difference in HSC-T6 cell migration between 62 µmol/L MD-1 and 128 µmol/L MT was not significant, but both significantly reduced motility compared to control (Fig. 2B; P < 0.05). However, after 48 h, the inhibitory effect of MD-1 on cell migration was significantly stronger than that of MT (P < 0.05, n = 3, Fig. 2A and 2B). Comparing the difference between 24 h and 48 h groups, the results showed that there was a significant different increase in MT-treated cells at 48 h compared with that at 24 h (P < 0.01), but not in MD-1-treated cells, suggesting a stronger effect of MD-1 on inhibiting HSC-T6 cell migration.Figure 2 Inhibitory effects of MD-1 on the motility of HSC-T6 cells. (A) Top: Representative picture to show the migration ability of HSC-T6 cells at the same cell viability (50%) after 24 h of drug action (MD-1, MT, and control). Bottom: After 48 h of drug action. (B) Statistics for the migration ability of HSC-T6 cells. *P < 0.05; **P < 0.01 Effects of MD-1 on cell cycle and apoptosis in HSC-T6 cells The effects of MD-1 on cell cycle and apoptosis in HSC-T6 cells were detected by flow cytometry. Compared to the control group, MD-1- and MT-treated HSC-T6 cells had an increased percentage of cells in G0/G1 phase and a decreased percentage of cells in the S phase (Fig. 3A). The change in the cell cycle induced by MD-1 was even more significant than the change induced by MT (G0/G1: 47.2% ± 3.2% in MD-1 group, 38.6% ± 2.5% in MT group, and 25.2% ± 3.0% in control; S: 33.6% ± 4.5% in MD-1 group, 36.6% ± 2.6% in MT group, and 58.9% ± 2.2% in control; G2/M: 18.5% ± 6.0% in MD-1 group, 24.9% ± 4.5% in MT group, and 17.4% ± 1.4% in control; n = 3 in each group). These results suggest that MD-1 could induce G0/G1 arrest in HSC-T6 cells and decrease the number of cells entering mitosis (Fig. 3A, right). The apoptosis rate in control HSC-T6 cells was 2.5% ± 1.3% (n = 3). After MD-1 or MT treatment for 48 h, the apoptosis rate increased to 25.6% ± 4.8% (n = 3; P = 0.001) or 7.6% ± 2.5% (n = 3; P < 0.05), respectively. The effect of MD-1 on the induction of apoptosis in HSC-T6 cells was significantly stronger than that of MT (P = 0.005; Fig. 3B, right).Figure 3 Inhibition of cell cycle and induction of apoptosis by MD-1 in HSC-T6 cells. (A) Left: The cell cycle distribution of HSC-T6 cells in the presence of MD-1 or MT at 62 μmol/L was examined by flow cytometry. Right: Statistics for cell cycle distribution. *P < 0.05; **P < 0.01; ***P < 0.001. (B) Left: The cell apoptosis rate was detected with flow cytometer in MD-1, MT and control group. *P < 0.05; **P < 0.01; ***P < 0.001 Effects of MD-1 on the EGFR signaling pathway After biotin-labeled MD-1 was prepared, the localization of MD-1 in HSC-T6 cells and its relationship with EGFR were examined by immunofluorescence labeling and co-immunoprecipitation assay. The localization of MD-1 was consistent with EGFR on the surface of the cell membrane (Fig. 4A). After knocking down the expression of EGFR in HSC-T6 cells using an EGFR-shRNA expression plasmid, immunofluorescence labeling showed that MD-1 binding to cell membranes decreased accordingly (Fig. 4B). Co-immunoprecipitation blots also showed a positive band of MD-1-Biotin was precipitated by EGFR in the MD-1-treated cells, but not in the parental cells (Fig. 4C). These results suggest that the target molecule of MD-1 in HSC-T6 cells was EGFR and that the drug exerted its function by binding to EGFR.Figure 4 Inhibitory effects of MD-1 on the EGFR signaling pathway in HSC-T6 cells. (A) Immunostaining of EGFR-positive HSC-T6 cells showed MD-1 localization. Nuclei are stained with DAPI. Scale bar: 10 µm. (B) Similar to (A), but in EGFR-knockdown HSC-T6 cells. Scale bar: 10 µm. (C) Co-immunoprecipitation blots showed MD-1-Biotin was precipitated by EGFR in the MD-1-treated HSC-T6 cells, but not in the parental cells. (D) Representative Western blots to show the expression of EGFR, AKT, cyclin D1, and p-Smad. GAPDH was used as the loading control. Densitometry was performed to determine the relative expression levels normalized to GAPDH. *P < 0.05; **P < 0.01; ***P < 0.001 We also detected the expression of EGFR downstream signaling proteins. After HSC-T6 cells were treated with MD-1, the phosphorylation levels of cell proliferation- and motility-associated EGFR and Akt significantly decreased (p-EGFR: P < 0.001; t-EGFR: P < 0.05; p-AKT: P < 0.01; n = 3 in each group, Fig. 4D), and the expression levels of cell cycle regulatory protein cyclin D1 and apoptosis-associated protein p-Smad also decreased (P < 0.01 and P < 0.05, respectively, Fig. 4D). These results suggest that the EGFR and Akt signaling pathways are key factors in regulating the molecular mechanisms by which MD-1 inhibited HSC-T6 cell proliferation, migration, and cell cycle arrest and apoptosis, and that cyclin D1 and p-Smad were also involved. Effects of MD-1 on ECM synthesis and secretion in HSC-T6 cells After activation, HSCs transdifferentiated into MFs, which expressed α-SMA and secreted ECM. qRT-PCR demonstrated that MD-1-treated HSC-T6 cells expressed significantly decreased levels of α-SMA mRNA (P < 0.01, n = 3, Fig. 5A). The expression of intracellular type I and type III collagens was also downregulated (P < 0.01 and P < 0.05, respectively, Fig. 5A), and the expression of tissue inhibitor of metalloproteinase 1 (TIMP-1) was slightly decreased (P = 0.20, n = 3, Fig. 5A). ELISA showed that the levels of α-SMA, type I and III collagen, TIMP-1 in the MD-1-treated cells were significantly decreased (Fig. 5B). These results indicated that HSC activation could be significantly suppressed by MD-1, resulting in decreased ECM synthesis and secretion.Figure 5 Inhibitory effects of MD-1 on ECM synthesis and secretion in HSC-T6 cells. (A) Relative expression levels of ECM components, including α-SMA, collagen I and III, and TIMP-1, as detected by real-time quantitative PCR (qRT-PCR). *P < 0.05; **P < 0.01. (B) Expression of ECM components in the collected cell supernatant as determined by ELISA. *P < 0.05; ***P < 0.001 Effects of MD-1 on DMN-induced hepatic fibrosis in rats A hepatic fibrosis model was established in rats by repeated intraperitoneal injection of DMN to induce hepatic fibrosis. During the initiation and progression of hepatic fibrosis, the rats were continuously treated with MD-1 or MT for 4 weeks. Two months after completing the treatment, blood and liver tissues were collected to examine liver function and hepatic fibrosis indicators. Van Gieson collagen staining of the liver tissue sections showed that DMN successfully induced hepatic fibrosis, proliferated collagen fibers segmented and surrounded pseudolobules, and hepatocytes had spotty necrosis and lymphocyte infiltration (Fig. 6A, positive control). Livers from MD-1-treated rats had significantly lower collagen fiber proliferation than controls, as well as narrow septae and incomplete segmentation of pseudolobules. However, the hepatocyte necrosis was not significant (Fig. 6A, MD-1 group). Based on pathological changes, the curative effect of MT was significantly less than that of MD-1 (Fig. 6A). Immunohistochemistry showed that the expression of p-EGFR, p-AKT, α-SMA, and cyclin D1 was significantly decreased in livers from the MD-1 group compared to the positive control group, but the expression was still higher than that of the normal control group (negative control group, Fig. 6B and 6C). Compared to the positive control group, p-EGFR, p-AKT, and α-SMA were downregulated in the MT group, whereas the reductions in cyclin D1 and p-Smad were not significant (Fig. 6B). ELISA showed that serum ALT, AST, and type I collagen levels were increased in the positive control group but significantly decreased in the MD-1 group while still being higher than the levels in the negative control group (Fig. 6C). In the MT group, ALT and AST were significantly decreased, but the reduction in type I collagen was not significant (Fig. 6C). Our results suggest a better inhibition of hepatic fibrosis in the MD-1 group than the MT group, both of which were more effective than the positive control.Figure 6 Inhibitory effects of MD-1 on DMN-induced hepatic fibrosis in rats. (A) The sectioned rat liver tissues were stained with Van Gieson staining reagent to show the proliferated collagen fibers (stained with red) segmented and surrounded pseudolobules in the positive control group, and MD-1 treatment significantly reduced the collagen fibers. Original magnification: 200×. (B) Top: Expression of p-EGFR, p-AKT, α-SMA, cyclin D1, and p-Smad in sectioned rat liver was examined by immunohistochemistry. The percentages of positive cells (stained with brown) were counted within five fields of view for each section using a 20× objective lens. Original magnification: 100×. Bottom: Statistics for p-EGFR, p-AKT, α-SMA, cyclin D1, and p-Smad. *P < 0.05; **P < 0.01; ***P < 0.001. (C) Top: Detection of serum ALT and AST levels on fully automated biochemical analyzer. Bottom: Detection of collagen I by ELISA. *P < 0.05; **P < 0.01; *** P < 0.001 DISCUSSION Hepatic cirrhosis resulting from chronic liver disease is difficult to reverse and effective clinical treatments are lacking (Tsochatzis et al., 2014; Crossan et al., 2015). Because early-stage hepatic fibrosis can be reversed, active screening of new treatment methods for early interventions for hepatic fibrosis can effectively reduce the incidence of hepatic cirrhosis, or even hepatic carcinoma, improving the quality of life for many people. Among cells involved in hepatic fibrosis, HSCs (also known as Ito cells or fat-storing cells) in the perisinusoidal space play a critical role in the initiation and development of hepatic fibrosis (Page et al., 2015; Koo et al., 2016). HSCs are activated by a variety of cytokines to proliferate and transdifferentiate into MFs and then secrete a large amount of ECM, such as collagen fibers. The interaction among cytokines, cells, and ECM promotes the initiation and progression of hepatic fibrosis, finally resulting in hepatic cirrhosis and cancer (Iwaisako et al., 2014a; Iwaisako et al., 2014b; Wallace and Friedman, 2014; Tang et al., 2015). Because the causes of the majority of liver diseases, such as chronic hepatitis B and autoimmune hepatitis, are difficult to remove, targeting the pathogenic mechanism underlying hepatic fibrosis may be an effective measure to intervene in the disease development process and treat hepatic fibrosis. If the occurrence of hepatic fibrosis can be delayed or prevented, the pathological process of hepatic injury can be attenuated or treated. Lamivudine and entecavir are antiviral drugs used for the clinical treatment of hepatitis B that can effectively inhibit HBV replication, with stronger binding and inhibitory effects on DNA polymerase. Therefore, these drugs can inhibit the development of hepatitis and hepatic fibrosis (Papachrysos et al., 2015). Inhibition of the phosphoinositide 3-kinase (PI3K)-related signal transduction pathway using PI3K-specific inhibitor LY294002 can significantly inhibit HSC proliferation (Wang et al., 2013). Gene therapy is also an effective measure for anti-hepatic fibrosis. Treatment of a rat thioacetamide-induced hepatic fibrosis model with a human MMP-1 adenovirus plasmid was shown to significantly attenuate the degree of hepatic tissue fibrosis, decrease the hydroxyproline level and the number of active HSCs, promote hepatocyte proliferation, and improve liver tissue structure (Iimuro et al., 2003). As a novel potential method, bone marrow mesenchymal stem cell (BMSC) transplantation has become a focus of attention (Jang et al., 2015). Studies in a murine hepatic fibrosis model showed that BMSCs injected into the tail vein can reach the liver and reduce the degree of hepatic fibrosis (Sakaida et al., 2004). However, the specific mechanism underlying the differentiation of BMSCs into hepatocytes and the reduction in hepatic fibrosis is still not clear, and an influence of the local microenvironment could not be fully excluded. Because intrahepatic inflammation is persistent in the majority of patients, BMSCs circulate to the liver, where they are activated by a variety of cytokines and induced to differentiate into MFs, which become a factor in the promotion of hepatic fibrosis. Therefore, more systemic in-depth studies on gene regulation and the microenvironment in BMSC differentiation are needed to facilitate the effective control of BMSC proliferation and accurately directed differentiation, allowing BMSCs to be effectively applied in clinical treatment. MT is a quinolizidine compound with the chemical formula is C15H24N20. Its molecular backbone is a heterodimeric quinolizidine ring. MT has anti-inflammatory, antiviral, anti-fibrotic, anti-arrhythmic, and immune-suppressive functions. The drug is mainly used for the treatment of viral hepatitis, hepatic fibrosis, and arrhythmia, but it also has certain efficacy in the prevention and treatment of tumors (Zhang et al., 2001; Zhang et al., 2011; Liu et al., 2014). MT can relieve pathological injury to liver parenchymal cells and non-parenchymal cells and has significant inhibitory effects on the proliferation of HSCs and fibroblasts. The extensive pharmacological activities of MT suggest that its mechanism of action is complex. A thio derivative of MT, MASM, can act directly on ribosomal protein S5 (RPS5) in vitro and in rats, reducing the phosphorylation levels of Ser473 and Thr308 in Akt and inhibiting HSC activation (Xu et al., 2014). However, the pharmacological activity of MT is not high, making it necessary to modify its structure to screen for derivatives that have high activities and low toxicity. The aforementioned MASM study confirmed that the pharmacological activity of a thio derivative of MT is greatly increased to that of MT and the toxicities equivalent (Hu et al., 2010; Xu et al., 2014). We performed modifications on the basis of thio derivatives of MT. The methylamino group at position 18 was acetylated to improve its stability, and an amino side chain was introduced to increase it activity. This produced the novel MT derivatives MD-1, MD-2, and MD-3, which had good stability and strong activity. MD-1, MD-2, and MD-3 had large differences in their activities based on the different side chain groups. At 100 μmol/L, MD-1 reduced the survival rate of HSC-T6 cells to around 20%, MD-2 reduced the cell survival rate to 27.4%, and the inhibitory effect of MD-3 (47.2% cell survival rate) was equivalent to that of MT (62.9% cell survival rate, Fig. 1). These results suggest that MD-1 and MD-2 had inhibitory effects on HSC-T6 cells, whereas MD-3 did not significantly improve upon the inhibitory effect of MT on HSC-T6 cells. MD-1 significantly inhibited the proliferation and migration of HSC-T6 cells and induced G0/G1 arrest and apoptosis. Although the mechanism underlying hepatic fibrosis is complex, the multi-functional transmembrane glycoprotein EGFR specifically interacts with EGF and TGF-β1, causing its dimerization and regulating cell growth, proliferation, and differentiation (Voon et al., 2013). The EGFR-related signal transduction pathways are activated in HSCs in liver injury and chronic liver disease to promote the development and progression of hepatic fibrosis. Therefore, we focused on studying the effect of MD-1 on the EGFR-related signal transduction pathways. Immunofluorescence showed that the target molecule of MD-1 in HSC-T6 cells was EGFR. MD-1 interacted with EGFR on the surface of cell membranes, inhibiting EGFR phosphorylation. Inhibition of the phosphorylation of downstream protein kinases, such as Akt, affected the expression and activity of target proteins that regulate cell proliferation, migration, cell cycle, and apoptosis, such as cyclin D1 and p-Smad, finally changing the biological behaviors of cells. MD-1 reduced the synthesis and secretion of ECM components, such as type I collagen and type III collagen, in HSC-T6 cells, thereby exerting its anti-hepatic fibrosis activity. In the DMN-induced hepatic fibrosis model, MD-1 treatment delayed the development and progression of hepatic fibrosis, protected liver parenchymal cells, and improved liver function. Although the present study focused on the effect of MD-1 by inhibiting EGFR activation, other signaling pathways, such as the Ras/ERK pathway, may also be involved in hepatic fibrosis. Therefore, there are further studies needed to be carried out on the mechanisms of MT derivatives. In summary, the present study reports a novel synthesized MT derivative, MD-1, that can significantly inhibit HSC activity, induce HSC apoptosis, and decrease the secretion of ECM components by HSCs. The drug has a protective effect on liver parenchymal cells in a rat DMN-induced hepatic fibrosis model. The possible mechanism by which MD-1 exerts its biological functions may be via EGFR binding on the cell surface, inhibiting its function and blocking the EGFR-related downstream signaling pathways. Thus, MD-1 is a potential clinical drug for anti-hepatic fibrosis. MATERIALS AND METHODS Cell culture The rat HSC-T6 cell line was a gift from the Molecular Cancer Research Laboratory in the Eastern Hepatobiliary Surgery Hospital of Second Military Medical University (Xu et al., 2015). Cells were cultured in DMEM (GIBCO, New York, USA) containing 10% fetal bovine serum (FBS) in a 5% CO2 atmosphere at 37°C. MT and its derivatives were synthesized by the School of Pharmacy, Second Military Medical University. The powder form of each compound (2 mg) was added to 200 µL DMSO until completely dissolved, then 1800 µL ddH2O added to obtain a working solution of 1 mg/mL for future use. Cell proliferation HSC-T6 cells were cultured to the logarithmic phase and then inoculated onto 96-well plates (104 cells/well) for 24 h. Different gradient concentrations of MT and its derivatives MD-1, MD-2, and MD-3 were added. Each concentration group had eight replicate wells. After cells were cultured for another 24 h, cell proliferation was detected using the Cell Counting Kit-8 (CCK-8) reagent kit (Dojindo Molecular Technologies, Inc., Shanghai, China). Based on the IC50 values of the three drugs, MD-1, the one with the strongest activity, was selected for other experiments. Cell motility Transwells were placed in 24-well plates and HSC-T6 cells placed in the top chamber (2 × 105 cells/200 µL). The bottom chamber contained 500 µL culture medium containing 10% FBS. Cells in the top chambers were treated with MD1 or MT at the corresponding IC50 (62 µmol/L or 128 µmol/L, respectively). A control group without drug was used for comparison. After 48 h of culture, cells in the top layer of the Transwell were wiped and stained with 0.1% crystal violet for 15 min. Three fields were randomly selected under a light microscope (200× magnifications) for cell counting and photography. Experiments were performed in three biologically independent replicates. Detection of cell cycle and cell apoptosis HSC-T6 cells were inoculated onto a 6-well plate at 5 × 105 cells/well. MD-1 or MT was added at 62 μmol/L. After 48 h, the cells were collected and washed twice with pre-cooled PBS. Some cells were fixed in pre-cooled 75% ethanol in a 4°C refrigerator overnight, washed with PBS twice, and stained with propidium iodide (PI) containing RNase in the dark for 30 min. Cell cycle phase was detected using a flow cytometer (FACS420, BD Biosciences, San Jose, CA). The other cells were used for annexin V/PI staining. The apoptosis rate was also determined by flow cytometry. Binding and inhibiting EGFR The shRNA plasmid pGenensil-shEGFR, targeting the rat EGFR gene, and a negative control vector, pGenesil-shMC, were constructed by Wuhan Genesil Biotechnology Co., Ltd. (Wuhan, China). The 21-nt sequence of EGFR-shRNA targeted base pairs 2268–2290 (5′-gga tat taa agg aaa cag aat-3′) of the EGFR gene (GenBank: M37394.2). A mock control shRNA vector (shMC: 5′-gac ttc ata agg cgc atg cat-3′) was concomitantly constructed. HSC-T6 cells were inoculated on a 6-well plate at 5 × 105 cells/well. The shRNA plasmids were transfected into HSC-T6 cells using Lipofectamine 2000 (Invitrogen Corporation Shanghai Representative Office, Shanghai, China). After 48 h of continuous culture, cells were screened using G418 (400 μg/mL). The obtained cells were named HSC-T6-shEGFR and HSC-T6-shMC cells. HSC-T6, HSC-T6-shEGFR, and HSC-T6-shMC cells were inoculated onto 6-well plates at 5 × 105 cells/well. Biotin-labeled MD-1 (MD-1-Biotin) was added at a concentration of 62 μmol/L. After culturing for another 48 h, the cells were collected. A part of cells were lysed in RIPA protein lysis buffer and the harvested proteins subjected to Western blot for detection of EGFR expression. A part of cells were smeared onto slides for immunofluorescence double labeling to localize MD-1-Biotin and EGFR. The working concentration of FITC-conjugated goat anti-rat EGFR and TRITC-conjugated anti-biotin antibodies was 1:1000 (Cell Signaling Technology, Inc., Beverly, MA). The rest of cells were prepared for the co-immunoprecipitation assay using the corresponding specific antibodies for MD-1-Biotin and EGFR. Detection of EGFR signal transduction pathway-associated proteins HSC-T6 cells were inoculated onto 6-well plates at 5 × 105 cells/well. MD-1 was added at a concentration of 62 μmol/L. After 48 h, the cells were collected and lysed in RIPA protein lysis buffer (Life Technologies Corporation, New York, USA). Total protein was extracted for Western blot detection of downstream proteins in the EGFR signal transduction pathway. The primary antibodies included EGFR, p-EGFR, AKT, p-AKT (Cell Signaling Technology, Danvers, MA), Survivin, cyclin D1 (Abcam Inc., Cambridge, MA), and p-Smad (Santa Cruz Biotech Inc., CA). Detection of ECM proteins HSC-T6 cells were inoculated onto 6-well plates at 5 × 105 cells/well. MD-1 was added at a concentration of 62 μmol/L. After 48 h, the cells and culture supernatant were collected. Cells were treated with TRIzol (Life Technologies Corporation, New York, USA) to extract total RNA for real-time quantitative RT-PCR (qRT-PCR) of ECM indicators using the PrimeScript RT Reagent Kit (TaKaRa Inc., Dalian, China). The primer sequences were: β-actin (537 bp), upstream 5′-ACC CAC ACT GTG CCC ATC TAT G-3′ and downstream 5′-AGA GTA CTT GCG CTC AGG A-3′; α-SMA (120 bp), upstream 5′-CCG AGA TCT CAC CGA CTA CC-3′ and downstream 5′-TCC AGA GCG ACA TAG CAC AG-3′; type III collagen (438 bp), upstream 5′-AGG CCA ATG GCA ATG TAA AG-3′ and downstream 5′-TAT TGG GTG AA A CAG CA-3′; TIMP-1 (250 bp), 5′-TCC CCA GAA ATC GAG AC-3′ and downstream 5′-TCA GAT TAT GCC AGG GAA CC-3′. The cell culture supernatant was used to detect ECM indicators by ELISA (Shanghai Westang Bio-Tech Co., Ltd, Shanghai, China). Construction of the rat hepatic fibrosis model A total of 20 male SD rats (Shanghai SLAC Laboratory Animal Co., Ltd, Chinese Academy of Sciences, Shanghai, China) at 4 weeks of age and a body weight of 80–100 g were randomly divided into four groups of five animals each: normal control group, model control group, MT group, and MD-1 group. Except for the normal control group, the rats were intraperitoneally injected with 1% DMN solution (10 μg/kg). Injections were performed three times each week (Monday, Wednesday, and Friday) for four consecutive weeks. The normal control group received an equal volume of normal saline. Starting from week 5, rats in the MT and MD-1 groups were treated with the drug of interest via intragastric administration (62 μmol/L/kg) three times each week (Monday, Wednesday, and Friday) for four consecutive weeks. The blank control group and model control group received of an equal volume of normal saline via intragastric administration. After the drug treatment was finished, animals were continuously fed for 2 months. Rats were sacrificed after anesthesia. Blood and liver tissues were collected for evaluation of liver function and liver fibrosis indicators. The liver was fixed in 10% formalin for 6 h, embedded in paraffin, and sectioned. Staining was performed using the Van Gieson staining reagent kit (Maxim Biotechnology Development Co. Ltd, Fuzhou, China) according to the manufacturer’s instructions. Fibrosis was observed under a microscope. p-EGFR, p-AKT, α-SMA, cyclin D1, and p-Smad were detected by immunohistochemistry. Serum ALT, AST, and type I collagen were detected by ELISA. All animal experiments were approved by the Animal Ethics Committee of Second Military Medical University (Shanghai, China). Statistical analysis Experimental data are presented as mean ± SD. One-way ANOVA was performed using SPSS (version 18.0). The least significant differences (LSD) test was performed if variances were homogeneous, and Dunnett’s T3 method was used if variances were heterogeneous. A P value < 0.05 was considered significant. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary material 1 (PDF 560 kb) ACKNOWLEDGEMENTS This work was sponsored by the National Natural Science Foundation of China (Grant Nos. 81270518, 81500463, and 81470858). ABBREVIATIONS α-SMA, α-smooth muscle actin; BMSC, bone marrow mesenchymal stem cell; CTGF, connective tissue growth factor; DMN, dimethylnitrosamine; ECM, extracellular matrix; EGF, epidermal growth factor; EGFR, epidermal growth factor receptor; EMT, epithelial-mesenchymal transition; ET, endothelin; FBS, fetal bovine serum; FGF, fibroblast growth factor; HSCs, hepatic stellate cells; IC50, half-maximal inhibitory concentration; MFs, myofibroblasts; MT, matrine; PDGF, platelet-derived growth factor; PI, propidium iodide; PI3K, phosphoinositide 3-kinase; TGF-β, transforming growth factor β. COMPLIANCE WITH ETHICAL STANDARDS Yi Feng, Hai-yan Ying, Ying Qu, Xiao-bo Cai, Ming-yi Xu, and Lun-gen Lu declare that they have no conflict of interest. All institutional and national guidelines for the care and use of laboratory animals were followed. AUTHOR CONTRIBUTION YF., M.Y.X., and L.G.L. designed the research; Y.Q. helped with cellular experiments; H.Y.Y. and X.B.C. helped with animal experiments; Y.F. and L.G.L. wrote the paper; L.G.L. supervised the project. ==== Refs References Crossan C, Tsochatzis EA, Longworth L, Gurusamy K, Davidson B, Rodriguez-Peralvarez M, Mantzoukis K, O’Brien J, Thalassinos E, Papastergiou V et al (2015) Cost-effectiveness of non-invasive methods for assessment and monitoring of liver fibrosis and cirrhosis in patients with chronic liver disease: systematic review and economic evaluation. Health Technol Assess 19:1–409, v–vi Cuevas MJ Tieppo J Marroni NP Tunon MJ Gonzalez-Gallego J Suppression of amphiregulin/epidermal growth factor receptor signals contributes to the protective effects of quercetin in cirrhotic rats J Nutr 2011 141 1299 1305 10.3945/jn.111.140954 21562239 Fagone P Mangano K Mammana S Pesce A Pesce A Caltabiano R Giorlandino A Portale TR Cavalli E Lombardo GA Identification of novel targets for the diagnosis and treatment of liver fibrosis Int J Mol Med 2015 36 747 752 26135677 Fausther M Lavoie EG Dranoff JA Contribution of myofibroblasts of different origins to liver fibrosis Curr Pathobiol Rep 2013 1 225 230 10.1007/s40139-013-0020-0 23997993 Handy JA Fu PP Kumar P Mells JE Sharma S Saxena NK Anania FA Adiponectin inhibits leptin signalling via multiple mechanisms to exert protective effects against hepatic fibrosis Biochem J 2011 440 385 395 10.1042/BJ20102148 21846328 Hu H Wang S Zhang C Wang L Ding L Zhang J Wu Q Synthesis and in vitro inhibitory activity of matrine derivatives towards pro-inflammatory cytokines Bioorg Med Chem Lett 2010 20 7537 7539 10.1016/j.bmcl.2010.09.075 21036613 Iimuro Y Nishio T Morimoto T Nitta T Stefanovic B Choi SK Brenner DA Yamaoka Y Delivery of matrix metalloproteinase-1 attenuates established liver fibrosis in the rat Gastroenterology 2003 124 445 458 10.1053/gast.2003.50063 12557150 Iwaisako K Jiang C Zhang M Cong M Moore-Morris TJ Park TJ Liu X Xu J Wang P Paik YH Origin of myofibroblasts in the fibrotic liver in mice Proc Natl Acad Sci USA 2014 111 E3297 E3305 10.1073/pnas.1400062111 25074909 Iwaisako K Taura K Koyama Y Takemoto K Asagiri M Strategies to detect hepatic myofibroblasts in liver cirrhosis of different etiologies Curr Pathobiol Rep 2014 2 209 215 10.1007/s40139-014-0057-8 25401051 Jang YO Jun BG Baik SK Kim MY Kwon SO Inhibition of hepatic stellate cells by bone marrow-derived mesenchymal stem cells in hepatic fibrosis Clin Mol Hepatol 2015 21 141 149 10.3350/cmh.2015.21.2.141 26157751 Koo JH Lee HJ Kim W Kim SG Endoplasmic reticulum stress in hepatic stellate cells promotes liver fibrosis via PERK-mediated degradation of HNRNPA1 and up-regulation of SMAD2 Gastroenterology 2016 150 181–193 e188 Liu Y Liu H Meyer C Li J Nadalin S Konigsrainer A Weng H Dooley S ten Dijke P Transforming growth factor-beta (TGF-beta)-mediated connective tissue growth factor (CTGF) expression in hepatic stellate cells requires Stat3 signaling activation J Biol Chem 2013 288 30708 30719 10.1074/jbc.M113.478685 24005672 Liu Y Xu Y Ji W Li X Sun B Gao Q Su C Anti-tumor activities of matrine and oxymatrine: literature review Tumour Biol 2014 35 5111 5119 10.1007/s13277-014-1680-z 24526416 Page A Paoli PP Hill SJ Howarth R Wu R Kweon SM French J White S Tsukamoto H Mann DA Alcohol directly stimulates epigenetic modifications in hepatic stellate cells J Hepatol 2015 62 388 397 10.1016/j.jhep.2014.09.033 25457206 Papachrysos N Hytiroglou P Papalavrentios L Sinakos E Kouvelis I Akriviadis E Antiviral therapy leads to histological improvement of HBeAg-negative chronic hepatitis B patients Ann Gastroenterol 2015 28 374 378 26126929 Pinzani M Pathophysiology of liver fibrosis Dig Dis 2015 33 492 497 10.1159/000374096 26159264 Sakaida I Terai S Yamamoto N Aoyama K Ishikawa T Nishina H Okita K Transplantation of bone marrow cells reduces CCl4 -induced liver fibrosis in mice Hepatology 2004 40 1304 1311 10.1002/hep.20452 15565662 Seki E Brenner DA Recent advancement of molecular mechanisms of liver fibrosis J Hepatobiliary Pancreat Sci 2015 22 512 518 10.1002/jhbp.245 25869468 Shimada H Staten NR Rajagopalan LE TGF-beta1 mediated activation of Rho kinase induces TGF-beta2 and endothelin-1 expression in human hepatic stellate cells J Hepatol 2011 54 521 528 10.1016/j.jhep.2010.07.026 21087804 Tang WP Akahoshi T Piao JS Narahara S Murata M Kawano T Hamano N Ikeda T Hashizume M Basic fibroblast growth factor-treated adipose tissue-derived mesenchymal stem cell infusion to ameliorate liver cirrhosis via paracrine hepatocyte growth factor J Gastroenterol Hepatol 2015 30 1065 1074 10.1111/jgh.12893 25639333 Tsochatzis EA Bosch J Burroughs AK Liver cirrhosis Lancet 2014 383 1749 1761 10.1016/S0140-6736(14)60121-5 24480518 Voon DC Wang H Koo JK Chai JH Hor YT Tan TZ Chu YS Mori S Ito Y EMT-induced stemness and tumorigenicity are fueled by the EGFR/Ras pathway PLoS One 2013 8 e70427 10.1371/journal.pone.0070427 23950932 Wallace MC Friedman SL Hepatic fibrosis and the microenvironment: fertile soil for hepatocellular carcinoma development Gene Expr 2014 16 77 84 10.3727/105221614X13919976902057 24801168 Wang FP Li L Li J Wang JY Wang LY Jiang W High mobility group box-1 promotes the proliferation and migration of hepatic stellate cells via TLR4-dependent signal pathways of PI3K/Akt and JNK PLoS One 2013 8 e64373 10.1371/journal.pone.0064373 23696886 Xu WH Hu HG Tian Y Wang SZ Li J Li JZ Deng X Qian H Qiu L Hu ZL Bioactive compound reveals a novel function for ribosomal protein S5 in hepatic stellate cell activation and hepatic fibrosis Hepatology 2014 60 648 660 10.1002/hep.27138 24668691 Xu Y Peng Z Ji W Li X Lin X Qian L Li X Chai X Wu Q Gao Q A novel matrine derivative WM130 inhibits activation of hepatic stellate cells and attenuates dimethylnitrosamine-induced liver fibrosis in rats Biomed Res Int 2015 2015 203978 26167476 Zhang B Liu ZY Li YY Luo Y Liu ML Dong HY Wang YX Liu Y Zhao PT Jin FG Antiinflammatory effects of matrine in LPS-induced acute lung injury in mice Eur J Pharm Sci 2011 44 573 579 10.1016/j.ejps.2011.09.020 22019524 Zhang JP Zhang M Zhou JP Liu FT Zhou B Xie WF Guo C Antifibrotic effects of matrine on in vitro and in vivo models of liver fibrosis in rats Acta Pharmacol Sin 2001 22 183 186 11741525

PMC005xxxxxx/PMC5003785.txt

==== Front Protein CellProtein CellProtein & Cell1674-800X1674-8018Higher Education Press Beijing 28910.1007/s13238-016-0289-yResearch ArticleAn unusual UMP C-5 methylase in nucleoside antibiotic polyoxin biosynthesis http://orcid.org/0000-0003-2557-8960Chen Wenqing 1Li Yan 2Li Jie 2Wu Lian 2Li Yan 2Wang Renxiao 2Deng Zixin zxdeng@sjtu.edu.cn 13Zhou Jiahai jiahai@mail.sioc.ac.cn 21 Key Laboratory of Combinatorial Biosynthesis and Drug Discovery, Ministry of Education, and School of Pharmaceutical Sciences, Wuhan University, Wuhan, 430071 China 2 State Key Laboratory of Bioorganic and Natural Products Chemistry, Shanghai Institute of Organic Chemistry, Chinese Academy of Sciences, Shanghai, 200032 China 3 State Key Laboratory of Microbial Metabolism, and School of Life Sciences & Biotechnology, Shanghai Jiao Tong University, Shanghai, 200030 China 14 7 2016 14 7 2016 9 2016 7 9 673 683 14 4 2016 14 6 2016 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.Polyoxin is a group of structurally-related peptidyl nucleoside antibiotics bearing C-5 modifications on the nucleoside skeleton. Although the structural diversity and bioactivity preference of polyoxin are, to some extent, affected by such modifications, the biosynthetic logic for their occurence remains obscure. Here we report the identification of PolB in polyoxin pathway as an unusual UMP C-5 methylase with thymidylate synthase activity which is responsible for the C-5 methylation of the nucleoside skeleton. To probe its molecular mechanism, we determined the crystal structures of PolB alone and in complexes with 5-Br UMP and 5-Br dUMP at 2.15 Å, 1.76 Å and 2.28 Å resolutions, respectively. Loop 1 (residues 117–131), Loop 2 (residues 192–201) and the substrate recognition peptide (residues 94–102) of PolB exhibit considerable conformational flexibility and adopt distinct structures upon binding to different substrate analogs. Consistent with the structural findings, a PolB homolog that harbors an identical function from Streptomyces viridochromogenes DSM 40736 was identified. The discovery of UMP C5-methylase opens the way to rational pathway engineering for polyoxin component optimization, and will also enrich the toolbox for natural nucleotide chemistry. Electronic supplementary material The online version of this article (doi:10.1007/s13238-016-0289-y) contains supplementary material, which is available to authorized users. KEYWORDS polyoxinnucleoside antibioticsbiosynthesisUMP C5-methylasethymidylate synthaseissue-copyright-statement© HEP and Springer 2016 ==== Body INTRODUCTION Nucleoside natural products are a large family of microbial secondary metabolites with diverse bioactivities and unusual structural features (Isono, 1988; Niu and Tan, 2015; Chen et al., 2016). They have played distinguished roles in the treatment of the infections for mammalians and plants (Isono, 1988). Normally, the biosynthesis of nucleoside antibiotics follows a distinct logic via sequential modifications of the simple building blocks including nucleoside and nucleotide from primary metabolisms (Isono, 1988). Polyoxin, a group of structurally-related peptidyl nucleoside antibiotics, is produced by Sreptomyces cacaoi var. asoensis (S. cacaoi hereafter) and Streptomyces aureochromogenes (Chen et al., 2009). As the chemical structure of polyoxin mimics UDP-N-acetyl glucosamine, a building block for fungal chitin biosynthesis, it functions as a potent chitin synthetase inhibitor by targeting fungal cell wall biosynthesis (Endo and Misato, 1969; Endo et al., 1970). Polyoxin has therefore been widely used as an agricultural fungicide to control phytopathogenic fungi due to its distinctive action mode (Chen et al., 2009). Polyoxin is composed of three moieties involving a nucleoside skeleton and two non-proteinogenic amino acids, carbamoylpolyoxamic acid and polyoximic acid (Fig. 1B) (Chen et al., 2009). The C-5 modifications within the nucleoside skeleton confer not only structural diversity but also possess a bioactivity preference for polyoxin (Isono et al., 1967; Isono and Suzuki, 1968; Isono et al., 1975; Zhai et al., 2012). Previous labeling studies indicated that the C-5 methylation originated from C-3 of serine and is catalyzed by a new enzyme independent of thymidylate synthase (Isono and Suhadolnik, 1976; Isono, 1988), however, the molecular mechanism for such modification remained elusive for decades.Figure 1 Gene cluster and structure of polyoxin as well as the dual functions of PolB. (A) The genetic organization of the polyoxin gene cluster. (B) The difference of polyoxin A, F, H and K is the C5 modification in nucleoside skeleton. (C) PolB catalyzes the UMP methylation in polyoxin biosynthesis. (D) ThyX and PolB could catalyze the dTMP biosynthesis. POL: Polyoxin We have previously identified the polyoxin biosynthetic gene cluster (Fig. 1A) from S. cacaoi, and tentatively proposed a pathway for the C5-methylation on the nucleoside skeleton (Chen et al., 2009). By analyzing the polyoxin biosynthetic gene cluster, we found that PolB, a thymidylate synthase (ThyX) homolog (Myllykallio et al., 2002; Graziani et al., 2004), is likely to be responsible for catalyzing the C5-methylation (Fig. 1C) (Chen et al., 2009). To dissect the function of PolB, we carried out a series of biochemical and crystallographic analysis which confirm PolB is an unusual flavin-dependent UMP/dUMP methylase (Fig. 1C and 1D). We solved the crystal structures of PolB as well as its complex structures with two substrate analogues 5-Br UMP or 5-Br dUMP. We found that the structure of PolB shares high similarity with its homologs ThyX proteins. However, the sequence identity is only 38%. Two special Loops, Loop 1 (residues 117–131) and Loop 2 (residues 192–201) which are highly conserved in primary sequence with ThyXs but not structurally existed in them are identified in PolB structure. Additional mutagenesis studies further reveal that residues Tyr124, Tyr126 and Tyr99 on Loop 1, Loop 2 and substrate recognition peptide (residues 94–102) are crucial for the catalytic activity and substrate selectivity of PolB. These results suggest that Loop 1 and Loop 2 cooperatively play a vital role in catalysis of UMP and dUMP methylation. Moreover, the findings of PolB as the first UMP methylase will shed light on the occurrence of C-5 modification for polyoxin biosynthesis and enrich the toolbox for thymidylate synthases. RESULTS polB isresponsible for the C5-methylation in polyoxin biosynthesis Bioinformatic analysis of polyoxin biosynthetic gene cluster revealed that polB is a potential candidate involved in performing C5 modification of the nucleoside skeleton (Fig. S1). To determine its involvement and the corresponding function, we first performed a gene knockout of polB in S. cacaoi. This was achieved via conjugation of a polB disruption vector pJTU2183 (Table S1) and a double-crossover replacement of the corresponding region in the chromosome of S. cacaoi. The sample of the resulting mutant CY5 was found to display higher bioactivity against the indicator strain Trichosporon cutaneum (Fig. S2A), implicating that the metabolites might be different. LC-MS analysis showed that CY5 was unable to produce polyoxin A (5-hydoxymethyl), polyoxin F (5-carboxyl) and polyoxin H (5-methyl), but instead accumulated polyoxin K (without C5 modification) (Figs. 2A and S2B–D). Complementation of the polB mutant CY5 restored the ability to produce polyoxin A, F and H, suggesting that polB is the target gene directly responsible for the C5-methylation of the nucleoside skeleton in polyoxin biosynthesis (Fig. 2A).Figure 2 Genetic characterization of the polB function. (A) HPLC profiles of the metabolites produced by wild-type and mutant S. cacaoi strains. ST, polyoxin authentic standards; I, metabolites from wild-type S. cacaoi; II, metabolites from CY5; III, metabolites from CY5 containing pIB139 as negative control; IV, metabolites from CY5 complemented by polB. (B) polB is capable of restoring growth phenotype for the thyX mutant of S. cacaoi. 1: CY5; 2: CY6, the thyX and polB double mutant of S. cacaoi; 3: CY6/pIB139, CY6 containing pIB139 as negative control; 4: CY6/polB, CY6 complemented by polB; 5: CY6/thyX, CY6 complemented by thyX. The left plate contains thymidine, whereas which is absent in the right one PolB harbors an alternative thymidylate synthase function A BLAST search for PolB homologs yielded sequences of several Streptomyces thymidylate synthases (ThyXs) (Fig. S1), suggesting that PolB may carry ThyX function. In bacteria, ThyXs catalyze the biosynthesis of dTMP by C5-methylation of dUMP, as dTMP is the key building block for DNA synthesis, deletion of thyX usually causes a lethal effect on cell growth in minimal medium. To evaluate the function of polB, we performed targeted gene knockout of thyX in the chromosome of S. cacaoi. Interestingly, CY3, the thyX mutant of S. cacaoi, survived and exhibited sporadic growth phenotype (Fig. S2E), implicating that PolB might complement the dTMP biosynthesis of ThyX. Indeed, when thyX and polB of S. cacaoi were both deleted, the resulting mutant CY6 could not grow at all in minimal medium (Fig. 2B). Biochemical characterization of PolB as a FAD-dependent UMP/dUMP methylase To elucidate its biochemical role, we expressed S. cacaoi PolB as a N-terminal His6-tagged protein in E. coli BL21(DE3)/pLysE (Fig. 3A) and test its activity in vitro. The purified recombinant protein displayed bright yellow color, a characteristic of flavoprotein (Fig. S3A and S3B). By incubating PolB with dUMP, NADPH and CH2H4folate in vitro, we observed the target product of dTMP using LC-MS analysis (Figs. 3B, 3C and S3C–E). Further investigation revealed that PolB possessed UMP C5-methylase activity in the presence of NADPH and CH2H4folate (Figs. 3B, 3D, S3C, S3F and S3G). In comparison to PolB, recombinant S. cacaoi ThyX could catalyze the methylation of dUMP but not UMP when incubating with NADPH and CH2H4folate (Fig. 3C and 3D). These results demonstrated PolB as an unusual FAD-dependent UMP C5-methylase with thymidylate synthase activity.Figure 3 In vitro characterization of PolB as a UMP/dUMP methylase. (A) SDS-PAGE analysis of PolB. M, molecular weight marker; lanes 1 and 2, purified His6-tagged PolB. LC-MS analysis of the reaction products of dUMP by ThyX or PolB. (B) Scheme of PolB catalyzed UMP/dUMP methylation reactions. (C) LC-MS analysis of the reaction products of ThyX or PolB using dUMP as substrate. Negative control, no PolB or ThyX was added into the reaction mixture. (D) LC-MS analysis of the reaction products of ThyX or PolB using UUMP as substrate. Negative control, no PolB or ThyX was added into the reaction mixture Next, we measured the kinetic parameters of PolB to substrates UMP and dUMP (Fig. S3H). Although dUMP exhibits higher affinity (Km = 12.96 ± 0.89 μmol/L) than UMP (Km = 19.48 ± 4.15 μmol/L), the kcat of UMP (kcat = 3.09 ± 0.17 min−1) is 78% higher than that of dUMP (kcat = 1.74 ± 0.05 min−1). Structural comparison of PolB and ThyX To investigate the molecular mechanism of PolB, we crystallized the protein and determined its crystal structure (abbreviated as apo-PolB in this work) using molecular replacement (McCoy et al., 2007) and refined at resolution of 2.15 Å (Table S3). Apo-PolB was crystallized in the P21 space group with two tetramers in an asymmetric unit. Although there is only 38% sequence identity between PolB and ThyX from Thermotoga maritima (abbreviated as TMAThyX, PDB code: 1O2A) (Mathews et al., 2003), structure comparison revealed that the homotetramer of apo-PolB strongly resembles ThyX with an r.m.s.d less than 0.871 Å for 969 aligned Cα atoms. Especially, for the cofactor FAD binding pocket, little conformational change was observed between the two proteins (Fig. S4A). Nevertheless, a notable difference was observed in Loop 1 (residues 117–131) and Loop 2 (residues 192–201), which was not detected in TMAThyX, but presented high sequence conservation in other ThyXs (Fig. 4A and 4B). To probe the functional role of Loop 1 and Loop 2 in catalysis, we first generated PolB mutants by displacing the residues of these regions with the counterparts of S. cacaoi ThyX protein. The activity assay indicated that replacement of Loop 1 did not affect the enzymatic activity of PolB, while replacement of Loop 2 decreased the UMP methylation activity to 10% and the dUMP methylation activity to about 30%. When Loop 1 and Loop 2 were both replaced by the counterparts of S. cacaoi ThyX protein, the mutant PolB lost the methylation activity for both UMP and dUMP (Fig. 4C). This suggested that Loop 1 and Loop 2 are involved in regulating the catalytic activity of this methylase.Figure 4 Comparison of PolB and ThyX proteins. (A) Superposition of PolB and TMAThyX (PDBID: 1O2A). PolB is shown in green, and TMAThyX is shown in purple. The N-and C-terminus are indicated. (B) Alignment of Loop 1 and Loop 2 between PolB and homologous ThyXs from Streptomyces. The residues in Loop 1 and Loop 2 of PolB and SVIThyX2 are boxed in magenta. (C) Comparison of the catalytic activities of ThyX, PolB and PolB variants for UMP (left panel) and dUMP (right panel). The relative activity was calculated on the basis of 3 repeats, and the error was all under control of ±5% Other major differences between PolB and TMAThyX were in the N-terminus and the C-terminus (Figs. 4A and S1). In the PolB structure, the N-terminus adopts a flexible conformation while the C-terminus within the interior structure protrudes outside; in the TMAThyX structure (Mathews et al., 2003), the N-terminus forms a pair of anti-parallel β-sheets while the C-terminus extends outside. Comparison of UMP and dUMP binding in the active site of PolB To obtain the precise mechanism of PolB in UMP/dUMP methylation, we solved the structures in complex with substrate analogs 5-Br dUMP and 5-Br UMP at individual resolution of 2.28 Å and 1.76 Å (Table S3). Because the C5-position is substituted by the Br atom, these two structures should mimic the state of substrate binding. Both structures can be superimposed with the apo-PolB structure within rmsd of 1.26 Å over all the Cα atoms (Figs. 5A and S4B). In the tetrameric structure of PolB/5-Br dUMP, the thymine ring of the substrate 5-Br dUMP poses strong π-π interaction with the isoalloxazine moiety of FAD, and its phosphate group forms hydrogen bonds or salt bridges with the side chains of Phe79, Arg82, His83 from one monomer and Ser96’, Ala97’, Arg98’, Arg166’ from another neighboring monomer. Besides these interactions, the ribose O3’ is hydrogen bond to Glu94’ and Arg86; the pyrimidine O2 makes hydrogen bond to Arg193; the pyrimidine O4 has hydrogen bond to Arg98’ and water-mediated hydrogen-bond to Arg193 and Gln206 (Fig. 5D). The PolB/5-Br UMP complex structure is nearly identical in substrate binding to the PolB/5-Br dUMP complex structure except for an additional water-mediated hydrogen bond between ribose O3’ and Arg82 (Fig. 5E). Interestingly, the characteristic ribose O2’ in 5-Br UMP does not contact with any residues of PolB or water molecules. Therefore, little difference was observed in the active pocket of the structures of the two complexes.Figure 5 Comparison of two structures of PolB complex. (A) Superposition of structures of PolB (green), PolB/5-Br UMP (blue) and PolB/5-Br UMP (yellow). The N-and C-terminus are indicated. (B) and (C) Electron density for 5-Br UMP and 5-Br dUMP respectively. The 2F o-F c map is contoured at 2.0 σ. (D and E) Schematic representations of the interactions between the active site residues of PolB and (D) 5-Br dUMP or (E) 5-Br UMP. The hydrogen bonds were labeled as dashed lines. Residues in the white and grey box are from two separate PolB monomers Although the two PolB complex structures displayed almost identical catalytic mechanism, we perceived that Loop 1 and Loop 2 exhibit considerable conformational flexibility and are structurally distinct in two different complexes (Fig. 5A). Loop 1 undergoes dramatic conformational changes upon binding of either 5-Br UMP or 5-Br dUMP. This region becomes structurally ordered to form three short tandem β-sheets when the substrate analog 5-Br UMP binds to the active sites of PolB. Loop 2 undergoes obvious shift and adopts a stable conformation when either 5-Br UMP or 5-Br dUMP binds to PolB. We next screened a serial of site-directed mutants in Loop 1 and Loop 2 and measured their catalytic activities towards UMP and dUMP (Figs. 6A, 6B and S5A). We found that the conserved Tyr124 in Loop 1 was essential for catalysis while Tyr126 was necessary for substrate specificity. Mutation of Tyr124 to Phe did not affect the activity of PolB for either UMP or dUMP; however, replacement of this residue by Ala or Ser led to more than 90% activity loss for each substrate, indicating that the aromatic ring of Tyr124 is essential for catalysis (Fig. 6C and 6D). Tyr126 is only found in PolB while the corresponding residue in ThyX proteins of Streptomyces is phenylalanine (Fig. 4B). The Y126F mutant of PolB retained its full activity for dUMP methylation but lost over 60% of activity for catalyzing UMP, suggesting that this residue might be important for UMP methylation.Figure 6 Analysis of the roles of Loop 1 and Loop 2 in PolB catalysis. (A) Comparison of the catalytic activities of PolB and its variants for UMP. (B) Comparison of the catalytic activities of PolB and its variants for dUMP. The relative activity was calculated on the basis of 3 repeats, and the error was all under control of ±5%. (C) Conformational change of Try99 in PolB-5-Br dUMP (salmon) and PolB-5-Br UMP (gray). (D) Conformation change of key residues Tyr124 and Tyr126 in Loop 1 around access while bound with 5-Br UMP (salmon) or 5-Br dUMP (gray) The third region varied between the structures of apo-PolB and two complexes was located in residues 94–102, whose counterpart in ThyX proteins was identified as the substrate recognition peptide (SRP). In apo-PolB, the electron densities of SRP were weak or difficult to observe. However, upon substrate binding, Glu94, Ser96, Ala97 and Arg98 in SRP are able to form hydrogen bonds to the substrate, which renders SRP ordered (Fig. S5B). We also discerned that the side chain of Arg98 in SRP forms hydrogen bond to Q206 in Loop 2. Notably, the side chain of Tyr99 in SRP adopts different rotamer structures in the two complex structures (Fig. 6C). In the PolB/5-Br dUMP structure, the hydroxyl group of the side chain points to the center of the active site and is close to the substrate. In contrast, for the structure of PolB/5-Br UMP, the hydroxyl group of the side chain points to the exterior of protein and is distant from the substrate. This suggested that the hydroxyl group of Tyr99 might be essential for substrate specificity. Indeed, the Tyr99F mutant kept 80% activity for dUMP but only 15% activity for UMP (Fig. 6A). As a control, the Tyr99A mutant almost abolished its activity for both UMP and dUMP. Genome mining of PolB-like UMP methylase To firmly validate the role of Loop 1 and Loop 2 in dual-substrate specificities, we used them as probes for the mining of PolB-like proteins capable of producing 5-methyl UMP. BLAST search hits a putative thymidylate synthase (ID: ZP_07305627, designated as SVIThyX2) in S. viridochromogenes DSM 40736 with 71% identity to PolB (Fig. S1). The Loop 1 and Loop 2 regions in SVIThyX2 were different from those of ThyX proteins but highly conserved with the corresponding parts of PolB (Fig. 4B). The purified recombinant SCIThyX2 was incubated with NADPH, CH2H4folate and dUMP or UMP in vitro. LC-MS data showed that SVIThyX2 was able to catalyze the methylation for both UMP and dUMP (Figs. 6A, 6B and S6). These results demonstrated the PolB-like SVIThyX2 as the UMP methylase with thymidylate synthase activity, and further unambiguously confirmed the essential roles of Loop 1 and Loop 2 in UMP methylation. DISCUSSION In this study, we demonstrated that PolB, combined with in vivo and in vitro assays, was able to catalyze the C5-methylation of both UMP and dUMP while the classic ThyXs only possess the dUMP methylase activity (Myllykallio et al., 2002). The crystal structures of PolB alone and in complex with the substrate analogs 5-Br UMP and 5-Br dUMP showed that the methylation mechanism of UMP and dUMP might be similar because they adopt the same binding pattern in the active site of PolB. The characteristic ribose O2’ in 5-Br UMP does not contact with any residues of PolB or water molecules. The structures indicate that Arg82, His83, Arg86, Glu94, Ser96, Arg98, Arg166, Arg193 and Gln206 play essential roles in PolB catalysis (Fig. 5D and 5E). Although UMP is the naturally preferred substrate of PolB, the kinetic studies and competitive binding experiments showed that dUMP rather than UMP exhibits higher affinity to PolB (Fig. S7). This is consistent with Frank Maley’s early reports on chick embryo thymidylate synthase half a century ago (Maley, 1960; Lorenson et al., 1967). The affinity difference between UMP and dUMP could be explained by steered molecular dynamics (SMD) simulation that dUMP rather than UMP need more external energy to dissociate from the protein (Fig. S8). Analysis of the crystal structures of PolB indicates that three regions including Loop 1, Loop 2 and the substrate recognition peptide are crucial for the methylation of UMP/dUMP. They exhibit considerable conformational flexibility and became ordered to form a “closed” conformation by interacting with the substrate. Mutational studies uncovered that the phenyl group of Tyr99 in the substrate recognition peptide and Tyr124 in Loop 1 are essential for catalysis, consistent with the structural information that the benzene groups of Tyr124 and Tyr99 likely made π-π stacking interaction with the uracil ring of the substrate. Further mutational studies also demonstrated that the hydroxyl groups of Tyr99 in the substrate recognition peptide and Tyr126 in Loop 1 were important for substrate specificity. This is in full agreement with the SMD simulation results that showed the Y126F mutant decreases the success rate of UMP dissociation trajectories by 50% and increased the success rate of dUMP dissociation trajectories by 100% (Fig. S8). We proposed that Loop 1, Loop 2 and the substrate recognition peptide constituted the gate-keeper for substrate entrance and cooperatively regulated the catalysis of PolB on UMP/dUMP methylation. Our hypothesis that Loop 1 and Loop 2 regulated the substrate specificity of PolB was further validated by identification of SVIThyX2 as the second thymidylate synthase with UMP methylase activity. The findings of PolB as a unique UMP methylase elucidated the origin of nucleoside skeleton C5-modification in polyoxin biosynthesis. Based on this work, we proposed that four different groups of polyoxins could be synthesized starting from UMP and its derivatives (5-methyl UMP, 5-hydroxymethyl UMP and 5-carboxyl UMP) via a potential pyrimidine salvage pathway. This hypothesis is supported by our previous report that when the biosynthetic gene cluster of polyoxin was heterologously expressed in S. lividians TK24, only the polyoxin H components were detected (Zhao et al., 2010). Co-expression of the polyoxin gene cluster of with sav_4805 (encoding a thymine-7-hydroxylase homologous protein) from S. avermitilis (Omura et al., 2001) lead to the production of polyoxin A in S. lividians TK24. This indicated that genes responsible for the reaction from 5-methyl UMP to 5-hydroxymethyl UMP and 5-carboxyl UMP are not all adjacent to the polyoxin gene cluster. The S. cacaoi cells may also evolve or hijack a decarboxylase to convert 5-carboxyl UMP to normal UMP and complete the UMP salvage pathway. Further investigation of the component diversity of polyoxins will require all related genes in the metabolic pathway for 5-methyl UMP to be cloned. In summary, we have reported the identification and structural basis of an unprecedented C5 methylase that employs FAD-dependent reductive mechanism for the methylation of UMP/dUMP. We also revealed that Loop 1, Loop 2 and substrate recognition peptide of the protein collectively constitute a gate-keeper for substrate selective-entrance and preferred-catalysis. The present data will provide insights for ThyXs evolution and enrich the chemical diversity of natural nucleotides. MATERIALS AND METHODS Materials, methods and procedures All chemicals were from Sigma-Aldrich (IL, USA) unless otherwise indicated. 5-Br UMP, 5-Br dUMP and 5-methyl UMP were purchased from Hongene Biotechnology Ltd. (Shanghai, China). CH2H4folate was a gift from Merck. Materials and primers were individually listed in Table S1 and Table S2, and general methods and procedures were described by Kieser et al. (2000) and Sambrook et al. (1989). Expression and purification of PolB and SVIThyX2 All constructs and point mutations were generated using a standard PCR-based cloning strategy and verified through DNA sequencing. The recombinant PolB from S. cacaoi and SVIThyX2 from S. viridochromogens DSM40736 (Blodgett et al., 2005) were overexpressed at 30°C in E. coli BL21(DE3) as N-terminally His6-tagged proteins. The soluble fraction of the cell lysate was first purified using nickel affinity column (GE Healthcare) and further purified by gel-filtration chromatography (Superdex 75, HiLoad 16/60, GE Healthcare). Activity assay All tests were performed in triplicates in 2 mL centrifuge tubes. A typical methylase activity assay (200 µL) contained 2.0 mmol/L NADPH, 0.2 mmol/L CH2H4folate, 0.2 mmol/L UMP (or dUMP), 50 mmol/L Tris-HCl (pH 8.0), and 10 μg of protein (PolB, ThyX or SVIThyX2). The reaction was terminated by adding TCA with final concentration 10% (v/v) and further analyzed by LC-MS using a ZORBA SB-C18 Column (5.0 μm, 4.6 × 250 mm, Agilent). The LC conditions were as follows. The elution buffer for LC was 10% methanol (v/v) contained 0.1% aqueous trifluoroacetic acid (v/v). The flow rate was 0.3 mL/min and the eluted fraction was monitored at 260 nm with a DAD detector. The parameters for MS analysis are 10 l/mL of drying gas flow, 30 psi of nebulizer pressure, and 325°C of drying gas temperature. Crystallization and data collection All crystallization experiments were performed at 20°C using the sitting-drop vapor-diffusion method. Protein samples (12.5 mg/mL, 1 µL) stored in 25 mmol/L Tris-HCl, pH 8.0, 150 mmol/L NaCl and 5 mmol/L β-mercaptoethanol were mixed with well solution (1 µL) and equilibrated against the well solution (75 µL) in 96-well plates (HR3-271, Hampton Research). The crystal of apo-PolB was grown under the condition of 17% (w/v) PEG4000, 0.2 mol/L Li2SO4 and 0.1 mol/L Tris-HCl, pH 8.5. Crystals of the PolB/5-BrdUMP complex were obtained by mixing PolB (11.4 mg/mL) with 4 mmol/L of 5-Br dUMP and growing under the conditions of 1.4 mol/L sodium acetate and 0.1 mol/L sodium cacodylate, pH 6.5. Crystals of the PolB/5-Br UMP complex were obtained by mixing PolB (11.4 mg/mL) with 4 mmol/L of 5-Br UMP and growing under the conditions of 10% (w/v) PEG4000, 5% isopropanol and 0.1 mol/L HEPES, pH 7.5. Prior to data collection, all crystals were flash-cooled in liquid nitrogen using Paratone-N (HR2-463, Hampton Research) as cryo-protectants. Diffraction data were collected on a Mar225 detector at 100 K on the beamline BL17U1 at Shanghai Synchrotron Radiation Facility (Shanghai, China). The data sets were integrated and scaled with HKL2000. Structure determination and refinement The structure of the PolB/5-Br dUMP complex was solved by molecular replacement using Phaser and the structure of TMAThyX (PDB code: 1O2A) as the search molecule. The structure of the PolB/5-Br UMP complex was solved by molecular replacement using Phaser and the PolB/5-Br dUMP complex as the search molecule. The structure of the PolB was solved by molecular replacement using Phaser and the PolB/5-Br UMP complex as the search molecule. Manual model building was performed with COOT (Emsley et al., 2010). Multiple rounds of refinement were carried out with Refmac5, CNS, and PHENIX. Noncrystallographic restraints were applied for one round of refinement. The overall quality of the final models was assessed by MolProbility and PROCHECK. Data collection and final refinement statistics are summarized in Table S3. All graphics were generated using PyMol. Accession codes The crystal structures of PolB, the PolB/5-Br dUMP and the PolB/5-Br UMP have been deposited in the Protein Data Bank under accession number of 4P5C, 4P5B and 4P5A. ACKNOWLEDGMENTS We are sincerely grateful to J. He, Q. Wang and S. Huang at SSRF BL17U1 beamline for data collection, and Prof. Zhihong Guo, Prof. Zong-Xiang Xia and Prof. Zhaohui Xu were appreciated for critical reading the manuscript. We’d also like to acknowledge Dr. Neil Price for assistance with MS data analysis, Prof. Shuangjun Lin for kinetic analysis of PolB and Xu-Dong Kong for help with figure preparation. This work was supported by grants from the National Basic Research Program (973 Program) (No. 2012CB721004 to W.C., No. 2011CB710800 to J.Z.), the National Grand Project for Medicine Innovation (2012ZX10002006 to J.Z.), the National Natural Science Foundation of Chin (Grant No. 31270100 to W.C.), Wuhan Youth Chenguang Program of Science and Technology (201507040401018 to W.C.). ABBREVIATIONS SMD, steered molecular dynamics; SRP, substrate recognition peptide; ThyX, thymidylate synthase. COMPLIANCE WITH ETHICS GUIDELINES The authors declare that they have no conflicts of interest pertaining to the contents of this article. This article does not contain any studies with human subjects performed by any of the authors. AUTHOR CONTRIBUTIONS W.C., Y.L., J.Z. and Z.D. conceived the project. W.C. performed genetics experiments. W.C., Y.L. and S.L. performed biochemical experiments. Y.L. L.W. and J.L carried out crystallographic studies, Y.L. and R.X. calculated the molecular modeling data. W.C., Y.L., J.Z. and Z.D. analyzed data and wrote the manuscript. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary material 1 (PDF 2243 kb) Wenqing Chen and Yan Li contributed equally to this work. ==== Refs REFERENCES Blodgett JA Zhang JK Metcalf WW Molecular cloning, sequence analysis, and heterologous expression of the phosphinothricin tripeptide biosynthetic gene cluster from Streptomyces viridochromogenes DSM 40736 Antimicrob Agents Chemother 2005 49 230 240 10.1128/AAC.49.1.230-240.2005 15616300 Chen W Huang T He X Meng Q You D Bai L Li J Wu M Li R Xie Z Characterization of the polyoxin biosynthetic gene cluster from Streptomyces cacaoi and engineered production of polyoxin H J Biol Chem 2009 284 10627 10638 10.1074/jbc.M807534200 19233844 Chen W Qi J Wu P Wan D Liu J Feng X Deng Z Natural and engineered biosynthesis of nucleoside antibiotics in Actinomycetes J Ind Microbiol Biotechnol 2016 43 401 417 10.1007/s10295-015-1636-3 26153500 Emsley P Lohkamp B Scott WG Cowtan K Features and development of Coot Acta Crystallogr D 2010 66 486 501 10.1107/S0907444910007493 20383002 Endo A Kakiki K Misato T Mechanism of action of the antifungal agent polyoxin D J Bacteriol 1970 104 189 196 5473886 Endo A Misato T Polyoxin D, a competitive inhibitor of UDP-N -acetylglucosamine:chitinN -acetylglucosaminyltransferase in Neurospora crassa Biochem Biophys Res Commun 1969 37 718 722 10.1016/0006-291X(69)90870-5 5353899 Graziani S Xia Y Gurnon JR Van Etten JL Leduc D Skouloubris S Myllykallio H Liebl U Functional analysis of FAD-dependent thymidylate synthase ThyX from Paramecium bursaria Chlorella virus-1 J Biol Chem 2004 279 54340 54347 10.1074/jbc.M409121200 15471872 Isono K Nucleoside antibiotics: structure, biological activity, and biosynthesis J Antibiot (Tokyo) 1988 41 1711 1739 10.7164/antibiotics.41.1711 3061990 Isono K Funayama S Suhadolnik RJ Biosynthesis of the polyoxins, nucleoside peptide antibiotics: a new metabolic role for L -isoleucine as a precursor for 3-ethylidene-L -azetidine-2-carboxylic acid (polyoximic acid) Biochemistry 1975 14 2992 2996 10.1021/bi00684a031 1156577 Isono K Suhadolnik RJ The biosynthesis of natural and unnatural polyoxins by Streptomyces cacaoi Arch Biochem Biophys 1976 173 141 153 10.1016/0003-9861(76)90244-7 769694 Isono K Suzuki S The structures of polyoxins A and B Tetrahedron Lett 1968 9 1133 1137 10.1016/S0040-4039(01)98906-3 5640102 Isono K Nagutsu J Kobinata K Sasaki K Suzuki S Studies on polyoxins antifungal antibiotics, Part V: isolation and characterization of polyoxins C, D, E, F, G, H and I Agric Biol Chem 1967 31 190 199 Kieser T Bibb MJ Chater KF Butter MJ Hopwood DA Practical Streptomyces genetics. A laboratory manual 2000 Norwich John Innes Foundation Lorenson MY Maley GF Maley F The purification and properties of thymidylate synthetase from chick embryo extracts J Biol Chem 1967 242 3332 3344 6067595 Maley F The synthesis of 5-methyluridine 5′-phosphate in rat embryo extracts Proc Natl Acad Sci USA 1960 46 632 636 10.1073/pnas.46.5.632 16590651 Mathews II Deacon AM Canaves JM McMullan D Lesley SA Agarwalla S Kuhn P Functional analysis of substrate and cofactor complex structures of a thymidylate synthase-complementing protein Structure 2003 11 677 690 10.1016/S0969-2126(03)00097-2 12791256 McCoy AJ Grosse-Kunstleve RW Adams PD Winn MD Storoni LC Read RJ Phaser crystallographic software J Appl Crystallogr 2007 40 658 674 10.1107/S0021889807021206 19461840 Myllykallio H Lipowski G Leduc D Filee J Forterre P Liebl U An alternative flavin-dependent mechanism for thymidylate synthesis Science 2002 297 105 107 10.1126/science.1072113 12029065 Niu G Tan H Nucleoside antibiotics: biosynthesis, regulation, and biotechnology Trends Microbiol 2015 23 110 119 10.1016/j.tim.2014.10.007 25468791 Omura S Ikeda H Ishikawa J Hanamoto A Takahashi C Shinose M Takahashi Y Horikawa H Nakazawa H Osonoe T Genome sequence of an industrial microorganism Streptomyces avermitilis : deducing the ability of producing secondary metabolites Proc Natl Acad Sci USA 2001 98 12215 12220 10.1073/pnas.211433198 11572948 Sambrook J Fritsch EF Maniatis T Molecular cloning: a laboratory manual 1989 2 Cold Spring Harbor Cold Spring Harbor Zhai L Lin S Qu D Hong X Bai L Chen W Deng Z Engineering of an industrial polyoxin producer for the rational production of hybrid peptidyl nucleoside antibiotics Metab Eng 2012 14 388 393 10.1016/j.ymben.2012.03.006 22465029 Zhao C Huang T Chen W Deng Z Enhancement of the diversity of polyoxins by a thymine-7-hydroxylase homolog outside the polyoxin biosynthesis gene cluster Appl Environ Microbiol 2010 76 7343 7347 10.1128/AEM.01257-10 20817795

PMC005xxxxxx/PMC5003786.txt

==== Front Protein CellProtein CellProtein & Cell1674-800X1674-8018Higher Education Press Beijing 29210.1007/s13238-016-0292-3Research ArticleSubcellular redistribution and sequential recruitment of macromolecular components during SGIV assembly Yuan Yongming Hong Yunhan dbshyh@nus.edu.sg Department of Biological Sciences, National University of Singapore, Science Drive 4, Singapore, 117543 Singapore 18 7 2016 18 7 2016 9 2016 7 9 651 661 18 5 2016 20 6 2016 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.Virus infection consists of entry, synthesis of macromolecular components, virus assembly and release. Understanding of the mechanisms underlying each event is necessary for the intervention of virus infection in human healthcare and agriculture. Here we report the visualization of Singapore grouper iridovirus (SGIV) assembly in the medaka haploid embryonic stem (ES) cell line HX1. SGIV is a highly infectious DNA virus that causes a massive loss in marine aquaculture. Ectopic expression of VP88GFP, a fusion between green fluorescent protein and the envelope protein VP088, did not compromise the ES cell properties and susceptibility to SGIV infection. Although VP88GFP disperses evenly in the cytoplasm of non-infected cells, it undergoes aggregation and redistribution in SGIV-infected cells. Real-time visualization revealed multiple key stages of VP88GFP redistribution and the dynamics of viral assembly site (VAS). Specifically, VP88GFP entry into and condensation in the VAS occurred within a 6-h duration, a similar duration was observed also for the release of VP88GFP-containing SGIV out of the cell. Taken together, VP088 is an excellent marker for visualizing the SGIV infection process. Our results provide new insight into macromolecular component recruitment and SGIV assembly. Electronic supplementary material The online version of this article (doi:10.1007/s13238-016-0292-3) contains supplementary material, which is available to authorized users. Keywords ES cellmedakaorf088SGIVviral assemblyissue-copyright-statement© HEP and Springer 2016 ==== Body INTRODUCTION In human, viral infection causes severe infectious diseases and death such as SARS and H5N1 (Li et al., 2004; Ge et al., 2013). In animal husbandry, viral infection perhaps represents the greatest challenge that results in a massive or complete loss in fish and shellfish aquaculture (Walker and Winton, 2010). Viral infection involves several major steps of viral entry, replication, assembly and release (Dimmock et al., 2007). Understanding of these processes and underlying molecular mechanisms is necessary to develop antiviral drugs and approaches in human healthcare and animal production. More than 25 virus species have been described in diverse fish species of aquaculture importance, 9 of which are listed by the Office of International Epizootic as highly infectious and notifiable viruses (http://www.oie.int). The highly infectious Singapore grouper iridovirus (SGIV) was first isolated in Singapore from the diseased brown spotted grouper (Epinephelus tauvina) as a novel member of genus Ranavirus in the family Iridoviridae (Qin et al., 2001). In natural and farmed habitats, SGIV infection causes serious systemic diseases and massive death in wild and farmed groupers as well as many other marine teleosts (Qin et al., 2003). In cell cultures, SGIV infection induces paraptosis in its natural host species but apoptosis in non-natural hosts (Huang et al., 2011; Yuan et al., 2013). The SGIV genome is a circular double-stranded DNA of 140,131 bp and predicts 162 protein-encoding genes or open reading frames (ORFs) (Song et al., 2004). According to the timing of expression after entry into the host cells, SGIV genes fall into three major groups, which are immediate-early (IE), delay early (DE) and late groups (Williams et al., 2005; Teng et al., 2008). Generally speaking, IE and DE genes are thought to encode regulatory proteins and key catalytic enzymes involved in cellular immune response, cell-cycle control and apoptosis, whereas late genes often code for viral structural proteins that participate in virion formation in a particular cellular compartment called the viral assembly site (VAS) (Chen et al., 2006; Xia et al., 2009). In order to study the viral protein subcellular distribution, the SGIV IE genes of orf086 and orf162 (Xia et al., 2009; Xia et al., 2010), DE gene orf136 (Huang et al., 2008) and late gene orf019 (Huang et al., 2013) were overexpressed in host cells with fused EGFP tag respectively. It revealed that the forced expressed protein encoded by these IE and DE genes was distributed in the cytoplasm, but the protein encoded by the late gene condensed in the VAS at the late stage of SGIV infection. These distribution patterns were further verified with immunofluorescent staining, which demonstrated the reliability of using the ectopically expressed fusion protein to analyze the subcellular location of the viral protein. We make use of medaka (Oryzias latipes) as a model to study virus-host interactions in fish. Medaka is a laboratory fish and holds many genetic resources and toolboxes to study their functions in cellular processes and viral infection. This fish has many stem cell lines (Hong et al., 1996, 1998; Hong et al., 2004b; Yi et al., 2009), which are susceptible to infection by aquaculture-important viruses such as the spring viremia of carp virus (SVCV), viral nervous necrosis virus (VNNV) and SGIV (Yuan et al., 2013). More specifically, medaka has given rise to haploid embryonic stem (ES) cell lines that are capable of whole animal production by semicloning (Yi et al., 2009) and susceptible to SVCV, VNNV and SGIV (Yuan et al., 2013). Therefore, medaka represents a unique organism for haploid genetic screening for host factors that viruses use for infection in fish, as has recently been demonstrated for the identification of human genes essential for influenza A infection in a near-haploid cell line (Carette et al., 2009). SGIV assembly in VAS is a key step in the viral infection cycle. The assembled viruses are subsequently matured and released as infectious pathogens. This study was aimed to identify a molecule that is suitable for visualizing dynamic processes of SGIV assembly and release in a living cell. The SGIV late gene encoded protein VP088 was identified as a putative myristylated envelope protein (Zhou et al., 2011). We overexpressed this protein in host cells and show that the transgene expression of VP88GFP, a fusion between VP088 and green fluorescent protein (GFP), does not compromise the ES cell properties and susceptibility to SGIV. More importantly, VP88GFP shows the dynamic distribution in subcellular compartments. Specifically, the fusion protein disperses evenly in the cytoplasm and undergoes aggregation and redistribution after SGIV infection, which allows for real-time visualization of VAS dynamics in living cells. These results suggest that VP088 plays an important role in SGIV assembly and represents a suitable fusion partner for the production of GFP-tagged recombinant SGIV towards screening for drugs and host factors that control SGIV infection in medaka haploid ES cells. RESULTS Production of transgenic HX1 cells The SGIV VP088 encoded by orf088 was chosen as a marker to visualize viral infection in cell culture. This protein is expressed late during SGIV infection in grouper cells and represents a putative SGIV envelope protein (Zhou et al., 2011), which is conserved in several iridoviruses (Fig. S1). The gene of orf088 was inserted in frame ahead of green fluorescent protein (GFP) in pGFP, resulting in p88GFP (Fig. 1A), which expresses VP88GFP, a fusion protein of 746 amino acid residues between VP088 and GFP (Fig. S2).Figure 1 VP88GFP expression retains the ES cell properties. HX1 cells at 48 h post mock, pGFP and p88GFP transfection were analyzed by microscopy. (A) Schematic structures of pGFP and p88GFP, which express GFP alone and the fusion protein VP88GFP between VP088 and GFP. CV, the human cytomegalovirus early promoter; E, H and X, sites for EcoRI, HindIII and XhoI. (B) Western blot analysis. Cell lysate from HX1, GFP-HX1 and 88GFP-HX1 were analyzed by using αGFP. GFP is seen as a 27-kDa band, and VP88GFP as an 80-kDa band. β-Actin was detected by using as a loading control (β-actin). (C–E’) Fluorescent and phase-contrast micrographs. ni, nucleolus; nu, nucleus. Scale bars, 10 µm. (F) The growth of GFP-HX1 and 88GFP-HX1 cells. Data are means ± S.D. (error bars) from two independent experiments in triplicates. (G) RT-PCR analysis of gene expression. Blastula and gastrula embryos were used as the positive control. Numbers of PCR cycles are given in parenthesis The medaka haploid ES cell line HX1 was chosen as a cellular model to study the SGIV infection process, because it offers a unique opportunity for genetic screening for molecular virus-host interactions and readily detectable cellular properties such as the ES cell phenotype, pluripotency expression and stable growth (Yi et al., 2009, 2010). The vectors of pGFP and p88GFP were separately transfected into HX1, producing stable transgenic clones of 88GFP-HX1 and GFP-HX1. On a Western blot analysis, GFP was detected as a band of about 27-kDa, and VP88GFP as a band of about 80-kDa (Fig. 1B). VP088 expression retains cellular properties Upon transgenic expression in HX1 cells, VP88GFP was evenly distributed in the cytoplasm (Fig. 1C), which is not different from GFP (Fig. 1D). Expression of either VP88GFP or GFP did not alter the ES cell phenotype, since the transgenic cells displayed a round shape, little cytoplasm and prominent nuclei with large nucleoli (Fig. 1C’–E’), as has been reported for medaka diploid ES cell lines (Hong et al., 1996) and haploid ES cell lines including HX1 (Yi et al., 2009). Furthermore, stable growth of VP88GFP-expressing cells 88GFP-HX1 was similar to that of GFP-expressing cells GFP-HX1 (Fig. 1F). Moreover, VP88GFP expression did not change pluripotency expression in HX1 cells, as the parental HX1 cells and their derivatives transgenic for VP88GFP and GFP all expressed pluripotency genes nanog and oct4, but not ntl, a differentiation marker (Fig. 1G). Dynamics of subcellular distribution of VP88GFP As mentioned above, VP88GFP exhibited an even distribution in the cytoplasm, which became more evident at large magnification, where VP88GFP was found to be distributed almost evenly in the cytoplasm (Fig. 2A), and the virus inoculation procedure (at 0 hpi) did not alter the even distribution. After SGIV infection, VP88GFP altered subcellular distribution depending on intervals post infection. At 48 hpi, VP88GFP was highly condensed in the VAS, which resided close to the nucleus and was intensely stained by Hoechst 33342 due to the viral DNA (Fig. 2B and 2C). For comparison, GFP and mitochondria did not exhibit redistribution and localization (Fig. 2A–C). Taken together, VP88GFP shows uniform cytoplasmic distribution on its own and undergoes redistribution into the VAS in SGIV-infected HX1 cells.Figure 2 Subcellular distribution of VP88GFP. HX1 cells were stained with Hoechst 33342 (blue) plus MitoTracker (red) and analyzed by fluorescent microscopy at indicated hpi with SGIV. (A and B) 88GFP-HX1 cells before (A) and 48 h post SGIV infection (B) showing prominent nucleus (nu), well organized mitochondria and wide distribution of VP88GFP (green) in cytoplasm in the absence of SGIV infection (A) and VP88GFP localization in the virus assembly site (VAS; circle), in which most of the mitochondria are excluded. (C) GFP-HX1 cells at 48 hpi with SGIV, showing the lack of GFP localization in VAS. nu, nucleus; ni, nucleolus. Scale bars, 5 µm SGIV infection elicits cell death (Huang et al., 2011; Yuan et al., 2013). It has remained unknown when cell death occurs during SGIV infection. We examined this issue by using fluorescent nuclear dyes Hoechst 33342 and PI. Hoechst 33342 stains both live and dying/dead cells, while PI stains dying/dead cells only. HX1 cells shortly after SGIV infection were positive for Hoechst 33342 but negative for PI (Fig. 3A), suggesting they were living cells as expected. A similar staining was seen also in cells until 24 hpi with SGIV, when VP88GFP formed aggregates and the VAS became visible as a Hoechst 33342-stainable DNA-rich area near the nucleus (Fig. 3B), implying cellular viability at this stage. Interestingly, VP088 remained predominantly in the cytoplasm when VAS was already visible, which demonstrates that VAS formation is independent on VP088. Apparently, the cells became positive for both Hoechst 33342 and PI at 36 hpi, when some VP88GFP aggregates were seen in the VAS (Fig. 3C), indicating the onset of cell death detectable by this staining procedure. Staining with Hoechst 33342 and PI became more confined to the nucleus at 48 (Fig. 3D) and 60 hpi (Fig. 3E). By flow cytometry analyses, SGIV infection caused massive cell death in HX1 cells, with a minority of dead cells being necrotic (7.8%) and a majority being apoptotic (64.4%). Similar values were obtained in GFP-HX1 (8.3% and 66.1%) and 88GFP-HX1 cells (8.6% and 65.4%) (Fig. 4), which indicates that the VP88GFP expression does not alter the cell death profile and pathways. Taken together, VP88GFP localizes into the VAS of SGIV-infected cells and the overexpressed VP88GFP does not change the cell death type of host.Figure 3 Dynamics of VP88GFP distribution and cell death. 88GFP-HX1 cells were infected with SGIV, stained for nuclei with Hoechst 33342 (blue for live cells) plus propidium iodide (PI; red for dead/dying cells) and analyzed by fluorescent microscopy. (A) 88GFP-HX1 cells, showing normal nuclei and even distribution of VP88GFP. (B) 88GFP-HX1 cells at 24 hpi with SGIV, showing viral DNA concentration within the VAS (blue; circle) near the nucleus and VP88GFP aggregation. (C) 88GFP-HX1 at 36 hpi with SGIV, showing VP88GFP distribution into the VAS and cellular death with PI-stainable nuclei (red). (D and E) 88GFP-HX1 at 48 and 60 hpi with SGIV, showing VP88GFP concentration (D) and condensation (E) in the VAS. nu, nucleus; ni, nucleolus. Scale bars, 3 µm Figure 4 SGIV causes medaka cell apoptosis. HX1, GFP-HX1and 88GFP-HX1cells at 48 hpi with SGIV were stained with Annexin-V pacific blue and PI for flow cytometric analysis. Double negative indicates viable cell population, Annexin positive indicates the apoptotic population, PI positive or PI-Annexin double positive indicates the necrotic population. (A–C) HX1, GFP-HX1and 88GFP-HX1cells without SGIV infection. (A’–C’) HX1, GFP-HX1and 88GFP-HX1cells infected with SGIV at 48 hpi, showing increased percentage of apoptotic and necrotic cells. Early apoptotic cells exhibit Annexin (+)/PI (−); late apoptotic cells exhibit Annexin (+)/PI (+); necrotic cells are Annexin (−)/PI (+) Visualization of viral assembly We wanted to analyze the subcellular redistribution of VP88GFP in SGIV-infected HX1 cells. To this, VP88GFP-expressing cells were stained with Hoechst 33342 for DNA in the nucleus and VAS, and they were continuously imaged for a period of 6 h starting at 48 hpi with SGIV. This revealed that VP88GFP became fully localized and highly condensed in the VAS within 6 h (Fig. 5). In the same time, the fully developed VAS disappeared as evidenced by a remarkable decrease in its viral DNA content and VP88GFP concentration (Fig. 5H’). The dynamic process of VP88GFP redistribution is more evident in a time-lapse movie (Movie S1A and B). Taken together, VP88GFP shows biphasic redistribution relative to VAS formation and subsequent viral release from VAS, as summarized in schematic diagrams (Fig. 6). Therefore, VP88GFP offers a marker to visualize the redistribution and recruitment of macromolecular components for SGIV assembly and SGIV release.Figure 5 Visualization of viral assembly site formation and disassembly. 88GFP-HX1 cells were infected with SGIV, stained for nuclei and VAS with Hoechst 33342 (blue). Time-lapse images represented stages of VP88GFP redistribution at intervals from 48 hpi onwards (For more details, see Movie S1 in the supplemental material). Upper panel, merged signals of VP88GFP and Hoechst 33342, showing colocalization of VP88GFP and VAS. Lower panel, signal of Hoechst 33342 showing the nuclei and VAS. Asterisk, the ongoing formation of VAS; triangle, earlier formed VAS containing VP88GFP; hash, release of cellular components containing VP88GFP. Scale bar, 5 µm Figure 6 Representative stages of VAS formation and disassembly. VP88GFP distributes evenly in the endoplasm. In SGIV-infected cells, VAS forms in a perinuclear area rich in viral DNA, which exhibits strong staining with Hoechst 33342 (blue) as the nucleus (nu). VP88GFP is recruited into VAS and condensed there for assembly and maturation. VAS disassembly occurs when matured virus particles are released out of the host cell. SGIV infection induces host cell death, resulting in nuclear fragmentation and cell membrane leakage permissive to staining by DNA dye PI (red) DISCUSSION VAS also known as the “viral factory” is a dynamic cellular structure that forms late during the viral infection cycle and functions in key process of viral replication and/or assembly, and thus represents a target for intervention of viral infection (Novoa et al., 2005; Williams et al., 2005). The VAS of SGIV is a unique subcellular component containing DNA and protein for virus assembly. This DNA gathering point locates at the perinuclear region of the host cell, which can be recognized with DNA staining (Xia et al., 2009; Huang et al., 2013). Understanding of VAS dynamics and underlying mechanisms is pivotal for basic research in host-virus interactions and for the control of viral infection diseases in wild and farmed animals. In this study, we present several independent lines of evidence that the SGIV gene orf088 offers an excellent molecular marker for visualizing VAS dynamics. First, VP088 expression does not alter the cellular property including an ES cell phenotype, self-renewal, pluripotency gene expression, SGIV susceptibility and host cell response to SGIV infection at molecular and cellular (flow cytometry) levels. Second, VP088 shows subcellular redistribution at various stages of SGIV infection, allowing for real-time visualization of VAS dynamics in a host cell. Finally, real-time imaging reveals that VP088 becomes fully localized to, and condensed in VAS within 6 h, and that fully condensed VP088 disappears together with the viral DNA as a consequence of virion releasing, which establishes, for the first time to our knowledge, a 12-h process for VAS formation and SGIV release in a host cell. Macromolecular assembly into complexes and cellular structures operates widely in the living system, ranging from viruses to higher eukaryotic organisms including plants and animals. The mechanisms underlying macromolecular assembly in normal and abnormal processes have attracted considerable attention. In this regard, VAS represents an excellent system to elucidate macromolecular assembly, because many, if not all macromolecular components for the VAS formation and ultimate assembly into virions within VAS are of viral origin and thus exogenous to host cells for clear identification. In this study, GFP-tagged VP088 serves an excellent marker for VAS visualization. Prior to SGIV infection, we revealed that GFP-tagged VP088 on itself is a cytoplasmic protein as intracellular expressed VP88GFP distributes evenly in the cytoplasm of a host cell. However, in SGIV-challenged cells, we have demonstrated that VAS formation is initiated before the condensation of VP088. Redistribution of VP88GFP is triggered by viral invasion and exhibits a distinct pattern by condensing in VAS, which is similar to the VAS related distribution of SGIV envelope protein VP19 (Huang et al., 2013), providing direct evidence that this protein is not required for VAS formation. On the contrary, reports indicated the distribution of non-structural proteins encoded by orf086 or orf162 has no colocalization with the VAS of SGIV (Xia et al., 2009; Xia et al., 2010). These observations suggest that VP088 is indeed a structural protein of SGIV particles and this notion is also supported by a previous report that the VP088 has three putative transmembrane domains and located as a viral envelope protein (Zhou et al., 2011). A closer inspection leads to a striking finding, which is the redistribution and sequential component recruitment for SGIV assembly in a host cell. After condensation in the VAS, VP88GFP becomes hardly detectable by fluorescence. This allows for two alternative explanations. One is its degradation after its involvement in maturation. The other is VAS disassembly due to the release of matured virions as its content. We prefer to the second possibility because the disappearance of VP88GFP from VAS accompanies the disappearance of viral DNA and the appearance of the VP88GFP signal in cytoplasmic areas other than VAS. This is also in accordance with VP088 as an envelope protein (Zhou et al., 2011). Visualization of VP88GFP in this study reveals the dynamic processes of VAS formation and disassembly, which may be described in eight representative stages (Fig. 6). VP88GFP disperses evenly in the cytoplasm, and the SGIV infection procedure does not alter this distribution pattern (Fig. 6A). Upon SGIV infection, VAS formation occurs in the absence of VP88GFP, when VP88GFP undergoes aggregation (Fig. 6B). During subsequent infection, this protein is first seen in the VAS (Fig. 6C), which demarcates the onset of its entry into VAS and suggests sequential recruitment of VP88GFP for SGIV assembly. Meanwhile, the dead cell can be detected by PI staining. When the infection proceeds, VP88GFP becomes concentrated (Fig. 6D) and condensed in the VAS (Fig. 6E). VP88GFP starts to appear outside the VAS (Fig. 6F), suggesting VAS disassembly and virion release into the nearby cytoplasm. Ultimately, VP88GFP-containing SGIV virus particles spread throughout the cytoplasm (Fig. 6G) and finally release out of the cell membrane (Fig. 6H), completing the infectious cycle. The host cells become dead by apoptosis and necrosis as evidenced by PI staining and nuclear fragmentation (Fig. 6D–H). Viral infection brings about two major events, namely virus propagation and host cell response. SGIV causes host cell death by two pathways: One is non-apoptotic programmed cell death (PCD) as has been reported in its natural host species (Huang et al., 2011), the other is apoptosis as has been reported in non-natural host species and medaka HX1 cells in culture (Huang et al., 2011; Yuan et al., 2013). In this study, we have observed that SGIV induces not only apoptosis as a major death pathway but also necrosis at a detectable level. More importantly, one interesting observation is that the distribution of ectopic expressed VP88GFP changed after virus infection by aggregation and condensation into the VAS. However, our results here do not reveal the distribution process of viral genome itself encoded VP088 throughout the infection cycle. The gene encoding VP88GFP is inserted into the host genome together with a CMV promoter, but the VP088 is encoded by the genome of the infected virus. The gene copy numbers of them are different, and the expression of each protein is driven by a different promoter. Additionally, the timing of protein expression varied from each other. The VP88GFP is expressed before the virus infection, but the expression of VP088 is activated only after virus infection. Considering the above concerns, generation of a recombinant SGIV containing a GFP-tagged VP088 will resolve this issue in the future. Successful visualization of VAS dynamics with fluorescent protein tagged virus has been reported (Heath et al., 2001). Future work is needed to elucidate the mechanisms underlying programmed aggregation and cell death commencement as well as the mechanism underlying SGIV infection-dependent redistribution of VP088 and the precise role that VP088 plays in SGIV assembly and release. The recently published study has illustrated the assembly and budding of SGIV with electron miscopy (Liu et al., 2016) and the details of how SGIV entry into host cells by labeling the SGIV particles with chemical dye (Wang et al., 2014). In summary, VP088 is not cytotoxic and does not compromise the ES cell property, viral susceptibility and host-virus interactions. This protein undergoes SGIV-dependent subcellular redistribution and shows sequential recruitment into the VAS for viral assembly. These features make VP88GFP an excellent marker for generating GFP-tagged recombinant SGIV for the experimental analysis and real-time visualization of SGIV infection. MATERIALS AND METHODS Fish Work with fish followed the guidelines on the Care and Use of Animals for Scientific Purposes of the National Advisory Committee for Laboratory Animal Research in Singapore and approved by this committee (permit number 27/09). Medaka was maintained under an artificial photoperiod of 14-h/10-h light/darkness at 26°C as described (Li et al., 2009; Hong et al., 2010). Plasmids Plasmid p88GFP that encodes the fusion protein VP88GFP between VP088 and GFP was constructed by three-component ligation. Briefly, the orf088 coding sequence (CDS) was amplified by using primers orf088Eco (aagaattcaccATGGGCGCAGCGC) plus orf088Hind (gcaagcttCTTTGCAGCTTC) from SGIV, and the gfp CDS was PCR-amplified by using primers GFPHind (gcaagcttGTGAGCAAGGGCGAG) plus GFPXho (gactcgagTCACTTGTACAGCTCG) from pEGFP-N1 (Clontech). The PCR products were digested with EcoRI plus HindIII (orf088 fragment) or HindIII plus XhoI (gfp fragment) and combined with EcoRI-XhoI double-digested pcDNA3.1 for ligation. Control plasmid pGFP was generated with an insertion gene of gfp between restriction sites of EcoRI and XhoI in pcDAN3.1. Correct constructs were confirmed by sequencing. Plasmid DNA was prepared with a Midiprep kit (Qiagen, Valencia, CA, USA). Cell culture and transfection The medaka haploid ES cell line HX1 was maintained at 28°C in the medium of ESM4 as previously described (Hong and Schartl, 2006; Yi et al., 2010). The grouper spleen cell line GS was maintained at 25°C in L15-medium (Leibovitz) containing 10% fetal bovine serum (Huang et al., 2009). Cell transfection was performed by using DNAfectin reagent (Applied Biological Materials, Richmond, BC, Canada) essentially as described (Hong et al., 2004a). Briefly, 2 µg of plasmid DNA (p88GFP or pGFP) and 8 µL of DNAfectin reagent were mixed in 200 µL of pure DMEM. After incubation at room temperature for 20 min, the transfection mixture was added dropwise to cells in a 6-well plate containing 2 mL of DMEM. After incubation for 6 h at 28°C, the cells were grown in ESM4 for 48 h and subcultured in 10-cm dishes for clonal growth in the presence of 0.5 mg/mL of G418 (Hong et al., 1996). The medium was changed every 5–7 days. Single colonies comprising GFP-positive cells were picked with 200-µL tips into 96-well plates and serially expanded into 88GFP-HX1 cells (p88GFP transfectants) and GFP-HX1 cells (pGFP transfectants) as described (Hong et al., 1996). Virus preparation and inoculation SGIV (strain A3/12/98) originally isolated from the diseased brown-spotted grouper (E. tauvina) was propagated in GS cells as described (Qin et al., 2003). Briefly, SGIV was inoculated onto confluent GS cells at a multiplicity of infection (MOI) of ∼0.1. Upon the appearance of apparent cytopathic effect, cells were harvested and centrifuged at 3000 ×g for 10 min at 4°C, the cell debris together with partial supernatant were collected and stored at −80°C until use. HX1 cells were infected similarly. RT-PCR analysis RNA isolation from cell culture and RT-PCR analyses were performed as described (Hong et al., 2004b; Yuan et al., 2013). PCR was run in a 20-µL volume containing 10 ng of cDNA reaction for 25 (β-actin as a loading control) and 35 cycles (95°C for 30 s, 60°C for 20 s and 72°C for 1 min; other genes). PCR products were separated on 2% agarose gels. Primers used are listed in Table S1. Cell growth assay Cell growth was analyzed as described (Hong et al., 1996; Yi et al., 2009). Briefly, 105 of 88GFP-HX1 and GFP-HX1 cells were seeded into the 6-well plate and counted in triplicates every 24 h until 8 days of culture. Cell staining Growing cells in culture were co-stained with Hoechst 33342 and propidium iodide (PI) before fluorescent microscopic observation. In detail, the culture medium containing Hoechst 33342 (1 μg/mL) plus PI (1 μg/mL) were added carefully into the culture containing virus-infected cells and incubated at 28°C for 10 min. To reduce the fluorescence background, the cells were carefully rinsed in phosphate buffered saline (PBS) and refed with fresh medium. Nuclear staining in living cells (Hoechst 33342) and dying/dead cells (PI) was visualized by fluorescent microscopy. Flow cytometric assay HX1, GFP-HX1 and 88GFP-HX1 cells at 48 hpi with SGIV (MOI of 0.1) were trypsinized into single cell suspension and 105 cells were stained with 5 μL of Annexin V/pacific blue (Invitrogen, USA) in 100 μL of binding buffer for 15 min at room temperature and counterstained with PI at 50 μg/mL. SGIV infected cells and mock control cells were analyzed on the BD LSR Fortessa (Becton Dickinson, San Jose, CA, USA). Microscopy Observation and photography on Zeiss Axiovert invert microscope with a Zeiss AxioCam M5Rc digital camera (Zeiss Corp., Germany) were done as described (Yi et al., 2009; Yuan et al., 2014). Confocal microscopic observation and time-lapse imaging were performed on the UltraView VoX (PerkinElmer, Waltham, MA, USA) using an Olympus water-immersion 40× objective lens (NA = 1.15; Olympus, Tokyo, Japan) by using software Volocity 6.2.1 (PerkinElmer) setting for sequential record modes at 3 channels of laser lines at 405, 488 and 561 nm. Statistical analysis The Dunnett’s test was conducted by using GraphPad Prism v4.0. Data are presented as means ± S.D, and P < 0.05 were calculated by using Student’s t-test and considered as significant differences as described (Yi et al., 2010). Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary material 1 (MOV 663 kb) Supplementary material 2 (JPEG 168 kb) Supplementary material 3 (JPEG 125 kb) Supplementary material 4 (PDF 181 kb) Supplementary material 5 (PDF 113 kb) ACKNOWLEDGMENTS We thank J. Deng for fish breeding, CM. Foong for laboratory management. This work was supported by the National Research Foundation Singapore (NRF-CRP7-2010-03). ABBREVIATIONS ES, embryonic stem; hpi, hours post infection; SGIV, Singapore grouper iridovirus; VAS, viral assembly site COMPLIANCE WITH ETHICS GUIDELINES Yongming Yuan and Yunhan Hong declare that they have no conflict of interest. This article does not contain any studies with human subjects performed by the any of the authors. Work with fish followed the guidelines on the Care and Use of Animals for Scientific Purposes of the National Advisory Committee for Laboratory Animal Research in Singapore and approved by this committee (permit number 27/09). ==== Refs References Carette JE Guimaraes CP Varadarajan M Park AS Wuethrich I Godarova A Kotecki M Cochran BH Spooner E Ploegh HL Haploid genetic screens in human cells identify host factors used by pathogens Science 2009 326 1231 1235 10.1126/science.1178955 19965467 Chen LM Wang F Song W Hew CL Temporal and differential gene expression of Singapore grouper iridovirus J Gen Virol 2006 87 2907 2915 10.1099/vir.0.82219-0 16963749 Dimmock NJ Easton AJ Leppard K Introduction to modern virology 2007 6 Malden Blackwell Ge XY Li JL Yang XL Chmura AA Zhu G Epstein JH Mazet JK Hu B Zhang W Peng C Isolation and characterization of a bat SARS-like coronavirus that uses the ACE2 receptor Nature 2013 503 535 538 10.1038/nature12711 24172901 Heath CM Windsor M Wileman T Aggresomes resemble sites specialized for virus assembly J Cell Biol 2001 153 449 455 10.1083/jcb.153.3.449 11331297 Hong Y Schartl M Isolation and differentiation of medaka embryonic stem cells Methods Mol Biol 2006 329 3 16 16845980 Hong Y Winkler C Schartl M Pluripotency and differentiation of embryonic stem cell lines from the medakafish (Oryzias latipes ) Mech Dev 1996 60 33 44 10.1016/S0925-4773(96)00596-5 9025059 Hong Y Winkler C Schartl M Production of medakafish chimeras from a stable embryonic stem cell line Proc Natl Acad Sci USA 1998 95 3679 3684 10.1073/pnas.95.7.3679 9520425 Hong Y Chen S Gui J Schartl M Retention of the developmental pluripotency in medaka embryonic stem cells after gene transfer and long-term drug selection for gene targeting in fish Transgenic Res 2004 13 41 50 10.1023/B:TRAG.0000017172.71391.fa 15070074 Hong Y Liu T Zhao H Xu H Wang W Liu R Chen T Deng J Gui J Establishment of a normal medakafish spermatogonial cell line capable of sperm production in vitro Proc Natl Acad Sci USA 2004 101 8011 8016 10.1073/pnas.0308668101 15141090 Hong N Li M Zeng Z Yi M Deng J Gui J Winkler C Schartl M Hong Y Accessibility of host cell lineages to medaka stem cells depends on genetic background and irradiation of recipient embryos Cell Mol Life Sci 2010 67 1189 1202 10.1007/s00018-009-0247-4 20238480 Huang X Huang Y Gong J Yan Y Qin Q Identification and characterization of a putative lipopolysaccharide-induced TNF-alpha factor (LITAF) homolog from Singapore grouper iridovirus Biochem Biophys Res Commun 2008 373 140 145 10.1016/j.bbrc.2008.06.003 18554501 Huang X Huang Y Sun J Han X Qin Q Characterization of two grouper Epinephelus akaara cell lines: application to studies of Singapore grouper iridovirus (SGIV) propagation and virus-host interaction Aquaculture 2009 292 172 179 10.1016/j.aquaculture.2009.04.019 Huang X Huang Y Ouyang Z Xu L Yan Y Cui H Han X Qin Q Singapore grouper iridovirus, a large DNA virus, induces nonapoptotic cell death by a cell type dependent fashion and evokes ERK signaling Apoptosis 2011 16 831 845 10.1007/s10495-011-0616-y 21656148 Huang X Gong J Huang Y Ouyang Z Wang S Chen X Qin Q Characterization of an envelope gene VP19 from Singapore grouper iridovirus Virol J 2013 10 354 10.1186/1743-422X-10-354 24341864 Li KS Guan Y Wang J Smith GJ Xu KM Duan L Rahardjo AP Puthavathana P Buranathai C Nguyen TD Genesis of a highly pathogenic and potentially pandemic H5N1 influenza virus in eastern Asia Nature 2004 430 209 213 10.1038/nature02746 15241415 Li M Hong N Xu H Yi M Li C Gui J Hong Y Medaka vasa is required for migration but not survival of primordial germ cells Mech Dev 2009 126 366 381 10.1016/j.mod.2009.02.004 19249358 Liu Y Tran BN Wang F Ounjai P Wu J Hew CL Visualization of assembly intermediates and budding vacuoles of Singapore Grouper Iridovirus in Grouper embryonic cells Sci Rep 2016 6 18696 10.1038/srep18696 26727547 Novoa RR Calderita G Arranz R Fontana J Granzow H Risco C Virus factories: associations of cell organelles for viral replication and morphogenesis Biol Cell 2005 97 147 172 10.1042/BC20040058 15656780 Qin QW Lam TJ Sin YM Shen H Chang SF Ngoh GH Chen CL Electron microscopic observations of a marine fish iridovirus isolated from brown-spotted grouper, Epinephelus tauvina J Virol Methods 2001 98 17 24 10.1016/S0166-0934(01)00350-0 11543880 Qin QW Chang SF Ngoh-Lim GH Gibson-Kueh S Shi C Lam TJ Characterization of a novel ranavirus isolated from grouper Epinephelus tauvina Dis Aquat Organ 2003 53 1 9 10.3354/dao053001 12608562 Song WJ Qin QW Qiu J Huang CH Wang F Hew CL Functional genomics analysis of Singapore grouper iridovirus: complete sequence determination and proteomic analysis J Virol 2004 78 12576 12590 10.1128/JVI.78.22.12576-12590.2004 15507645 Teng Y Hou Z Gong J Liu H Xie X Zhang L Chen X Qin QW Whole-genome transcriptional profiles of a novel marine fish iridovirus, Singapore grouper iridovirus (SGIV) in virus-infected grouper spleen cell cultures and in orange-spotted grouper, Epinephulus coioides Virology 2008 377 39 48 10.1016/j.virol.2008.04.011 18555886 Walker P Winton JR Emergin viral diseases of fish and shrimp Vet Res 2010 41 24 10.1051/vetres/2010022 19941812 Wang S Huang X Huang Y Hao X Xu H Cai M Wang H Qin Q Entry of a novel marine DNA virus, Singapore grouper iridovirus, into host cells occurs via clathrin-mediated endocytosis and macropinocytosis in a pH-dependent manner J Virol 2014 88 13047 13063 10.1128/JVI.01744-14 25165116 Williams T Barbosa-Solomieu V Chinchar VG A decade of advances in iridovirus research Adv Virus Res 2005 65 173 248 10.1016/S0065-3527(05)65006-3 16387197 Xia L Cao J Huang X Qin Q Characterization of Singapore grouper iridovirus (SGIV) ORF086R, a putative homolog of ICP18 involved in cell growth control and virus replication Arch Virol 2009 154 1409 1416 10.1007/s00705-009-0457-y 19629635 Xia L Liang H Huang Y Ou-Yang Z Qin Q Identification and characterization of Singapore grouper iridovirus (SGIV) ORF162L, an immediate-early gene involved in cell growth control and viral replication Virus Res 2010 147 30 39 10.1016/j.virusres.2009.09.015 19800375 Yi M Hong N Hong Y Generation of medaka fish haploid embryonic stem cells Science 2009 326 430 433 10.1126/science.1175151 19833967 Yi M Hong N Hong Y Derivation and characterization of haploid embryonic stem cell cultures in medaka fish Nat Protoc 2010 5 1418 1430 10.1038/nprot.2010.104 20671725 Yuan Y Huang X Zhang L Zhu Y Huang Y Qin Q Hong Y Medaka haploid embryonic stem cells are susceptible to Singapore grouper iridovirus as well as to other viruses of aquaculture fish species J Gen Virol 2013 94 2352 2359 10.1099/vir.0.054460-0 23828270 Yuan Y Li M Hong N Hong Y Correlative light and electron microscopic analyses of mitochondrial distribution in blastomeres of early fish embryos FASEB J 2014 28 577 585 10.1096/fj.13-233635 24136588 Zhou S Wan Q Huang Y Huang X Cao J Ye L Lim TK Lin Q Qin Q Proteomic analysis of Singapore grouper iridovirus envelope proteins and characterization of a novel envelope protein VP088 Proteomics 2011 11 2236 2248 10.1002/pmic.200900820 21538879

PMC005xxxxxx/PMC5003787.txt

==== Front Protein CellProtein CellProtein & Cell1674-800X1674-8018Higher Education Press Beijing 29710.1007/s13238-016-0297-yLetterObacunone activates the Nrf2-dependent antioxidant responses Xu Shengmei 1Chen Weimin 1Xie Qingfeng 1Xu Yang yangxu@ucsd.edu 121 Center for Regenerative and Translational Medicine, Guangdong Provincial Academy of Chinese Medical Sciences, the Second Affiliated Hospital of Guangzhou University of Chinese Medicine, Guangzhou, 510632 China 2 Division of Biological Sciences, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093 USA 16 8 2016 16 8 2016 9 2016 7 9 684 688 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.issue-copyright-statement© HEP and Springer 2016 ==== Body Dear Editor, The transcription factor nuclear factor erythroid 2-related factor 2 (Nrf2) plays a crucial role in human antioxidant defense response against environmental insults. Small molecular chemical activators of Nrf2 can confer protection against oxidative insults and inhibit the progression of diseases related to oxidative stress. Here, we identified obacunone as a novel activator of Nrf2 by decreasing Nrf2 ubiquitination and increasing its stability. In support of these findings, the systemic administration of obacunone strongly inhibits bleomycin-induced lung fibrosis in mice. Therefore, obacunone may also provide antioxidant protection for humans against tissue damage caused by oxidative insults. The nuclear factor-erythroid 2-related factor (Nrf2) is a redox-sensitive transcriptional factor, activity of which is tightly regulated by cytosolic protein Kelch-like ECH-associated protein 1 (Keap1) (Chen et al., 2009). The keap1-Nrf2 signaling pathway is a master antioxidant defense mechanism against environmental insults. In the absence of oxidative stress, Nrf2 is bound to Keap1 and ubiquitinated by a cul3-based E3 ligase, leading to proteasome-mediated degradation (Cullinan et al., 2004). In response to oxidative stress, the E3 ubiquitin ligase is inactivated through the modification of critical cysteine residues of Keap1, thereby allowing the accumulation of Nrf2 and its translocation into the nucleus (Zhang et al., 2004; Villeneuve et al., 2010). The activated Nrf2 exerts an intracellular antioxidant defense response through transcriptional activation of many genes including phase II detoxifying enzymes, drug transporters, antioxidant enzymes, and proteins involved in the repair and removal of damaged macromolecules (Villeneuve et al., 2010), thereby neutralizing reactive oxygen species (ROS), detoxifying harmful chemicals, and maintaining redox homeostasis. Therefore, the Nrf2 activation with small molecular chemicals can enhance body’s defense against oxidative insults. Several small molecular chemical activators of Nrf2 have already been identified, most of which are plant-derived chemicals such as sulforaphane, tanshinone I, curcumin, BHA (Jeong et al., 2006; Nishinaka et al., 2007; Zhang and Hannink, 2003). Using in vivo mouse models, the Nrf2 activation induced by these activators has been shown to inhibit the progression of diseases related to oxidative stress, including diabetes, cancer, cardiovascular diseases, neurodegenerative diseases, pulmonary fibrosis, and inflammatory diseases (Chapple et al., 2012). Conversely, the Nrf2-knockout mice are prone to acute damages and tumor formation induced by various harmful chemicals (Ramos-Gomez et al., 2001; Wang et al., 2008). All these data demonstrate that Nrf2 can be used as a drug target for prevention and therapy of diseases related to oxidative stress. Therefore, the identification of novel and non-toxic Nrf2 activators is important for the development of effective dietary supplements or therapeutic drugs that protect humans against oxidative damage. Employing the MDA-MB-231 cells stably transfected with an ARE-luciferase reporter as a screening platform (Chen et al., 2016), we identified that obacunone could induce the expression of the ARE-dependent luciferase gene in a dose-dependent manner (Fig. S1). Obacunone is a small molecular compound derived from citrus fruits with a higher content in seeds. Published studies indicate that obacunone can inhibit cancer proliferation and has some therapeutic effects on cardiovascular diseases (Poulose et al., 2006; Yoon et al., 2014). As expected, the protein levels of Nrf2 were increased in various cell types after obacunone treatment, while Keap1 levels remained constant in MDA-MB-231 cells (Fig. 1A–C). Consistent with this finding, the mRNA levels of Nrf2 target genes such as NQO1, Mrp2, and HO-1 were increased in these cells after being treated with increasing doses of obacunone (Fig. 1D and 1E). To evaluate the impact of obacunone on Nrf2-dependent antioxidant effect, MDA-MB-231 cells were pretreated with obacunone for 8 h, and then challenged with H2O2 for an additional 12 h. Pretreatment with obacunone reduced ROS levels significantly (Fig. 1F). Therefore, obacunone can effectively protect cells from oxidative stress by activating the Nrf2 pathway.Figure 1 Obacunone activates the Nrf2 pathway. (A) MDA-MB-231 cells were treated with the indicated dose of obacunone for 4 h, and cell lysate was subject to immunoblot analysis with anti-Nrf2, -Keap1, -GAPDH antibodies. (B) RAW264.7 cells were treated with the indicated dose of obacunone for 4 h, and cell lysate was subject to immunoblot analysis. (C) LO2 cells were treated with the indicated dose of obacunone for 4 h, and cell lysate was subject to immunoblot analysis. 5 μmol/L SF treatment was included as a positive control in (A), (B), and (C). (D) MDB-MB-231 cells were either mock treated or treated with OC or SF for 24 h, then total RNAs were extracted. Relative amounts of NQO1, MRP2, and HO-1 mRNAs were measured by qRT-PCR. The standard deviations were calculated from triplicate samples. * P < 0.05, ** P < 0.01 compared with its control. (E) RAW264.7 cells were either mock treated or treated with OC or SF for 24 h, then total RNAs were extracted. Relative amounts of NQO1 and HO-1 mRNAs were measured by qRT-PCR. The standard deviations were calculated from triplicate samples. * P < 0.05, ** P < 0.01 compared with its control. (F) MDA-MB-231 cells were mock treated or pretreated with 80 μmol/L OC or 5 μmol/L SF for 8 h, then challenged with 0.5 mmol/L H2O2 for another 12 h. Cellular ROS level was detected by DCF staining and subsequent flow cytometry. **P < 0.01, compared with its control. ## P < 0.01, compared with only H2O2-treated cells. (G) MDB-MB-231 cells were either mock treated or treated with 40 μmol/L OC for 4 h. 50 μmol/L cycloheximide was added and cells were lysed at the indicated time points. Cell lysates were subjected to immunoblot analysis using anti-Nrf2 and anti-GAPDH antibodies. The intensity of the bands was quantified using Quantity One software and plotted against the time after cycloheximide treatment. (H) 293T cells were co-transfected with the plasmids encoding the indicated proteins. Cells were then treated with either 5 μmol/L SF or 60 μmol/L OC along with 10 μmol/L MG132 for 4 h before cell lysates were collected for ubiquitination assay. Anti-Nrf2 immunoprecipitates were analyzed by immunoblot with anti-Nrf2 antibodies for detection of ubiquitin-conjugated Nrf2 To investigate the mechanism how obacunone increases the protein levels of Nrf2, we measured the half-life of Nrf2 protein in the absence or presence of obacunone. At the indicated time points after the addition of cycloheximide (CHX), the protein levels of Nrf2 in the absence or presence of obacunone were examined. While the half-life of Nrf2 protein was 16.9 min in the absence of obacunone, the half-life of Nrf2 protein in the presence of obacunone was increased to 35.9 min, indicating that obacunone stabilizes Nrf2 protein (Fig. 1G). To examine the impact of obacunone on the ubiquitination of Nrf2, the cells were either mock treated or treated with obacunone and the known Nrf2 activator SF. As expected, obacunone inhibits the ubiquitination of Nrf2 similarly to the positive control SF (Fig. 1H). These data support the notion that obacunone activates the Nrf2 pathway by stabilizing Nrf2 through inhibiting its ubiquitination and activating Nrf2-dependent responses. Bleomycin is a chemotherapeutic drug used clinically for a variety of human malignancies. It has been reported that the administration of bleomycin often leads to lung injury and fibrosis in human patients. Bleomycin-induced lung injury in mice is a well-established in vivo model of human pulmonary fibrosis (Hoshino et al., 2009). Numerous studies have shown that the disturbance of the alveolar oxidant-antioxidant balance may play a role in the pathogenesis of chronic fibrosis (Gasse et al., 2007). Therefore, considering that the activation of Nrf2 results in antioxidant defense response, bleomycin-induced lung injury can be used to evaluate the therapeutic effects of Nrf2 activators, We first evaluated the conditions (dose and injection frequency) that result in the activation of the Nrf2-dependent response in the lung of mice. Forty-eight hours after systemic delivery of obacunone (10 mg/kg, i.p.) in B6 mice, the protein levels of Nrf2 in the lung were upregulated (Fig. 2A). Consistent with this data, the mRNA levels of Nrf2 target genes NQO1 and HO-1 were also increased (Fig. 2B).Figure 2 Obacunone activates the pulmonary Nrf2 and inhibits bleomycin-induced lung fibrosis. (A) Lung tissue lysates from mice were subjected to immunoblot analysis with anti-Nrf2, -GAPDH antibodies. Lanes 1–3 indicate control samples, lanes 4–6 indicate samples from mice subjected to OC administration. (B) Total RNAs were extracted from freshly isolated lung tissue, and relative amounts of NQO1, HO-1 mRNAs were measured by qRT-PCR. The standard deviations were calculated from triplicate samples. ** P < 0.01, compared with its control. (C) Each group shows a representative image of the lung tissue for HE staining respectively. (D) The representative images of the lung tissue for Masson’s trichrome staining were shown. (E) HYP level in different lung tissues was compared. * P < 0.05, control group vs. BLM-administrated froup; # P <0.05, BLM-administrated group vs. OC + BLM-administrated group. (F) Relative mRNA expression of TGF-β was measured by real-time RT-PCR. ** P < 0.01, control group compared with BLE-administrated group; # P < 0.05, BLM-administrated group compared with OC + BLM-administrated group. (G) Obacunone inhibits the BLM-induced lung fibrosis as revealed by IHC analysis with anti α-SMA antibodies Bleomycin (BLM) induced lung inflammation and fibrosis (Fig. 2C). For the analysis of lung fibrosis, the collagen content in the lung tissue was examined. When compared with that in PBS-treated mice, collagen deposition (stained blue by masson’s staining) was obviously increased in the lung tissue of BLM-treated mice. Treatment of mice with obacunone significantly reduces the collagen deposition in the lung of BLM-administrated mice (Fig. 2D). To further quantify for the levels of collagen deposition, we measured the hydroxyproline content as an index of collagen, further demonstrating that obacunone treatment significantly reduces the collagen deposition in BLM-administered mice (Fig. 2E). As additional indicators of tissue fibrosis, we examined the expression of α-SMA and TGF-β. We show that TGF-β expression was significantly upregulated in BLM-damaged lung tissue, while obacunone treatment suppressed BLM-induced TGF-β expression in the lung tissue (Fig. 2F). Consistent with this finding, the number of α-SMA+ myofibroblasts was significantly increased in the lung tissue of bleomycin-administered mice, obacunone treatment reduces the number of α-SMA+ myofibroblasts in the lung tissues of bleomycin administered mice (Fig. 2G). In summary, these findings demonstrate that obacunone can effectively protect mice against BLM-induced lung fibrosis. BLM-induced damage causes inflammation in the lung, leading to fibrosis (Gasse et al., 2007; Hoshino et al., 2009). To understand the mechanism how obacunone protects mice against BLM-induced lung fibrosis, we examined the impact of obacunone on the BLM-induced inflammation in the lung of B6 mice. We examined the mRNA levels of several cytokines associated with pulmonary inflammation and fibrosis. In support of the notion that obacunone suppresses the BLM-induced inflammation, obacunone treatment suppressed the BLM-induced expression of inflammatory cytokines IL-6, IL-17, MCP-1, but increased the mRNA levels of IFN-γ in the lung (Fig. S2). In summary, these data support the notion that obacunone suppresses BLM-induced inflammation and fibrosis of the lung by activating Nrf2 pathway. With accumulating knowledge of the roles of Nrf2 in protecting against environmental insults, it becomes increasingly important to identify and optimize Nrf2 activators that can protect humans from various environmental insults. We discovered a novel Nrf2 activator, a natural product obacunone. In this context, we demonstrate that obacunone can stabilize and activate Nrf2 by inhibiting Nrf2 ubiquitination. In addition, we demonstrated that oxidative stress-induced lung injury including inflammation and fibrosis could be effectively inhibited by systemic administration of obacunone. Similar to another known Nrf2 activator sulforaphane, obacunone is able to block ubiquitination and degradation of Nrf2, thus resulting in the prolonged half-life of Nrf2. It remains technically challenging to identify the direct target of obacunone in cells. Future effort will be devoted to understand the mechanism how obacunone stabilizes Nrf2. One possibility is that obacunone interferes with the function of Keap1, the E3 ligase of Nrf2, leading to the stabilization of Nrf2. Another possibility is that obacunone can disrupt the interaction between Nrf2 and Keap1, leading to the stabilization of Nrf2. Identification of obacunone as a novel Nrf2 activator broadens the choice for new chemopreventive compounds against tissue damage or disease caused by various environmental insults. Compared with other small molecule Nrf2 activators, obacunone is a natural product with an advantage in its safety. In support of this notion, obacunone exhibits no apparent cytotoxicity in vitro and in vivo for the limited time duration tested (Fig. S3). Taken together, our experiments demonstrate the feasibility of preventing bleomycin-induced lung injury by systemic administration of obacunone, a novel potent Nrf2 activator. The obacunone-mediated intervention may also be efficacious for other types of environmental insults and confer protection against tissue damage in other organs. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary material 1 (PDF 845 kb) FOOTNOTES This study is supported by a grant from the National Basic Research Program (973 Program) (No. 2013CB966900), Guangdong Provincial Key Laboratory of Cancer Immunotherapy and Guangzhou Key Laboratory of Tumor Immunology Research, and the National Natural Science Foundation of China (No. 81473294). SX and YX conceived and designed the experiments. SX WC, and QX performed the experiments. SX WC, and YX analyzed the data. SX WC, and YX wrote the manuscript. Shengmei Xu, Weimin Chen, Qingfeng Xie, and Yang Xu declare that they have no conflict of interest. All institutional and national guidelines for the care and use of laboratory animals were followed. ==== Refs References Chapple SJ Siow RC Mann GE Crosstalk between Nrf2 and the proteasome: therapeutic potential of Nrf2 inducers in vascular disease and aging Int J Biochem Cell Biol 2012 44 1315 1320 10.1016/j.biocel.2012.04.021 22575091 Chen W Sun Z Wang XJ Jiang T Huang Z Fang D Zhang DD Direct interaction between Nrf2 and p21(Cip1/WAF1) upregulates the Nrf2-mediated antioxidant response Mol Cell 2009 34 663 673 10.1016/j.molcel.2009.04.029 19560419 Chen, W, Li, S, Li, J, Zhou, W, Wu, S, Xu, S, Cui, K, Zhang, DD, Liu, B (2016) Artemisitene activates the Nrf2-dependent antioxidant response and protects against bleomycin-induced lung injury. FASEB J. fj.201500109R Cullinan SB Gordan JD Jin J Harper JW Diehl JA The Keap1-BTB protein is an adaptor that bridges Nrf2 to a Cul3-based E3 ligase: oxidative stress sensing by a Cul3-Keap1 ligase Mol Cell Biol 2004 24 8477 8486 10.1128/MCB.24.19.8477-8486.2004 15367669 Gasse P Mary C Guenon I Noulin N Charron S Schnyder-Candrian S Schnyder B Akira S Quesniaux VF Lagente V IL-1R1/MyD88 signaling and the inflammasome are essential in pulmonary inflammation and fibrosis in mice J Clin Invest 2007 117 3786 3799 17992263 Hoshino T Okamoto M Sakazaki Y Kato S Young HA Aizawa H Role of proinflammatory cytokines IL-18 and IL-1beta in bleomycin-induced lung injury in humans and mice Am J Respir Cell Mol Biol 2009 41 661 670 10.1165/rcmb.2008-0182OC 19265174 Jeong WS Jun M Kong AN Nrf2: a potential molecular target for cancer chemoprevention by natural compounds Antioxid Redox Signal 2006 8 99 106 10.1089/ars.2006.8.99 16487042 Nishinaka T Ichijo Y Ito M Kimura M Katsuyama M Iwata K Miura T Terada T Yabe-Nishimura C Curcumin activates human glutathione S-transferase P1 expression through antioxidant response element Toxicol Lett 2007 170 238 247 10.1016/j.toxlet.2007.03.011 17449203 Poulose SM Harris ED Patil BS Antiproliferative effects of citrus limonoids against human neuroblastoma and colonic adenocarcinoma cells Nutr Cancer 2006 56 103 112 10.1207/s15327914nc5601_14 17176224 Ramos-Gomez M Kwak MK Dolan PM Itoh K Yamamoto M Talalay P Kensler TW Sensitivity to carcinogenesis is increased and chemoprotective efficacy of enzyme inducers is lost in nrf2 transcription factor-deficient mice Proc Natl Acad Sci USA 2001 98 3410 3415 10.1073/pnas.051618798 11248092 Villeneuve NF Lau A Zhang DD Regulation of the Nrf2-Keap1 antioxidant response by the ubiquitin proteasome system: an insight into cullin-ring ubiquitin ligases Antioxid Redox Signal 2010 13 1699 1712 10.1089/ars.2010.3211 20486766 Wang XJ Sun Z Villeneuve NF Zhang S Zhao F Li Y Chen W Yi X Zheng W Wondrak GT Nrf2 enhances resistance of cancer cells to chemotherapeutic drugs, the dark side of Nrf2 Carcinogenesis 2008 29 1235 1243 10.1093/carcin/bgn095 18413364 Yoon J Park M Lee J Min BS Ryoo S Endothelial nitric oxide synthase 11 activation through obacunone-dependent arginase inhibition restored impaired endothelial function in ApoE-null mice Vascul Pharmacol 2014 60 102 109 10.1016/j.vph.2014.01.006 24509132 Zhang DD Hannink M Distinct cysteine residues in Keap1 are required for Keap1-dependent ubiquitination of Nrf2 and for stabilization of Nrf2 by chemopreventive agents and oxidative stress Mol Cell Biol 2003 23 8137 8151 10.1128/MCB.23.22.8137-8151.2003 14585973 Zhang DD Lo SC Cross JV Templeton DJ Hannink M Keap1 is a redox-regulated substrate adaptor protein for a Cul3-dependent ubiquitin ligase complex Mol Cell Biol 2004 24 10941 10953 10.1128/MCB.24.24.10941-10953.2004 15572695

PMC005xxxxxx/PMC5003788.txt

==== Front Protein CellProtein CellProtein & Cell1674-800X1674-8018Higher Education Press Beijing 29810.1007/s13238-016-0298-xLetterRab1A mediates proinsulin to insulin conversion in β-cells by maintaining Golgi stability through interactions with golgin-84 Liu Xiaojing 12Wang Zhenguo 1Yang Ying 1Li Qingrun 1Zeng Rong 1Kang Jiuhong 2Wu Jiarui wujr@sibs.ac.cn 1341 Key Laboratory of Systems Biology, Institute of Biochemistry and Cell Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai 200031, China 2 Clinical and Translational Research Center of Shanghai First Maternity and Infant Health Hospital, Shanghai Key Laboratory of Signaling and Disease Research, School of Life Science and Technology, Tongji University, Shanghai, 200092 China 3 School of Life Science and Technology, ShanghaiTech University, Shanghai, 200031 China 4 Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai, 201210 China 9 8 2016 9 8 2016 9 2016 7 9 692 696 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.issue-copyright-statement© HEP and Springer 2016 ==== Body Dear Editor, Insulin plays a critical role in mammalian glucose homeostasis, and dysfunctional insulin secretion results in diabetes. Insulin maturation in pancreatic β cells can be generally divided into three stages: the first stage is the biosynthesis of proinsulin in the endoplasmic reticulum (ER), the second stage is the transport of proinsulin from the ER into the Golgi, and the third stage is the cleavage of proinsulin into insulin and C peptide in immature insulin granules (Dodson and Steiner, 1998). Previous studies reported that Rab family, which belongs to the small GTPase family, is involved in the regulation of insulin maturation. For example, Rab3, Rab11, Rab27, and Rab37 were associated with insulin-containing secretory granules and regulation of their exocytosis (Yi et al., 2002; Sugawara et al., 2009; Ljubicic et al., 2013; Cazares et al., 2014). A recent report showed that Rab2A can promote either insulin secretion or ER-associated degradation of proinsulin (Sugawara et al., 2014). By re-analyzing a dataset of microarray data derived from islets of type 2 diabetes patients (Dominguez et al., 2011), we found that rab1a mRNA expression was significantly decreased (P < 0.001, Fig. S1A). To confirm this observation, we compared the Rab1A protein levels in islets isolated from Wistar rats with those from diabetic Goto-Kakizaki (GK) rats that have spontaneous type 2 diabetes mellitus. Western blotting analysis showed that Rab1A expression was significantly reduced in GK islets (Fig. S1B). Taken together, Rab1A expression is down-regulated in diabetic islets, implying that Rab1A has an important function in pancreatic β-cells. We examined the sub-cellular localization of exogenous Rab1A in rat insulinoma INS-1E cells by using confocal microscopy. It has been reported that the Rab1A is mainly localized at the endoplasmic reticulum (ER)-Golgi membranes (Allan et al., 2000; Moyer et al., 2001). In agreement with previous studies, we found that Rab1A is mainly located in the cis-Golgi and endoplasmic reticulum (ER)-Golgi intermediate compartment (ER-GIC) in INS-1E cells (Fig. S2). To investigate the function of rab1a gene, we used the CRISPR/Cas9 approach to delete the rab1a gene in INS-1E cells. Two cell clones (KO-2 and KO-3) were identified that contained additional T nucleotides in the rab1a gene, which led to a frameshift mutation (Fig. S3A). No detectable Rab1A was found in KO-2 or KO-3 cells by Western blotting (Fig. S3B) and RT-PCR (Fig. S3C). We examined glucose-stimulated insulin secretion (GSIS) in INS-1E cells with or without Rab1A function. The results showed that insulin secretion in KO-2 and KO-3 cells was significantly decreased compared with that in the control cells (Fig. 1A). Since the RT-PCR results showed no significant changes in ins mRNA expression levels in KO-2 cells (Fig. 1B), we propose that Rab1A is not involved in regulating ins transcription. Western-blotting results showed that the proinsulin content of KO-2 cells was not significantly different from that of the control cells, but the matured insulin in KO-2 cells was largely decreased compared with that of the control cells (Fig. 1C). In addition, using rat insulin and proinsulin ELISA kits, we showed that rab1a knockout resulted in a significantly decreased insulin content (Fig. 1D, right panel), whereas no significant proinsulin content changes were detected (Fig. 1D, left panel).Figure 1 Insulin content is decreased in rab1a knockout INS-1E cells. (A) Insulin secretion was detected in INS-1E cells (Control) and rab1a knockout cells (KO-2 and KO-3). Insulin secretion levels were measured using a rat insulin ELISA kit. (B) Measurements of ins and rab1a mRNA expression levels in INS-1E cells (Control), rab1a knockout cells (KO-2 and KO + GFP) and exogenous rab1a expression cells (KO + Rab1A) by RT-PCR. (C) Measurements of proinsulin and insulin contents in the cells as described in Fig. 1B by Western blotting. (D) Measurements of proinsulin and insulin contents in the cells as described in Fig. 1B using the rat proinsulin ELISA kit (left panel) and insulin ELISA kit (right panel), respectively. The results of (A) and (C and D) are presented as the mean ± S.E.M. (n = 3). *, P < 0.05; **, P < 0.01; ***, P < 0.001 We next expressed wild type Rab1A and its mutants Q70L (a GTP-restricted mutant, activated form) and S25N (a GDP-restricted mutant, inactivated form) by lentivirus infection to see whether the activated form of Rab1A is essential to INS-1E cells’ function. We observed that overexpression of Rab1A WT or Rab1A Q70L resulted in no apparent increase of basal insulin secretion (BSIS), GSIS and insulin content. This lack of effect was not surprising, given that endogenous activated form of Rab1A may be functionally enough. But overexpression of Rab1A S25N significantly reduced BSIS, GSIS and insulin content (Fig. S4), suggesting that the activated form of Rab1A is essential to insulin secretion and insulin content. To further confirm that Rab1A regulates proinsulin to insulin conversion, we established KO + Rab1A cells by infecting KO-2 cells with a Rab1A-lentivirus (Fig. 1B and 1C). The exogenous expression of Rab1A in KO + Rab1A cells recovered the insulin content to levels that were similar to those observed in the control INS-1E cells (Fig. 1C and right panel of 1D), whereas no significant increase in proinsulin content was detected in these KO + Rab1A cells (Fig. 1C and left panel of 1D). These results indicate that exogenous Rab1A expression could rescue decreased insulin content in Rab1A-knockout cells. Taken together, we concluded that Rab1A plays an important role in the conversion of proinsulin to insulin. Rab1 has two isoforms, Rab1A and Rab1B, which share 92% amino acid identity (Touchot et al., 1989). The present results showed that the rab1a gene was successfully knocked out in KO-2 and KO-3 cells without influencing Rab1B protein expression (Fig. S3B). To detect whether Rab1B plays the similar role as Rab1A, we transfected small interfering RNA (siRNA) to knockdown rab1b in INS-1E cells (Fig. S5A). The results showed that insulin content in the rab1b knockdown cells was the same as that in the control cells (Fig. S5B), indicating that Rab1B is not involved in proinsulin to insulin conversion. Using electron microscopy, we evaluated the ultrastructure of Golgi in the cells. The results showed that Golgi ribbon sizes were significantly reduced in rab1a knockout cells (KO-2, KO + GFP, Fig. 2A and B). Further analysis of Golgi cisternae indicated that large Golgi cisternae (>0.05 μm2) numbers were significantly decreased in rab1a knockout cells (Fig. 2C), whereas small Golgi cisternae (0.002–0.01 μm2) numbers were significantly increased (Fig. 2C). Importantly, exogenous Rab1A expression in rab1a knockout cells (KO + Rab1A) restored the Golgi ribbon structures to resemble those of the control cells (Fig. 2A–C), suggesting that Rab1A is required for Golgi stability.Figure 2 Rab1a knockout results in Golgi ribbon fragmentation. (A) Electron micrograph of the cells as described in Fig. 1B (G, Golgi; M, mitochondria; N, nucleus). All images are presented at 21,000× magnification. Scale bars, 0.5 μm. (B and C) Golgi (B) and Golgi cisternae (C) areas in the electron micrographic images were measured using Image Pro Plus (numbers of measured cells: Control, n = 21; KO-2, n = 20; KO + GFP, n = 14; KO + Rab1A, n = 14). (D) Immunoprecipitation of Rab1A interaction proteins by anti-HA antibody and analyzed with anti-golgin-84 antibody by Western blotting. IP: immunoprecipitation. (E) Insulin content in golgin-84 shRNAs infected INS-1E cells was determined. The results are presented as the mean ± S.E.M. (n = 3). *, P < 0.05; **, P < 0.01; *** P < 0.001 Since a recent report showed that Rab2A is involved in the regulation of ER stress within insulin-secreting cells (Sugawara et al., 2014), we wondered whether rab1a knockout would increase ER stress in INS-1E cells. Rough ER has normal flattened cisternae structure that is packed with polyribosomes. Dramatic ER distension is usually an indicative of severe ER stress (Wikstrom et al., 2013). We used electron microscopy to evaluate ultrastructure of ER in the cells. No obvious difference of ER morphology was observed between Ctr and rab1a knockout cells (Fig. S6A, compare KO-2 and KO + GFP with Ctr). The rough ER in the rab1a knockout cells have normal flattened cisternae structure, which are packed with polyribosomes. We also detected several ER-stress marker protein expression levels by Western bloting. Western blotting results showed that several ER-stress marker protein expression levels in rab1a knockout cells were similar to those of the control cells (Fig. S6B), suggesting that rab1a knockout does not induce ER stress in insulin-secreting cells. Because we did not observe proinsulin accumulation in rab1a knockout cells (Fig. 1C, left panel of 1D), we proposed that there may be a negative feedback loop regulating proinsulin production in insulin-secreting cells that may prevent proinsulin accumulation when the proinsulin conversion to insulin is impeded to avoid ER stress. Since a number of studies showed that Rab GTPases carry out their functions through the recruitment of various effector proteins (Allan et al., 2000; Sugawara et al., 2009; Ljubicic et al., 2013; Sugawara et al., 2014), we used mass spectrometry to identify Rab1A effectors that can maintain Golgi stability. Using rab1a knockout cells, we constructed cell lines that overexpressed HA-tagged wild-type (WT) or constitutively active (Q70L) Rab1A (Short et al., 2001). Then, both HA-tagged Rab1A WT and Q70L were immunoprecipitated and then subjected to mass spectrometry. We identified a total of 447 proteins, among which 18 were detected in both Rab1A-WT and -Q70L samples but not in Rab1A knockout samples (Table S2), suggesting that these 18 proteins were likely to interact with Rab1A. Previous studies showed that a Golgi membrane protein golgin-84, which is shown in Table S2, can interact with Rab1 to maintain Golgi structure (Diao et al., 2003; Satoh et al., 2003). Therefore, we speculated that golgin-84 interacts with Rab1A in INS-1E cells. To confirm this observation, we immunoprecipitated golgin-84 with anti-HA antibody. The result showed that golgin-84 was detected in both Rab1A-WT and -Q70L samples, but not in Rab1A knockout samples (Fig. 2D). Furthermore, the expression of golgin-84 in the Rab1A-Q70L sample was higher than that in the Rab1A-WT sample (Fig. 2D), implying that the constitutively active Rab1A-Q70L cells may generate stronger golgin-84-binding activity. We also found that exogenously expressed EGFP-Rab1A colocalized with golgin-84 in INS-1E cells (Fig. S7A). Taken together, Rab1A interacts with golgin-84 in INS-1E cells. Approximately 20 of the 70 known Rab proteins are associated with the Golgi apparatus (Liu and Storrie, 2015). In a recent review paper, Liu and Storrie proposed that there are two classes of Golgi-associated Rab proteins: in Class 1, Rab inactivation leads to Golgi ribbon disruption; in Class 2, Rab inactivation has little to no obvious effect on Golgi organization (Liu and Storrie, 2015). However, both Rab1A and Rab1B are categorized as Class 1 Rab proteins (Table 2 in reference (Liu and Storrie, 2015)). The present result that rab1a depletion disrupts Golgi ribbon organization (Fig. 2A–C) supports this hypothesis. Identification of golgin-84 as a Rab1A-associating protein (Fig. 2D) in our present study further suggests that this type of Golgi ribbon fragmentation may result from the loss of Rab1A and golgin-84 interactions. Based on the observation that rab1a knockout resulted in inhibition of the proinsulin to insulin conversion (Fig. 1C and 1D), we proposed that down-regulation of golgin-84 expression may also impede proinsulin to insulin conversion. Therefore, we knocked down golgin-84 expression using shRNAs in INS-1E cells (Fig. S7B and S7C) and observed that the insulin content decreased as we expected (Fig. 2E). But interestingly, knockdown of golgin-84 in INS-1E cells also resulted in decreased proinsulin content (Fig. S7D). This may because that knockdown of golgin-84 not only disrupts Rab1A and golgin-84 interaction but also influences other functions of golgin-84, which may need further studies. Furthermore, because golgin-84 depletion also induced Golgi ribbon fragmentation (Diao et al., 2003; Satoh et al., 2003), we conclude that the functional interaction between Rab1A and golgi-84 maintaining Golgi stability is critical to proinsulin to insulin conversion. In conclusion, our findings indicate that Rab1A interacts with golgin-84 to maintain the Golgi ribbon structure and required for converting proinsulin to insulin within insulin-secreting cells. The disruption of the interaction between Rab1A and golgin-84 results in inhibition of the proinsulin to insulin conversion due to Golgi ribbon fragmentation. Footnotes This work was supported by grants to Wu JR from the National Natural Science Foundation of China (Grant Nos. 31130034 and 31470808), and Strategic Priority Research Program of the Chinese Academy of Sciences (XDA12000000). Xiaojing Liu, Zhenguo Wang, Ying Yang, Qingrun Li, Rong Zeng, Jiuhong Kang, and Jiarui Wu declare that they have no conflict of interest. All institutional and national guidelines for the care and use of laboratory animals were followed. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary material 1 (PDF 8465 kb) ==== Refs References Allan BB Moyer BD Balch WE Rab1 recruitment of p115 into a cis-SNARE complex: programming budding COPII vesicles for fusion Science 2000 289 444 448 10.1126/science.289.5478.444 10903204 Cazares VA Subramani A Saldate JJ Hoerauf W Stuenkel EL Distinct actions of Rab3 and Rab27 GTPases on late stages of exocytosis of insulin Traffic 2014 15 997 1015 10.1111/tra.12182 24909540 Diao A Rahman D Pappin DJ Lucocq J Lowe M The coiled-coil membrane protein golgin-84 is a novel rab effector required for Golgi ribbon formation J Cell Biol 2003 160 201 212 10.1083/jcb.200207045 12538640 Dodson G Steiner D The role of assembly in insulin’s biosynthesis Curr Opin Struct Biol 1998 8 189 194 10.1016/S0959-440X(98)80037-7 9631292 Dominguez V Raimondi C Somanath S Bugliani M Loder MK Edling CE Divecha N da Silva-Xavier G Marselli L Persaud SJ Class II phosphoinositide 3-kinase regulates exocytosis of insulin granules in pancreatic beta cells J Biol Chem 2011 286 4216 4225 10.1074/jbc.M110.200295 21127054 Liu S Storrie B How Rab proteins determine Golgi structure Int Rev Cell Mol Biol 2015 315 1 22 10.1016/bs.ircmb.2014.12.002 25708460 Ljubicic S Bezzi P Brajkovic S Nesca V Guay C Ohbayashi N Fukuda M Abderrhamani A Regazzi R The GTPase Rab37 Participates in the Control of Insulin Exocytosis PLoS ONE 2013 8 e68255 10.1371/journal.pone.0068255 23826383 Moyer BD Allan BB Balch WE Rab1 interaction with a GM130 effector complex regulates COPII vesicle cis-Golgi tethering Traffic 2001 2 268 276 10.1034/j.1600-0854.2001.1o007.x 11285137 Satoh A Wang Y Malsam J Beard MB Warren G Golgin-84 is a rab1 binding partner involved in Golgi structure Traffic 2003 4 153 161 10.1034/j.1600-0854.2003.00103.x 12656988 Short B Preisinger C Korner R Kopajtich R Byron O Barr FA A GRASP55-rab2 effector complex linking Golgi structure to membrane traffic J Cell Biol 2001 155 877 883 10.1083/jcb.200108079 11739401 Sugawara K Shibasaki T Mizoguchi A Saito T Seino S Rab11 and its effector Rip11 participate in regulation of insulin granule exocytosis Genes Cells 2009 14 445 456 10.1111/j.1365-2443.2009.01285.x 19335615 Sugawara T Kano F Murata M Rab2A is a pivotal switch protein that promotes either secretion or ER-associated degradation of (pro)insulin in insulin-secreting cells Sci Rep 2014 4 6952 10.1038/srep06952 25377857 Touchot N Zahraoui A Vielh E Tavitian A Biochemical properties of the YPT-related rab1B protein. Comparison with rab1A FEBS Lett 1989 256 79 84 10.1016/0014-5793(89)81722-3 2509243 Wikstrom JD Israeli T Bachar-Wikstrom E Swisa A Ariav Y Waiss M Kaganovich D Dor Y Cerasi E Leibowitz G AMPK regulates ER morphology and function in stressed pancreatic beta-cells via phosphorylation of DRP1 Mol Endocrinol 2013 27 1706 1723 10.1210/me.2013-1109 23979843 Yi Z Yokota H Torii S Aoki T Hosaka M Zhao S Takata K Takeuchi T Izumi T The Rab27a/granuphilin complex regulates the exocytosis of insulin-containing dense-core granules Mol Cell Biol 2002 22 1858 1867 10.1128/MCB.22.6.1858-1867.2002 11865063

PMC005xxxxxx/PMC5003789.txt

==== Front Protein CellProtein CellProtein & Cell1674-800X1674-8018Higher Education Press Beijing 29910.1007/s13238-016-0299-9LetterAn episomal CRISPR/Cas9 system to derive vector-free gene modified mammalian cells Li Linlin Gao Fei Wu Sen swu@cau.edu.cn State Key Laboratory of Agrobiotechnology, College of Biological Sciences, China Agricultural University, Beijing, 100193 China 29 7 2016 29 7 2016 9 2016 7 9 689 691 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.issue-copyright-statement© HEP and Springer 2016 ==== Body Dear Editor, CRISPR and CRISPR-associated (Cas) proteins play their adaptive immunity role in degrading foreign nucleic acids in both bacteria and archaea. CRISPR/Cas has proved efficient in modifying mammalian genomes (Cong et al., 2013; Mali et al., 2013) and various delivery methods for Cas9:gRNA complex have been established. Among different delivery means, in vitro transcribed Cas9 mRNA/gRNA, purified Cas9 protein, and adeno-associated virus (AAV) driven seem promising toward clinical applications (Kouranova et al., 2016; Lin et al., 2014; Yin et al., 2016). Still, all the fore-mentioned methods for delivering CRISPR/Cas9 have certain shortcomings. In vitro transcription and protein purification are complicated and inconvenient for use, while AAV mediated system suffers from the side effects of immune response and insertional mutagenesis (Donsante et al., 2007; Li et al., 2011). Episomal vectors based on oriP-EBNA1 have been shown as an effective method to derive vector-free cells in reprogramming studies (Okita et al., 2011; Yu et al., 2009). Without selection pressure, EBV vectors are gradually lost in each generation due to their defects in synthesizing and partitioning thus resulting in transgene-free daughter cells (Nanbo et al., 2007). We thus set out to test if Cas9 protein and gRNA can be effectively delivered as episomes to generate vector-free mutations in mammalian cells. Here we report efficient human genome editing with CRISPR/Cas9 technology in the form of oriP-EBNA1-based episomes. To obtain oriP-EBNA1-based CRISPR/Cas9 vectors, we first assembled CMV-FLAG-hSpCas9-2A-GFP and U6-gRNA and oriP-EBNA1 fragments together. To further enrich successfully transfected cells, we added SV40-puro to form pCRISPR-S12 (Fig. 1), which hereafter is referred to as oriP-EBNA1-based CRISPR/Cas9 vector or system.Figure 1 Circular map of oriP-EBNA1-based CRISPR/Cas9 vector. CMV, cytomegalovirus promoter; 3× FLAG, FLAG epitope tag; hCas9, humanized Streptococcus pyogenes Cas9; F2A, 2A regions of foot-and-mouth disease virus; GFP, green fluorescent protein gene; BGH pA, bovine growth hormone polyA signal; U6, human U6 promoter; chimeric gRNA, synthetic gRNA scaffold; oriP, Epstein-Bar virus (EBV) latent origin of replication; EBNA1, EBV nuclear antigen-1, replication transactivator of EBV; SV40, SV40 promoter; Puro, puromycin resistant gene; Amp, ampicillin resistant gene; pUC ori, replication origin; TK promoter, thymidylate kinase promoter; Hygro, hygromycin resistant gene; pA, polyA signal. To test the oriP-EBNA1-based CRISPR/Cas9 system, we first examined its efficiency for disrupting single-copy tdTomato expression driven by CAG promoter in mouse induced pluripotent stem cells (miPSCs). We designed crtdTomato to target tdTomato coding sequence with the expectation of destructing an existing Sau3AI restriction enzyme site (Fig. 2A). The tdTomato-tagged-miPSCs were transfected with pCRISPR-S12-crtdTomato and manipulated according to the flowchart (Fig. 2B). We observed disappearance of tdToamto fluorescence in some of the miPSCs clones under a fluorescent microscope (Fig. 2C). Further, flow-activated cell sorting (FACS) showed a reduction of 26% tdTomato-labeled cells in crtdTomato treated group when compared with non-transfected group five days post transfection (Fig. 2D). These results suggested that our system works in inactivating reporter genes in miPSCs.Figure 2 oriP-EBNA1-based CRISPR/Cas9 disrupted fluorescent expression in miPSCs. (A) Schematic of crtdTomato targeting site (blue arrow), primers used (purple arrows) for amplifying genomic sequences flanking target sites (light gray boxes), CAG promoter (green box) drove the expression of tdTomato (red box), and Sau3AI restriction sites were used for RFLP assay. (B) Schematic of the experimental procedure. (C) Disappearance of red fluorescence in partial tdTomato-labelled-miPSCs clones. Scale bar, 200 μm. (D) FACS analysis of crtdTomato transfected and non-transfected miPSCs. The upper two panels are crtdTomato transfected group (tdTomato + with a portion of 67%) and the lower two panels are non-transfected group (tdTomato+ with a portion of 93%). (E) RFLP of non-red (NR) clones by Sau3AI. The green pentagrams indicated the uncut PCR amplicons by Sau3AI. Ladder, 1 kb plus ladder in all figures in this study. (F) Sanger sequencing confirmed deletion of 341 bp of NR1-d20 clone. The purple line between G/G showed the deleted site. (G) PCR confirmed the removal of EBNA1 fragment in clone NR1-d20, NR2-d20, NR5-d20 but not in NR3-d20. pCRISPR-S12 DNA was used as positive control (“+” in all figures in this study). “wt” stands for wild type mouse genomic DNA. Red pentagram shows visible EBNA1 residue in NR3-d20. (H) Western blot showed no hSpCas9 expression in clone NR1-d20, NR2- d20, NR3-d20, and NR5- d20. HeLa cells expressing FLAG-hSpCas9-2A-GFP stably was used as positive control (“+” in all figures in this study). The expected protein band is about 190 kDa. Next, we further detected the targeting efficiency of our system in 6 non-red (NR) single clones. We were able to PCR amplify the target regions in 4 out of the 6 NR clones. Clones NR4 and NR6 were not analyzed further since PCR failed to produce products (Fig. S1). PCR products of the remaining 4 clones were confirmed as expected mutants by restriction fragment length polymorphism (RFLP) assays (Fig. 2E and 2F). Among these, clone NR1 showed a truncated size (Fig. S1), which was further confirmed by Sanger sequencing as a 341 nucleotides deletion (Fig. 2F). Above data showed that our oriP-EBNA1-based CRISPR/Cas9 system could mediate efficient gene disruption. Finally, we examined whether these four successfully engineered miPSCs (NR1, NR2, NR3, NR5) were free of targeting vector. PCR analysis showed that 3 clones were free of vectors, but clone NR2-d20 still retained residual amount of vectors (Fig. 2G). However, Western blot of FLAG-hSpCas9-2A-GFP further verified no continuous expression of hSpCas9 protein of all four clones (Fig. 2H), suggesting that PCR detection was more sensitive in detecting foreign gene expression than Western blot assay. Together, these results demonstrated that our oriP-EBNA1-based CRISPR/Cas9 system inactivated reporter gene expression as expected, and more importantly, vector-free engineered cell lines can be readily obtained. We further attempted to examine whether our system works well in human cells. The basic functions of CRISPR/Cas9 were confirmed by deleting accurately more than 2 kb intervening nucleotides of human chemokine receptor 5 (CCR5) as well as editing multiplex genes (human TET1, TET2, and TET3) simultaneously (Figs. S3–S6 and Table S1). Moreover, our oriP-EBNA1-based CRISPR/Cas9 system was free of vector in both genetically modified miPSCs and human cells. Together, these results demonstrated the feasibility of manipulating mouse and human genomes with our oriP-EBNA1-based CRISPR/Cas9 system. In the current study, we have verified the functionality of our episomal CRIPSR/Cas9 system first by inactivating visible reporter gene, and further by specifically deleting a single genomic region and editing multiplex genes in a transgene-free manner. One potential risk for CRISPR/Cas9 in genome editing is its uncertain off-target effects which could confound genome-editing-based therapies. In present study, we detected certain frequencies (none for crTETs and 10% for crCCR5) of off-target events by T7EI assay (Fig. S7 and Table S2). In the future, combined with other progresses, such as mutant Cas9 or new homologues and optimized gRNA design, episomal CRISPR/Cas9 could further ameliorate, if not eliminate, the off-target influences, thereby broadening its application. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary material 1 (PDF 10699 kb) Linlin Li and Fei Gao contributed equally to this article. FOOTNOTES Linlin Li conceived and designed the experiments. Linlin Li and Fei Gao performed the experiments. Linlin Li, Fei Gao, and Sen Wu wrote the paper. We are grateful for continuous support from the Wu lab members. This work was supported by the National Natural Science Foundation of China (Grant No. 31271598) and The Project for Extramural Scientists of State Key Laboratory of Agrobiotechnology (2015SKLAB6-15). Linlin Li, Fei Gao, and Sen Wu declare no conflict of interest. This article does not contain any studies with human or animal subjects performed by the any of the authors. ==== Refs References Cong L Ran FA Cox D Lin S Barretto R Habib N Hsu PD Wu X Jiang W Marraffini LA Multiplex genome engineering using CRISPR/Cas systems Science 2013 339 819 823 10.1126/science.1231143 23287718 Donsante A Miller DG Li Y Vogler C Brunt EM Russell DW Sands MS AAV vector integration sites in mouse hepatocellular carcinoma Science 2007 317 477 10.1126/science.1142658 17656716 Kouranova E Forbes K Zhao G Warren J Bartels A Wu Y Cui X CRISPRs for optimal targeting: delivery of CRISPR components as DNA, RNA and protein into cultured cells and single-cell embryos Hum Gene Ther 2016 27 464 475 10.1089/hum.2016.009 27094534 Li H Malani N Hamilton SR Schlachterman A Bussadori G Edmonson SE Shah R Arruda VR Mingozzi F Wright JF Assessing the potential for AAV vector genotoxicity in a murine model Blood 2011 117 3311 3319 10.1182/blood-2010-08-302729 21106988 Lin S Staahl BT Alla RK Doudna JA Enhanced homology-directed human genome engineering by controlled timing of CRISPR/Cas9 delivery eLife 2014 3 e04766 25497837 Mali P Yang L Esvelt KM Aach J Guell M DiCarlo JE Norville JE Church GM RNA-guided human genome engineering via Cas9 Science 2013 339 823 826 10.1126/science.1232033 23287722 Nanbo A Sugden A Sugden B The coupling of synthesis and partitioning of EBV's plasmid replicon is revealed in live cells EMBO J 2007 26 4252 4262 10.1038/sj.emboj.7601853 17853891 Okita K Matsumura Y Sato Y Okada A Morizane A Okamoto S Hong H Nakagawa M Tanabe K Tezuka K A more efficient method to generate integration-free human iPS cells Nat Methods 2011 8 U409 U452 10.1038/nmeth.1591 Yin H Song CQ Dorkin JR Zhu LJ Li Y Wu Q Park A Yang J Suresh S Bizhanova A Therapeutic genome editing by combined viral and non-viral delivery of CRISPR system components in vivo Nat Biotechnol 2016 34 328 333 10.1038/nbt.3471 26829318 Yu JY Hu KJ Smuga-Otto K Tian SL Stewart R Slukvin II Thomson JA Human induced pluripotent stem cells free of vector and transgene sequences Science 2009 324 797 801 10.1126/science.1172482 19325077

PMC005xxxxxx/PMC5003790.txt

==== Front Protein CellProtein CellProtein & Cell1674-800X1674-8018Higher Education Press Beijing 30010.1007/s13238-016-0300-7Research ArticleElimination of the geomagnetic field stimulates the proliferation of mouse neural progenitor and stem cells http://orcid.org/0000-0002-1616-6916Fu Jing-Peng 13Mo Wei-Chuan 12http://orcid.org/0000-0003-0802-3832Liu Ying yingliu@ibp.ac.cn 13Bartlett Perry F. 2He Rong-Qiao herq@sun5.ibp.ac.cn 1341 State Key Laboratory of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences, Beijing, 100101 China 2 Queensland Brain Institute, University of Queensland, Brisbane, QLD 4072 Australia 3 University of the Chinese Academy of Sciences, Beijing, 100049 China 4 Alzheimer’s Disease Center, Beijing Institute for Brain Disorders, Capital Medical University, Beijing, 100069 China 3 8 2016 3 8 2016 9 2016 7 9 624 637 4 6 2016 7 7 2016 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.Living organisms are exposed to the geomagnetic field (GMF) throughout their lifespan. Elimination of the GMF, resulting in a hypogeomagnetic field (HMF), leads to central nervous system dysfunction and abnormal development in animals. However, the cellular mechanisms underlying these effects have not been identified so far. Here, we show that exposure to an HMF (<200 nT), produced by a magnetic field shielding chamber, promotes the proliferation of neural progenitor/stem cells (NPCs/NSCs) from C57BL/6 mice. Following seven-day HMF-exposure, the primary neurospheres (NSs) were significantly larger in size, and twice more NPCs/NSCs were harvested from neonatal NSs, when compared to the GMF controls. The self-renewal capacity and multipotency of the NSs were maintained, as HMF-exposed NSs were positive for NSC markers (Nestin and Sox2), and could differentiate into neurons and astrocyte/glial cells and be passaged continuously. In addition, adult mice exposed to the HMF for one month were observed to have a greater number of proliferative cells in the subventricular zone. These findings indicate that continuous HMF-exposure increases the proliferation of NPCs/NSCs, in vitro and in vivo. HMF-disturbed NPCs/NSCs production probably affects brain development and function, which provides a novel clue for elucidating the cellular mechanisms of the bio-HMF response. Electronic supplementary material The online version of this article (doi:10.1007/s13238-016-0300-7) contains supplementary material, which is available to authorized users. Keywords hypomagnetic fieldneural progenitor/stem cellsneurosphereproliferationstemnessmultipotencyissue-copyright-statement© HEP and Springer 2016 ==== Body Introduction Living organisms are exposed to the geomagnetic field (GMF, 35–70 µT) throughout their lifespan. The GMF is well-known for providing navigation information for migrating animals (Gould and Gould 2012; Nathan et al. 2014) or locomotion direction for magnetotactic bacteria (Jogler and Schuler 2009). The effects of disturbed environmental magnetic field have been concerned for long, and that of the hypomagnetic field (HMF, <5 µT), one of the key environmental risk factors for astronauts travelling in outer space, have been considered seriously with the need of manned mission to explore the deep space (Mo et al. 2012a). Recently, wide attention has been drawn on biological roles of the GMF when some animals, even human beings, who were thought to have no magnetic sense, were found to have a potential response to environmental magnetic fields (Lohmann 2010; Gegear et al. 2012). It has been established that the elimination of the GMF has adverse effects on living systems (Mo et al. 2012a). Animals continuously exposed to the HMF condition by shielding or compensatively eliminating the GMF exhibited dysfunction of central nervous system (CNS), with symptoms such as disturbed vocal behavior and circadian activity rhythm in birds (Bliss and Heppner 1976; Jiang et al. 1998); amnesia in chicken (Wang et al. 2002) and Drosophila (Zhang et al. 2004); and decreased general activity, altered circadian drinking rhythm and analgesia in mice (Prato et al. 2005; Mo et al. 2015). The HMF also markedly disturbs development processes, evidenced by delayed embryonic and nymphal development in planthoppers (Wan et al. 2014), increased embryo malformation in newt (Asashima and Shimada 1991) and Xenopus (Mo et al. 2012b), as well as inhibited early embryogenesis in mice (Fesenko et al. 2010). In particular, abnormalities of the head and spine were marked observed under the HMF condition (Mo et al. 2012b). These data reveal that the GMF is involved in the regulation of the brain functions and development. However, the roles of the GMF on animals are still far from clear, and the cellular mechanisms underlying these effects have not been clearly identified so far . Neural progenitor/stem cells (NPCs/NSCs) play critical roles in CNS development and maintainence of brain function (Lui et al. 2011; Gage and Temple 2013). Inhibited self-renewal or differentiation of NPCs/NSCs cause either behavioral disorders (Yau et al. 2014; Cameron and Glover 2015) or abnormal development (Merkle and Alvarez-Buylla 2006). It has been reported that exposure to applied magnetic fields affect the growth and fate of NPCs/NSCs (Di Lazzaro et al. 2013). Applied electromagnetic fields (EMFs) inhibit proliferation and promote differentiation of mouse bone marrow mesenchymal stem cells (Wu et al. 2005) and embryonic stem cells (Ventura et al. 2005), and stimulate the maturation and differentiation of cerebellar granule neurons in newborn rat (Lisi et al. 2005). Nakamichi and colleagues showed that exposure to a 100 mT static magnetic field (SMF) reduced proliferation of NPCs in the fetal rat brain (Nakamichi et al. 2009). Thus, disturbance of the GMF could modulate proliferation and differentiation of NPCs/NSCs. Recently, we found that proliferation of human neuroblastoma cells (SH-SY5Y) was accelerated in an HMF (<200 nT) by deep-shielding the GMF (Mo et al. 2013). Therefore, the GMF condition might also be necessary to maintain the homeostasis of NPCs/NSCs, and the HMF may serve as a physical stimulator to the proliferation of NPCs/NSCs. Investigating the HMF effect on the NSC/NPCs would provide useful clues for the cellular mechanism of biomagnetic interactions. In the present study, we evaluated the growth and differentiation of NPCs/NSCs under the HMF condition (<200 nT). Primary neurospheres (NSs) from the brains of neonatal, young, and adult mice were exposed to either the GMF or the HMF. Our results showed that the growth of the NSs was greatly accelerated in the HMF, and their capacity for self-renewal and multipotency was maintained. Moreover, the number of proliferative cells in the subventricular zone (SVZ) increased in the HMF-exposed adult mouse. Our findings suggest that the NPCs/NSCs can respond to the HMF, and that these bio-magnetic responses could contribute at a cellular level to the GMF’s necessary role in maintaining homeostasis of the CNS. Results Exposure to the HMF accelerates the growth of primary NSs from the neonatal mouse Primary cell suspensions of postnatal day 2 (P2) mouse brains were incubated in either the HMF or GMF environment for 7 days. The morphologies of the NSs exposed to the HMF were similar to those exposed to the GMF; however, the HMF-exposed NSs grew faster and larger (Figs. 1A, 1B and S1). After the final day of exposure to the magnetic fields (day 7), the diameter of each NS was measured, and those incubated in the HMF were found to be significantly larger than those incubated in the GMF (P < 0.0001, χ2 test; Fig. 1C). Significantly fewer NSs with diameters <100 μm were observed in the HMF (P = 0.014, 19.9% ± 1.7%, <50 μm; P = 0.0016, 41.9% ± 1.2%, 50–99 μm), when compared to the GMF controls (26.8% ± 2.7%, <50 μm; 54.1% ± 1.7%, 50–99 μm). A greater number of NSs exposed to the HMF had diameters between 100 and 200 μm (P < 0.0001, 26.6% ± 1.3%, 100–149 μm; P < 0.0001, 10.0% ± 1.1%, 150–199 μm) and also ≥200 μm (P = 0.0009, 1.51% ± 0.25%), when compared to the GMF controls (17.5% ± 1.2%, 100–149 μm; 15.2% ± 0.4%, 150–199 μm; 0.025% ± 0.025%, ≥200 μm). As magnetic intensity of the control GMF condition (~15 μT) was lower than the local GMF (~50 μT), we used a reference GMF (R-GMF: 56.6 ± 4.4 μT) in another incubator, as described previously (Mo et al. 2013). The NSs showed no difference in size between the GMF and R-GMF groups at day 6 (P = 0.566), while the HMF-exposed NSs were significantly larger than both the GMF and R-GMF groups (P < 0.0001, Fig. 1D). These results indicate that the HMF exposure accelerates growth of primary NSs, and the effect is attributed to elimination instead of partial shielding of the GMF.Figure 1 Exposure to the HMF accelerates the growth of primary NSs from neonatal mouse. Primary cells from whole brains of P2 mice were seeded in either 60 mm dishes (8.0 × 105 cells/dish for cell counting) or 96-well plates (1000 cells/well for NS counting and size analysis) and exposed to either the GMF or HMF. (A and B) Representative pictures of NSs at day 7. Those grown in the HMF appeared significantly larger. (C) Size distribution of day 7 NSs. A greater number of large NSs were counted in cultures exposed to the HMF. (D) Size distribution of day 6 NSs cultured in the GMF, R-GMF, and HMF conditions. Sizes of NSs in the GMF and R-GMF groups were similar, but smaller than those in the HMF group. (E) Total cell numbers of the day 7 NS cultures were significantly greater in the HMF group, compared to the control GMF group. (F) Cells exposed to the HMF underwent more divisions, as shown by the significantly decreased CFSE fluorescence and lower mean fluorescence. Data are shown as mean ± SEM (C and E) or SD (F). n is the number of animals (C–E) and the trials (F) used in the experiments. The P-value was calculated using a χ2 test for NS size distributions in (C) and (D), a one-way ANOVA for mean comparisons in (E). The two-tailed paired student’s t-test in (F) *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; n.s. P ≥ 0.05 The NSs formation assay was used to quantify NPCs/NSCs in suspension culture (Louis et al. 2008; Ahmed 2009). At day 7 of primary culture, the number of NSs produced in the HMF (44.0 ± 4.0 NSs per 1000 cells) were very similar with the GMF controls (43.0 ± 4.0 NSs per 1000 cells, P = 0.82), which suggests that the HMF-exposure did not activate latent cells or repress proliferation of active cells. The total number of cells after 7 days of culture in the HMF (10.62 ± 0.45 × 105 cells) was about thrice the number in the GMF control cultures (3.56 ± 0.45 × 105 cells; P < 0.0001; Fig. 1E), which indicates promoted proliferation of NPCs/NSCs in the HMF condition and is supported by distributions of NSs sizes above. For further evidences, we used CFSE staining assay to measure cell divisions. When cells undertook divisions, the CFSE was distributed to daughter cells equally and CFSE fluorescence decreased in daughter cells. Results showed that the fluorescence values of the NSs exposed to the HMF were significantly lower than the GMF control cells (P = 0.039; Fig. 1F), confirming that a higher amount of cell division occurred when cells were exposed to the HMF. These two assays confirm that the HMF-exposure had accelerated the proliferations of NPCs/NSCs. Continued exposure is required for HMF-promoted growth of NSs To test whether the accelerated growth of NSs requires continued exposure to the HMF, primary NSs exposed to the GMF or HMF for 7 days were passaged in either the GMF or the HMF, giving four experimental groups: GMF to GMF (G-to-G), GMF to HMF (G-to-H), HMF to GMF (H-to-G), and HMF to HMF (H-to-H) (Fig. 2A). To avoid overgrowth of the NSs that could be stimulated by HMF-exposure in the H-to-H group, the NS assay of the four first-passage cultures were conducted at day 6. The results revealed that the HMF cultures (G-to-H, H-to-H) produced a significantly greater total cell number (P < 0.0001), and a greater number of NSs (P < 0.01, NSs per 1000 cells) which were also larger (P < 0.0001) compared to the cultures exposed to the GMF (G-to-G; H-to-G) (Fig. 2B–D). A greater number of NSs per 1000 cells (P < 0.0001) was observed in the H-to-G group (146.40 ± 3.97 NSs per 1000 cells) than the G-to-G group (121.46 ± 3.97 NSs per 1000 cells) (Fig. 2C), confirming that more NSCs/NPCs are produced in the primary culture from the HMF group, which maintain their capacity for self-renewal. The total numbers of cells in the two groups were similar (G-to-G, 6.69 ± 0.71 × 105 cells; H-to-G, 5.32 ± 0.71 × 105 cells; P = 0.185) (Fig. 2B), as was the NS size distribution (P = 0.936) (Fig. 2D). These results confirm that the growth of the primary NSs in the HMF require continued exposure for accelerated growth of NSs.Figure 2 Continued exposure is required for HMF-promoted growth of NSs. (A) A schematic of the GMF recovery experiment. Primary cultures from the HMF/GMF condition were trypsinized and exposed to either the HMF or GMF for the first-passage culture in either 60 mm dishes (1.0 × 105 cells/dish for cell counting) or 96-well plates (1000 cells/well for NS counting and size analysis). (B) Numbers of NSs per 1000 seeded cells at day 6. 1000 cells from the HMF-exposed NSs (H-to-G) formed more NSs than the control cells (G-to-G), but the cells growing in HMF condition (G-to-H, H-to-H) formed more cells than these in the GMF condition (G-to-G, H-to-G). (C) Total cell numbers of first passage in the 6th day. More cells were yielded from the NSs growing under the HMF condition than these under the GMF condition, while the cell numbers were similar in the same condition. (D) NS size distributions at day 6. Compared to the NSs in the GMF, the growth of these NSs in the HMF was enhanced. The distributions of NSs size were similar between two GMF groups (G-to-G, H-to-G), while more large NSs and less small NSs were observed in the H-to-H than these in the G-to-H. (E) When HMF-exposed cells from the primary cultures were continuously passaged under the HMF condition (seeded at 1.0 × 105 cells/60 mm dish), they were also seeded in the GMF as controls. NSs maintained a significantly higher proliferation rates in the HMF condition in four passages. (F) The ratio of successfully passaged large NSs (diameter ≥ 150 μm) in single clone culture assays showed no significant difference between the two conditions. n is the numbers of experimental trials. Data are shown as mean ± SEM in (B–E), or mean ± SD in (F). P values were calculated using a one-way ANOVA in (B–E), and using a χ2 test for NS size distribution in (D), and two-tailed paired student’s t-text in (F). *P < 0.05; **P < 0.01; *** P < 0.001; ****P < 0.0001; n.s. P ≥ 0.05 The HMF-exposed NSs maintain their capacity for self-renewal To evaluate the proliferative capacity of the NPCs/NSCs exposed to the HMF, NSs from primary cultures were continuously passaged in the HMF; a subset of cells was transferred to the GMF condition at each passage to act as controls. The proliferation rate of each passage was recorded on day 6. NSs exposed to the HMF maintained significantly higher proliferation rates (>10 times the seeding cell number) than those returned to the GMF condition (~5 times the seeding cell number) (P1, P = 0.0026; P2, P = 0.0457; P3, P = 0.0024; P4, P = 0.0007) (Fig. 2E). The single clone culture assay during the passaging showed that the percentage of large NSs (≥150 μm in diameter) able to be passaged in the HMF (39.6% ± 8.0%) was similar to that in the GMF (31.3% ± 4.2%, P = 0.182) (Fig. 2F). The continuous passaging experiment showed that the large NSs from HMF group could be successfully passaged for nine rounds under the HMF or the GMF condition (Fig. S2). These results indicate that NPCs/NSCs can maintain their capacity for self-renewal during exposure to the HMF. Nestin and Sox2 were used as markers of NPCs/NSCs (Park et al. 2010; Graham et al. 2003). Using an immunofluorescence assay similar to the GMF group, NSs grown in the HMF were found to be both Nestin- and Sox2-positive, confirming that the NSs had kept their ‘stem cell’ identity (Fig. 3A–C/A’–C’; 3G–I/G’–I’ for the GMF; Fig. 3D–F/D’–F’; 3J–L/J’–L’ for the HMF). Interestingly, using a qPCR assay, Sox2 expression was observed to increase, albeit in a non-significant fashion, in the HMF group to the GMF group (P = 0.065; Fig. 4A). Nestin expression, on the other hand, had decreased significantly (P = 0.029; Fig. 4B). Expression of Neurod1 was significantly decreased in the NSs exposed to the HMF (P = 0.022; Fig. 4C), and no change between these two groups was detected in the expression of the neuronal marker ßIII-tubulin (P = 0.76; Fig. 4D) nor the glial cell marker Gfap (P = 0.85; Fig. 4E). No significant change was shown in the expression of Gapdh, which was used as an internal reference gene (in addition to Tubulin 5α) (P = 0.84; Fig. 4F). After 7 days of exposure to the HMF, the expression of both Cry1 and Cry2, putative magneto-sensing genes (Gegear et al. 2012; Xu et al. 2014), was significantly down-regulated (P = 0.044 for Cry1 gene; P = 0.048 for Cry2 gene; Fig. S3), suggesting that the magneto-sensing molecules in the NPCs/NSCs had responded to the HMF. For all the detected genes, the fold-changes in expression between the HMF and the GMF groups were no greater than two. These results indicate that following exposure to the HMF, the culture products maintained the properties of a true stem cell. In the HMF, reduced differentiation was observed, which is consistent with the observed increase in the production of NPCs/NSCs.Figure 3 The HMF-exposed NSs were positive of nestin and Sox2. Primary cultures of day 7 NSs from P2 mice were collected and immunostained with the neural stem cell markers, nestin (green) and Sox2 (red). Nuclei are stained with Hoechst (blue). Panels show representative large (A–L) and medium-sized (A’ –L’) NSs from the GMF (A–C/A’–C’; G–I/G’–I’) and HMF (D–F/D’–F’; J–L/J’–L’). The NSs of different sizes from both GMF and HMF condition were positive for nestin and Sox2 Figure 4 The HMF-exposed NSs showed altered genes expressions. When checked expressions of the NSCs markers (Nestin; Sox2), differentiated neuronal markers (Neurod1; ßIII-tubulin) and glial marker (Gfap), the expression of both Nestin (B) and Neurod1 (C) were significantly down-regulated following exposure to the HMF. Compared to the GMF groups, the Sox2 had a trend of increase in the HMF group (A).The ßIII-tubulin and Gfap were found no significant changes (D and E). Tubulin 5α was used as the internal reference gene, and Gapdh was used as another internal reference gene (F). Data are shown as mean ± SEM from three independent experiments, six animals per expeiment. P values were calculated by one-way ANOVA for mean comparisons. *P < 0.05; n.s. P ≥ 0.05 The HMF-exposed NSs maintain their multipotency To test whether NSs exposed to the HMF retained their multipotency, these NSs of different sizes from primary cultures of either the GMF or HMF groups were collected and made a differentiation assay in the GMF condition, and an immunofluorescence assay was subsequently used to detect their differentiation states. Both GFAP-positive and ßIII-tubulin-positive cells were observed in the HMF-exposed NSs as well as in those exposed to the GMF (Fig. 5A–D/A’ –D’ for the GMF; Fig. 4E–H/E’–H’ for the HMF), indicating that the NSs retained their ability to differentiate into both neurons and glial cells despite exposure to the HMF.Figure 5 The HMF-exposed primary NSs can differentiate into astrocytes/glia cells and neurons. HMF-exposed and GMF primary NSs from P2 mice (day 7) were collected and induced to differentiate under the GMF condition for 5 days. Representative differentiated large (A–H) and middle (A’–H’) size spheres from the GMF (A–D/A’–D’) and HMF (E–H/E’–H’) group show NS cells from HMF condition could differentiate into both neuron (ßIII-tubulin, green) and glial cells (GFAP, red), compared to those from the GMF group. Nuclei were stained with Hoechst (blue) The HMF counteracts the negative effects of the removal of growth factors on NS growth The growth factors bFGF and EGF are critical in the maintenance of NSs growth as well as the production of NSCs (Ciccolini and Svendsen 1998; Azari et al. 2010). To assess the effect of the HMF on the growth factor sensitivity of NS cells, the concentration of either EGF or bFGF in the culture medium was halved (10 ng/mL EGF or 5 ng/mL bFGF), and an NS assay conducted at day 6. Very few NSs with a diameter of >50 μm were observed in both the HMF and GMF primary cultures when both EGF and bFGF were reduced (Data not shown). As shown in Figure 6, when compared to those cultured in normal media (20 ng/mL EGF and 10 ng/mL bFGF), in the reduced EGF medium, the percentage of NSs with a diameter between 50–100 μm was significantly smaller in the GMF-exposed groups (P = 0.012), and a trend toward a decrease was observed in bFGF-halved culture media (P = 0.624). NSs with diameters ≥100 μm were observed almost exclusively in the HMF; very few were observed in the GMF cultures. However, no significant differences were observed in the HMF-exposed groups in the reduced growth factor medium and in the normal medium, which indicates that the HMF still stimulates NSs growth under shortage of growth factor condition. These results indicate that exposure to the HMF can counteract the negative effects on NS growth induced by shortage of exogenous growth factors.Figure 6 Exposure to the HMF counteracts the negative effects of the removal of growth factors on NS growth. Primary cells were seeded in complete medium, EGF-halved (1/2EGF) or bFGF-halved (1/2bFGF) medium, and cultured in the GMF/HMF condition for 6 days. Compared to cells in the complete medium, NSs significantly decreased in size when the concentration of EGF was halved in the GMF but not in the HMF, and the NS size of the three groups in HMF were significantly larger than those in the GMF. Three trials were completed with 18 repeated wells in each trial. Data are shown as mean ± SEM. P values were calculated using a χ2 test for NS size distribution comparison. *P < 0.05; ****P < 0.0001; n.s. P ≥ 0.05 The HMF promotes the proliferation of NPCs/NSCs from the adult mouse brain In mature mammalian brains, NPCs/NSCs are mainly restricted to the SVZ (Golmohammadi et al. 2008) and the hippocampus (Bull and Bartlett 2005). The growth of primary NSs from the SVZ and hippocampus in young (P15) and adult (2-month-old) mice was compared following exposure to either the HMF or GMF. Primary SVZ NS cultures from both young and adult mice were found to have similar numbers of NSs irrespective of the magnetic field to which they were exposed (Fig. 7A and 7C, left panels). A greater number of large NSs (diameter ≥ 100 μm) was observed following exposure to the HMF in both groups (P15, P < 0.0001; 2-month-old, P < 0.0001) (Fig. 7A and 7C, right panels).Figure 7 The HMF promotes the proliferation of NPCs/NSCs from the adult mouse brain. SVZ and hippocampal tissues from young (P15) (A and B) and adult (2-month-old) mice (C and D) were collected and cultured in either the HMF or GMF. The number of NSs per 1000 primary cells (left panels) and the NSs size distributions (right panels) were determined. A greater number of large NSs were observed in young and adult SVZ cultures, as well as young and adult hippocampal cultures. The number of NSs was significantly greater than the HMF only in the adult hippocampal cultures, but was found with no changes in other groups. n is the number of animals. Data are shown as mean ± SEM. P values were calculated using a one-way ANOVA for mean comparisons and using a χ2 test for NS size distribution comparison. *P < 0.05; ****P < 0.0001; n.s. P ≥ 0.05 Compared to the SVZ groups, the overall number of NSs formed from the primary hippocampus was very low in both the GMF and HMF conditions (<1.5 NSs per 1000 cells) (Fig. 7B and 7D, left panels). No significant difference was observed between the HMF and GMF groups in NSs numbers per 1000 cells from the P15 mice (Fig. 7B, left panel), but the 2-month-old mice showed a significantly greater number of NSs per 1000 cells following exposure to the HMF (0.29 ± 0.05 for the GMF group, 0.48 ± 0.05 for the HMF group; P = 0.023; Fig. 7D, left panel). As observed in the SVZ NSs, a greater number of large hippocampal-derived NSs (P15, P < 0.0001; 2-month-old, P = 0.046) were present in the NS cultures exposed to the HMF for both age groups (Fig. 7B and 7D, right panels). These results indicate that HMF exposure triggers an acceleration of NS growth in primary cultures from both young and adult mice brains, and promotes formation of NSs from the adult hippocampus. Exposure to the HMF increases cell proliferation in the adult mouse brain To investigate whether exposure to the HMF has a positive effect on proliferation of NPCs/NSCs in vivo, adult mice were reared in an environment in which they were exposed to the HMF for 30 days. As shown in Fig. 8, BrdU-positive cells were observed in sections of the SVZ after mice had been exposed to both the HMF and GMF conditions (Fig. 8A and 8B). Exposure to the HMF resulted in a significant increase in the total numbers of BrdU-positive cells (P < 0.0001; Fig. 8C), which suggests a greater degree of proliferation in the SVZ. BrdU-positive cells were also detected in the hippocampus (Fig. 8D and 8E), but no significant differences between the two groups were observed (Fig. 8F; P = 0.059). These data demonstrate that exposure to the HMF promotes the proliferation of adult NPCs/NSCs in the SVZ, but not the hippocampus, in vivo.Figure 8 HMF exposure promotes cell proliferation in the adult mouse brain. Adult male mice (4 to 6-week-old) were reared in either the HMF or GMF conditions for 30 days. Representative images of the SVZ are shown in panels (A) and (B), and the hippocampus in (D) and (E). Proliferative cells were immunostained with anti-BrdU antibody (green) and glia cells with anti-GFAP antibody (red). Nuclei were counterstained with Hoechst (blue). White dash lines outline the edge of the lateral ventricles (LV). (C and F) show number of BrdU-positive cells per section from the SVZ (C) and hippocampus (F). Exposure to the HMF increased the number of proliferative cells in the SVZ but not the hippocampus. n indicates the number of animals from at least three independent experiments. Data are shown as mean ± SEM. P values were calculated using a one-way ANOVA for mean comparisons. ****P < 0.0001; n.s. P ≥ 0.05 Discussion Our results provide the first evidence that elimination of the GMF affects the growth of NPCs/NSCs, and that the GMF is required for maintaining the homeostasis of stem cells in the CNS. Following exposure to the HMF condition: (1) The proliferation of NPCs/NSCs from newborn (P2), young (P15), and adult mice (2-month) is accelerated in vitro; (2) these NPCs/NSCs were positive for NSC markers (Nestin and Sox2), and could be continuously passaged and differentiated into neurons and astrocyte/glial cells; (3) the HMF-enhanced NS growth could be maintained during continuous passages and restored by the GMF recovery; (4) the number of proliferative cells in adult SVZ were increased in vivo. These results indicate that NPCs/NSCs in the CNS can respond to the HMF. Previously, the HMF exposure has been found to cause abnormal cognitive behaviors and disrupt embryonic development (Wang et al. 2002; Zhang et al. 2004; Asashima and Shimada 1991; Mo et al. 2012b; Fesenko et al. 2010), but the potential biological mechanism is rarely been investigated. The proliferation, differentiation, and quiescence of NPCs/NSCs are under strict control and kept in delicate balance in embryonic as well as in adult brains (Simons and Clevers 2011; Stine and Matunis 2013). Excessive proliferation of stem cells would lead to stem cell exhaustion and aging (Orford and Scadden 2008; Oh et al. 2014), probably resulting in abnormal development and dysfunction of the CNS. Hyper-proliferation of NPCs/NSCs was reported to cause brain overgrowth and autism-associated behaviors (Nordahl et al. 2013; Le Belle et al. 2014), and decreased neurogenesis in affected animals led to interrupted learning and memory (Cameron and Glover 2015). According to our results, the HMF accelerates the proliferation of NPCs/NSCs in vivo (Fig. 8), which could break the balance of cell activities in CNS. Besides, HMF induced pro-proliferation of NSCs could also result in hyper-proliferation of NPCs/NSCs in brains. Thus, our study suggests that disturbed NPCs/NSCs activities under the HMF condition could be a reason for the bio-HMF effects on animal development and behaviors. Usually, NSs formed by NPCs are smaller than those formed by NSCs under normal culture condition (Louis et al. 2008; Ahmed 2009). The growth of both small-sized and large-sized NSs from embryonic brains were significantly accelerated by the HMF-exposure (Figs. 1, 2 and 7), indicating a common pro-proliferation effect on both NPCs and NSCs. In addition, the self-renewal capacity is limited in NPCs but not in NSCs so that NSCs can be enriched during continuous passaging. The pro-proliferative effect on NSs in the HMF maintained during the continuous passages (Figs. 2E and S2), proving the growth of NSCs is accelerated in the HMF. On the other hand, both NSCs and NPCs are present in the adult SVZ (Golmohammadi et al. 2008), but only NPCs and very few NSCs are found in the hippocampus (Bull and Bartlett 2005). Our results showed that the growth of NSs from young and adult hippocampus tissues were enhanced in the HMF (Fig. 7), which confirmed that the exposure to HMF has a pro-proliferative effect on NPCs. However, the response of NPCs and NSCs to the HMF might be different. The in-vitro NS assay and in-situ BrdU assay showed that the proliferation of NSCs/NPCs from the adult SVZ was greatly enhanced after exposure to the HMF condition (Figs. 7C, 8A–8C); however, the effect on adult hippocampal NPCs was rather low in vitro, and no significant change was observed in vivo (Figs. 7D, 8D–8F). These results suggest that NPCs are less sensitive to the HMF exposure than NSCs, though the proliferation of both NSCs and NPCs can be accelerated in the HMF. The pro-proliferation effect of the HMF is sustainable and reversible. When NSs were exposed to the HMF for four continuous passages, higher proliferation rate of the HMF group was maintained (Fig. 2E), and cell numbers increased continuously during the nine passage rounds (Fig. S2), which confirmed that the HMF effect sustained during the exposure. The numbers of cells were increased in all four passages and theoretically yielded a total of 1.88 ± 0.45 × 1010 cells, 26.3 times that the GMF group (8.79 ± 3.17 × 108). Since huge amount of cells are required for stem cell transplantations (Tsukamoto et al. 2013), our study suggests that HMF exposure could serve as a novel method to produce donor cells for stem cell therapy. Meanwhile, the growth of the HMF-exposed NSs dropped to the same level as the GMF controls when they returned to the GMF condition; the reversible HMF effect on NS growth is consistent with the previously reported HMF effects on animal behavior. HMF-induced (10 generations of continuous HMF exposure) amnesia in Drosophila has been shown to be recoverable after six consecutive generations in the GMF (Zhang et al. 2004). Since HMF is one of the environmental factors of outer space, GMF recovery would be a good option to minimize the HMF-related health risks to astronauts participating with long-term and long-distance space missions (Mo et al. 2014). Interestingly, although the HMF-exposed NSs maintain self-renewal and differentiation capacity, the transcription of Nestin, the common NSC marker, and Neurod1, the terminal neuronal differentiation marker (Cho and Tsai 2004), were downregulated in the HMF-exposed NSs compared with that in the GMF (Fig. 4B and 4C). The decrease of the transcription of Neurod1 is consistent with the enrichment of NSCs and an overall decrease of differentiation level in the HMF-exposed cells. The reduced transcription of Nestin suggests a contrary direction of cell fate, as Nestin is required for proper self-renewal and survival of NSCs (Park et al. 2010), and is down-regulated after differentiation (Namiki et al. 2012). Thus, the molecular property of the HMF-exposed NPCs/NSCs might be changed, although their ‘stemness’ identity is still maintained. However, the expressions of Gfap and ßIII-tubulin were not changed in the HMF-exposed NSs after the 7 day culture, as compared with the GMF control group (Fig. 4D and 4E), indicating that the pluripotency of the HMF-exposed cells to neuronal cells and astrocyte/glia was maintained. The differentiation assay confirmed that the HMF-exposed NSs could differentiate into neurons and astrocytes/glia (Fig. 5). Therefore, the HMF-exposure maintains ‘stemness’ of NSs, but might reduce their potency of neuronal differentiation. It would be interesting to examine the cell properties after exposure to the HMF to investigate whether this magnetic field could be applied to regulate the fate of NSCs. Our data showed that the NSs under the HMF exposure could tolerate the shortage of EGF or bFGF (Fig. 6). EGF and bFGF play a pivotal role in maintaining the self-renewal capacity of NPCs/NSCs in primary culture, and their signals were transduced via EGF receptors (EGFR) and bFGF receptors (bFGFR). Recently, it has been established that EGFR could serve as the mediator of the biomagnetic effects. A 0.4 mT power frequency magnetic field could induce clustering of the EGFR (Jia et al. 2007) and act as a stimulation factor, similar as the EGF, to activate the EGFR signaling (Wu et al. 2014). Since identical concentrations of EGF and bFGF were provided in the GMF-control and HMF-exposed groups under the normal and growth factor shortage incubation situation, it is probably that the HMF had elevated the activity or signal transduction efficiency of the growth factor receptors, EGFR and bFGFR in the NSs. Additionally, based on our in-vivo results, the levels of EGF and bFGF in the brain of HMF-exposed mice are also worth to be measured in the further study, to extend our understanding on the HMF effects on the growth of NSCs/NPCs. Moreover, ROS is involved both in the regulation of EGFR signaling (Carreira et al. 2010) and in the biological responses to the HMF (Martino and Castello 2011; Portelli et al. 2012; Fu et al. 2016) and weak EMF (Usselman et al. 2014; Castello et al. 2014). Recently, we found that the HMF could increase cellular ROS levels in primary skeletal muscle cells (Fu et al. 2016); however, no significant changes on oxidative stress signaling were detected in the HMF-exposed mice (Ding et al. 2014). Therefore, the role of ROS in the HMF effects is complicated and further investigation is necessary to elucidate the function of ROS in the HMF-enhanced NPCs/NSCs proliferation. In terms of the molecular mechanism of the HMF effects, a number of recent publications have provided evidence to support the role of Cry genes (Cry1 and Cry2) as the key molecules in the bio-magnetic interaction in avian, Arabidopsis, Drosophila and human cells (Mo et al. 2013; Xu et al. 2014; Gegear et al. 2012). Although Crys are photoreceptors and the radical pair generated by photo-chemical reaction in Crys plays a key role in its magnetosensitive function (Biskup et al. 2009), Crys can also generate radical pairs by the light-independent dark reoxidation of the flavin cofactor (Müller and Ahmad 2011). Recently, it has been shown that Cry could form a magnetic protein complex with MagR and sensing the direction of the applied static magnetic field, in vitro (Qin et al. 2016). Therefore, Crys would probably acting as an important molecule in the sensation of the elimination of the GMF under the dark condition such as in a cell incubator. Accumulating evidence has shown that cells and tissues incubated in the HMF exhibited observable cellular responses which were light-independent (Fesenko et al. 2010; Mo et al. 2013; Martino and Castello 2011; Fu et al. 2016; Martino et al. 2010; Mo et al. 2016). Our previous study has screened HMF-responding genes from neuroblastoma cells and verified that Cry2 was significant up-regulated after an 18 h short-term HMF exposure and down-regulated after a 48 h continuous HMF exposure (Mo et al. 2014b). Reduced mRNA levels of both Cry1 and Cry2 were detected in the HMF-exposed NSs (Fig. S3). Thus, Crys in non-photo-sensing organs, such as NPCs/NSCs would also play a role in magnetosensation. Down-regulated Cry1 expression in cells was found to cause accelerated growth rate in primary fibroblasts (Destici et al. 2011), so it is probable that the HMF-induced Cry down-regulation has contributed to the accelerated NS growth. Investigating the role of Cry genes, would help to explain how stem cells respond to the HMF at the molecular level. In conclusion, our research proves that the GMF is required for the maintenance of stem cell homeostasis. Exposure to the HMF accelerates the proliferation of NPCs/NSCs in vitro and in vivo, but the self-renewal and pluripotency capacity of NPCs/NSCs is maintained. Disturbed NPCs/NSCs growth in brains would probably contribute to abnormal development and altered behaviors in HMF-exposed animals. Furthermore, stem cell culturing in the HMF, such a physical approach which stimulates the growth of NPCs/NSCs non-invasively, has great potency for clinical application in stem cell therapy. Materials and methods The HMF conditions The HMF condition for cell culture was established as described (Mo et al. 2013). The samples exposed to the HMF were cultured in a magnetic shielding chamber with a residual magnetic field <200 nT. The GMF control samples were cultured on a plastic shelf outside the magnetic shielding box with a local magnetic field of 15.1 ± 2.2 μT. The other conditions inside and outside the chamber were almost identical. The HMF environment for animal rearing was provided by a 3-axis Helmholtz coils system (HCS) as described (Mo et al. 2015). The HMF-exposed animals were reared in a residual magnetic field of 0.029 ± 0.029 μT (center) and 0.55 ± 0.3 μT (average value); the control animals were reared on a wooden table (49.88 ± 1.82 μT average value), 1.5 m away from the HCS, in the same room (Table S1). The alternating magnetic fields (AMFs) were measured using a CCG-1000 induction alternative magnetometer (National Institute of Metrology, Beijing, China) and the predominant AMF frequencies were checked from the output of signal using a Textronics TDS 2014 digital real-time oscilloscope (Tequipment.NET, Long Branch, NJ, USA). The AMFs of the HCS and the GMF control environments were identical. Animals C57BL/6 mice were provided by the animal experiment center of the Institute of Biophysics (IBP). P2 (male/female), P15 (male), and 2-month (male) littermate naïve mice were used for primary NS cultures. 4–6 week male mice were used for the in vivo assay. All animal experiments were approved by the Animal Care and Use Committee at the IBP, Chinese Academy of Sciences (CAS) (authorized No.: SYXK2014-31) and carried out in accordance with the national guidelines for the care and use of laboratory animals. For the HMF/GMF exposure assay, animals were firstly reared on the GMF for 7 days to adapt to the environment, and then randomly grouped as four mice per standard “shoebox” cage. The cages in the HMF group were aligned as described (Prato et al. 2005). Animals were reared under a 12 h/12 h light/dark cycle. Daily magnetic field fluctuation was recorded, and the room temperature and humidity was maintained at 22 ± 1°C and ~40%–60%, respectively (Fig. S4). Primary neurosphere (NS) culture The whole brain, SVZ and hippocampus of mice were obtained as described previously (Walker et al. 2008). Whole brain P2 mouse or adult SVZ samples were enzymatically digested (0.25% trypsin and 0.025% EDTA in PBS) for 7–10 min at 37°C, and the hippocampus were digested using papain [1 mg/mL papain (Sigma-Aldrich, St. Louis, MO) and 0.5 mg/mL DNase I in L-15 medium (Invitrogen/molecular probe, Grand Island, NY)] for 10–15 min at 37°C. The mixtures were then mechanically triturated and filtered through 40 μm sieves (BD Bioscience, San Jose, CA). Cells were collected by centrifugation and resuspended in NS culture media (NSA) [DMEM/F12 (Invitrogen) supplemented with 10% proliferation supplement (Stem Cell Technologies, Vancouver, British Columbia, Canada), 2% bovine serum albumin (BSA) (Roche, Basel, Switzerland), 2 μg/mL heparin (Sigma-Aldrich), 10 ng/mL fibroblast growth factor 2 (bFGF) (Roche), and 20 ng/mL epidermal growth factor (EGF) (BD Bioscience)]. Then the primary cells were plated in a 96-well plate (1000 cells/well with 200 μL NSA medium) for NS counting and size analysis, or in a 60 mm dish (8.0 × 105 cells/dish with 4 mL NSA medium) for cell counting. The number and size of mature NSs (diameter ≥ 40 μm) were determined with an inverted microscope. NS passaging The cultured NSs were collected and passaged at day 7. The NSs were washed with PBS, trypsinized and then mechanically triturated into dissociated cells in NSA medium. After cell counting (Ntotal d7), the cells were seeded at a density of 1000 cells per well with 200 μL NSA medium in a 96-well plate for NS counting and size analysis, or 1.0 × 105 cells (Nseed d0) per 60 mm dish with 4 mL NSA medium for counting total cell numbers (Ntotal d7). Cell proliferation rate (R) was calculated as: R = Ntotal d7/Nseed d0. The theoretical total number of cells obtained at a certain passage (NPi) was calculated as: NPi = Nseed P1*RP1*RP2*…*RPi. For single clone culture, individual large NS (diameter ≥ 150 μm) was collected and trypsinized, as described above. All cells were seeded in a 6-well plate with 2 mL NSA medium per well. Successful cultures were determined when large NSs (diameter ≥ 150 μm) were observed at day 7. At least six NSs were used in each trial. CFSE staining Cell division in the primary NS was measured with Carboxyfluorescein Diacetate Succinimidyl Ester (CFSE) (Sigma-Aldrich) staining as described (Quah and Parish 2010). Primary cells from P2 mouse brain were stained with 25 μmol/L CFSE (107 cells/mL) for 20 min at 37°C. After two washes with PBS (with 0.1% BSA), CFSE-stained cells were seeded in a 60 mm dish and incubated in the HMF or GMF condition, as described above. NSs were collected and trypsinized at day 7, CFSE fluorescence was measured using a FACS Calibur flowcytometer (BD Bioscience) and analyzed with the Cell Quest Pro software. The cells that were not stained with CFSE were used as blank control (Blank). Quantitative real-time PCR (qPCR) Total RNA of primary NSs was extracted using an RNA extraction kit (QIAGEN, Hilden, Germany). cDNA samples were synthesized using an EasyScript First-Strand cDNA Synthesis SuperMix (Transgen Biotech, Beijing, China). The gene-specific primers (Supplementary Table S2) were designed by PrimerBank (Wang et al. 2012). A TransStart Green qPCR SuperMix UDG kit (TransGen Biotech) was applied to prepare the qPCR samples, which were run in triplicate on a Rotorgen Q real-time PCR cycler (QIAGEN). Thermal cycling was performed at an initial UDG incubation step at 50°C and a UDG inactivation step at 94°C, and then subjected to 45 cycles of 15 s denaturing at 95°C, 30 s at annealing temperature, and a 30 s extension at 72°C. Quantitative gene expressions were referenced to Tubb5 and normalized to the GMF samples. NS differentiation assay NS differentiation was performed as described (Golmohammadi et al. 2008) under the GMF condition. Coverslips were pre-coated with 15% poly-ornithine (Sigma-Aldrich) and 2% laminin (Invitrogen) at 37°C overnight and then washed six times with PBS. At day 7, mature NSs (10–20 NSs per coverslip) were transferred onto the coverslip and incubated in 2 mL differentiation medium (90% DMEM/F12 and 10% proliferation supplement) in each well. After 5 days of incubation, the NSs became adhered to the coverslips. Immunofluorescent staining Immunofluorescent staining of mature or differentiated NSs was performed as described (Golmohammadi et al. 2008). The NSs were fixed with 4% paraformaldehyde (Amresco, Solon, OH) for 30 min at RT. After one PBS wash, they were blocked in blocking solution [PBS with 0.1% triton100 (Amresco), 5% FBS, and 5% goat serum (ZSGB-Bio, Beijing, China)] for 60 min at 37°C. After that, the NSs were firstly stained with primary antibody and then with secondary antibody in blocking solution, the nuclei were stained with 10 μg/mL Hoechst (1:1000; Beyotime, Jiangsu, China). Primary antibodies: mouse-anti-nestin (1:200; 307501, R&D, Minneapolis, MN), rabbit-anti-SOX2 (1:200; L1D6A2, Cell Signal Technology (CST), Boston, MA), mouse-anti-GFAP (1:200; GA5, CST) and rabbit-anti-ßIII-tubulin (1:200; D65A4, CST); secondary antibodies: Alexa Fluor 568 donkey-anti-mouse (1:1000; Invitrogen/molecular probe) and Alexa Fluor 488 donkey-anti-rabbit (1:1000; Invitrogen/molecular probe). The coverslips were then mounted using GVA mounting medium (ZSGB-Bio). BrdU assay Animals were administered with BrdU (10 mg/mL in physiological saline; Sigma-Aldrich) twice per day by intraperitoneal injection at a dose of 350 mg per kilogram body weight, 3 days before sacrifice. After ether anesthesia, animals were transcardially perfused with 50 ml normal saline and 50 mL 4% PFA. The brains were fixed in 4% PFA overnight at 4°C, then cryoprotected in 30% sucrose and embedded in embedding medium (Tissue-Tek; Sakura Finetek, Torrance, CA). Transverse sections were cut using a cryostat (10 µm) (Leica, Wetzlar, Germany). For immunofluorescence staining, sections were re-hydrated (three PBS washes and 2 mol/L HCl for 1 h at RT). After four PBS washes, sections were treated with blocking solution (PBST and 10% goat serum) for 1 h at RT, then incubated in sheep-anti-BrdU antibody (ab1893, Abcam, Cambridge, MA) solution (1:200 in blocking solution) at 4°C overnight. After three PBS washes, sections were incubated in Alexa Fluor 488 donkey-anti-sheep antibody (1:300, Invitrogen/Molecular probe) for 2 h at RT. Following three PBST washes, the nuclei were stained with Hoechst (10 µg/mL) at RT for 20 min. The sections were mounted with GVA mounting medium. The number of BrdU-positive cells was determined from five consecutive sections (Wojtowicz and Kee 2006). Statistical methods Each experiment was repeated at least twice in triplicate each time, if not otherwise specified. A one-way ANOVA was used for mean comparison. A χ2 test was used for the comparison of NS size distribution p < 0.05. Microscopy Phase contrast images of alive NSs in the culture medium were taken at RT using an inverted microscope with UPlanFl 10×PH/0.3 objective lens (Olympus IX71, Japan) and a cooled EMCCD (Andor iXon DV897, UK, 512 × 512 pixels). Fluorescent images of immunostained NSs and sections were taken using a fluorescent microscope with Plan APO 10×/0.45 or 20×/0.75 objective lens (Nikon FXA, Japan), and a cooled CCD (Olympus DP71, Japan, 1360x1024 or 4080× 3072 pixels). G-2A (EX510-560/DM575/BA590), B-2A (EX450-490/DM505/BA520), UV-2A (EX330-308/DM400/BA420) filters (Nikon) were applied for Alexa Fluor 568, Alexa Fluor 488 and Hoechst, respectively. Images were merged with Adobe Photoshop CS4. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary material 1 (PDF 642 kb) Jing-Peng Fu and Wei-Chuan Mo have contributed equally to this work. Acknowledgements We sincerely thank Chief Engineer Wen-Kui Jia and Engineer Qiang Shi [Center of Space Science and Applied Research (SCCAR), Chinese Academy of Sciences (CAS)], for their assistance in the construction of the HMF animal rearing system and the measurement of the magnetic fields; Dr. Daniel Blackmore and Dr. Di Xia [Queensland Brain Institute (QBI), the University of Queensland (UQ)], for their kind instruction on neurosphere culture; M.M. Haiming Ding (Beijing University of Chinese Medicine) for her assistance in animal rearing; Miss Ashley Cooper (QBI, UQ) and Dr. Yinghao Zhang [Institute of Biophysics (IBP), CAS] for their assistance on the writing of the manuscript; Dr. Jianbo Yue [City University of Hong Kong] and Prof. Yue Ma [IBP, CAS] for their kind gift of the bFGF protein. This work is funded by: The External Cooperation Program of BIC, Chinese Academy of Sciences (Grant No. GJHZ201302), the project of Chinese Academy of Sciences for the development of major scientific research equipment (Grant No. YZ201148), the National Natural Science Foundation of China (Grant Nos. 31200628 and 31271387), and the Queensland-Chinese Academy of Sciences Biotechnology Fund (Grant No. GJHZ1131). Abbreviations CNS, central nervous system; EMFs, electromagnetic fields; GMF, geomagnetic field; HMF, hypogeomagnetic field; NPCs/NSCs, neural progenitor/stem cells; NSs, neurospheres; SMF, static magnetic field; SVZ, subventricular zone. Compliance with ethical standards Jing-Peng Fu, Wei-Chuan Mo, Ying Liu, Perry F. Bartlett, and Rong-Qiao He declare that they have no conflict of interest. All institutional and national guidelines for the care and use of laboratory animals were followed. ==== Refs References Ahmed S The culture of neural stem cells J Cell Biochem 2009 106 1 6 10.1002/jcb.21972 19021147 Asashima M Shimada K Pfeiffer CJ Magnetic shielding induces early developmental abnormalities in the newt, Cynops pyrrhogaster Bioelectromagnetics 1991 12 215 224 10.1002/bem.2250120403 1930306 Azari H Isolation and expansion of the adult mouse neural stem cells using the neurosphere assay JoVE 2010 45 2393 21113123 Biskup T Direct observation of a photoinduced radical pair in a cryptochrome blue-light photoreceptor Angew Chem Int Ed Engl 2009 48 404 407 10.1002/anie.200803102 19058271 Bliss VL Heppner FH Circadian activity rhythm influenced by near zero magnetic field Nature 1976 261 411 412 10.1038/261411a0 934271 Bull ND Bartlett PF The adult mouse hippocampal progenitor is neurogenic but not a stem cell J Neurosci 2005 25 10815 10821 10.1523/JNEUROSCI.3249-05.2005 16306394 Cameron HA Glover LR Adult neurogenesis: beyond learning and memory Annu Rev Psychol 2015 3 53 81 10.1146/annurev-psych-010814-015006 25251485 Carreira BP Nitric oxide stimulates the proliferation of neural stem cells bypassing the epidermal growth factor receptor Stem Cells 2010 28 1219 1230 20506358 Castello PR Inhibition of cellular proliferation and enhancement of hydrogen peroxide production in fibrosarcoma cell line by weak radio frequency magnetic fields Bioelectromagnetics 2014 35 598 602 10.1002/bem.21858 25251337 Cho JH Tsai MJ The role of BETA2/NeuroD1 in the development of the nervous system Mol Neurobiol 2004 30 35 47 10.1385/MN:30:1:035 15247487 Ciccolini F Svendsen CN Fibroblast growth factor 2 (FGF-2) promotes acquisition of epidermal growth factor (EGF) responsiveness in mouse striatal precursor cells: Identification of neural precursors responding to both EGF and FGF-2 J Neurosci 1998 18 7869 7880 9742155 Destici E Mammalian cryptochromes impinge on cell cycle progression in a circadian clock-independent manner Cell Cycle 2011 10 3788 3797 10.4161/cc.10.21.17974 22033214 Di Lazzaro VF A consensus panel review of central nervous system effects of the exposure to low-intensity extremely low-frequency magnetic fields Brain Stimul 2013 6 469 476 10.1016/j.brs.2013.01.004 23428499 Ding H The hematopoietic system responses to one-month continuous hypomagnetic field exposure in adult mice Prog Mod Biomed 2014 26 5001 5004 Fesenko EE Effect of the “zero” magnetic field on early embryogenesis in mice Electromagn Biol Med 2010 29 1 8 10.3109/15368371003627290 20230271 Fu JP Decline of cell viability and mitochondrial activity in mouse skeletal muscle cell in a hypomagnetic field Bioelectromagnetics 2016 37 212 222 10.1002/bem.21968 27003876 Gage FH Temple S Neural stem cells: generating and regenerating the brain Neuron 2013 80 588 601 10.1016/j.neuron.2013.10.037 24183012 Gegear RJ Animal cryptochromes mediate magnetoreception by an unconventional photochemical mechanism Nature 2012 463 804 807 10.1038/nature08719 20098414 Golmohammadi MG Comparative analysis of the frequency and distribution of stem and progenitor cells in the adult mouse brain Stem Cells 2008 26 979 987 10.1634/stemcells.2007-0919 18203672 Gould JL Gould CG Nature’s compass: the mystery of animal navigation 2012 Princeton Princeton University Press Graham V SOX2 functions to maintain neural progenitor identity Neuron 2003 39 749 765 10.1016/S0896-6273(03)00497-5 12948443 Jia C EGF receptor clustering is induced by a 0.4 mT power frequency magnetic field and blocked by the EGF receptor tyrosine kinase inhibitor PD153035 Bioelectromagnetics 2007 28 197 207 10.1002/bem.20293 17019730 Jiang JC Effect of magnetic free field space (MFFS) on vocal behavior in melop sittacus undulafus Acta Seismol Sin 1998 20 421 426 Jogler C Schuler D Genomics, genetics, and cell biology of magnetosome formation Annu Rev Microbiol 2009 63 501 521 10.1146/annurev.micro.62.081307.162908 19575557 Le Belle JE Maternal inflammation contributes to brain overgrowth and autism-associated behaviors through altered redox signaling in stem and progenitor cells Stem Cell Rep 2014 3 725 734 10.1016/j.stemcr.2014.09.004 Lisi A Exposure to 50 Hz electromagnetic radiation promote early maturation and differentiation in newborn rat cerebellar granule neurons J Cell Physiol 2005 204 532 538 10.1002/jcp.20322 15754325 Lohmann KJ Animal behavior: magnetic-field perception Nature 2010 464 1140 1142 10.1038/4641140a 20414302 Louis SA Enumeration of neural stem and progenitor cells in the neural colony-forming cell assay Stem Cells 2008 26 988 996 10.1634/stemcells.2007-0867 18218818 Lui JH Hansen DV Kriegstein AR Development and evolution of the human neocortex Cell 2011 146 18 36 10.1016/j.cell.2011.06.030 21729779 Martino CF Castello PR Modulation of hydrogen peroxide production in cellular systems by low level magnetic fields PLoS One 2011 6 e22753 10.1371/journal.pone.0022753 21887222 Martino CF Reduction of the Earth’s magnetic field inhibits growth rates of cancer cells Bioelectromagnetics 2010 31 649 655 10.1002/bem.20606 20830734 Merkle FT Alvarez-Buylla A Neural stem cells in mammalian development Curr Opin Cell Biol 2006 18 704 709 10.1016/j.ceb.2006.09.008 17046226 Mo WC Liu Y He RQ A biological perspective of the hypomagnetic field: from definition towards mechanism Prog Biochem Biophys 2012 399 835 842 Mo WC Altered development of Xenopus embryos in a hypogeomagnetic field Bioelectromagnetics 2012 33 238 246 10.1002/bem.20699 21853450 Mo WC Magnetic shielding accelerates the proliferation of human neuroblastoma cell by promoting G1-phase progression PLoS One 2013 8 e54775 10.1371/journal.pone.0054775 23355897 Mo W Liu Y He R Hypomagnetic field, an ignorable environmental factor in space? Sci China Life Sci 2014 57 726 728 10.1007/s11427-014-4662-x 24824587 Mo W Transcriptome profile of human neuroblastoma cells in the hypomagnetic field Sci China Life Sci 2014 57 448 461 10.1007/s11427-014-4644-z 24777382 Mo WC Hypomagnetic field alters circadian rhythm and increases algesia in adult male mice Prog Biochem Biophys 2015 42 7 639 646 Mo W Shielding of the geomagnetic field alters actin assembly and inhibits cell motility in human neuroblastoma cells Sci Rep 2016 6 22624 10.1038/srep22624 27029216 Müller P Ahmad M Light-activated cryptochrome reacts with molecular oxygen to form a flavin-superoxide radical pair consistent with magnetoreception J Biol Chem 2011 286 21033 21040 10.1074/jbc.M111.228940 21467031 Nakamichi N Possible promotion of neuronal differentiation in fetal rat brain neural progenitor cells after sustained exposure to static magnetism J Neurosci Res 2009 87 2406 2417 10.1002/jnr.22087 19382241 Namiki J Nestin protein is phosphorylated in adult neural stem/progenitor cells and not endothelial progenitor cells Stem Cells Int 2012 2012 430138 10.1155/2012/430138 23028390 Nathan FP An inherited magnetic map guides ocean navigation in Juvenile Pacific Salmon Curr Biol 2014 24 446 450 10.1016/j.cub.2014.01.017 24508165 Nordahl CW Maternal autoantibodies are associated with abnormal brain enlargement in a subgroup of children with autism spectrum disorder Brain Behav Immun 2013 30 61 65 10.1016/j.bbi.2013.01.084 23395715 Oh J Lee YD Wagers AJ Stem cell aging: mechanisms, regulators and therapeutic opportunities Nat Med 2014 20 870 880 10.1038/nm.3651 25100532 Orford KW Scadden DT Deconstructing stem cell self-renewal: genetic insights into cell-cycle regulation Nat Rev Genet 2008 9 115 128 10.1038/nrg2269 18202695 Park D Nestin is required for the proper self-renewal of neural stem cells Stem Cells 2010 28 2162 2171 10.1002/stem.541 20963821 Portelli LA Reduction of the earth’s magnetic field inhibits Drosophila melanogaster ability to survive ionizing radiation Bioelectromagnetics 2012 33 706 709 10.1002/bem.21720 22532126 Prato FS Daily repeated magnetic field shielding induces analgesia in CD-1 mice Bioelectromagnetics 2005 26 109 117 10.1002/bem.20056 15672364 Qin S A magnetic protein biocompass Nat Mater 2016 15 217 226 10.1038/nmat4484 26569474 Quah BJC Parish CR The use of Carboxyfluorescein Diacetate Succinimidyl Ester (CFSE) to monitor lymphocyte proliferation JoVE 2010 44 2259 2261 20972413 Simons BD Clevers H Strategies for homeostatic stem cell self-renewal in adult tissues Cell 2011 145 851 862 10.1016/j.cell.2011.05.033 21663791 Stine RR Matunis EL Stem cell competition: finding balance in the niche Trends Cell Biol 2013 23 357 364 10.1016/j.tcb.2013.03.001 23597843 Tsukamoto A Clinical translation of human neural stem cells Stem Cell Res Ther 2013 4 102 10.1186/scrt313 23987648 Usselman R Spin biochemistry modulates reactive oxygen species (ROS) production by radio frequency magnetic fields PLoS One 2014 9 e93065 10.1371/journal.pone.0093065 24681944 Ventura C Turning on stem cell cardiogenesis with extremely low frequency magnetic fields FASEB J 2005 19 155 157 15507470 Walker TL Latent stem and progenitor cells in the hippocampus are activated by neural excitation J Neurosci 2008 28 5240 5247 10.1523/JNEUROSCI.0344-08.2008 18480280 Wan GJ Bio-effects of near-zero magnetic fields on the growth, development and reproduction of small brown planthopper, Laodelphax striatellus and brown planthopper, Nilaparvata lugens J Insect Physiol 2014 68 7 15 10.1016/j.jinsphys.2014.06.016 24995837 Wang XB Long-term memory was impaired in one-trial passive avoidance task of day-old chicks hatching from hypomagnetic field space Chin Sci Bull 2002 48 2042 2045 10.1360/02tb9442 Wang X PrimerBank: a PCR primer database for quantitative gene expression analysis, 2012 update Nucleic Acids Res 2012 40 1144 11499 10.1093/nar/gkr1013 Wojtowicz JM Kee N BrdU assay for neurogenesis in rodents Nat Protoc 2006 1 1399 1405 10.1038/nprot.2006.224 17406427 Wu H Effect of electromagnetic fields on proliferation and differentiation of cultured mouse bone marrow mesenchymal stem cells J Huazhong Univ Sci Technol Med Sci 2005 25 185 187 10.1007/BF02873572 16116968 Wu X Weak power frequency magnetic field acting similarly to EGF stimulation, induces acute activations of the EGFR sensitive actin cytoskeleton motility in human amniotic cells PLoS One 2014 9 e87626 10.1371/journal.pone.0087626 24505297 Xu C Blue light-dependent phosphorylations of cryptochromes are affected by magnetic fields in Arabidopsis Adv Space Res 2014 53 1118 1124 10.1016/j.asr.2014.01.033 Yau SY Physical exercise-induced hippocampal neurogenesis and antidepressant effects are mediated by the adipocyte hormone adiponectin Proc Natl Acad Sci USA 2014 111 15810 15815 10.1073/pnas.1415219111 25331877 Zhang B Exposure to hypomagnetic field space for multiple generations causes amnesia in Drosophila melanogaster Neurosci Lett 2004 371 190 195 10.1016/j.neulet.2004.08.072 15519755

PMC005xxxxxx/PMC5003791.txt

==== Front Protein CellProtein CellProtein & Cell1674-800X1674-8018Higher Education Press Beijing 30110.1007/s13238-016-0301-6Research ArticleCellular model of neuronal atrophy induced by DYNC1I1 deficiency reveals protective roles of RAS-RAF-MEK signaling Liu Zhi-Dong 1Zhang Su 12Hao Jian-Jin 1Xie Tao-Rong 1http://orcid.org/0000-0002-2603-9718Kang Jian-Sheng jskang@sibs.ac.cn 121 Key Laboratory of Nutrition and Metabolism, Institute for Nutritional Sciences, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai, 200231 China 2 University of Chinese Academy of Sciences, Beijing, 100049 China 10 8 2016 10 8 2016 9 2016 7 9 638 650 17 6 2016 7 7 2016 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.Neuronal atrophy is a common pathological feature occurred in aging and neurodegenerative diseases. A variety of abnormalities including motor protein malfunction and mitochondrial dysfunction contribute to the loss of neuronal architecture; however, less is known about the intracellular signaling pathways that can protect against or delay this pathogenic process. Here, we show that the DYNC1I1 deficiency, a neuron-specific dynein intermediate chain, causes neuronal atrophy in primary hippocampal neurons. With this cellular model, we are able to find that activation of RAS-RAF-MEK signaling protects against neuronal atrophy induced by DYNC1I1 deficiency, which relies on MEK-dependent autophagy in neuron. Moreover, we further reveal that BRAF also protects against neuronal atrophy induced by mitochondrial impairment. These findings demonstrate protective roles of the RAS-RAF-MEK axis against neuronal atrophy, and imply a new therapeutic target for clinical intervention. Electronic supplementary material The online version of this article (doi:10.1007/s13238-016-0301-6) contains supplementary material, which is available to authorized users. Keywords RAS-RAF-MEK pathwayatrophydynein intermediate chainmitochondriahippocampal neuronautophagyissue-copyright-statement© HEP and Springer 2016 ==== Body Introduction For normal aging, age-related brain atrophy is a mild process, responsible for increased risk of memory decline with increasing age (Pakkenberg et al., 2003; Fox and Schott, 2004), whereas the highly accelerated rate of brain atrophy has a major or sole role in neurodegenerative diseases (Regeur et al., 1994; Schott et al., 2003). Loss of neuronal architecture is the main contributor to brain atrophy (Swaab et al., 1994; Freeman et al., 2008), thus atrophy of brain and neuron can be well defined as the shortening or shrinkage of neurites. Neuritic atrophy is a common pathological feature of many neurodegenerative disorders including amyotrophic lateral sclerosis (ALS), Alzheimer’s disease (AD), Parkinson’s disease (PD), and Huntington’s disease (HD). A neuron consists of a soma and long neurites including one axon and multiple dendrites. The physiological structures of neurons render them particularly vulnerable to motor protein malfunction, protein aggregation, and mitochondrial dysfunction. Particularly, abnormalities to dynein and mitochondria are linked to ALS (Soo et al., 2011). Cytoplasmic dynein is the main driving force for minus-end-directed transport of cargos (Holzbaur and Vallee, 1994). Cytoplasmic dynein is a large protein complex (~1.5 MDa) containing heavy chains, intermediate chains, light intermediate chains, and light chains (Pfister et al., 2005). Mutations and 9-bp deletion of cytoplasmic dynein heavy chain (Dync1h1) are sufficient to cause neuron degeneration (Hafezparast et al., 2003; Chen et al., 2007; Banks and Fisher, 2008; Lipka et al., 2013). Mutations of cytoplasmic dynein light intermediate chain 2 (Dync1li2) and dynein intermediate chain (Dync1i) result in reduction of dendrite arborization of Drosophila neurons (Zheng et al., 2008; Boylan and Hays, 2002). However, less is known about the physiologic role of Dync1i in higher animals, especially the isoforms of Dync1i1, which are neuron-specific and not expressed in glia in the brain (Myers et al., 2007). Two genes (Dync1i1 and Dync1i2) in rodents encode cytoplasmic dynein intermediate chains, in which Dync1i1 is a neuron-specific gene (Fig. 1A). In line with previous reports (Myers et al., 2007), most mRNAs of the intermediate chain isoforms are expressed in rat brain, while the mRNA of Dync1i2C and Dync1i2B are expressed in most tissues (Fig. 1A). Here, we demonstrate that the knockdown of Dync1i1 causes neuronal atrophy and decreases mitochondrial motility in rat primary hippocampal neurons.Figure 1 Knockdown of cytoplasmic dynein 1 intermediate chain 1 ( Dync1i1 ) causes dendritic atrophy in primary hippocampal neurons. (A) The quantitative-PCR results show the expression levels of cytoplasmic dynein 1 intermediate chains, including the isoforms of Dync1i1 (1A, 1B, 1C, and 1D) and the isoforms of Dync1i2 (2A, 2B, and 2C) in P0 rat hippocampus, cortex, heart, kidney, liver, and lung tissues. (B–G) Representative neurons are transfected with shRNA1-6 (B–D, a specific shRNA of Dync1i1, see also Fig. S1) or Ctrl (E–G) at DIV6 and immunostained for the dendritic marker MAP2 (red in D and G) at DIV11. Green neurons are shRNA1-6 or Ctrl transfected neurons (green in D and G). The scale bars represent 20 μm. (H) Sholl analysis for dendritic complexity of neurons transfected with control vector (Ctrl, n = 75) and Dync1i1 shRNA1-6 (n = 101). In control neurons, the maximum length of dendrite branches is found between 300 and 400 µm from the cell body. The shRNA1-6 transfected neurons show marked shortened dendrites such that the majority of dendrites were located within 200 µm of the soma. Data are represented as mean ± SEM. (I) Scatterplots with boxplots show that knockdown of DYNC1I1 expression causes dendritic atrophy. Primary hippocampal neurons are transfected with control vector (Ctrl, n = 26, gray box), control shRNA (shRNA1-4, n = 26, blue box) or shRNA1-6 (n = 26, red box) of Dync1i1 at DIV6. Following transfection, neurons are cultured additional 1, 3 or 5 days before imaging and quantification of total dendritic length. The total dendritic lengths of control neurons slightly increase with increasing days in vitro. There is almost no length change of shRNA1-4 transfected neurons. However, the total dendritic lengths of shRNA1-6 transfected neurons are gradually and dramatically reduced at 3 and 5 days after transfection. *, P < 0.001 Using the cellular model of neuronal atrophy caused by DYNC1I1 deficiency, we are able to identify that RAS-RAF-MEK signaling, but not PI3K-AKT signaling, protects neurons against dendritic atrophy in primary hippocampal neurons, and reveals that RAS-RAF signaling activates MEK-dependent protective autophagy. Moreover, we further demonstrate that BRAF can also protect dendrites from atrophy arisen from mitochondrial dysfunction. These findings of the RAS-RAF-MEK pathway for neuronal atrophy protection may provide a therapeutic target against the on-sets of neuronal atrophy. Results Knockdown of DYNC1I1 expression causes dendritic atrophy of primary hippocampal neurons Considering DYNC1I2 may play housekeeping function (Myers et al., 2007) and neurites of cultured neuron grow fast before 7 days in vitro (DIV7) (Dotti et al., 1988), we selectively knockdown DYNC1I1 expression at DIV6 in rat primary hippocampal neurons. Neurons transfected with shRNA1-6, a specific shRNA of Dync1i1 (Fig. S1A–F), show reduced dendritic complexity and shortened dendritic length (Fig. 1B–D) compared to control neurons (Fig. 1E–G) at DIV11. For dendritic branches, the length of single branch seldom distributes beyond 350 μm in control neurons or 200 μm in DYNC1I1-knockdown neurons (Fig. 1H). Compared to the total dendritic lengths of control neurons (1100 ± 453 μm, n = 75, mean ± SD), the total dendritic lengths of DYNC1I1 knockdown neurons (303 ± 307 μm, n = 101, mean ± SD) are dramatically decreased (P < 0.001, t test) at DIV11 (Fig. 1I). Knockdown-resistant isoforms of Dync1i1 can partially rescue the phenotype caused by shRNA1-6 (Fig. S1C–H). The data suggest that the functions of DYNC1I1 in neuron are nonredundant and necessary for the maintenance of neuronal architecture. The potential effects of shRNA1-6 on dendritic development or dendritic atrophy can both explain the reduced total dendritic lengths of DYNC1I1 knockdown neurons. To determine the effect of shRNA1-6, we have checked and quantified total dendritic lengths at different time points (1, 3, and 5 days) after shRNA1-6 transfection. As shown in Fig. 1F, the distributions of total dendritic lengths show large variance. For 1, 3, and 5 days after transfection, the total dendritic lengths (in μm, mean ± SD) of control neurons are 1075 ± 330, 1130 ± 391, and 1307 ± 420, respectively; for control shRNA1-4 transfected neurons, the lengths are 1037 ± 230, 1010 ± 234, and 1006 ± 310; while the lengths of shRNA1-6 transfected neurons are 1041 ± 340, 535 ± 119, and 279 ± 113, respectively. These results are in line with reports that the dendritic architecture of neurons are mature at stage 4 (DIV4-7), while the maturation of synapses proceeds for following 2 weeks (stage 5) in vitro (Lalli, 2014). The data demonstrate that the DYNC1I1 knockdown by shRNA1-6 results in dendritic atrophy rather than dendritic-development retardation (Fig. 1F). Thus, to address the question of whether there is an intrinsic signaling pathway for protecting neuron from atrophy, dendritic atrophy caused by DYNC1I1 knockdown in primary hippocampal neuron is a good cellular model of neuronal atrophy arisen by motor protein dynein malfunction. BRAF protects against dendritic atrophy caused by DYNC1I1 knockdown in primary hippocampal neurons For signaling pathways, we focus on the RAS-RAF-MEK-ERK mitogen activated protein kinase (MAPK) cascade and phosphoinositide 3-kinase (PI3K)-Protein kinase B (PKB, also known as AKT) pathways since they are both downstream signaling of neurotrophic factors (Chao, 2003). We overexpress AKT or BRAF in rat primary hippocampal neurons, where BRAF is the dominant functional RAF homolog in MAPK cascade and brain among three mammalian RAF proteins (ARAF, BRAF, and CRAF) (Galabova-Kovacs et al., 2006; Zhong et al., 2007). Consistent with the role in the signaling of neurotrophic factors, we have found that both BRAF and AKT overexpression can promote dendritic growth in control neurons (Fig. 2). The total dendritic lengths (in μm, mean ± SD) of control (Fig. 2A) and BRAF overexpressed neurons (Fig. 2B) are 1007 ± 342 and 1491 ± 540, respectively (Fig. 2F, gray and green boxes, P < 0.001, t test), and the lengths of control (Fig. 2G) and AKT overexpressed neurons (Fig. 2H) are 1021 ± 323 and 1581 ± 567, respectively (Fig. 2L, gray and green boxes, P < 0.001, t test).Figure 2 BRAF but not AKT overexpression protects against dendritic atrophy caused by DYNC1I1 deficiency in primary hippocampal neurons. (A–D) BRAF overexpression protects against dendritic atrophy caused by DYNC1I1 knockdown. Representative neurons are transfected with control vector (Ctrl) or shRNA1-6 co-transfected with or without BRAF at DIV6, and sequentially imaged at DIV11. The scale bars represent 20 μm. (E) Sholl analysis for dendritic complexity of neurons transfected with control vector combined without (Ctrl, n = 37, black) or with BRAF (Ctrl + BRAF, n = 48, green) or Dync1i1 shRNA1-6 combined without (n = 60, red) or with BRAF (shRNA-6 + BRAF, n = 74, purple). Data are represented as mean ± SEM. (F) Scatterplots with boxplots show that BRAF overexpression promotes the dendritic growth of Ctrl neurons and protects dendritic atrophy caused by DYNC1I1 knockdown with shRNA1-6. (G–J) AKT overexpression has no effect on dendritic atrophy caused by DYNC1I1 knockdown. Representative neurons are transfected with control vector (Ctrl) or shRNA1-6 co-transfected with or without AKT at DIV6, and imaged at DIV11. The scale bars represent 20 μm. (K) Sholl analysis for dendritic complexity of neurons transfected with control vector combined without (Ctrl, n = 26, black) or with AKT (Ctrl + AKT, n = 28, green) or Dync1i1 shRNA1-6 combined without (n = 28, red) or with AKT (shRNA-6 + AKT, n = 33, purple). Data are represented as mean ± SEM. (L) Scatterplots with boxplots show that AKT overexpression can also enhance the dendritic growth of Ctrl neurons, but fail to protect from dendritic atrophy caused by Dync1i1 knockdown with shRNA1-6. *, P < 0.001 However, only BRAF can partially rescue dendritic complexity and protect against dendritic atrophy caused by DYNC1I1 knockdown (Fig. 2D–F), and the total dendritic lengths (in μm, mean ± SD) of DYNC1I1-knockdown neurons without (Fig. 2C) or with BRAF overexpression (Fig. 2D) are 264 ± 135 and 776 ± 607, respectively (Fig. 2F, red and purple boxes, P < 0.001, t test). Meanwhile, AKT overexpression has no effect on dendritic complexity and atrophy (Fig. 2J–L), and the total dendritic lengths (in μm, mean ± SD) of DYNC1I1-knockdown neurons without (Fig. 2I) or with AKT overexpression (Fig. 2J) are 330 ± 197 and 347 ± 242, respectively (Fig. 2L, P = 0.78, t test). Thus, the results suggest that RAF signaling may have important role in protecting against neuronal atrophy caused dynein malfunction. RAS-RAF-MEK signaling protects against dendritic atrophy by activating MEK-dependent autophagy To determine the signaling pathway for protecting dendritic atrophy, we use H-RAS mutants that activate RAF-MEK or/and PI3K-AKT signaling selectively or non-selectively. Dominant active mutant RAS-G12V activates both RAF-MEK and PI3K-AKT signaling, while RAS-G12V-T35S mutant selectively activates RAF-MEK and the RAS-G12V-Y40C mutant selectively stimulates PI3K-AKT (Fiordalisi et al., 2001). In line with the above results of BRAF and AKT (Fig. 2), we have found that RAS-G12V-T35S mutant but not RAS-G12V-Y40C mutant has similar protective effect of BRAF on dendritic atrophy caused by DYNC1I1 knockdown (Fig. 3A–E). The total dendritic lengths (in μm, mean ± SD) of control (Fig. 3A), shRNA1-6 (Fig. 3B), RAS-G12V-T35S (Fig. 3C), and RAS-G12V-Y40C (Fig. 3D) transfected neurons are 1188 ± 228, 377 ± 148, 623 ± 309, and 351 ± 234, respectively (Fig. 3E). Compared to the length of shRNA1-6 group, the lengths of RAS-G12V-T35S group are improved (P < 0.01, t test); RAS-G12V also plays protective role in dendritic atrophy by DYNC1I1 knockdown (562 ± 97 μm, mean ± SD, P < 0.01, t test), while RAS-G12V-Y40C has no effect (P = 0.64, t test). These results suggest RAS-RAF pathway rather than PI3K-AKT signaling can play protective role in neuronal atrophy caused by motor protein malfunction.Figure 3 RAS-RAF-MEK signaling pathway protects against dendritic atrophy caused by DYNC1I1 deficiency. (A–D) Dominant active mutant G12V of RAS and selective RAF-activator mutant (G12V, T35S) of RAS protect against dendritic atrophy caused by DYNC1I1 knockdown, while selective AKT-activator mutant (G12V, Y40C) of RAS has no effect. Primary hippocampal neurons are transfected with control vector (Ctrl) or shRNA1-6 co-transfected with RAS-G12V, RAS-G12V-T35S or RAS-G12V-Y40C at DIV6, and imaged at DIV11. The scale bars represent 20 μm. (E) Scatterplots with boxplots show the total dendritic length distribution of neurons transfected with control vector (Ctrl, n = 18, gray), shRNA1-6 (n = 25, red), shRNA1-6 with RAS-G12V (n = 7, light blue), RAS-G12V-T35S (RAS-T35S, n = 25, green) or RAS-G12V-Y40C (RAS-Y40C, n = 25, purple). (F and G) Trametinib, a specific inhibitor of MEK kinase in RAS-RAF-MEK signaling, inhibits the protective role of BRAF in dendritic atrophy caused by DYNC1I1 deficiency with shRNA1-6. Neurons are co-transfected with BRAF and shRNA1-6 at DIV6, and imaged at DIV11. The scale bars represent 20 μm. (H) Scatterplots with boxplots show the total dendritic length distribution of neurons transfected with control vector treated without (n = 69, gray) or with 100 nmol/L of trametinib (n = 40, green), shRNA1-6 (n = 50, red), shRNA1-6 co-transfected with BRAF treated without (n = 50, purple) or with 100 nmol/L of trametinib (n = 50, blue). *, P < 0.001 To confirm the role of RAS-RAF signaling in neuronal atrophy, we inhibit RAS-RAF-MEK signaling with pharmacological inhibitor trametinib, a specific inhibitor of MEK kinase (Yamaguchi et al., 2011). As shown in Fig. 3F–H, 100 nmol/L of trametinib blocks the protective effect of BRAF in dendritic atrophy induced by DYNC1I1 knockdown (P < 0.001, t test). For DYNC1I1-knockdown neurons at DIV11, the total dendritic lengths (μm, mean ± SD) of BRAF overexpressed neurons without (Fig. 3F) or with 100 nmol/L of trametinib treatment (Fig. 3G) are 740 ± 265 and 234 ± 112, respectively (Fig. 3H, purple and blue boxes), where the length of trametinib treated neurons with BRAF overexpression does not differ from the length of shRNA1-6 transfected neurons (224 ± 102 μm, mean ± SD, P = 0.67, t test). Meanwhile, 100 nmol/L of trametinib slightly reduces the total dendritic lengths of control neurons (Fig. 3H, gray and green boxes), where the lengths (in μm, mean ± SD) of neurons treated with or without trametinib are 1016 ± 268 and 1220 ± 355, respectively (P < 0.01, t test). Together, these results indicate that RAS-RAF-MEK signaling can play protective a role in dendritic atrophy caused by dynein malfunction. BRAF protects against DYNC1I1 deficiency induced dendritic atrophy by activating MEK-dependent autophagy Since dynein is required for autophagic clearance (Maday et al., 2012; Kimura et al., 2008; Ravikumar et al., 2005), dynein malfunction may cause protein aggregation and mitochondrial dysfunctions, and result in neuronal atrophy. Thus, we have checked whether RAS-RAF-MEK signaling can enhance autophagic function against atrophy in neuron. Indeed, the number of autophagosomes labeled with GFP-LC3 is dramatically increased in DYNC1I1-knockdown neurons with BRAF overexpression (32 ± 19, Fig. 4C and 4E), which is blocked by MEK inhibitor trametinib (9 ± 14, P < 0.001, t test, Fig. 4D and 4E). Whereas, the numbers of autophagosomes in control neurons treated without (Fig. 4A) or with trametinib, and DYNC1I1-knockdown neurons (Fig. 4B) are 7 ± 6, 5 ± 5, and 6 ± 5, respectively (P = 0.25, one-way ANOVA test, Fig. 4E). These results suggest BRAF can activate MEK-dependent autophagy (Fig. 4C), which is necessary for the protective role of BRAF in DYNC1I1 deficiency caused neuronal atrophy (Fig. 3F–H). Moreover, the decreased number of autophagosomes in the presence of MEK inhibitor trametinib indicates that BRAF increases MEK-dependent autophagic influx and activity.Figure 4 BRAF overexpression protects against DYNC1I1 deficiency induced dendritic atrophy by activating MEK-dependent autophagy. (A–D) BRAF overexpression enhances protective autophagy, which is MEK dependent. Primary hippocampal neurons are co-transfected GFP-LC3 with blank vector, shRNA1-6, shRNA1-6 and BRAF treated without or with 100 nmol/L of MEK inhibitor trametinib at DIV6, cultured additional 5 days and imaged at DIV11. Autophagosomes (green puncta) are labeled with GFP-LC3. All photos of GFP-LC3 are imaged with confocal microscope and further deconvolved for clarity. The scale bars represent 2 μm. (E) Scatterplots with boxplots show the number distribution of GFP-LC3 puncta in the soma of neurons transfected with control vector treated without (n = 41) or with 10 nmol/L of trametinib (n = 22), shRNA1-6 (n = 48), shRNA1-6 co-transfected with BRAF treated without (n = 39) or with 100 nmol/L of trametinib (n = 44). (F–I) Lysosomal protease inhibitors (E64D and pepstatin A) can inhibit the protective role of BRAF in dendritic atrophy caused by DYNC1I1 knockdown with shRNA1-6. Neurons are transfected and treated without or with 1 μmol/L of E64D and pepstatin A at DIV6, cultured additional 5 days and imaged at DIV11.The scale bars represent 20 μm. (J) Scatterplots with boxplots show the total dendritic length distribution of neurons transfected with control vector treated without (n = 29, gray) or with 1 μmol/L of E64D and pepstatin A (n = 30, green), shRNA1-6 (n = 30, red), shRNA1-6 co-transfected with BRAF treated without (n = 30, purple) or with 1 μmol/L of E64D and pepstatin A (n = 30, blue). *, P < 0.001 Additionally, to test whether the activity of increased autophagosomes is necessary for BRAF to play protective role in neuronal atrophy, we used lysosomal protease inhibitors (E64D and pepstatin A) to block autophagic clearance (Klionsky et al., 2012). As shown in Fig. 4F–J, blocking autophagic clearance inhibits the protective role of BRAF in neuronal atrophy. Again, BRAF shows robust role of protection in dendritic atrophy caused by DYNC1I1 knockdown. The total dendritic lengths (in μm, mean ± SD) of DYNC1I1-knockdown neurons with or without BRAF overexpression are 910 ± 322 and 233 ± 101, respectively (Fig. 4F–J). The protective role of BRAF in neuronal atrophy is blocked by 1 μmol/L of E64D and pepstatin A treatment after transfection (218 ± 135 μm, mean ± SD, P < 0.001 compared to BRAF protected neurons, P = 0.62 compared to shRNA1-6 transfected neurons, t test, Fig. 4I and 4J). Thus, the results demonstrate that the activity of increased autophagosomes is necessary for the protective role of BRAF in neuronal atrophy. 1 μmol/L of inhibitors alone (E64D and pepstatin A) slightly reduces the total dendritic lengths of control neurons (Fig. 4J, green boxes), where the lengths (in μm, mean ± SD) of neurons treated with or without inhibitors are 1097 ± 312 and 1314 ± 317, respectively (P = 0.01, t test, Fig. 4J). The protective role of enhancing autophagic activity is also demonstrated with mTOR inhibitor rapamycin (Fig. S2), which also increase the number of autophagosomes in neurons (Fig. S3). Lysosomal inhibitors can also increase the number of autophagosomes by blocking the function of lysosomal protease (Fig. S3F and S3G), but result in slightly reduced dendritic lengths (Fig. 4J). In addition, lysosomal inhibitors can slightly but not significantly increase the number of autophagosomes in BRAF protected shRNA1-6 transfected neurons (Fig. S3G), and block the protective role of BRAF in neuronal atrophy, which supports that lysosomal inhibitors block the activity of lysosomal enzymes to increase the number of autophagosomes in neurons. Whereas, BRAF increases the number of autophagosomes (Fig. 4C and 4E), enhances the activity of autophagy, and protects neuronal atrophy (Figs. 2D, 3F, and 4F), which are both MEK dependent and blocked by MEK inhibitor trametinib (Figs. 4D and 3G). Together, we reveal the protective function of RAS-RAF signaling in neuronal atrophy is mediated by activating MEK-dependent autophagy, which is protective and helpful against neuronal atrophy by cleaning protein aggregations and dysfunctional organelles, such as mitochondria. BRAF also protects against dendritic atrophy caused by mitochondrial dysfunction in primary hippocampal neurons Considering dynein is the major motor protein for cargo transport in dendrites (Kapitein et al., 2010), especially for mitochondrial transport, we have checked the motility of dendritic mitochondria (Fig. 5A–C). As shown in the kymograph of Fig. 5A (See also supplemental video 1), dendritic mitochondria show active motility in control neuron, while only a few dendritic mitochondria are motile in DYNC1I1-knockdown neuron (Fig. 5B and supplemental video 2). To quantify the motility of dendritic mitochondria, motile and stationary dendritic mitochondria are separated with fast Fourier transform (FFT) algorithm (Fig. 5A and 5B). Considering motile mitochondria is prone to be ambiguous and overestimated due to the vague definition of motile mitochondria and the effect of photobleaching, we have quantified and compared the absolutely stationary mitochondria (The third row in Fig. 5A and 5B). The results demonstrate that stationary pool of dendritic mitochondria (75% ± 6%, mean ± SD in percentage) in DYNC1I1-knockdown neurons is significantly increased (versus 68% ± 9% in control neurons, P < 0.001, t test, Fig. 5A–C).Figure 5 BRAF overexpression also protects against dendritic atrophy caused by mitochondrial dysfunction in primary hippocampal neurons. (A and B) Dync1i1 Knockdown decreases dendritic mitochondrial motility in primary hippocampal neurons. For clarity and quantification, motile and stationary dendritic mitochondria are separated with fast Fourier transform (FFT) algorithm. Kymographs show active mitochondrial motility in control neuron (A, Ctrl) and a few motile dendritic mitochondria in DYNC1I1-knockdown neuron (B, shRNA1-6). In kymographs, vertical lines represent stationary mitochondria, and slant lines or curves indicate motile mitochondria. Neurons are co-transfected at DIV6 with DsRed-mito and control vector (A) or shRNA1-6 (B), imaged at DIV11 and sequentially analyzed. The scale bars represent 10 μm (x axis) and 100 seconds (y axis). (C) Scatterplots with boxplots show the percentage distribution of stationary dendritic mitochondria in neurons transfected with control vector (n = 32) or shRNA1-6 (n = 32). (D–G) BRAF overexpression also protects against dendritic atrophy caused by mitochondrial dysfunction. Neurons are transfected with control vector (Ctrl) or BRAF treated with control solution or with 50 nmol/L of specific mitochondrial function inhibitor TMRE at DIV6, cultured additional 5 days and imaged at DIV11. The scale bars represent 20 μm. (H) Scatterplots with boxplots show the total dendritic length distribution of neurons transfected with control vector treated with vehicle (Ctrl, n = 53, gray), with 50 nmol/L of TMRM (n = 31, blue) or 50 nmol/L of TMRE (n = 63, red), or transfected with BRAF treated without (n = 24, green) or with 50 nmol/L of TMRE (n = 32, purple). *, P < 0.001 Illustrated in Fig. 5A–C, DYNC1I1 knockdown decreases mitochondrial motility, which may result in accumulation of dysfunctional mitochondria (Maday et al., 2012; Xie et al., 2015). In turn, the accumulation of dysfunctional mitochondria might be an important or major factor to cause neuronal atrophy. To inspect this, we used a specific inhibitor (tetramethylrhodamine ethyl ester, TMRE) of mitochondrial function (Scaduto and Grotyohann, 1999), which is a frequently used mitochondrial marker. As shown in Fig. 5D–J, compared to the total dendritic lengths (in μm, mean ± SD) of control neurons treated with relevant DMSO solution (944 ± 278, Fig. 5D and 5H) or a non-inhibitory mitochondrial marker (tetramethylrhodamine methyl ester, TMRM) at 50 nmol/L (Scaduto and Grotyohann 1999) (818 ± 234, Fig. 5H, blue box), neurons treated with 50 nmol/L of TMRE show shortened dendritic lengths (421 ± 229, P < 0.001, t test, Fig. 5E and 5H, red box). Thus, mitochondrial dysfunction is sufficient to result in neuronal atrophy. Moreover, since both dynein and mitochondria are therapeutic targets in neurodegeneration (Eschbach and Dupuis, 2011; Banks and Fisher, 2008; Moreira et al., 2010), we are also wondering whether BRAF can protect against dendritic atrophy induced by mitochondrial dysfunction. For the total dendritic lengths (in μm, mean ± SD) of BRAF overexpressed neurons, BRAF promotes the dendritic growth of non-treated neurons (1570 ± 229, Fig. 5F and 5H), similar as shown in Fig. 2B and 2F; BRAF overexpressed neurons are somewhat resistant to TMRE treatment, and protect dendrites from atrophy induced by functional inhibition of mitochondria (775 ± 607, Fig. 5G and 5H, purple box). The total dendritic lengths of BRAF overexpressed neurons are significantly improved (P < 0.001, t test) compared to the lengths of TMRE treated control neurons (Fig. 5H red box). Together, these results based on above cellular models indicate that BRAF has a general protective role in neuronal atrophy caused by dynein malfunction or mitochondrial impairment. Discussion Dynein is not only the major motor protein for dendritic cargo transport (Kapitein et al., 2010) and dendritic mitochondrial motility (Fig. 5), but also plays important roles in axonal structures and polarities (Song et al., 2009; Zheng et al., 2008). DYNC1I1 knockdown causes abnormal ER distribution in axon (Fig. S4), which is in line with the reported gatekeeper function of dynein (Song et al., 2009; Zheng et al., 2008). As an important cargo binding subunit of dynein (Ha et al., 2008), dynein intermediate chains (DYNC1I1 and DYNC1I2) are necessary for the functional integrity of dynein (Figs. 5 and S4) and the maintenance of neuronal architecture (Fig. 1). In addition, functions of both DYNC1I1 and DYNC1I2 are regulated by phosphorylation. The phosphorylation of DYNC1I1 serine 83 or DYNC1I2 serine 84 inhibits dynein intermediate chain binding to dynactin or paxillin (Vaughan et al., 2001; Rosse et al., 2012); while the phosphorylation of DYNC1I1 serine 80 or DYNC1I2 serine 81 is MAP kinase ERK1/2 dependent and can strengthen dynein activity in signaling cargos transport (Mitchell et al., 2012). In our study, we cannot exclude that the protective role of RAS-RAF-MEK signaling in neuronal atrophy may be partially due to upregulating the phosphorylation of DYNC1I2 serine 81 and thus partially compensate the dynein malfunction by DYNC1I1 knockdown, although we fail to detect any change of phosphorylation of DYNC1I2 serine 81 for control and DYNC1I1 knockdown (data not shown) in primary cultured hippocampal neuron, which may be explained by the low transfection efficiency of primary neuron and the expression of DYNC1I2C in glia. Protective role of RAS-RAF-MEK axis in neuronal atrophy caused by dynein malfunction In corroboration with our results, transgenic activation of RAS in neurons promotes neuronal growth and protects from lesion-induced degeneration (Heumann et al., 2000), but the mechanism is unknown; in addition, selective activation of BRAF can provide neuroprotection both in vitro and in vivo although it is not MEK dependent (Chin et al., 2004). Here, we identify that RAS-RAF signaling protects neurons against dendritic atrophy arisen from dynein malfunction, which relies on MEK-dependent autophagy (Fig. 4). MEK-dependent autophagy can be either protective or destructive autophagy (Wang et al., 2009) in non-neuronal cells. The role of MEK-dependent autophagy in neuron is unknown yet. Trametinib treatment and BRAF in control neurons doesn’t affect the numbers of autophagosomes (Figs. 4 and Fig. S3E), which suggests that MEK-dependent autophagy has minor role in neuron under normal condition. However, we demonstrate that RAS-RAF pathway activates protective autophagy in primary hippocampal neuron with DYNC1I1 knockdown (Fig. 4), which is favorable for structural and functional integrity of neuron. Both RAS-RAF and PI3K-AKT pathways are necessary for dendritic morphogenesis (Kumar et al., 2005) and neuron survival (Mazzoni et al. 1999). Additionally, PI3K-AKT signaling has a major role in antiapoptotic function (Brunet et al., 2001). Here, we demonstrate that only RAS-RAF-MEK pathway has the protective role in dendritic atrophy cause by DYNC1I1 knockdown (Fig. 6). Interestingly, it is recently reported that inhibition of RAS-MAPK pathway has a role in longevity of Drosophila (Slack et al., 2015). However, to avoid compromising brain function of higher animals (Fig. 2–4), our results raise caution about inhibiting RAS-RAF-MEK signaling for longevity. Thus, it deserves further studies on the roles of RAS-RAF-MEK signaling in longevity and neuronal atrophy with models of higher animals, such as rodents or monkeys.Figure 6 Schematic summary of general roles of RAS-RAF-MEK pathway in protecting dendritic atrophy caused by dynein malfunction or mitochondrial dysfunction A more general role of RAS-RAF-MEK signaling in neuronal atrophy In the past decades, dominated researches and drug developments have focused or based on cholinergic hypothesis or the amyloid cascade hypothesis for Alzheimer’s disease (Becker et al., 2008; Craig et al., 2011; Karran et al., 2011). However, the continual failures of clinical trials for neurodegenerative disorders suggest that it is important and necessary to think about new models and strategies, such as motor protein dynein malfunction and mitochondrial dysfunction, both of which are in the spotlight of neurodegeneration therapy, ALS in particular (Eschbach and Dupuis, 2011; Banks and Fisher, 2008; Moreira et al., 2010; Payne and Chinnery, 2015). Here, based on the cellular models of neuronal atrophy, we demonstrate that BRAF has a general protective role in neuronal atrophy caused by dynein malfunction or mitochondrial impairment (Fig. 6). Dynein malfunction by DYNC1I1 knockdown decreases mitochondrial motility (Fig. 4), which may augment mitochondrial pathology. This study provides some missing linkages among dynein, mitochondria, and atrophy/degeneration. These findings about the RAS-RAF-MEK pathway for neuronal atrophy protection provide a therapeutic intervention signaling against the on-sets of neuronal atrophy caused by dynein malfunction or mitochondrial impairment. Importantly, neuronal atrophy is not only the hallmark of neurodegeneration, such as ALS, but also a continuous process in adult brain with increasing age (Pakkenberg et al., 2003; Fox and Schott, 2004). Therefore, more speculatively, it might even imply a potential target to ameliorate memory decline due to age-related brain atrophy. Materials and Methods Additional information can be found in the supplemental materials and methods. Plasmids and shRNA The isoforms of Dync1i1 are cloned into the vector pEGFP-N1 with restriction sites EcoRI and BamHI (NEB). The shRNA1-6 oligonucleotide for Dync1i1 (5′-GCATGGAGCTGGTGTACAA-3′) and control shRNA1-4 oligonucleotide (5′-GCTGGAGCCAACCTTTCTT-3′) are constructed into the modified pSUPER vector (Kanr, EGFP expression) with BglII and HindIII restriction sites. The calcium phosphate method is used for transfection. Data and statistical analysis Imaging is performed using an Olympus FV1000 confocal microscope with a 40×/0.95 objective (Olympus) for dendritic length imaging and a 60×/1.2w objective (Olympus) for high-resolution imaging at room temperature or mitochondrial motility study at 37°C. The sholl analysis of neuritic morphology and complexity is performed using Fiji/ImageJ software with Simple Neurite Tracer. Photos of GFP-LC3 labeled autophagosomes are imaged with confocal microscope and further deconvolved using the SharpStack Total Deconvolution function of Image-Pro Plus (Media Cybernetics). The numbers of autophagosomes are counted by triple-blinded analysis. Data sets in sholl-analysis graphs are presented as mean ± SEM from repeats in at least three independent experiments, while data sets in text are presented as mean ± SD. Scatterplots with boxplots are plotted with R software. For mitochondrial motility study, motile and stationary dendritic mitochondria are separated with an immobile filter, which is computed with FFT algorithm. When comparing multiple samples in a group, one-way ANOVA test is used. When comparing two samples, two-tailed Student’s t test is used. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary material 1 (PDF 634 kb) Supplementary material 2 (MOV 1419 kb) Supplementary material 3 (MOV 779 kb) Zhi-Dong Liu and Su Zhang have contributed equally to this work. Acknowledgements We thank the following people for their help: Dr. Huai-Bin Cai for critical reading of the manuscript; Dr. Kevin Pfister, Dr. Yan Chen, Dr. Zheng Li, and Dr. Zun-Ji Ke for sharing experimental materials. This work was supported by the National Natural Science Foundation of China (Grant No. 31171369), and partially supported by the National Basic Research Program (973 Program) (Nos. 2011CB910903 and 2010CB912001), Chinese Academy of Sciences (Hundred Talents Program and 2009OHTP10). Abbreviations AD, Alzheimer’s disease; ALS, amyotrophic lateral sclerosis; HD, Huntington’s disease; MAPK, mitogen activated protein kinase; PD, Parkinson’s disease; PI3K, phosphoinositide 3-kinase; PKB/AKT, Protein kinase B; TMRE, tetramethylrhodamine ethyl ester; TMRM, tetramethylrhodamine methyl ester. Compliance with Ethical Standards Zhi-Dong Liu, Su Zhang, Jian-Jin Hao, Tao-Rong Xie, and Jian-Sheng Kang declare that they have no conflict of interest. All institutional and national guidelines for the care and use of laboratory animals were followed. Author Contributions Z.-D. L and S. Z designed and conducted the experiments and manuscript writing. J.-J. H and T.-R. X did biochemistry experiments and data analysis. J.-S. K developed the idea, directed the study and wrote the paper. All authors participated in discussions. ==== Refs References Banks GT Fisher EM Cytoplasmic dynein could be key to understanding neurodegeneration Genome Biol 2008 9 214 10.1186/gb-2008-9-3-214 18373888 Becker RE Greig NH Giacobini E Why do so many drugs for Alzheimer’s disease fail in development? Time for new methods and new practices? J Alzheimers Dis 2008 15 303 325 18953116 Boylan KLM Hays TS The gene for the intermediate chain subunit of cytoplasmic dynein is essential in Drosophila Genetics 2002 162 1211 1220 12454067 Brunet A Datta SR Greenberg ME Transcription-dependent and -independent control of neuronal survival by the PI3K–Akt signaling pathway Curr Opin Neurobiol 2001 11 297 305 10.1016/S0959-4388(00)00211-7 11399427 Chao MV Neurotrophins and their receptors: a convergence point for many signalling pathways Nat Rev Neurosci 2003 4 299 309 10.1038/nrn1078 12671646 Chen X-J Levedakou EN Millen KJ Wollmann RL Soliven B Popko B Proprioceptive sensory neuropathy in mice with a mutation in the cytoplasmic dynein heavy chain 1 gene J Neurosci 2007 27 14515 14524 10.1523/JNEUROSCI.4338-07.2007 18160659 Chin PC Liu L Morrison BE Siddiq A Ratan RR Bottiglieri T D’Mello SR The c-Raf inhibitor GW5074 provides neuroprotection in vitro and in an animal model of neurodegeneration through a MEK-ERK and Akt-independent mechanism J Neurochem 2004 90 595 608 10.1111/j.1471-4159.2004.02530.x 15255937 Craig LA Hong NS McDonald RJ Revisiting the cholinergic hypothesis in the development of Alzheimer’s disease Neurosci Biobehav Rev 2011 35 1397 1409 10.1016/j.neubiorev.2011.03.001 21392524 Dotti CG Sullivan CA Banker GA The establishment of polarity by hippocampal neurons in culture J Neurosci 1988 8 1454 1468 3282038 Eschbach J Dupuis L Cytoplasmic dynein in neurodegeneration Pharmacol Ther 2011 130 348 363 10.1016/j.pharmthera.2011.03.004 21420428 Fiordalisi JJ Johnson RL II Ülkü AS Der CJ Cox AD Der CJ Balch WE Mammalian expression vectors for Ras family proteins: generation and use of expression constructs to analyze Ras family function Methods in enzymology 2001 San Diego Academic Press 3 36 Fox NC Schott JM Imaging cerebral atrophy: normal ageing to Alzheimer’s disease Lancet 2004 363 392 394 10.1016/S0140-6736(04)15441-X 15074306 Freeman SH Kandel R Cruz L Rozkalne A Newell K Frosch MP Hedley-Whyte ET Locascio JJ Lipsitz L Hyman BT Preservation of neuronal number despite age-related cortical brain atrophy in elderly subjects without Alzheimer disease J Neuropathol Exp Neurol 2008 67 1205 1212 10.1097/NEN.0b013e31818fc72f 19018241 Galabova-Kovacs G Kolbus A Matzen D Meissl K Piazzolla D Rubiolo C Steinitz K Baccarini M ERK and beyond: insights from B-Raf and Raf-1 conditional knockouts Cell Cycle Georget. Tex 2006 5 1514 1518 10.4161/cc.5.14.2981 Ha J Lo KW-H Myers KR Carr TM Humsi MK Rasoul BA Segal RA Pfister KK A neuron-specific cytoplasmic dynein isoform preferentially transports TrkB signaling endosomes J Cell Biol 2008 181 1027 1039 10.1083/jcb.200803150 18559670 Hafezparast M Klocke R Ruhrberg C Marquardt A Ahmad-Annuar A Bowen S Lalli G Witherden AS Hummerich H Nicholson S Mutations in dynein link motor neuron degeneration to defects in retrograde transport Science 2003 300 808 812 10.1126/science.1083129 12730604 Heumann R Goemans C Bartsch D Lingenhöhl K Waldmeier PC Hengerer B Allegrini PR Schellander K Wagner EF Arendt T Transgenic activation of Ras in neurons promotes hypertrophy and protects from lesion-induced degeneration J Cell Biol 2000 151 1537 1548 10.1083/jcb.151.7.1537 11134081 Holzbaur ELF Vallee RB Dyneins: molecular structure and cellular function Annu Rev Cell Biol 1994 10 339 372 10.1146/annurev.cb.10.110194.002011 7888180 Kapitein LC Schlager MA Kuijpers M Wulf PS van Spronsen M MacKintosh FC Hoogenraad CC Mixed microtubules steer dynein-driven cargo transport into dendrites Curr Biol 2010 20 290 299 10.1016/j.cub.2009.12.052 20137950 Karran E Mercken M Strooper BD The amyloid cascade hypothesis for Alzheimer’s disease: an appraisal for the development of therapeutics Nat. Rev. Drug Discov. 2011 10 698 712 10.1038/nrd3505 21852788 Kimura S Noda T Yoshimori T Dynein-dependent movement of autophagosomes mediates efficient encounters with lysosomes Cell Struct Funct 2008 33 109 122 10.1247/csf.08005 18388399 Klionsky DJ Abdalla FC Abeliovich H Abraham RT Acevedo-Arozena A Adeli K Agholme L Agnello M Agostinis P Aguirre-Ghiso JA Guidelines for the use and interpretation of assays for monitoring autophagy Autophagy 2012 8 445 544 10.4161/auto.19496 22966490 Kumar V Zhang M-X Swank MW Kunz J Wu G-Y Regulation of dendritic morphogenesis by Ras–PI3K–Akt–mTOR and Ras–MAPK signaling pathways J Neurosci 2005 25 11288 11299 10.1523/JNEUROSCI.2284-05.2005 16339024 Lalli G Regulation of neuronal polarity Exp Cell Res 2014 328 267 275 10.1016/j.yexcr.2014.07.033 25107381 Lipka J Kuijpers M Jaworski J Hoogenraad CC Mutations in cytoplasmic dynein and its regulators cause malformations of cortical development and neurodegenerative diseases Biochem Soc Trans 2013 41 1605 1612 10.1042/BST20130188 24256262 Maday S Wallace KE Holzbaur ELF Autophagosomes initiate distally and mature during transport toward the cell soma in primary neurons J Cell Biol 2012 196 407 417 10.1083/jcb.201106120 22331844 Mazzoni IE Saïd FA Aloyz R Miller FD Kaplan D Ras regulates sympathetic neuron survival by suppressing the p53-mediated cell death pathway J Neurosci 1999 19 9716 9727 10559381 Mitchell DJ Blasier KR Jeffery ED Ross MW Pullikuth AK Suo D Park J Smiley WR Lo KW-H Shabanowitz J Trk activation of the ERK1/2 kinase pathway stimulates intermediate chain phosphorylation and recruits cytoplasmic dynein to signaling endosomes for retrograde axonal transport J Neurosci 2012 32 15495 15510 10.1523/JNEUROSCI.5599-11.2012 23115187 Moreira PI Zhu X Wang X Lee H Nunomura A Petersen RB Perry G Smith MA Mitochondria: a therapeutic target in neurodegeneration Biochim Biophys Acta 2010 1802 212 220 10.1016/j.bbadis.2009.10.007 19853657 Myers KR Lo KW-H Lye RJ Kogoy JM Soura V Hafezparast M Pfister KK Intermediate chain subunit as a probe for cytoplasmic dynein function: biochemical analyses and live cell imaging in PC12 cells J Neurosci Res 2007 85 2640 2647 10.1002/jnr.21213 17279546 Pakkenberg B Pelvig D Marner L Bundgaard MJ Gundersen HJG Nyengaard JR Regeur L Aging and the human neocortex Exp Gerontol 2003 38 95 99 10.1016/S0531-5565(02)00151-1 12543266 Payne BAI Chinnery PF Mitochondrial dysfunction in aging: much progress but many unresolved questions Biochim. Biophys. Acta BBA - Bioenerg. 2015 1847 1347 1353 10.1016/j.bbabio.2015.05.022 Pfister KK Fisher EMC Gibbons IR Hays TS Holzbaur ELF McIntosh JR Porter ME Schroer TA Vaughan KT Witman GB Cytoplasmic dynein nomenclature J Cell Biol 2005 171 411 413 10.1083/jcb.200508078 16260502 Ravikumar B Acevedo-Arozena A Imarisio S Berger Z Vacher C O’Kane CJ Brown SDM Rubinsztein DC Dynein mutations impair autophagic clearance of aggregate-prone proteins Nat Genet 2005 37 771 776 10.1038/ng1591 15980862 Regeur L Badsberg Jensen G Pakkenberg H Evans SM Pakkenberg B No global neocortical nerve cell loss in brains from patients with senile dementia of Alzheimer’s type Neurobiol Aging 1994 15 347 352 10.1016/0197-4580(94)90030-2 7936059 Rosse C Boeckeler K Linch M Radtke S Frith D Barnouin K Morsi AS Hafezparast M Howell M Parker PJ Binding of dynein intermediate chain 2 to paxillin controls focal adhesion dynamics and migration J Cell Sci 2012 125 3733 3738 10.1242/jcs.089557 22553211 Scaduto RC Jr Grotyohann LW Measurement of mitochondrial membrane potential using fluorescent rhodamine derivatives Biophys J 1999 76 469 477 10.1016/S0006-3495(99)77214-0 9876159 Schott JM Fox NC Frost C Scahill RI Janssen JC Chan D Jenkins R Rossor MN Assessing the onset of structural change in familial Alzheimer’s disease Ann Neurol 2003 53 181 188 10.1002/ana.10424 12557284 Slack C Alic N Foley A Cabecinha M Hoddinott MP Partridge L The Ras-Erk-ETS-signaling pathway is a drug target for longevity Cell 2015 162 72 83 10.1016/j.cell.2015.06.023 26119340 Song A Wang D Chen G Li Y Luo J Duan S Poo M A selective filter for cytoplasmic transport at the axon initial segment Cell 2009 136 1148 1160 10.1016/j.cell.2009.01.016 19268344 Soo KY Farg M Atkin JD Molecular motor proteins and amyotrophic lateral sclerosis Int J Mol Sci 2011 12 9057 9082 10.3390/ijms12129057 22272119 Swaab DF Hofman MA Lucassen PJ Salehi A Uylings HBM Neuronal atrophy, not cell death, is the main hallmark of Alzheimer’s disease Neurobiol Aging 1994 15 369 371 10.1016/0197-4580(94)90037-X 7936066 Vaughan PS Leszyk JD Vaughan KT Cytoplasmic dynein intermediate chain phosphorylation regulates binding to dynactin J Biol Chem 2001 276 26171 26179 10.1074/jbc.M102649200 11340075 Wang J Whiteman MW Lian H Wang G Singh A Huang D Denmark T A non-canonical MEK/ERK signaling pathway regulates autophagy via regulating beclin 1 J Biol Chem 2009 284 21412 21424 10.1074/jbc.M109.026013 19520853 Xie Y Zhou B Lin M-Y Wang S Foust KD Sheng Z-H Endolysosomal deficits augment mitochondria pathology in spinal motor neurons of asymptomatic fALS mice Neuron 2015 87 355 370 10.1016/j.neuron.2015.06.026 26182418 Yamaguchi T Kakefuda R Tajima N Sowa Y Sakai T Antitumor activities of JTP-74057 (GSK1120212), a novel MEK1/2 inhibitor, on colorectal cancer cell lines in vitro and in vivo Int J Oncol 2011 39 23 31 21523318 Zheng Y Wildonger J Ye B Zhang Y Kita A Younger SH Zimmerman S Jan LY Jan YN Dynein is required for polarized dendritic transport and uniform microtubule orientation in axons Nat Cell Biol 2008 10 1172 1180 10.1038/ncb1777 18758451 Zhong J Li X McNamee C Chen AP Baccarini M Snider WD Raf kinase signaling functions in sensory neuron differentiation and axon growth in vivo Nat Neurosci 2007 10 598 607 10.1038/nn1898 17396120

PMC005xxxxxx/PMC5003792.txt

==== Front Protein CellProtein CellProtein & Cell1674-800X1674-8018Higher Education Press Beijing 30510.1007/s13238-016-0305-2RecollectionOn the ground in Western Africa: from the outbreak to the elapse of Ebola Liu William J. liujun@ivdc.chinacdc.cn National Institute for Viral Disease Control and Prevention, Chinese Center for Disease Control and Prevention, Beijing, China 9 8 2016 9 8 2016 9 2016 7 9 621 623 © The Author(s) 2016 Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.issue-copyright-statement© HEP and Springer 2016 ==== Body I was sitting in the chair of the airplane back to Beijing when I started writing the memoirs to recollect my three journeys to Western Africa. All of the trips were related to the Ebola epidemic and covered different stages, from the outbreak peak to the post-Ebola era. Each journey was taken for a different task, but all of them represented the urgent and specific support from the Chinese Center for Disease Control and Prevention (China CDC) during the Ebola epidemic in Western African. On the August 8th, 2014, the World Health Organization (WHO) declared the Ebola outbreak in West Africa a Public Health Emergency of International Concern (PHEIC) (http://who.int/mediacentre/news/statements/2014/ebola-20140808/en/). Three days later, together with two colleagues, I was on the ground of Conakry (Fig. 1), the capital of Guinea, in which country the first Ebola case was traced back to the end of 2013 (Baize et al., 2014).Figure 1 Training of the local health workers in Guinea to use the freely offered sanitary and biological protection supplies from China As the first international support force during this epidemic, China rapidly reacted to help the three major affected countries, Guinea, Sierra Leone and Liberia, to fight against Ebola. Three experts were sent to each of these countries, respectively, to train the local health workers to use freely offered sanitary and biological protection supplies. The training went smoothly under the active cooperation of local health workers, and the supplies were soon delivered to the Ebola Treatment Units (ETUs) and the diagnostic laboratories in Guinea. This was the first time that China has sent public health specialists to countries outside of Asia to support disease control. In September, 2014, as the Ebola-related support to Western Africa went deeper, additional Chinese specialists were sent to Sierra Leone. Clinical treatment teams in ETUs, Ebola test teams based in a mobile biological safety level 3 laboratory (BSL-3 or P3 lab) and public health training teams working in the field were included. The Deputy Director-General of the China CDC, Dr. George F. Gao, worked in Sierra Leone as the co-team leader of the first team of the mobile laboratory for 2 months (Gao and Feng, 2014). He represented an idol to recruit more young health experts to work on the ground in Africa (Fig. 2), and I was one of them.Figure 2 The Deputy Director-General of the China CDC, Dr. George F. Gao, worked in Sierra Leone as the co-team leader of the first team of the mobile laboratory from September to November, 2014 Aside from the mobile laboratory, taking into consideration a long-term goal to support disease control in Western Africa, the Chinese government has built the first fixed P3 lab in West Africa (Jui, Freetown, Sierra Leone). In the middle of April, 2015, I came back to Africa with the second team for the fixed P3 lab, working as the main operator in the core area for two and a half months. Every time that I opened specimen packages in the hood, I reminded myself that the tube in my hand might contain the Ebola virus at a sky-high titer. Indeed, all of my colleagues recognized that biosafety is the priority of the laboratory testing for Ebola. During this period, Ebola virus genomes were quickly assembled and analyzed by the Chinese deep sequencing platform in Western Africa (Tong et al., 2015). This platform in the field makes it possible to trace the origin and transmission chain of the Ebola viruses soon after any new case is reported (Fig. 3).Figure 3 Ebola virus genomes were assembled and analyzed by the Chinese deep sequencing platform in Freetown. The lead of the second team, Professor Yuelong Shu was discussing the sequencing project with the colleagues With the elapse of new Ebola cases, the Ebola diagnostic laboratories run on international aid were closed one by one. In contrast, China CDC strengthened the laboratory work force to avoid a potential flare-up of the Ebola epidemic. In the end of 2015, I came back to Freetown as the lead of the fourth team for the fixed P3 lab to perform Ebola testing together with nine other Chinese colleagues. The Ebola surveillance strategy in Sierra Leone was a community-based swab test of dead individuals, together with hospital-based blood testing of suspected cases. As the number of specimens to be tested increased, the human resources in the laboratory became insufficient. However, our team accomplished its Ebola testing job and established good coordination with Dr. Abdul Kamara, who is the National Laboratory Services Manager of the Ministry of Health and Sanitation (MOHS), Sierra Leone (Fig. 4).Figure 4 Photograph of Dr. George F. Gao (middle) with Dr. Abdul Kamara (left) and myself (right) during Dr. Gao’s second journey to Sierra Leone In the end of January, 2016, we have helped to diagnose the latest and hopefully the last Ebola case in Sierra Leone. In the meantime, aside from the swab tests of the dead for routine Ebola surveillance, the laboratory have taken part in a project focused on virus persistence in the survivors (Virus Persistence Study, VPS), collaborating with MOHS of Sierra Leone, the WHO and U.S. Centers for Disease Control and Prevention (US-CDC) (Deen et al., 2015). During the post Ebola era, the proper management and care of the Ebola survivors is one of the most important tasks for the control of Ebola. The VPS yielded a good scientific reference for survivor counseling, and this project is also a representative example for international cooperation for disease control in Africa. In the post-Ebola era, the international support of public health to Western Africa should not be diminished but strengthened. The risk of a flare-up of Ebola still exists (Wong et al., 2016), considering the unknown reservoir of Ebola viruses that may persist in animal hosts. Furthermore, the disease surveillance capacity of many African countries is yet unbelievable  weak, and the risk of importing yellow fever and Zika virus also exists. Thus, China CDC quickly reacted to this global health situation and empowered the fixed P3 lab with the capacity to test for yellow fever virus and Zika virus in West Africa in March, 2016. In addition to the direct support for disease control, the fixed P3 lab is responsible for training local specialists for disease control, which is the foundation of long-term support for Africa. In the past year, several Sierra Leonian specialists have received training in the laboratory, including theory and practice in the fixed P3 lab and field training in China. They now work as the team members together with Chinese colleagues, which facilitated the establishment of a long-term co-work strategy in the laboratory. When I was invited to the wedding of one of my Sierra Leonian colleagues, Mr. Gerald Bagura, he mentioned that the team in the laboratory worked like a family (Fig. 5), which was the highest praise for the teamwork in the fixed P3 lab.Figure 5 The fourth Ebola test team from China working in the fixed P3 lab together with their Sierra Leonian colleagues From its commencement in February 2015, the fixed P3 lab has been playing an important role in virus detection and training the local health workforce. Based on the previous contribution and the current capacity of the laboratory, the MOHS of Sierra Leone authorized the designation of the laboratory as the “National Reference Laboratory for Viral Hemorrhagic Fevers” and the “National Training Center for Virus Detection and Biosafety” in the end of June, 2016, before our team left the country. Dr. Gao was invited back to Freetown to attend the unveiling ceremony (Fig. 4). He gave a speech to emphasize the developing orientation of the P3 lab. Dr. Gao also talked with the president of Sierra Leone, Ernest Bai Koroma, about the possibility for China CDC to strengthen our support of the research on and pre-warning of tropical diseases. Supporting the people of Western Africa during the Ebola epidemic is the first step of the action ‘moving the disease control frontline onto the “battlefield” anywhere in the world’ of the China CDC. In the future, additional young scientists and specialists in disease control are needed to work on the ground of Africa. I am proud to have witnessed this process and contributed a little based on what I learned over the past few years. The trip from public health to global health has just begun. I would like to thank Dr. George F. Gao, Dr. Guizhen Wu, Dr. Abdul Kamara and Dr. Rong Wei for their great support during my three visits of Western Africa. I am also grateful for the kind help from Dr. Ming Chen and Dr. Hao Cheng on the preparation of this manuscript. ==== Refs References Baize S Pannetier D Oestereich L Rieger T Koivogui L Magassouba N Soropogui B Sow MS Keita S De Clerck H Emergence of Zaire Ebola virus disease in Guinea N Engl J Med 2014 371 1418 1425 10.1056/NEJMoa1404505 24738640 Deen GF, Knust B, Broutet N, Sesay FR, Formenty P, Ross C, Thorson AE, Massaquoi TA, Marrinan JE, Ervin E et al (2015) Ebola RNA persistence in semen of ebola virus disease survivors—preliminary report. N Engl J Med. doi:10.1056/NEJMoa1511410 Gao GF Feng Y On the ground in Sierra Leone Science 2014 346 666 10.1126/science.346.6209.666 25359978 Tong YG Shi WF Liu D Qian J Liang L Bo XC Liu J Ren HG Fan H Ni M Genetic diversity and evolutionary dynamics of Ebola virus in Sierra Leone Nature 2015 524 93 96 10.1038/nature14490 25970247 Wong G Gao GF Qiu X Can Ebola virus become endemic in the human population? Protein Cell 2016 7 4 6 10.1007/s13238-015-0231-8 26676470

PMC005xxxxxx/PMC5003802.txt

==== Front Cell Stress ChaperonesCell Stress ChaperonesCell Stress & Chaperones1355-81451466-1268Springer Netherlands Dordrecht 71010.1007/s12192-016-0710-8Original PaperCopy number variations of genes involved in stress responses reflect the redox state and DNA damage in brewing yeasts Adamczyk Jagoda 1Deregowska Anna 12Skoneczny Marek 3Skoneczna Adrianna 4Natkanska Urszula 3Kwiatkowska Aleksandra 1Rawska Ewa 1Potocki Leszek 1Kuna Ewelina 1Panek Anita 1Lewinska Anna +48-1778-55403+48-1787-21265alewinska@o2.pl 5Wnuk Maciej +48-1787-23711+48-1787-23711mawnuk@gmail.com 11 Department of Genetics, University of Rzeszow, Rejtana 16C, 35-959 Rzeszow, Poland 2 Postgraduate School of Molecular Medicine, Medical University of Warsaw, Warsaw, Poland 3 Department of Genetics, Institute of Biochemistry and Biophysics, Polish Academy of Sciences, Warsaw, Poland 4 Laboratory of Mutagenesis and DNA Repair, Institute of Biochemistry and Biophysics, Polish Academy of Sciences, Warsaw, Poland 5 Department of Biochemistry and Cell Biology, University of Rzeszow, Zelwerowicza 4, 35-601 Rzeszow, Poland 14 6 2016 14 6 2016 9 2016 21 5 849 864 19 3 2016 26 5 2016 3 6 2016 © The Author(s) 2016 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.The yeast strains of the Saccharomyces sensu stricto complex involved in beer production are a heterogeneous group whose genetic and genomic features are not adequately determined. Thus, the aim of the present study was to provide a genetic characterization of selected group of commercially available brewing yeasts both ale top-fermenting and lager bottom-fermenting strains. Molecular karyotyping revealed that the diversity of chromosome patterns and four strains with the most accented genetic variabilities were selected and subjected to genome-wide array-based comparative genomic hybridization (array-CGH) analysis. The differences in the gene copy number were found in five functional gene categories: (1) maltose metabolism and transport, (2) response to toxin, (3) siderophore transport, (4) cellular aldehyde metabolic process, and (5) L-iditol 2-dehydrogenase activity (p < 0.05). In the Saflager W-34/70 strain (Fermentis) with the most affected array-CGH profile, loss of aryl-alcohol dehydrogenase (AAD) gene dosage correlated with an imbalanced redox state, oxidative DNA damage and breaks, lower levels of nucleolar proteins Nop1 and Fob1, and diminished tolerance to fermentation-associated stress stimuli compared to other strains. We suggest that compromised stress response may not only promote oxidant-based changes in the nucleolus state that may affect fermentation performance but also provide novel directions for future strain improvement. Electronic supplementary material The online version of this article (doi:10.1007/s12192-016-0710-8) contains supplementary material, which is available to authorized users. Keywords Brewing yeastsGenomeArray-CGHDNA damageRedox equilibriumRegional Operational Programme of Subcarpathia VoivodeshipWND-RPPK-01.03.00-18-038/13Wnuk Maciej issue-copyright-statement© Cell Stress Society International 2016 ==== Body Introduction Beer, one of the most common alcoholic beverages consumed worldwide, is produced by the fermentation of sugars into alcohol by the action of yeasts. Brewing yeasts are typically divided into two groups, ale brewing yeasts and lager brewing yeasts, according to their use for the production of ale and lager beers, respectively (Gibson and Liti 2015; Kodama et al. 2006; Wendland 2014). They differ in sugar utilization, flocculation, off-flavor production, and overall fermentation performance (Gibson and Liti 2015; Kodama et al. 2006; Wendland 2014). Ale yeasts are called top-fermenting yeasts because of their tendency to become buoyant after flocculation and rise to the surface of fermenting wort, whereas lager yeasts are cryotolerant and have a tendency to sediment after flocculation and sink to the bottom of the fermentation vessel and are named bottom-fermenting yeasts (Gibson and Liti 2015; Kodama et al. 2006; Wendland 2014). Ale beer has been produced for thousands of years and is usually brewed at 20–30 °C (Legras et al. 2007). However, nowadays, the more popular is lager beer that has been originated in the fifteenth century in Bavaria (Germany) and is usually brewed at low temperatures (5–15 °C) (Gibson and Liti 2015). The nomenclature of brewing yeasts is not consistent that also reflects their genomic diversity (Querol and Bond 2009). Ale strains of diploid nature are thought to be varieties of Saccharomyces cerevisiae, but also there are evidences that they may be hybrids, e.g., hybrids between S. cerevisiae and Saccharomyces kudriavzevii (Gonzalez et al. 2008; Querol and Bond 2009). Lager yeasts have been already classified as Saccharomyces carlsbergensis, Saccharomyces pastorianus, or Saccharomyces uvarum, but actually, they are not species of their own (Wendland 2014). They are interspecific hybrids between S. cerevisiae and the newly discovered and cold-tolerant Saccharomyces eubayanus (Caesar et al. 2007; Dunn and Sherlock 2008; Naumova et al. 2005). The cerevisiae part of lager yeast genome has been postulated to be an ale yeast, e.g., Fosters O-like ale yeast and/or stouts yeast (Dunn and Sherlock 2008; Monerawela et al. 2015). Lager yeasts can be further divided into two genetic groups (group I, Saaz/Carlsberg group and group II, Frohberg group) that differ in fermentation performance, sugar utilization and adaptation to growth at low temperature (Dunn and Sherlock 2008; Gibson et al. 2013). There are conflicting results on the ploidy state of these two groups of lager yeasts (Dunn and Sherlock 2008; Walther et al. 2014). Group I has been reported to be 2n or 3n and group II to be 3n or 4n (Dunn and Sherlock 2008; Querol and Bond 2009; Walther et al. 2014). However, more recently, it has been shown that S. carlsbergensis (group I) is essentially triploid, whereas Weihenstephan 34/70 strain (group II) is (allo)tetraploid (Walther et al. 2014). Aneuploidy and regions with copy number variations have been also revealed in lager yeasts (Bond et al. 2004). The hybrid and dynamic nature of lager yeast genome may provide adaptive potential but could also result in genomic instability adversely modulating fermentation performance and finally affecting the quality of beer. Thus, it seems worthwhile to monitor genetic and genomic features of brewing yeast strains, especially that their genomes are dynamic and may undergo rearrangements and gene amplification in response to stresses experienced during the brewing process (James et al. 2008). In the present study, we have investigated the chromosome profiles of 30 industrial yeast strains used in brewing (n = 29) and cider production (n = 1) and subjected four strains with distinct chromosome patterns to array-based comparative genomic hybridization (array-CGH)-based genome-wide analysis. Regions with variations in gene copy number were revealed, and we found that aryl-alcohol dehydrogenase gene dosage correlated with intracellular redox equilibrium, genetic stability, and the nucleolar state, which may modulate tolerance to stress stimuli during fermentation conditions in brewing yeasts. Materials and methods Chemicals All reagents were obtained from Sigma (Poznan, Poland) unless otherwise specified. Yeast strains and growth conditions All brewing yeast strains used in this work are listed in Table 1.Table 1 Brewing yeast strains used in this study No. Trade name Company 1 Safale S-04 Fermentis 2 Safale US-05 Fermentis 3 Safbrew S-33 Fermentis 4 Safbrew T-58 Fermentis 5 Safbrew WB-06 Fermentis 6 Saflager W-34/70 Fermentis 7 Saflager S-23 Fermentis 8 Belle Saison Belgian Ale Yeast Lallemand 9 Windsor British Ale Yeast Lallemand 10 Munich German Wheat Beer Yeast Lallemand 11 BRY-97 American West Coast Yeast Lallemand 12 Gozdawa American West Coast Yeast Gozdawa 13 Gozdawa Pure Ale Yeast 7 (PAY7) Gozdawa 14 Gozdawa Porter & Kvass (POK V) Gozdawa 15 Gozdawa Bavarian Wheat 11 (BW11) Gozdawa 16 Gozdawa Old German Altbier 9 (OGA9) Gozdawa 17 Gozdawa Classic Belgian Witbier (CBW) Gozdawa 18 Gozdawa Czech Pilsner 18 (CP18) Gozdawa 19 Mauribrew Ale Y514 Mauribrew 20 Mauribrew Weiss Y1433 Mauribrew 21 Coobra Allround Yeast CBF Drinkit 22 Muntons Premium Gold Yeast Muntons 23 Brewferm Top Brewferm 24 Gozdawa Belgian Fruit and Spicy Ale Yeast (BFSAY) Gozdawa 25 Gozdawa Fruit Blanche B1 (FBG1) Gozdawa 26 Gozdawa French Cider G1 (FCG1) Gozdawa 27 Brewferm Blanche Brewferm 28 Wyeast 1056 American Ale Wyeast Laboratories 29 Wyeast 2308 Munich Lager Wyeast Laboratories 30 Wyeast 3068 Weihenstephan Weizen Wyeast Laboratories According to the suppliers’ information provided, all ale and lager brewing yeasts (top- and bottom-fermenting strains, respectively) were classified as Saccharomyces cerevisiae. One cider yeast strain was provided for comparison (no. 26) Yeast from one single colony was grown either on liquid yeast extract peptone dextrose (YPD) medium (1 % w/v Difco Yeast Extract, 2 % w/v Difco Yeast Bacto-Peptone, 2 % w/v dextrose) or on solid YPD medium containing 2 % w/v Difco Bacto agar, at 30 °C. Growth rate and cell viability For the kinetics of growth assay (Lewinska et al. 2011), cells at the logarithmic phase of growth were washed, diluted, suspended in YPD medium, and cultured at 30 °C. Their growth was monitored turbidimetrically at 600 nm in a microplate reader every 2 h during a 10-h period. Cell viability was estimated with a LIVE/DEAD® Yeast Viability Kit (Thermo Fisher Scientific, Poland) using the standard protocol according to the manufacturer’s instructions as described elsewhere (Lewinska et al. 2014a). Briefly, cells at the logarithmic phase of growth were washed and stained with a mixture of FUN® 1 and Calcofluor® White M2R and inspected under an Olympus BX61 fluorescence microscope equipped with a DP72 CCD camera and Olympus CellF software. Typically, a total of 200 cells were used for the analysis. Fluorescence-activated cell sorting-based ploidy analysis The DNA content was measured via flow cytometry as previously described (Deregowska et al. 2015b). Pulsed-field gel electrophoresis Preparation of agarose-embedded yeast DNA and pulsed-field gel electrophoresis (PFGE) separation of yeast DNA were conducted as described elsewhere (Lewinska et al. 2014b). The dendrogram of chromosomal DNA-based similarity was created using Free-Tree software using unweighted pair group method with arithmetic mean (UPGMA) algorithm, Jaccard similarity coefficient, and Java TreeView 1.1.6.r2 (http://jtreeview.sourceforge.net/) (Deregowska et al. 2015b). Array-based comparative genomic hybridization Genomic DNA of selected brewing strains (4, 6, 8, and 9) was labeled with SureTag DNA Labeling Kit and either Cy3- or Cy5-dUTP as previously described (Deregowska et al. 2015b). Briefly, equal amounts of labeled DNA of tested and of the reference laboratory strain (BY4741) were combined and hybridized to Yeast (V2) Gene Expression Microarray, 8x15K using Oligo aCGH Hybridization Kit. All components were supplied by Agilent Technologies Inc. (Santa Clara, CA, USA), and all steps of the experiment were performed according to the manufacturer’s protocols. Following hybridization and washing, the slides were scanned using Axon GenePix 4000B. Feature extraction was conducted using GenePix Pro 6.1 and normalization using Acuity 4.0 (Molecular Devices, Sunnyvale, CA, USA). CGH profiles with superimposed piecewise regression plots to highlight aneuploidies were generated using CGH-Explorer v3.2 (Lingjaerde et al. 2005). The original CGH profiles obtained after the comparison of analyzed strains to BY4741 gave consistently high noise due most probably to genomic DNA sequence differences between BY4741 and brewing strains that influenced the hybridization strength of individual probes. Therefore, to obtain final CGH profiles, the data for each strain were compared to the average of all strains analyzed. Gene analysis after array-based comparative genomic hybridization The analysis of over-representation of functional categories was performed using Cytoscape v. 2.8.2 with BiNGO v. 2.44 plug-in and hypergeometric test using Benjamini and Hochberg false discovery rate (FDR) correction and significance level of 0.05. Cluster analysis The array-CGH data were subjected to complete linkage clustering with Cluster 3.0 software using Euclidean distance similarity metrics (de Hoon et al. 2004) as previously described (Deregowska et al. 2015b). Quantitative reverse transcriptase real-time PCR Yeast cells at the logarithmic phase of growth were harvested and stored at −80 °C until needed. RNA was extracted using a hot acid phenol method. The removal of genomic DNA contamination and complementary DNA (cDNA) synthesis (with 500 ng of total RNA as a template) was performed using RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific) according to the manufacturer’s protocol. To ensure that RNA had no genomic DNA contamination, control PCR reactions were done without prior reverse transcription. Reactions were done using LightCycler 480 Sybr Green Master Mix and LightCycler Real Time PCR System (Roche) according to the manufacturer’s protocol. Quantitative reverse transcriptase real-time PCR (qRT-PCR) primers (Table 2) were analyzed for specificity and efficiency.Table 2 Primers used for qRT-PCR Gene Forward primer Reverse primer ALD2 TGTTACCGTT CCTTTTGGCG TGTGAACTGCTTTTGTTTGAAGATAGGT FIT3 GCAGCAGCAA CACCTGGTCT CAGCAGTGGTGGTTGCAGTG AAD10 CATTGAGGCTTTGAGCATTAAA ACATTTCGGTGAGAAATGAAGG MAL13 TGAAATTAGAAGCATGGAAT AGGTTTAGAAATGGGCAGAG ALG9 CTACCATCAGAACCGCATTC TCCATGATACAGGAGCAAGC ALG9 gene was used as a housekeeping gene. The crossing thresholds (CT) were calculated by the second derivative method using the LightCycler Relative Quantification Software and corrected for PCR efficiency, which was between 1.9 and 2.0. Comet assay Yeast spheroplasts were obtained (Lewinska et al. 2014b), and DNA double-strand breaks (DSBs) were assessed by neutral single-cell microgel electrophoresis (comet assay) as described elsewhere (Dworak et al. 2014). The percentage of tail DNA was used as a parameter of DNA damage. Oxidative stress parameters Intracellular reactive oxygen species (ROS) production was measured using 2′,7′-dichlorodihydrofluorescein diacetate (H2DCF-DA), and superoxide production was measured using dihydroethidium as described elsewhere (Lewinska et al. 2014a). Oxidative DNA damage as a level of 8-hydroxy-2′-deoxyguanosine (8-OHdG, 8-oxo-dG) was measured using Epigentek EpiQuik 8-OHdG DNA Damage Quantification Direct Kit (Gentaur, Poland) as previously described (Deregowska et al. 2015b). Western blotting For WB analysis, whole-cell extracts were prepared according to Lewinska et al. (2014a). The following primary antibodies were used: anti-Rad1p (1:400), anti-Rap1p (1:400), anti-Fob1p (1:200), anti-Nop1p (1:400), anti-Act1p (1:1000), and anti-Tub1p (1:400) (Santa Cruz, Abcam). The respective proteins were detected after incubation with one of the horseradish peroxidase-conjugated secondary antibodies (1:80,000, 1:100,000 or 1:125,000) (Sigma). The chemiluminescence signals were detected with a Clarity™ Western ECL Blotting Substrate (BIORAD) and a G:BOX imaging system (Syngene, Cambridge, UK). Fluorescence in situ hybridization with whole-chromosome painting probes To detect chromosome I, XI, and XII signals, whole-chromosome painting probes (WCPPs) were used as previously and comprehensively described (Wnuk et al. 2015). Briefly, cells were synchronized, fixed, treated with zymolyase, and subjected to fluorescence in situ hybridization (FISH) procedure using WCPPs. Chromosome-specific signals were counted and presented as a percentage of 100 total cell scores. Moreover, to analyze the nucleolar ribosomal DNA (rDNA) content (chromosome XII-specific signals), ImageJ software http://rsbweb.nih.gov/ij/ was used as described elsewhere (Lewinska et al. 2014b). Briefly, the integrated fluorescence density (green channel) that is the sum of all pixel values within the marked area of each cell analyzed and equivalent to the product of the area and mean gray value was evaluated. The integrated fluorescence density is presented in relative fluorescence units (RFUs). Utilization of non-fermentable carbon sources and tolerance to fermentation-associated stress stimuli The spot assay (the semiquantitative measurement of growth/survival) (Lewinska et al. 2011) was used. To analyze the utilization of non-fermentable carbon sources, several dilutions (1 × 107, 1 × 106, 1 × 105, 1 × 104, and 1 × 103 cells/ml) of a yeast exponential phase culture in a volume of 2 μl were used, inoculated on solid YPG medium (1 % w/v Difco Yeast Extract, 2 % w/v Difco Yeast Bacto-Peptone, 2 % v/v glycerol) and YPE medium (1 % w/v Difco Yeast Extract, 2 % w/v Difco Yeast Bacto-Peptone, 2 % v/v ethanol) containing 2 % w/v agar, at 30 °C, and inspected after 48 h. For stress tolerance analysis, yeast cells were grown on standard solid YPD medium in the presence of NaCl, KCl, and sorbitol (0.5, 1, and 1.5 M) and high glucose concentrations (5, 10, and 20 %), ethanol (2.5, 5, and 10 %) or at different temperature conditions (4, 20, 30, 37, and 55 °C). Hydrogen peroxide toxicity was analyzed after 40-min incubation of cells (1 × 107 cells/ml) in the presence of 2, 5, and 10 mM H2O2 and transfer to solid YPD medium. Typically, the cell growth was inspected after 48 h. Yeast cells grown at 4 °C were inspected after 120 h. Statistical analysis The results represent the mean ± SD from at least three independent experiments. Statistical significance was assessed by one-way ANOVA using GraphPad Prism 5 and with the Tukey’s multiple comparison test. Results Electrophoretic karyotyping of brewing yeast strains reveals chromosome pattern variability Firstly, we have evaluated the chromosome profiles of 29 brewing yeast strains both top-fermenting ale strains and bottom-fermenting lager strains that were purchased from multiple suppliers and classified as S. cerevisiae (Table 1, Fig. 1).Fig. 1 The dendrogram of chromosome band-based similarity created after electrophoretic karyotyping of 30 brewing yeast strains (lanes from 1 to 30). The yeast S. cerevisiae chromosome marker YNN295 (BIORAD) is also presented (lane M). Lager strains (strains 6, 7, 18, and 29) are denoted as L, whereas cider strain (strain 26) is denoted as C. Strains 4, 6, 8, and 9 with various chromosome profiles were selected for further analysis (in red frames) (Color figure online) Indeed, after PFGE separation, we were able to observe S. cerevisiae-like chromosome profiles (Fig. 1). In general, the chromosome number of analyzed strains is 16 (Fig. 1). However, some additional bands can be also observed; e.g., an additional band between chromosomes IV and VII was shown for strains 6, 7, 18, and 29 that is a characteristic feature of Saccharomyces bayanus karyotype (Naumov et al. 1992). Perhaps, some of analyzed strains may be considered as hybrids between S. cerevisiae and S. bayanus. Indeed, strains 6, 7, 18, and 29 are bottom-fermenting lager strains, and in general, lager yeasts are natural hybrids between S. cerevisiae and S. eubayanus (Dunn and Sherlock 2008; Walther et al. 2014; Wendland 2014). Some other chromosome variabilities may also reflect increased level of translocations and aneuploidy events and dynamic nature of industrial yeast genomes (Bond et al. 2004; Querol and Bond 2009). Similar chromosome profiles among analyzed strains were revealed using UPGMA clustering; e.g., lager strains were grouped together (Fig. 1). Four strains, namely strains 4, 6, 8, and 9, representing different karyotype profiles were then selected for further analysis (Fig. 1, in red frames). Additionally, one cider yeast strain (strain 26) was included for chromosome comparison (Fig. 1). Differences in growth rate, viability, and ploidy state Secondly, the kinetics of growth of selected brewing yeasts both top-fermenting strains (strains 4, 8, and 9) and bottom-fermenting strain (strain 6) was inspected using standard yeast growth medium (YPD medium) (Fig. 2a).Fig. 2 Growth rate, viability, and ploidy state of selected brewing yeast strains. a Yeast growth was monitored turbidimetrically at 600 nm in a microplate reader every 2 h during a 10 h. Bars indicate SD, n = 6. b Cell viability was estimated with a LIVE/DEAD® Yeast Viability Kit using the standard protocol according to the manufacturer’s instructions. The percentage of live and dead cells is shown. n = 200. c Fluorescence-activated cell sorting (FACS)-based analysis of DNA content of strains 4, 6, 8, and 9. Representative histograms are shown. Diploid, triploid, and tetraploid reference strains are also presented The growth rate of strains 4 and 9 was accelerated compared to the growth rate of strains 6 and 8 in the first 6 h of experiment (Fig. 2a). However, the delay of growth of strains 6 and 8 was overcome in the next 2 h, and after 10 h, the growth yield was comparable between all analyzed strains (Fig. 2a). The viability of cells at the logarithmic phase of growth was similar and ranging from 99.5 to 98 % (Fig. 2b). Fluorescence-activated cell sorting (FACS)-based analysis of DNA content revealed diverse ploidy states of analyzed strains (Fig. 2c). Strains 4 and 9 are tetraploid and strain 8 is diploid. The histogram for strain 6 is more ambiguous and shows some features of the tetraploid reference strain histogram (Fig. 2c). Intracellular redox equilibrium is shifted toward more oxidation state in Saflager W-34/70 strain and, to lesser degree, in Windsor British ale strain We then analyzed intracellular reactive oxygen species (ROS) production and superoxide production in the control growth conditions (Fig. 3).Fig. 3 Intracellular redox state of selected brewing yeast strains. Reactive oxygen species (ROS) production was assessed using H2DCF-DA fluorogenic probe (a), and superoxide production was monitored using dihydroethidium fluorogenic probe (b). The results are presented as relative fluorescence units per minute (RFU/min). Bars indicate SD, n = 4. c The level of 8-hydroxy-2′-deoxyguanosine (8-oxo-dG) was analyzed using ELISA-based assay. Bars indicate SD, n = 3, *p < 0.05 compared to strain 6 (ANOVA and Tukey’s a posteriori test) Redox imbalance was observed in strains 6 and 9 compared to strains 4 and 8 (Fig. 3). Higher level of ROS production of approximately 3.5-fold and 1.8-fold was shown in strains 6 and 9, respectively (Fig. 3a). In contrast, augmented superoxide production was only observed in strain 6 (Fig. 3b). Intracellular superoxide production was higher approximately 2-fold in strain 6 compared to other strains (Fig. 3b). Statistically significant higher level of ROS and superoxide production was observed in strain 6 compared to other strains analyzed (p < 0.05) (Fig. 3a, b). Redox disequilibrium resulted in elevated levels of oxidative DNA damage (the level of 8-hydroxy-2′-deoxyguanosine (8-oxo-dG)) in strains 6 and 9 (Fig. 3c). Increased 8-oxo-dG levels of approximately 60 % were observed in strains 6 and 9 compared to strains 4 and 8 (Fig. 3c). Statistically significant higher level of oxidative DNA damage was noticed in strain 6 compared to strains 4 and 8 (p < 0.05) (Fig. 3c). Subtelomeric regions are sites of the most accented differences in the gene copy number and loci-specific gains and losses The genome of selected strains was further characterized using array-based comparative genomic hybridization (array-CGH) (Figs. 4 and 5).Fig. 4 Analysis of the variability in the gene copy number of strains 4, 6, 8, and 9 using array-CGH. a Array-CGH profiles are shown. Each gray dot represents the value of the log2 ratio for an individual gene. Blue lines were provided to emphasize the most accented differences (DNA losses and gains). b The relatedness of strains analyzed using cluster analysis. Similarity tree is shown (Color figure online) Fig. 5 The divergence of relative abundance of genes as determined by array-CGH represented by standard deviation (SD) of log2 ratio values for each gene in strains 4, 6, 8, and 9. a The summary plot for the whole genome. b Individual plots for each chromosome. Blue dots indicate the SD values for individual genes, the red lines denote the smoother trend calculated by moving average of SD values to expose the genome regions of higher log2 ratio divergence, and green triangles indicate centromere position. Individual plots for mitochondrial genome are also presented (Mt) (Color figure online) The genome of strain 4 of a euploid nature was characterized by decreased number of subtelomeric genes (Fig. 4a). The gains of chromosomes I, III, and VI were observed in strain 8, whereas the losses of chromosomes I, III, VI, and IX were shown in strain 9 (Fig. 4a). The most affected genome was strain 6 compared to the other analyzed genomes with the majority of its chromosomes containing gains and/or losses (Fig. 4a). The gains of fragments of chromosomes II, III, VII, VIII, and XI and the losses of fragments of chromosomes III, V, and X were observed (Fig. 4a). Moreover, some other losses were much more accented, e.g., gross deficiencies of chromosomes I, VI, XII, and XVI (Fig. 4a). In the case of chromosome XII, the lack of whole chromosome and, in the case of chromosome XVI, the lack of distal part of right arm were observed (Fig. 4a). Additionally, array-CGH profiles were used to estimate the level of similarity (relatedness) between selected strains analyzed on the basis of observed genomic diversity (Fig. 4b). As expected, the most variable was lager strain (strain 6) with its own category (Fig. 4b). All three ale strains were grouped together (Fig. 4b). Array-CGH data allowed us also to analyze the diversity of the copy number of individual genes among all brewing yeast strains studied. As a measure of this diversity, we used standard deviation (SD) of log2 ratio values. Figure 5 plots SD of log2 ratio values for all genes. Figure 5a gives an overview of the whole yeast genome, and Fig. 5b shows an expanded view of the same data divided into individual chromosomes. The data points for single genes (blue dots) are overplayed with the red line representing the moving average of the individual data points to visualize greater regions of high diversity. According to these plots, the most evident diversity in the gene copy number was revealed within subtelomeric regions in almost all analyzed chromosomes and also within short intrachromosomal regions of chromosomes IV, IX, XII, and XIII (Fig. 5). Also, highly diverse is the whole mitochondrial chromosome (Fig. 5b, Mt). Gene ontology overrepresentation profiles vary between strains As the observed differences in the gene copy number and loci-specific gains and losses may affect the functional properties of brewing strains, the genes that were most divergent according to array-CGH-based analysis (showing log2 ratio values higher than 2 or lower than −2 for at least one of analyzed strains) were then subjected to gene ontology overrepresentation analysis (Fig. 6).Fig. 6 A heat map generated from array-CGH data. Functional categories overrepresented in the group of genes that were the most divergent among analyzed strains are shown. The strains were ordered according to the result of clustering analysis (Fig. 4b), and the selected genes were grouped according to their functional assignment. Positive and negative log2 ratio values represent higher and lower than average abundance of the gene, as determined by array-CGH analysis Five functional categories overrepresented in the group of selected genes were revealed, namely (1) maltose metabolism and transport, (2) response to toxin, (3) siderophore transport, (4) cellular aldehyde metabolic process, and (5) L-iditol 2-dehydrogenase activity (p < 0.05) and are presented as a heat map in Fig. 6. Within two functional categories of genes involved in response to toxin and cellular aldehyde metabolic process, the loss of aryl alcohol dehydrogenase (AAD) genes was observed in strains 6 and 9 with unbalanced redox equilibrium (Figs. 3 and 6). The effects were statistically significant for AAD3, AAD6, AAD10, and AAD15 genes (p < 0.05) (Fig. 6), whereas similar but weak tendency for AAD4 and AAD14 genes was not significant (Supplementary Material). Moreover, the most accented loss of genes involved in siderophore transport was also shown in strain 6 (p < 0.05) (Fig. 6). In contrast, the most evident loss of genes involved in maltose metabolism was observed in strain 8 (p < 0.05) (Fig. 6). A heat map generated from array-CGH data reflecting the variability in the gene copy number of the whole genome of brewing strains analyzed is also presented in Supplementary Material. Validation of array-based comparative genomic hybridization data To test if the variations in the gene copy number are reflected by the levels of mRNA for those genes, qRT-PCR assay was employed for FIT3, MAL13, AAD10, and ALD2 genes representing major functional categories depicted in Fig. 6. The qRT-PCR results are presented in Table 3.Table 3 The relative mRNA levels for genes selected from the set shown in Fig. 6 Strain Gene FIT3 MAL13 AAD10 ALD2 4 0.0180 ± 0.0010 1.0668 ± 0.0693 0.2257 ± 0.0417 0.4143 ± 0.0423 6 0.0004 ± 0.0001 0.9243 ± 0.0067 0.0003 ± 0.0001 0.5779 ± 0.0418 8 0.0558 ± 0.0021 0.0004 ± 0.0001 0.4108 ± 0.0084 0.5524 ± 0.0461 9 0.0264 ± 0.0018 0.9239 ± 0.0435 1.3551 ± 0.0274 0.0004 ± 0.0003 The numbers represent the levels of respective transcripts normalized to the data for a housekeeping gene ALG9, relative to the normalized levels of transcript in BY4741 strain. The data represent the mean ± SD from at least three independent experiments Moreover, the comparison of gene copy number and mRNA levels for these genes is shown in Fig. 7.Fig. 7 Validation of gene copy number data obtained using array-CGH analysis with mRNA levels determined by qRT-PCR. The comparisons to the array-CGH results were made for genes selected from the set shown in Fig. 6. The qRT-PCR data from Table 3 were normalized with the average of the data for each gene and converted to log2 values to bring them to the same format as array-CGH data. The correlation coefficient between both sets of data is 0.89. a Log2 values obtained with both methods. b Graphical representation of these data To make this comparison possible, the qRT-PCR data had to be processed in the same way as array-CGH data. Therefore, the individual data points for each gene were divided by the average of the values for that gene in all strains. The resulting normalized data were converted to log2 values. As seen in Fig. 7, the results obtained with both methods correlate very well, with correlation coefficient of 0.89. The negative values obtained for the genes that were absent in some strains are lower using qRT-PCR method than using array-CGH assay, but this is because the former method is more sensitive. Yet, array-CGH result for a gene that is below −2 means that this gene is absent in that strain. Genomic stability and nucleolus state are affected in Saflager W-34/70 strain We were then interested if strains with redox disequilibrium and decreased dosage of genes involved in stress responses, e.g., strain 6, may be susceptible to DNA breaks and changes in nucleolus state. Indeed, strain 6 was found to be the most affected by DNA double-strand breaks (DSBs) in the control growth conditions (p < 0.05) (Fig. 8a).Fig. 8 Evaluation of genomic instability and nucleolus state in selected brewing yeast strains in the control growth conditions. a The susceptibility to DNA double-strand breaks (DSBs). DSBs were assessed using neutral comet assay. As a DNA damage marker, the % tail DNA was used. The bars indicate SD, n = 150, *p < 0.05 compared to strain 6 (ANOVA and Tukey’s a posteriori test). The typical micrographs are shown (right). DNA was visualized using YOYO-1 staining (green). b Western blot analysis of Nop1, Fob1, Rad1, and Rap1 contents. Anti-Tub1 antibody served as a loading control. Anti-Act1 antibody was ruled out as a loading control because analyzed strains are characterized by different levels of beta-actin. c Analysis of chromosome I, XI, and XII signals using fluorescence in situ hybridization and whole-chromosome painting probes (WCPPs). Chromosome-specific signals were scored in 100 nuclei and presented as a percentage, n = 100. Three categories were considered, i.e., cells with one, two, and more than two chromosome specific signals. d Analysis of rDNA content. rDNA was visualized using WCPP specific to chromosome XII that contains rDNA locus in yeast. Fluorescence signals of chromosome XII were quantified using ImageJ software. The integrated fluorescence density is presented in relative fluorescence units (RFUs). Box-and-whisker plots are shown, n = 100. The typical micrographs are shown (right). The cells were labeled with FITC to detect chromosome XII-specific signals (green). DNA was visualized using DAPI staining (blue) (Color figure online) We have then compared the levels of protein involved in DNA damage repair, namely Rad1p, but its level was not lower in strain 6 compared to other strains analyzed in the control growth conditions (Fig. 8b). In contrast, the levels of nucleolar proteins Nop1 and Fob1 and transcription regulator Rap1 were lower in strain 6 compared to other strains analyzed in the control growth conditions (Fig. 8b). Interestingly, anti-Act1p antibody cannot be considered as a loading control in strain 6 because strain 6 is characterized by very low level of beta-actin compared to other analyzed strains (Fig. 8b). We selected then three chromosomes of different size, namely small chromosome I, medium-sized chromosome XI, and large chromosome XII to analyze their signal variability in brewing strains using FISH with WCPPs (Fig. 8c). One should remember that array-CGH method is a population-scale approach and is not designed to study the discrete cellular observations, whereas FISH, here single-cell analysis of chromosome instability, can be used to address cellular heterogeneity. Some of our FISH data are in agreement with array-CGH results, especially on genomic diversity observed in strain 6 (Figs. 4 and 8c). Gross deficiencies of chromosomes I and XII (Fig. 4) were revealed using array-CGH that may reflect low frequency of signals of chromosomes I and XII observed using WCPPs (Fig. 8c). Analogically, the gains of chromosome XI (Fig. 4) were correlated with higher frequency of signals of chromosome XI in strain 6 compared to other strains (Fig. 8c). Interestingly, higher frequency of signals of chromosome XII was observed in strain 8 (Fig. 8c, d). Because similar effect was not revealed using array-CGH not detecting rDNA sequences, higher frequency of signals of chromosome XII that contains rDNA locus in yeast may suggest chromosome XII fragmentation and/or nucleolus (rDNA) fragmentation in strain 8. Chromosome XII (rDNA) signals were also quantified (Fig. 8d). However, except of strain 4, rDNA content was comparable among analyzed strains (Fig. 8d). Perhaps, chromosome XII fragmentation does not affect rDNA levels in strain 8 (Fig. 8d). Tolerance to fermentation-associated stress stimuli is diminished in Saflager W-34/70 strain We then asked the question of whether lower copy number of aryl-alcohol dehydrogenase (AAD) genes, imbalanced redox homeostasis, and genetic instability in strain 6 compared to other strains analyzed may also affect fermentation performance in strain 6. First, the utilization of non-fermentable carbon sources, namely glycerol and ethanol, was investigated (Fig. 9a).Fig. 9 Analysis of the utilization of non-fermentable carbon sources (a) and tolerance to fermentation-associated stress stimuli in selected brewing yeast strains (strains 4, 6, 8, and 9) (b) using spot assay. a Yeast cells at the logarithmic phase of growth were diluted (1 × 107, 1 × 106, 1 × 105, 1 × 104, and 1 × 103 cells/ml), and growth on solid YPG and YPE media was inspected after 48 h. The growth of strain 6 was improved in the presence of 0.1 % glucose in YPG medium. b Yeast cells at the logarithmic phase of growth were diluted (1 × 107, 1 × 106, 1 × 105, 1 × 104, and 1 × 103 cells/ml), and growth on solid YPD medium in the presence of different stress stimuli was inspected after 48 h. In the case of hydrogen peroxide, cells were incubated with hydrogen peroxide for 40 min and then transferred to solid YPD medium. The growth of cells incubated at 4 °C was inspected after 120 h. Representative photographs are shown The growth capacity of strain 6 was diminished in YPG and YPE media compared to control YPD medium and also to growth of other strains analyzed (Fig. 9a). The growth of strain 6 was improved when YPG medium was supplemented with 0.1 % glucose (Fig. 9a). Second, the tolerance to fermentation-associated stress stimuli was considered, namely salt, osmotic, oxidative, ethanol, high glucose, and cold/heat stresses (Fig. 9b). In general, diminished resistance to stress stimuli of strain 6 was observed compared to other strains analyzed (Fig. 9b). Strain 6 was found to be more sensitive to NaCl, KCl, sorbitol, hydrogen peroxide, ethanol and high-glucose treatments, and heat stress (Fig. 9b). Strain 6 was unable to grow at 37 °C (Fig. 9b). In contrast, cryotolerant lager strain 6 grew better at 4 °C compared to ale strains (strains 4, 8, and 9) (Fig. 9b). Discussion In the present study, we found that the naturally occurring variations in the gene copy number within two functional gene categories: response to toxin and cellular aldehyde metabolic process, mainly the loss of AAD genes, may affect intracellular redox equilibrium (elevated ROS production and oxidative DNA damage), the nucleolus state, and the tolerance to fermentation-associated stress stimuli in brewing yeasts of economic importance. We used 30 commercially available yeast industrial strains, namely 25 ale brewing strains, four lager brewing strains, and one cider strain that, according to the suppliers’ information, belong to S. cerevisiae species. Molecular karyotyping revealed the variability of S. cerevisiae-like chromosome profiles that reflects genetic and genomic diversity of industrial yeasts (Kodama et al. 2006; Querol and Bond 2009). In the case of lager strains, S. bayanus-like chromosomes were also shown that confirm a hybrid nature of lager yeast genome composed of two subgenomes of S. cerevisiae and S. eubayanus (Dunn and Sherlock 2008; Walther et al. 2014). As ale strains are less studied (U’Ren et al. 2015), we selected three ale brewing yeasts with distinct chromosome patterns for further genomic analysis. One lager strain (Saflager W-34/70), the prototype of lager yeast group II (Frohberg group), was also included for comparison. Although ale yeasts are mostly diploid in nature (Legras et al. 2007), two analyzed ale strains were found to be tetraploid, whereas one strain (strain 8) was denoted as diploid with high variability of chromosome-specific signals (this study). This again confirms a wide range of genomic diversity of brewing yeasts including ale strains (Gonzalez et al. 2008). Whole-genome sequencing revealed that the reference lager yeast group II strain Weihenstephan (WS) 34/70 has approximately 64 chromosomes (4n ploidy with 36 distinct chromosome structures including eight chromosomes with translocations between the two subgenomes) (Nakao et al. 2009; Walther et al. 2014) that is in agreement with our FACS-based analysis of ploidy. Moreover, the analysis of array-CGH profiles showed that the most accented loci-specific gains and losses were observed in Saflager W-34/70 strain compared to other brewing strains. Statistically significant differences in the gene copy number were revealed in five functional gene categories, namely (1) maltose metabolism and transport, (2) response to toxin, (3) siderophore transport, (4) cellular aldehyde metabolic process, and (5) L-iditol 2-dehydrogenase activity, and analyzed strains varied in gene ontology overrepresentation profiles. We found that decreased dosage of AAD genes was correlated with intracellular redox disequilibrium and oxidative DNA damage. In the Saflager W-34/70 (strain 6), the losses of AAD3, AAD6, AAD10, and AAD15 genes were the most accented and were accompanied by elevated production of reactive oxygen species (ROS) and superoxide, oxidative DNA damage, and diminished tolerance to fermentation-associated stress stimuli, whereas in the Belle Saison Belgian ale strain (strain 8) characterized by the lowest ROS production and the levels of 8-hydroxy-2′-deoxyguanosine (8-oxo-dG), the AAD gene set dosage was the highest among analyzed strains. In the S. cerevisiae genome, there are seven telomeric open reading frames (ORFs) (YNL331c, YDL243c, YCR107w, YJR155w, YFL056c, YFL057c, and YOL165c) and one non-telomeric ORF (YPL088w) whose protein products show high amino acid sequence similarity to the aryl alcohol dehydrogenase (AAD) of the lignin-degrading fungus Phanerochaete chrysosporium (Delneri et al. 1999a). AAD is an enzyme-converting aromatic aldehydes, e.g., veratraldehyde or anisaldehyde, into their corresponding alcohols, and the budding yeast, although not being a lignin degrader, is able to metabolize veratraldehyde into veratryl alcohol (Delneri et al. 1999a). It has been suggested that the telomere-associated members (AAD3, AAD4, AAD6, AAD10, AAD14, and AAD15) form six-member AAD gene family and AAD16 is more distantly related (Delneri et al. 1999b). The expression of the AAD genes was shown to be elevated after diamide and diethyl maleic acid ester treatment that promoted oxidative stress by glutathione (GSH) depletion (Delneri et al. 1999b). AAD-mediated oxidative stress response was found to be redox-sensitive transcription factor Yap1 dependent (Delneri et al. 1999b). The genetic analysis using single and multiple AAD disruptants revealed that only AAD4 (YDL243c) and AAD6 (YFL056/57c) may take part in the oxidative stress response and the contribution of other members to cellular stress responses should be further elucidated (Delneri et al. 1999b). Redox imbalance-mediated susceptibility to oxidative DNA damage of Windsor British ale strain (strain 9) (this study) may also reflect decreased dosage of ALD2 gene. It has been reported that Ald2p and Ald3p are stress-inducible aldehyde dehydrogenases in S. cerevisiae (Navarro-Avino et al. 1999). The expression of ALD2 and ALD3 genes is dependent on the general stress transcription factors Msn2,4 (Navarro-Avino et al. 1999). ALD3 gene expression is induced by osmotic shock, heat shock, glucose exhaustion, oxidative stress, and drugs, whereas ALD2 gene expression is stimulated by osmotic stress and glucose exhaustion (Navarro-Avino et al. 1999). The double-mutant cells, namely Δald2Δald3, are sensitive to ethanol treatment (Navarro-Avino et al. 1999). Moreover, Ald2p and Ald3p are important during acetaldehyde stress in the budding yeast (Aranda and del Olmo 2003). Taken together, the enzymes that participate in NAD+/NADH balancing, here AADs and aldehyde dehydrogenases, may be important for the maintenance of intracellular redox homeostasis in brewing yeasts, especially during stresses associated with industrial brewery handling (Gibson et al. 2007). The other genes characterized by pronounced variations in the gene copy number were genes involved in carbohydrate metabolism and iron transport that may also modulate the fermentation performance. The dosage of maltose utilization genes was affected, namely, MAL11, MAL13, MAL33, MPH2, and MPH3 genes. To metabolize maltose, at least one of five unlinked polymeric (MAL) loci located in the telomeric regions of the different chromosomes (MAL1-MAL4, and MAL6) is required (Chow et al. 1989). Decreased dosage of MAL genes may not only indicate a defect in maltose fermentation but may also suggest a modulation of drug response pathways as deletions of MAL11 lead to nystatin sensitivity (Giaever et al. 2002). Analyzed strains varied also in the copy number of SOR1 and SOR2 genes that encode a NAD-dependent sorbitol dehydrogenase, a member of the polyol dehydrogenase branch of the medium-chain dehydrogenase/reductase (MDR) superfamily of enzymes (Sarthy et al. 1994). Despite S. cerevisiae is a non-xylose-utilizing microorganism, in the presence of sorbitol or xylose, the expression of SOR1 gene is upregulated (Sarthy et al. 1994; Toivari et al. 2004). Redox imbalance and lower dosage of AAD genes were correlated with DNA breaks and affected nucleolus state (lower levels of Nop1p and Fob1p) in Saflager W-34/70 strain. The nucleolus is suggested to be a guardian of cellular homeostasis and genome integrity acting as a central hub in coordinating the cellular stress response (Grummt 2013). Shifts in the levels and the relocation of nucleolar proteins have been shown during stress conditions both in mammalian and yeast cells that may result in the inhibition of rRNA synthesis saving the energy required to maintain cellular homeostasis during stress (Lewinska et al. 2010; Mayer et al. 2005; Olson 2004; Rubbi and Milner 2003). In contrast, higher levels of Nop1p were correlated with nucleolus (rDNA) fragmentation in strain 8. It has been recently shown that overexpression of other nucleolar protein Nop2 resulted in nucleolus fragmentation (de Beus et al. 1994; Lewinska et al. 2014a). However, rDNA pools were not diminished. Perhaps, living in stressful conditions, industrial strains optimized their growth rate via the control of rDNA level to preserve genome integrity (Deregowska et al. 2015a). Conclusions In summary, we show that strain-specific variability in the gene copy number, especially variations in the dosage of AAD genes, may modulate redox homeostasis and susceptibility to DNA damage in brewing yeasts that may be particularly important during fermentation processes when industrial strains are subjected to stress conditions. Electronic supplementary material ESM 1 (XLS 2.20 mb) Jagoda Adamczyk, Anna Deregowska, and Marek Skoneczny have contributed equally as first authors. Anna Lewinska and Maciej Wnuk have contributed equally as last authors. The authors would like to thank the two anonymous reviewers for their valuable comments and suggestions, which were helpful in improving the paper. This work was supported by European Union, within Regional Operational Programme of Subcarpathia Voivodeship (2007–2013), Priority 1: Competitive and Innovative Economy, Action 1.3, Regional Innovation System, grant WND-RPPK-01.03.00-18-038/13. Author contributions MW conceived and designed the experiments. JA, AD, MS, AS, UN, AK, ER, LP, EK, AP, AL, and MW performed the experiments. AL, MS, AS, and MW analyzed the data. AL and MW contributed reagents/materials/analysis tools. AL and MW wrote the paper. Compliance with ethical standards Conflict of interest The authors declare that they have no conflicts of interest. ==== Refs References Aranda A del Olmo MM Response to acetaldehyde stress in the yeast Saccharomyces cerevisiae involves a strain-dependent regulation of several ALD genes and is mediated by the general stress response pathway Yeast 2003 20 747 759 10.1002/yea.991 12794936 Bond U Neal C Donnelly D TC J Aneuploidy and copy number breakpoints in the genome of lager yeasts mapped by microarray hybridisation Curr Genet 2004 45 360 370 10.1007/s00294-004-0504-x 15103502 Caesar R Palmfeldt J Gustafsson JS Pettersson E Hashemi SH Blomberg A Comparative proteomics of industrial lager yeast reveals differential expression of the cerevisiae and non-cerevisiae parts of their genomes Proteomics 2007 7 4135 4147 10.1002/pmic.200601020 17994632 Chow TH Sollitti P Marmur J Structure of the multigene family of MAL loci in Saccharomyces Mol Gen Genet 1989 217 60 69 10.1007/BF00330943 2549370 de Beus E Brockenbrough JS Hong B Aris JP Yeast NOP2 encodes an essential nucleolar protein with homology to a human proliferation marker J Cell Biol 1994 127 1799 1813 10.1083/jcb.127.6.1799 7806561 de Hoon MJ Imoto S Nolan J Miyano S Open source clustering software Bioinformatics 2004 20 1453 1454 10.1093/bioinformatics/bth078 14871861 Delneri D Gardner DC Bruschi CV Oliver SG Disruption of seven hypothetical aryl alcohol dehydrogenase genes from Saccharomyces cerevisiae and construction of a multiple knock-out strain Yeast 1999 15 1681 1689 10.1002/(SICI)1097-0061(199911)15:15<1681::AID-YEA486>3.0.CO;2-A 10572264 Delneri D Gardner DC SG O Analysis of the seven-member AAD gene set demonstrates that genetic redundancy in yeast may be more apparent than real Genetics 1999 153 1591 1600 10581269 Deregowska A Shifts in rDNA levels act as a genome buffer promoting chromosome homeostasis Cell Cycle 2015 14 3475 3487 10.1080/15384101.2015.1093705 26566866 Deregowska A Genome-wide array-CGH analysis reveals YRF1 gene copy number variation that modulates genetic stability in distillery yeasts Oncotarget 2015 6 30650 30663 26384347 Dunn B Sherlock G Reconstruction of the genome origins and evolution of the hybrid lager yeast Saccharomyces pastorianus Genome Res 2008 18 1610 1623 10.1101/gr.076075.108 18787083 Dworak N Wnuk M Zebrowski J Bartosz G Lewinska A Genotoxic and mutagenic activity of diamond nanoparticles in human peripheral lymphocytes in vitro Carbon 2014 68 763 776 10.1016/j.carbon.2013.11.067 Giaever G Functional profiling of the Saccharomyces cerevisiae genome Nature 2002 418 387 391 10.1038/nature00935 12140549 Gibson B Liti G Saccharomyces pastorianus : genomic insights inspiring innovation for industry Yeast 2015 32 17 27 25088523 Gibson BR Lawrence SJ Leclaire JP Powell CD KA S Yeast responses to stresses associated with industrial brewery handling FEMS Microbiol Rev 2007 31 535 569 10.1111/j.1574-6976.2007.00076.x 17645521 Gibson BR Storgards E Krogerus K Vidgren V Comparative physiology and fermentation performance of Saaz and Frohberg lager yeast strains and the parental species Saccharomyces eubayanus Yeast 2013 30 255 266 10.1002/yea.2960 23695993 Gonzalez SS Barrio E Querol A Molecular characterization of new natural hybrids of Saccharomyces cerevisiae and S. kudriavzevii in brewing Appl Environ Microbiol 2008 74 2314 2320 10.1128/AEM.01867-07 18296532 Grummt I The nucleolus-guardian of cellular homeostasis and genome integrity Chromosoma 2013 122 487 497 10.1007/s00412-013-0430-0 24022641 James TC Usher J Campbell S Bond U Lager yeasts possess dynamic genomes that undergo rearrangements and gene amplification in response to stress Curr Genet 2008 53 139 152 10.1007/s00294-007-0172-8 18183398 Kodama Y, Kielland-Brandt MC, Hansen J (2006) Lager brewing yeast. In: Comparative genomics. Springer, Berlin Heidelberg, pp 145–164 Legras JL Merdinoglu D Cornuet JM Karst F Bread, beer and wine: Saccharomyces cerevisiae diversity reflects human history Mol Ecol 2007 16 2091 2102 10.1111/j.1365-294X.2007.03266.x 17498234 Lewinska A Wnuk M Grzelak A Bartosz G Nucleolus as an oxidative stress sensor in the yeast Saccharomyces cerevisiae Redox Rep 2010 15 87 96 10.1179/174329210X12650506623366 20500990 Lewinska A Macierzynska E Grzelak A Bartosz G A genetic analysis of nitric oxide-mediated signaling during chronological aging in the yeast Biogerontology 2011 12 309 320 10.1007/s10522-011-9329-4 21424154 Lewinska A Miedziak B Kulak K Molon M Wnuk M Links between nucleolar activity, rDNA stability, aneuploidy and chronological aging in the yeast Saccharomyces cerevisiae Biogerontology 2014 15 289 316 10.1007/s10522-014-9499-y 24711086 Lewinska A Miedziak B Wnuk M Assessment of yeast chromosome XII instability: single chromosome comet assay Fungal Genet Biol 2014 63 9 16 10.1016/j.fgb.2013.12.003 24333410 Lingjaerde OC Baumbusch LO Liestol K Glad IK Borresen-Dale AL CGH-explorer: a program for analysis of array-CGH data Bioinformatics 2005 21 821 822 10.1093/bioinformatics/bti113 15531610 Mayer C Bierhoff H Grummt I The nucleolus as a stress sensor: JNK2 inactivates the transcription factor TIF-IA and down-regulates rRNA synthesis Genes Dev 2005 19 933 941 10.1101/gad.333205 15805466 Monerawela C, James TC, Wolfe KH, Bond U (2015) Loss of lager specific genes and subtelomeric regions define two different Saccharomyces cerevisiae lineages for Saccharomyces pastorianus group I and II strains. FEMS Yeast Res:15 Nakao Y Genome sequence of the lager brewing yeast, an interspecies hybrid DNA Res 2009 16 115 129 10.1093/dnares/dsp003 19261625 Naumov GI Naumova ES Lantto RA Louis EJ Korhola M Genetic homology between Saccharomyces cerevisiae and its sibling species S. paradoxus and S. bayanus : electrophoretic karyotypes Yeast 1992 8 599 612 10.1002/yea.320080804 1441740 Naumova ES Naumov GI Masneuf-Pomarede I Aigle M Dubourdieu D Molecular genetic study of introgression between Saccharomyces bayanus and S. cerevisiae Yeast 2005 22 1099 1115 10.1002/yea.1298 16240458 Navarro-Avino JP Prasad R Miralles VJ Benito RM Serrano R A proposal for nomenclature of aldehyde dehydrogenases in Saccharomyces cerevisiae and characterization of the stress-inducible ALD2 and ALD3 genes Yeast 1999 15 829 842 10.1002/(SICI)1097-0061(199907)15:10A<829::AID-YEA423>3.0.CO;2-9 10407263 Olson MO Sensing cellular stress: another new function for the nucleolus? Sci STKE 2004 2004 pe10 15026578 Querol A Bond U The complex and dynamic genomes of industrial yeasts FEMS Microbiol Lett 2009 293 1 10 10.1111/j.1574-6968.2008.01480.x 19175410 Rubbi CP Milner J Disruption of the nucleolus mediates stabilization of p53 in response to DNA damage and other stresses EMBO J 2003 22 6068 6077 10.1093/emboj/cdg579 14609953 Sarthy AV Schopp C KB I Cloning and sequence determination of the gene encoding sorbitol dehydrogenase from Saccharomyces cerevisiae Gene 1994 140 121 126 10.1016/0378-1119(94)90741-2 8125328 Toivari MH Salusjarvi L Ruohonen L Penttila M Endogenous xylose pathway in Saccharomyces cerevisiae Appl Environ Microbiol 2004 70 3681 3686 10.1128/AEM.70.6.3681-3686.2004 15184173 U’Ren JM, Wisecaver JH, Paek AL, Dunn BL, Hurwitz BL (2015) Draft genome sequence of the ale-fermenting Saccharomyces cerevisiae strain GSY2239. Genome Announc 3 Walther A Hesselbart A Wendland J Genome sequence of Saccharomyces carlsbergensis , the world’s first pure culture lager yeast G3 (Bethesda) 2014 4 783 793 10.1534/g3.113.010090 24578374 Wendland J Lager yeast comes of age Eukaryot Cell 2014 13 1256 1265 10.1128/EC.00134-14 25084862 Wnuk M Single-cell analysis of aneuploidy events using yeast whole chromosome painting probes (WCPPs) J Microbiol Methods 2015 111 40 49 10.1016/j.mimet.2015.01.022 25639739

PMC005xxxxxx/PMC5003813.txt

==== Front Mol CellMol. CellMolecular Cell1097-27651097-4164Cell Press S1097-2765(16)30361-610.1016/j.molcel.2016.07.008ArticleMolecular Principles of Gene Fusion Mediated Rewiring of Protein Interaction Networks in Cancer Latysheva Natasha S. natashal@mrc-lmb.cam.ac.uk1∗Oates Matt E. 2Maddox Louis 1Flock Tilman 1Gough Julian 2Buljan Marija 1Weatheritt Robert J. 13Babu M. Madan madanm@mrc-lmb.cam.ac.uk1∗∗1 MRC Laboratory of Molecular Biology, Francis Crick Avenue, Cambridge CB2 0QH, UK2 Department of Computer Science, University of Bristol, Bristol BS8 1UB, UK3 The Donnelly Centre, University of Toronto, Toronto, ON M5S 3E1, Canada∗ Corresponding author natashal@mrc-lmb.cam.ac.uk∗∗ Corresponding author madanm@mrc-lmb.cam.ac.uk18 8 2016 18 8 2016 63 4 579 592 12 4 2016 14 6 2016 14 7 2016 © 2016 The Authors2016This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).Summary Gene fusions are common cancer-causing mutations, but the molecular principles by which fusion protein products affect interaction networks and cause disease are not well understood. Here, we perform an integrative analysis of the structural, interactomic, and regulatory properties of thousands of putative fusion proteins. We demonstrate that genes that form fusions (i.e., parent genes) tend to be highly connected hub genes, whose protein products are enriched in structured and disordered interaction-mediating features. Fusion often results in the loss of these parental features and the depletion of regulatory sites such as post-translational modifications. Fusion products disproportionately connect proteins that did not previously interact in the protein interaction network. In this manner, fusion products can escape cellular regulation and constitutively rewire protein interaction networks. We suggest that the deregulation of central, interaction-prone proteins may represent a widespread mechanism by which fusion proteins alter the topology of cellular signaling pathways and promote cancer. Graphical Abstract Highlights • Parents of fusion proteins occupy central positions in protein interaction networks • Parents are rich in interaction-mediating features, which are often lost via fusion • Fusions preferentially join proteins with no previous connection in protein networks • Fusion proteins escape regulation by losing post-translational modification sites The molecular mechanisms of fusion-mediated interactome disruption are currently unclear. Latysheva et al. find that fusion-forming proteins occupy central positions in interaction networks. They lose much of their extensive interaction-mediating ability and capacity for regulation upon fusion. These findings provide insights into how fusion proteins could rewire networks in cancer. Keywords gene fusionfusion proteincancer genomicsprotein interaction networksPublished: August 18, 2016 ==== Body Introduction Fusion genes are hybrid genes formed from two previously independent parent genes. Historically, gene fusions have been viewed as common driver mutations in malignancies associated with blood, lymph, and bone marrow tissue, but are becoming increasingly recognized as important players in solid tumors (Mertens et al., 2015a, Mertens et al., 2015b, Yoshihara et al., 2015). For example, translocation-induced gene fusions are found in about 90% of all lymphomas and over half of all leukemias (Lobato et al., 2008), and the TMPRSS2-ERG fusion is the most frequent genetic aberration in prostate cancer (Nam et al., 2007). In accord with their important role in oncogenesis, fusion transcripts and proteins have been utilized in many areas of clinical care, from biomarker development and diagnostics to acting as therapeutic targets (Kumar-Sinha et al., 2015, Mertens et al., 2015b). Yet, aside from a relatively small number of well-studied fusions, the functions of fusion proteins and the cellular context in which they operate remain unclear. A variety of mechanisms can lead to the fusion of two genes, such as insertions, deletions, inversions, and translocations. Continuous transcription of neighboring genes (Varley et al., 2014) or trans- and cis-splicing of pre-mRNAs (Jividen and Li, 2014, Zhang et al., 2012) can also generate fusion transcripts and proteins. If fusion transcripts are translated, the resulting fusion proteins have the potential to redirect cellular signaling pathways and act as principal oncogenic drivers (see Watson et al., 2013, Yoshihara et al., 2015). Despite some concerns over whether certain putative fusion mRNAs may be artifacts of the sequencing procedure (Yu et al., 2014), the widespread finding of recurrent gene fusions in tumor samples, the clinical utility of an increasing number of gene fusions, and a growing body of literature on fusion protein functionality adds support to their potential for significant biological impact. There are now approximately 10,000 known gene fusions, most of which have only recently been discovered using deep sequencing technology (Mertens et al., 2015a). The molecular functions of gene fusions, and the fusion proteins they encode, remain relatively poorly understood. Recent bioinformatics work on gene fusions (reviewed in Latysheva and Babu, 2016) has examined fusion protein domain content and recombination, reading frame conservation, intrinsic disorder at fusion junctions, and expression properties. However, the molecular principles of fusion-mediated rewiring of protein networks and how fusion proteins could disrupt native protein interactions remain unclear. Here, we devise a genome-scale computational data analysis framework to investigate the molecular principles by which fusion proteins affect protein interactions (Figures 1A and 1B). Understanding the structural features of fusion proteins, as well as the interactions that are recurrently disrupted or created as a result of fusion, will help clarify how fusions contribute to specific cellular phenotypes and influence cancer initiation and progression. Results To compose a set of human fusion proteins, a list of fusion transcripts from the ChiTaRS v1 database (Frenkel-Morgenstern et al., 2013) was acquired and mapped onto Ensembl protein sequences (Experimental Procedures; Figure 1C). In this study, only fusions affecting protein-coding regions were examined. In total, we mapped 2,699 distinct fusion proteins derived from 3,279 genes (Table S1; fusion protein mappings are available as a web resource at http://fusion.d2p2.pro, integrated into the D2P2 database; Oates et al., 2013). Genes that form fusions (“parent genes”) are enriched for functions related to translation, mRNA splicing, and the cell cycle, and for protein classes related to translation, acetyltransferase activity, and the binding of actin, chromatin, and RNA (Table S1). Parent genes that form multiple fusions, especially five or more, are further enriched for functions relating to translation, RNA binding, and nucleic acid binding. Gene fusion events can be summarized as a network, in which nodes indicate genes and a link between nodes indicates the occurrence of a fusion between genes. Our resulting network of gene fusions involving 3,209 genes (as gene symbols; Figure S1A) expands upon previous networks of ∼300 gene fusions (Höglund et al., 2006, Mitelman et al., 2007); we confirm the presence of several major hubs, i.e., nodes with many edges (e.g., MLL, ETV6, NUP98, EWSR1, and ALK), and highlight novel fusion hubs (e.g., COL1A1, HSP90AA1, MT1A, NCL, and AFF1; Table S1; Figure S1A). The number of fusions formed for each gene follows a power law distribution (Figure S1B), with most parent genes forming few fusions (e.g., only 21 genes form ten or more fusion proteins). Over a third of known oncogenes (OGs) and a quarter of known tumor suppressor genes (TSGs) form fusions in this data set (Figure S1). Parent Proteins Have More Central Roles in Protein Interaction Networks and Are Expressed at Higher Levels To examine whether parent genes encode proteins with central positions in the human interactome, a high-confidence data set of human protein-protein interactions (PPIs) (Wang et al., 2012) was analyzed. In addition to a much higher number of interaction partners (node degree; Figure 2A), parent proteins have a significantly higher tendency to interconnect interaction clusters, as quantified by betweenness centrality, which measures the extent to which a given node in a network lies on the shortest paths between all other nodes (Figure 2B). Furthermore, parent proteins have higher Kleinberg’s hub scores (see Experimental Procedures), which measure a protein’s connection to network hubs (Figure 2C). Compared to central non-parents, the most central parent proteins were more likely to be involved in functions such as mRNA splicing, cell proliferation, DNA replication, and repair (Table S2). We observed that parent mRNAs and proteins are more abundant compared to non-parents (∼3-fold difference between averages; Figures S2A and S2B) in medulloblastoma cell lines (Vogel et al., 2010). Additionally, parent proteins have very similar half-lives to non-parent proteins (Figure S2C). Further, by integrating data on 12 oncogenic signaling blocks (Cui et al., 2007), we find that parent proteins are over twice as likely to be involved in signaling processes implicated in oncogenesis (χ2 = 29.5, df = 1, and p = 5.7 × 10−8) (Figure S2D) and are over 2.5 times as likely to be genes essential for cellular viability (χ2 = 396.8, df = 1, and p < 2.2 × 10−16) (Figure S2E). Although these trends need to be analyzed in different tissues, these results suggest that altering parent proteins could have a major effect on critical cellular functions and for a sustained period of time. Parent genes were grouped into OG parent genes, TSG parent genes, and all other parent genes (Figure S3A). Parent genes that are neither OGs nor TSGs possess significantly higher network centrality than non-parent genes, indicating that centrality is a feature of parent genes more broadly and not simply reflective of the centrality of OGs and TSGs. Further, parent OGs and TSGs tend toward higher centrality than non-parent OGs and TSGs, respectively (Figure S3B). For example, average centrality measures for parent TSGs are approximately 30% higher than non-parent TSGs. Replicate network centrality calculations on two additional PPI data sets—the consensus network used in further analyses (see below; Bossi and Lehner, 2009) (Figure S3C) and an unbiased interaction network derived using mass spectrometry (Huttlin et al., 2015) (Figure S3D)—were consistent with those described above. Parent Proteins Have Higher Centrality in the Interaction Networks of Cancer-Associated Cell and Tissue Types Next, the role of parent proteins in tissue-specific protein interaction networks (Bossi and Lehner, 2009) was examined. PPIs involving parent proteins are present in more human tissues (median of 64 of 79 tissues, compared to 52 of 79 for non-parents; p < 2.2 × 10−16; Figure 2D), indicating that fusion events do not only affect tissue-specific interactions. Parent proteins consistently have on average ∼5 additional interaction partners across most tissues (Figure 2E). Interestingly, the tissues and cell types with the highest degrees for parent proteins—e.g., B and T cells, bone marrow cells, and blood cells—are cell types often associated with fusion-induced cancers (gold dots, Figure 2E). Furthermore, parent proteins in the five cancer cell types in the data set (teal dots) have on average 9.1% higher degree than non-cancer cells and 12.1% higher degree than the set of non-cancer and non-blood/bone/lymph cell types (Table S3). This trend is not observed for betweenness (Figure S3E), but is for hub scores (Figure S3F), which may indicate that gene fusions in cancer may preferentially affect nodes of high degree (either directly or indirectly) rather than alter global network cohesion. Fusions could therefore be especially disruptive in tissues with interaction networks containing proteins with unusually high degree. Finally, fusion transcripts detected in cell lines of metastatic tumor origin may have parent genes with higher centrality compared to those from primary tumors (Figures S4A–S4D; Supplemental Information), suggesting a possible connection between cancer aggressiveness and parent centrality. Although this trend was not observed in the mass spectrometry PPI data set (Huttlin et al., 2015; data not shown), the concept of a link between cancer stage and the roles of parent proteins in PPI networks may be relevant in specific contexts (e.g., certain cancer types). Parent Proteins Are Unstructured and Enriched for Interaction-Mediating Domains, which Are Preferentially Excluded from Fusion Proteins The structural features of parent proteins and their retention within fusion proteins were investigated (Figures 3A–3L and S5A–S5K). In agreement with a previous study (Hegyi et al., 2009), parent proteins in our expanded data set (3,279 parent proteins versus 406) have significantly higher intrinsic structural disorder scores than non-parents (Figure S5A): OG parents have on average 1.27 × (0.39 versus 0.31; p = 2.8 × 10−4; and pairwise Wilcoxon rank-sum tests with Holm multiple testing correction), TSG parents 1.15 × (p = 1.5 × 10−3), and other parents 1.13 × (p < 2 × 10−16) higher disorder compared to non-parents. Parent OGs and TSGs are approximately equally disordered as non-parent OGs and TSGs (Figure S5B), as are included versus excluded fusion protein segments (Figure S5C). This suggests that any observed enrichment of linear motifs and post-translational modifications (PTMs) in included segments (see below), which are features correlated with disorder (Davey et al., 2012), are not simply due to included segments being more disordered. Throughout the structural feature calculations, densities instead of counts are used to control for protein length. Using a database of PPIs defined at the structurally resolved level of domains (Meyer et al., 2013), we investigated parent versus non-parent densities of interaction-mediating domains (IMDs). Parent proteins, especially OG and TSG parent proteins, have higher densities of IMDs (Figure 3A). On average, compared to non-parent proteins, OG parents have 4.6×, TSG parents 2.7×, and other parents 1.5× the IMD densities (all corrected p values: <2.2 × 10−16). There is a slight tendency for parent OGs to have higher IMD densities than non-parent OGs (on average 1.3×; p = 9.1 × 10−3; Figure S5D). Hence, although parent proteins are generally more intrinsically disordered, they are also enriched in structured domains that mediate protein interactions. IMDs tend to largely be excluded from fusion proteins (Figure 3B; Table S4). OG parent proteins, in contrast to TSG and other parent proteins, tend to retain IMDs upon fusion. Overall, the most frequently retained IMDs include RNA-recognition, tyrosine kinase, pleckstrin homology (signaling and cytoskeleton), and SH3 and SH2 signaling domains (Table S4). The average level of domain truncation upon transfer varies significantly by domain type, and the most intact IMDs which occur ≥10 times include ubiquitin conjugating domains, the ubiquitin-like PB1 domain (a specificity adaptor to kinases), and the proliferation modulating S_100 domain. Parents that repeatedly donate large portions of IMDs are enriched for functions in translation, cell structure morphogenesis, and cell cycle and protein modification (Table S4). Transfer of IMDs Can Create Novel Interactions and Preserve Important Natural Interactions The repeated inclusion of large portions of specific IMDs in fusion proteins is interesting for two reasons (Figure 4A). First, it can point to the importance of a particular domain-domain interaction (DDI) for a fusion protein’s function. Second, as a result of the fusion, a novel interaction-like link can occur between the interaction partner of the included domain and the fusion partner. We map which domain-mediated PPIs are repeatedly conserved in fusion proteins (Figures 4B and S6A; Table S5). We find that 192 IMD-mediated PPIs are recurrently retained in fusion proteins and comment on the most frequently conserved DDIs (see the Figure S6A legend). We also map novel protein links that are created through IMD transfer (Figures 4C and S6B; Table S5). A protein interaction “link” was drawn between proteins A and B if there existed some fusion protein B-C, where C normally interacts with A and at least 90% of C’s IMD was retained (Figure 4A). Of the 126 novel links, 116 (92%) do not normally occur in the cell. The most frequent novel links include many connections for BCR, with the newly linked proteins being enriched for functions in cell proliferation and cellular component movement (Table S5), and 11 new connections for the nuclear trafficking protein TPR, including eight tyrosine protein phosphatases (Figures 4C and S6B). Certain fusion-induced novel links are recurrent, e.g., fusion proteins involving both EML4 and TFG lead to the gain of similar links (i.e., connections to receptor-type protein tyrosine phosphatases PTPRB, PTPRG, and PTPRJ). Fusion-Generated Novel Links Disproportionately Connect Proteins that Are Distant in the Interaction Network We examined the distance between the protein pairs in the novel links set in a non-diseased PPI network. Where a path existed between the novel links pairs, the distance was overall slightly shorter than in other protein pairs in the network (Figure S6C). However, fusion was found to disproportionately connect proteins which normally reside in separate sections of the interactome, whereas only 10.7% of protein pairs in the PPI network had no connecting path, 29.3% of protein pairs in the novel links set had no previous connecting path (Fisher’s exact test on contingency table, odds ratio = 3.47, p = 3.0 × 10−8) (Figure S6D). We examine the 34 newly connected protein pairs in Figure S6E (see the legend). Independent Structural Evidence Supports the Potential of Fusion Proteins to Disrupt PPI, Protein-RNA Interactions, and Protein-DNA Interactions Structural interfaces in fusion proteins were identified by analyzing the Protein Interfaces, Surfaces, and Assemblies (PISA) database, which houses macromolecular interfaces (involving proteins, RNA, and DNA) in the Protein Data Bank (PDB). Parent proteins in the PDB contain more interface-forming residues (Figure 3C). On average, 1.5% of residues in non-parents form interfaces, and OG parents have on average 4.5×, TSG parents 2.1×, and other parents 2.0× this PISA residue density. Parent OGs have 2.4× the average interface residue density of non-parent OGs (p = 3.6 × 10−5; Figure S5E). Interface residue densities on included and excluded segments of parent proteins are similar (Figure 3D), though the distribution is skewed toward exclusion (Figure S5F). The 302 parent proteins which donate ten or more interface-forming residues to fusion proteins are enriched for functions relating to cell cycle signaling, carbohydrate and lipid metabolism, cellular component morphogenesis, and cell death (Table S6). Parent Proteins Are Enriched in Interaction-Mediating Short Linear Motifs, which May Be Preferentially Excluded from Fusion Products Linear motifs (LMs) are short sequence motifs, usually <10 residues, often found in intrinsically disordered regions (Tompa et al., 2014). Using 1,410 experimentally validated LMs from the ELM database (Dinkel et al., 2014) and over a million putative LMs identified using the ANCHOR program (Dosztányi et al., 2009), we tested for enrichment of LMs within parent proteins compared to all other proteins. Parent proteins have more experimentally verified LMs on average (Figure 3E), with OG and TSG parents harboring more motifs. Although most parents have zero experimental LMs due to the small size of this data set, on average, OG parents have 10.1× (p < 2 × 10−16), TSG parents 7.1× (p < 2 × 10−16), and other parents 1.3× (p = 8.0 × 10−10) the LM density of non-parents. Parent TSGs have slightly higher LM densities compared to non-parent TSGs (Figure S5G). Fusion proteins tend to retain ELM LMs, as shown by higher mean LM densities in included segments (Figure 3F). Parent proteins, which donate ELM LMs, function in the regulation of cell death, the stress response, protein metabolism, and nucleic acid binding (Table S6). Similarly, the expanded ANCHOR data shows higher densities of LMs in parents (Figure 3G), though parent OGs and TSGs have similar densities to the non-parent categories (Figure S5H). Interestingly, the larger ANCHOR data set shows a strong trend toward the exclusion of LMs (Figure 3H). Either trend implies that fusion substantially disrupts transient interactions mediated by LMs. PTMs that Regulate Protein Interactions Are Enriched in Parent Proteins We mapped putative interaction-regulating PTMs (PTMcode v2 database; Minguez et al., 2015) onto proteins and found that compared to non-parents, OG parents have on average 4.6×, TSG parents 3.5×, and other parents 2.2× the PTM density (all corrected p < 2 × 10−16; Figure 3I). Parent TSGs have slightly more interaction-regulating PTMs compared to non-parent TSGs (1.5×, p = 0.03; Figure S5I). These PTM sites overall tend toward exclusion from fusion proteins (Figure S5J), though the retention and loss is comparable in OG and TSG parents. Parent Proteins Are Enriched in PTM Sites, and Fusion Proteins Tend to Selectively Escape Regulation by PTMs In addition to regulating protein interactions, post-translational and co-translational modification sites can regulate protein stability (e.g., by ubiquitination), subcellular localization (e.g., N-myristoylation), and protein function (e.g., acetylation). Parent proteins have significantly more PTMs (Figure 3J) compared to non-parents (on average 0.009 PTMs/residue): OG parents have 3.5×, TSG parents 3.5×, and other parents 2.3× the PTM densities of non-parents (all corrected p < 2 × 10−16). This suggests that the function, stability, and subcellular location of parent proteins are extensively regulated by PTMs. Further, on average, parent OGs have 1.5× (p = 7.5 × 10−3) the PTM content of non-parent OGs, and parent TSGs have 2.1× (p = 1.4 × 10−5) the PTM content of non-parent TSGs (Figure S5K). PTMs are generally excluded from fusion proteins, though not in OG parents (Figure 3K). The selective exclusion of PTM sites suggests that fusion proteins tend to escape regulation by signaling pathways. TSG parents experience the heaviest loss of PTMs, with excluded segments having over triple the median PTM density of included segments (excluded: 0.022 PTMs/residue; included: 0.007; p = 3.0 × 10−4; Figure 3K). Parent proteins which retain at least 90% of their PTM content are enriched for functions in translation, ion transport, and metabolism (Table S6), while parent proteins which lose at least 90% of their PTMs have a wide range of functions, including splicing and cell matrix adhesion. Next, we examined the PTM profiles in included and excluded fusion protein segments (Experimental Procedures; Figure 3L). Certain PTM types (e.g., S-Nitrosylation) occur in either parental segment more frequently than expected given the global frequencies of all PTMs in dbPTM, while other PTM types (e.g., methylation and acetylation) showed marked presence/absence patterns based on segment inclusion (Table S6). Fusion Can Lead to the Gain and Loss of Ubiquitination Sites, which May Deregulate the Activity of OGs and TSGs Ubiquitination (UB) sites are of particular interest since their loss and gain upon fusion could “upregulate” OG activity or “downregulate” TSG activity, due to the role of UB sites in mediating protein stability and degradation. We find 14 fusion proteins in which OGs lose ≥5 UB sites and ten fusion proteins in which a TSG gains ≥5 UB sites (Table 1). As an illustrative example, we profile the well-known EWSR1-FLI1 gene fusion from Ewing’s sarcoma (Figure 5A). The specific pattern of segment retention in EWSR1-FLI1 fusion proteins leads to UB site loss, which may confer increased stability onto the fusion product, adding to the known oncogenic mechanism of transcriptional deregulation. Notably, decreased UB-mediated degradation of ETS family transcription factors (e.g., FLI1) has been linked to cancer (Vitari et al., 2011). Conversely, one of the most extreme examples of UB site gain by a TSG occurs in the previously unstudied ATP50-TGFB1 fusion (Figure 5B), which results in the amalgamation of a heavily ubiquitinated segment with a short portion of the TGFB1 tumor suppressor domain, hinting at a fusion-mediated loss of TSG function. TGF-β signaling is known to inhibit cell proliferation and is normally tightly regulated by UB (Huang and Chen, 2012). OG parents do not lose and TSG parents do not gain UB sites more often than expected (data not shown), but individual cases identified here (Table 1) could be of substantial biological interest for follow-up studies. Fusions Involving Transcription Factors Are Linked to Significant Alterations in Downstream Target Gene Expression Levels To investigate the potential downstream network rewiring effects due to fusion events, we investigated whether fusions involving transcription factors (TFs) are associated with downstream expression changes in the TFs’ regulatory targets. TCGA tumor samples with TF-containing fusion transcripts and paired normal controls were identified (Experimental Procedures). The regulatory target genes of TFs were acquired from the TRRUST database (Han et al., 2015). Differential gene expression (DGE) values were calculated (absolute log2 fold change between diseased and healthy samples). The targets of TFs had significantly (i.e., corrected p < 0.05) higher DGE values in five of the eight paired breast cancer samples when compared to all other genes (Figure S7). For example, four fusion transcripts containing TFs were detected in patient TCGA-GI-A2C9; these four TFs together affected 51 mapped regulatory targets, the mean (absolute log2) DGE of which is 2.0× the mean DGE of all other genes (Table S7; corrected p = 9.6 × 10−5). Across the eight available biospecimen pairs, the average DGE of TF targets is 1.41× (mean) and 1.45× (median) the DGE of all other genes. Discussion Many disease states result from altered dynamics of complex regulatory and signaling interactions. Representing interactions as networks provides a conceptual framework for understanding how mutations in proteins can affect entire cellular systems and cause disease (Wang et al., 2011, Wu et al., 2010), especially when combined with structural analyses of interacting proteins (Sudha et al., 2014, Wang et al., 2012). Here, we investigated the interaction properties and structural features of thousands of putative fusion proteins. Based on our observations, we delineate genome-scale molecular principles by which gene fusions can affect protein networks, rewire signaling pathways, and contribute to disease (Figure 6). These trends will be useful for setting novel gene fusions into context, building on the performance of previous driver gene fusion prioritization algorithms (Abate et al., 2014, Shugay et al., 2013), and interpreting studies of fusion protein functionality. Fusion Preferentially Affects Highly Central, Interaction-Prone Proteins Although it is likely that not all of the analyzed fusion proteins drive disease (e.g., genomic instability can produce passenger fusions; Mertens et al., 2015a), parent proteins are nonetheless enriched for a wide variety of interaction-prone elements, such as IMDs, interface-forming residues, LMs, and PTM sites that regulate PPIs. The observed density of interaction-mediating features in parent proteins is in accord with their centrality in interaction networks. These results are consistent with other computational work on disease mutations, which have shown that disease-related in-frame mutations (Wang et al., 2012) and disease-causing non-synonymous single nucleotide polymorphisms (David et al., 2012) are preferentially located on PPI interfaces. Finally, the finding that many parent genes are essential genes dovetails with the concept of “edgetic” perturbations in cancer, i.e., mutations that disrupt specific interactions (or edges) of proteins rather than the entire node (Charloteaux et al., 2011, Rolland et al., 2014, Wang et al., 2015), given that disrupting essential genes is associated with lethality, fusion may offer an opportunity to disrupt only a portion of an essential protein’s function, such as specific interactions. Network disruption may play a role in fusion proteins that first appear to have relatively simple mechanisms of oncogenesis (Figure 6A), for example, the concurrent rewiring of signaling pathways can be critical for BCR-ABL1 mediated transformation (Pawson and Warner, 2007). Importantly, targeting the interacting partners or downstream signaling of fusion proteins could be a fruitful area for therapeutic agent development (see Tognon et al., 2011). In this context, our observation that TF fusions significantly perturb target gene expression in breast cancer lends further weight to the signaling perturbation capabilities of fusion events. Fusion Results in a Loss of Parental Interaction-Mediating Features and Regulatory Sites Although parent proteins are enriched for interaction-mediating features, the segments of parents that are included within fusion proteins appear to be depleted of functional regions (though OG parents retain more of these features than other parents). Examining specific cases of fusion-mediated loss and gain of molecular features (Figures 3A–3L), as well as interaction preservation and creation (Figures 4A–4C), is a rich resource for hypothesis generation. For example, fusion proteins characterized by the repeated inclusion of largely complete tyrosine kinase domains (e.g., Figure 6B) could be promising targets for kinase inhibitors. Proteins dependent on the function of several distinct molecular features (such as the interface residues and nuclear import/export signal motifs in nucleophosmin; Figure 6B), as well as proteins sensitive to changes in PTM content (such as EWSR1; Figure 6C), may be especially disrupted by fusion events. Although we largely addressed each interaction-mediating and regulatory molecular feature of parent and fusion proteins separately, these entities are not independent. For instance, LMs tend to form interactions conditionally on PTM site status (Van Roey et al., 2013). For example, the retinoic receptor alpha gene (RARα) encodes a LM that acts as a phosphorylation-dependent switch for binding Pin1. RARα forms driver fusion proteins in acute promyelocytic leukemia, for which Pin1 suppression is used as a treatment (Gianni et al., 2009). We find a RARα fusion protein that excludes the LM in question (Figure 6C), which could correspond to a treatment resistant patient. Knowledge of the specific retained sequence of fusion proteins has previously been observed to be key to patient treatment (Robinson et al., 2011). Conclusions Our findings demonstrate that proteins that form fusions tend to be highly interactive and positioned in critical regions of PPI networks. Disruption of such proteins may alter the topology of signaling and regulatory pathways of cells and promote cancer. A detailed understanding of the molecular impact of the rewired network will be helpful for future drug discovery studies. For example, in cases where driver fusion proteins retain the ability to form interactions, their carcinogenic activity could be reduced by the targeted disruption of specific interaction interfaces with small molecules (Cierpicki and Grembecka, 2015, Jin et al., 2014, Kuenemann et al., 2015). Additionally, recent methodological advances in therapeutically degrading specific proteins in vivo (Bondeson et al., 2015, Winter et al., 2015) could be instrumental to targeting oncogenic fusion proteins that have escaped normal regulatory pathways. Experimental Procedures Database Identification, Processing, and Integration To compose a set of human fusion proteins, we acquired a database (ChiTaRS v1 database; Frenkel-Morgenstern et al., 2013) of 9,237 fusion mRNAs. The fusion transcripts were mapped onto known proteins in the Ensembl database using ChiTaRS genomic coordinates and segments that mapped to non-exonic regions (intronic, UTR, or intergenic sequences) were discarded. The resulting data set maps all fusion protein segments defined at the DNA/gene, mRNA, and protein levels (Table S1). We limit our analysis to fusion proteins in which both parents were mapped to known Ensembl proteins. Fusion protein mapping information is made available via a web server (http://www.fusion.d2p2.pro). A fusion network of all gene fusions was constructed using Cytoscape. Throughout this study, gene sets were tested for enrichments of GO-Slim molecular functions and protein classes using PantherDB (Mi et al., 2013). See the Supplemental Information for further methodological details. mRNA and Protein Abundance and Half-Lives of Parents Protein and mRNA abundances were acquired from a microarray and shotgun proteomics study performed on the Daoy medulloblastoma cell line (Vogel et al., 2010), and protein half-life data were taken from a SILAC study in HeLa cells (Boisvert et al., 2012). These data sets were overlapped onto parent and non-parent gene sets, and differences in distributions of abundance and half-life by category were quantified by non-parametric Wilcoxon rank-sum tests. Parent Gene Participation in Oncogenic Signaling Blocks Disproportionate parent protein participation in cancer signaling processes (Cui et al., 2007) was assessed using a contingency table and a chi-square test of independence. Parent Gene Essentiality 1,734 “core” essential genes shared between two cell lines (Blomen et al., 2015) were acquired and tested for enrichment among parent genes as above. PPI Network Centrality Network centrality calculations for both parent and non-parent genes/proteins were performed on a non-tissue specific PPI network (Wang et al., 2012) using the igraph R package. See the Supplemental Information for definitions of centrality measures. A tissue-specific PPI network (Bossi and Lehner, 2009) was acquired in order to calculate tissue-specific PPI metrics (Buljan et al., 2012). A more recent, expanded, and unbiased protein interaction data set from human cells (Huttlin et al., 2015) was also investigated. Intrinsic Structural Disorder in Parent Proteins Residue-by-residue predictions for disorder for each protein in the human proteome were generated using the IUPred program (Dosztányi et al., 2005; http://www.iupred.enzim.hu/). Scores range from 0 to 1, where higher scores indicate a higher propensity toward intrinsic disorder. Intrinsic disorder was calculated for genes (i.e., longest isoform Ensembl protein) and for specific included and excluded segments as an average over either the protein or segment length. Analysis of Interacting Domains within Proteins A data set of curated, structurally resolved PPIs was acquired (Meyer et al., 2013), and residues that form IMDs were mapped onto parent and non-parent proteins. IMD retention was quantified by calculating IMD residue densities on included and excluded segments. The frequency and completeness of retention of different domain types was summarized across the fusion protein set. Statistically significant differences between gene sets in the distributions of IMD residues were assessed as before. Parents which donate ≥20% of at least one IMD were analyzed for functional and protein class enrichments. Identifying Novel and Retained PPIs of Fusion Proteins The above set of domain-mediating PPIs was analyzed to identify which PPIs are recurrently (two or more times) retained in fusion proteins. DDIs were deemed to be “retained” if at least one fusion protein incorporated at least 90% of the IMD. Novel interactions created as a result of the transfer of IMDs were between protein A and B if there existed at least one fusion protein B-C, where C normally interacts with A and at least 90% of C’s IMD was retained. Novel links were those that did not appear in a set of known PPIs (Wang et al., 2012). Identifying Shortest Path Distances between Proteins Newly Linked by Fusion Pairwise shortest path lengths (geodesics) between all protein pairs in a PPI network (Wang et al., 2012) were calculated using igraph. The distribution of shortest path lengths in the novel link set was compared to the distribution of path lengths in 1,000 randomly sampled protein pairs from the complete geodesic matrix as before. Disconnected protein pairs had infinite shortest path lengths, reflecting the absence of a geodesic. A contingency table containing the counts of disconnected novel links versus other disconnected protein pairs was constructed and tested for independence using Fisher’s exact test. Analysis of Interaction Interfaces in Parents Structures of proteins in complex with proteins, DNA, or RNA molecules were obtained from the PDB and PISA database (http://www.ebi.ac.uk/pdbe/pisa/). Interface residues were identified and their positions converted into Ensembl protein coordinates. PISA residue densities were calculated by counting unique positions and dividing by protein lengths. Differences in the distributions of interface-forming PISA residue densities were analyzed as before. Biological process and protein class enrichments for parent genes that donate ten or more interface-forming residues to fusion proteins were calculated. Analysis of Short Linear Peptide Motifs in Parents A set of 1,410 experimentally validated (Dinkel et al., 2014) and 1,036,282 computationally predicted (Dosztányi et al., 2009) LMs were acquired and mapped onto proteins. LM densities were calculated by counting unique ELM accessions and dividing by protein length. Differences in LM density were assessed across parent gene sets and across included versus excluded segments. Due to the small sample size of experimentally verified LMs, functional enrichments were reported even if the number of genes in an enriched category was less than ten. Parent proteins that donate LMs to fusion proteins were assessed for functional enrichments. Analysis of PTM Sites PTM sites, which are candidate sites for regulating protein interactions, were acquired from the PTMcode v2 data set (Minguez et al., 2015). Differences in PTMcode site densities per gene were assessed for different parent gene sets and across included versus excluded segments. Further, we obtained and cleaned a data set of experimentally validated PTMs (dbPTM 3.0 database; Lu et al., 2013). PTM densities were analyzed as before at the whole protein and fusion segment level. Enrichments of specific types of modification sites were quantified in included and excluded segments. Analysis of TF Fusions and the Expression Levels of Target Genes Fusion transcripts in TCGA samples (Yoshihara et al., 2015) were filtered to identify fusions involving TFs (n = 1,131) (Table S7). The TCGA database (Tomczak et al., 2015) was queried to identify matched RNaseq data for TF fusion containing samples (n = 29). Normalized expression counts for each matched sample pair were extracted, genes with extremely small read counts (n < 10) removed, and DGE calculated as the absolute log2 fold change between the diseased and healthy samples. The regulated target genes of TFs were acquired from the TRRUST database (Han et al., 2015). DGE values for the TF targets were compared against all other genes using non-parametric Wilcoxon rank-sum tests in cases where sufficient regulatory targets (n ≥ 20) were available (n = 8). The resulting p values were corrected for multiple testing using Holm’s procedure. Author Contributions Study Conception and Design: N.S.L., R.J.W., and M.M.B.; Acquisition of Data: N.S.L., L.M., M.E.O., J.G., and R.J.W.; Analysis and Interpretation of Data: N.S.L., L.M., M.B., R.J.W., T.F., and M.M.B.; Manuscript Writing: N.S.L. and M.M.B.; and Critical Inputs to Manuscript: N.S.L., L.M., T.F., M.E.O., R.J.W., M.B., and M.M.B. The project was led by N.S.L. and supervised by M.M.B. Supplemental Information Document S1. Supplemental Experimental Procedures and Figures S1–S7 Table S1. Fusion Proteins and Parent Functions, Related to Figure 1 The table shows a description of fusion proteins used in this study and biological process and protein class enrichments of parent genes. The gene symbols in the original ChiTaRS mapping can differ from gene names associated with mapped Ensembl proteins (see Supplemental Information), and we provide both gene name sets in the fusion protein listing (see the fusion protein mapping web server http://www.fusion.d2p2.pro/ for further details). Table S2. Functions of the Top Quartile Centrality Genes, Related to Figure 2 The table shows the biological process and protein class enrichments of parent and non-parent genes with the highest PPI network centralities. Table S3. Tissue-Specific Network Centrality, Related to Figure 2 The table shows the averaged network centrality measures for parent and non-parent proteins in tissue-specific interaction networks. Table S4. Interaction-Mediating Domains in Fusion Proteins and Parent Functions, Related to Figure 3 The table shows IMD residues incorporated into fusion proteins and biological process and protein class enrichments of parent genes that donate ≥20% of an IMD. Table S5. Retained and Novel PPI Arising from Fusion-Mediated Domain Recombination, Related to Figure 4 The table shows the retained and novel PPIs resulting from the transfer of largely intact (≥90% of the domain sequence) IMD into fusion proteins. Table S6. Interfaces, Linear Motifs, and PTMs in Fusion Proteins and Parent Functions, Related to Figure 3 The table shows structural interfaces of protein complexes that are incorporated into fusion proteins and biological process and protein class enrichments of parent genes that donate ten or more interface forming residues; experimentally validated short linear peptide motifs incorporated into fusion proteins and biological process and protein class enrichments of parent genes that donate at least one such linear motif; and experimentally validated PTMs incorporated into fusion proteins. Also, biological process and protein class enrichments for parent genes that either retain or lose ≥90% of their PTM content upon fusion. Certain PTM types were found to occur in both included and excluded segments more frequently than expected given the global frequencies of all known PTMs, such as S-Nitrosylation (1.7× enrichment in included segments and 1.6× in excluded segments). Other PTM types showed differential presence/absence patterns based on segment inclusion: methylation sites are more highly enriched in included segments (3.5× enrichment) than in excluded segments (2.5×), as are acetylation sites (1.8× included and 1.3× excluded). Interestingly, both N-linked and O-linked glycosylation, which are involved in protein folding and stability and cancer processes like migration and invasion, are generally depleted in parent proteins. Table S7. Fusion-Mediated Deregulation of TF Target Genes, Related to Figure 6 The table shows a differential expression analysis of breast cancer samples containing fusion transcripts composed of at least one TF parent. The differential gene expression values of the TF targets were compared to those of all other genes. Document S2. Article plus Supplemental Information Acknowledgments We thank A. Krishnan, A. Bateman, B. Luisi, C. Ravarani, G. Chalancon, and S. Chavali for helpful discussions and feedback on the manuscript and M. Frenkel-Morgenstern for providing genomic coordinates from the ChiTaRS database for fusion protein mapping. This work was supported by the Medical Research Council (MC_U105185859 to M.M.B., N.S.L., L.M., R.J.W., and T.F and MC-A025-5PK11-6801 to M.B.), the Human Frontier Science Program (RGY0073/2010 to M.B. and M.M.B.), the Boehringer Ingelheim Fond (to T.F.), the Canadian Institute of Health Research (to R.J.W.), the IOF Marie Curie Fellowship (to R.J.W.), and the Lister Institute Research Prize Fellowship (to M.M.B.). We apologize for not being able to cite several relevant papers on this topic due to space constraints. We have extensively discussed a number of important papers in Latysheva and Babu (2016). Supplemental Information includes Supplemental Experimental Procedures, seven figures, and seven tables and can be found with this article online at http://dx.doi.org/10.1016/j.molcel.2016.07.008. Figure 1 Study Outline (A) Investigating how gene fusions and fusion proteins could affect molecular interactions in cancer. (B) Summary of analyses employed. (C) Description of processing procedure applied to the ChiTaRS database of fusion (“chimeric”) mRNA sequences to obtain a data set of fusion proteins. See also Figures S1 and S2 and Table S1. Figure 2 Network Centrality of Parent Genes and Proteins (A–C) Parent genes possess more interaction partners in PPI networks (A), have higher betweenness centrality (B), and higher hub scores (C). (D) PPIs involving parent proteins occur in more human tissues than interactions not involving parent proteins. (E) The average number of interaction partners for parent proteins and all other proteins by tissue or cell type (gold = blood, bone marrow, and lymph tissues and teal = cancer cells). Throughout this study, distribution outliers are excluded from boxplots for presentation purposes, but included in statistical analyses. See also Figures S3 and S4 and Tables S2 and S3. Figure 3 Interaction-Mediating Molecular Features in Fusion Proteins (A and B) IMDs in parent proteins (A) and fusion proteins (B). (C and D) The PPI interface residues in parent proteins (C) and fusion proteins (D) are shown. (E and F) The ELM LMs in parent proteins (E) and fusion proteins (F) are shown. (G and H) The predicted ANCHOR LMs in parent proteins (G) and fusion proteins (H) are shown. (I) The putative interaction-regulating PTMs in parent proteins are shown. (J and K) Other PTM sites in parent proteins (J) and fusion proteins (K) are shown. (L) The PTM type enrichments in included and excluded parent protein segments are shown. Within each subplot, Holm’s sequential Bonferroni correction for multiple testing was applied. See also Figure S5 and Tables S4 and S6. Figure 4 Retained and Novel PPI in Fusion Proteins (A) The repeated inclusion of large portions of specific IMDs in fusion proteins can lead to the retention of domain-mediated interactions or the creation of novel interaction-like links between proteins. (B and C) Subsets of the recurrently retained domain-mediated PPIs (B) and novel links (C) are shown. See also Figure S6 and Table S5. Figure 5 Fusion-Induced UB Site Gain and Loss in Cancer-Associated Proteins Fusion proteins involving OGs and TSGs can lead to the loss or gain of ubiquitination sites. (A) Example of an OG losing UB sites upon fusion. (B) Example of a TSG gaining UB sites upon fusion. The protein structure cartoons are of EWSR1 (PDB: 2CPE), FLI1 (PDB: 1FLI), and TGFB1 (PDB: 1KLA). Figure 6 Molecular Principles by which Gene Fusions Can Alter Protein Interaction Networks in Cancer (A) Fusion tends to involve highly central proteins in interaction networks and can alter networks by several mechanisms. Rewiring effects can play key roles in seemingly straightforward fusion events, as in the constitutive kinase activation found in the BCR-ABL1 fusion. (B and C) More generally, fusion can affect molecular interactions of proteins by shuffling interaction-prone regions within ordered (B) and disordered (C) protein segments. See also Figure S7 and Table S7. Table 1 OGs Losing ≥5 UB Sites and Tumor Suppressor Genes Gaining ≥5 UB Sites as a Result of Fusion Events Fusion Accessiona OG Description Number of UB Sites Lost Length of OG Retained Segment Fusion Partner Description of Fusion Partner BF736842 EGFR epidermal growth factor receptor 17 25 SLC12A9 solute carrier family 12, member 9 AK098472 CTNNB1 catenin (cadherin-associated protein), beta 1, and 88 kDa 9 420 RP11-345J4.5 bolA-like protein 2 BE176861 COPS5 COP9 signalosome subunit 5 9 112 HNRNPH3 heterogeneous nuclear ribonucleoprotein H3 (2H9) BE176782 COPS5 COP9 signalosome subunit 5 9 112 HNRNPH3 heterogeneous nuclear ribonucleoprotein H3 (2H9) BG953255 CTTN cortactin 9 21 MYC v-myc avian myelocytomatosis viral OG homolog BP430745 CSE1L CSE1 chromosome segregation 1-like (yeast) 7 41 UGP2 UDP-glucose pyrophosphorylase 2 CN278368 TRIM32 tripartite motif-containing protein 32 7 36 DDX21 DEAD (Asp-Glu-Ala-Asp) box helicase 21 CV340327 ERBB2 v-erb-b2 avian erythroblastic leukemia viral OG homolog 2 6 21 NOMO1 NODAL modulator 1 BE273347 DCUN1D1 DCN1, defective in cullin neddylation 1, and domain containing 1 6 24 QTRT1 queuine tRNA-ribosyltransferase 1 BC001010 CDK4 cyclin-dependent kinase 4 6 30 RPL4 ribosomal protein L4 AW371253 ERBB2 v-erb-b2 avian erythroblastic leukemia viral OG homolog 2 5 49 RABGAP1 RAB GTPase activating protein 1 U08818 MET met proto-OG 5 380 MIR548F1 microRNA 548f-1 U19348 MET met proto-OG 5 380 MIR548F1 microRNA 548f-1 DA624159 TFG TRK-fused gene 5 90 GPR128 G protein-coupled receptor 128 Fusion Accession Tumor Suppressor Gene Description Number of UB Sites Gained Length of TSG Retained Segment Fusion Partner Description of Fusion Partner CD368725 TGFB1 transforming growth factor, beta 1 13 45 ATP50 ATP synthase, H+ transporting, mitochondrial F1 Complex, and O subunit DB041801 SMARCA4 SWI/SNF related, matrix associated, actin dependent regulator of chromatin, subfamily a, and member 4 9 78 UBB ubiquitin B BP213958 ARID1A AT rich interactive domain 1A (SWI-like) 6 34 DNAJA2 DnaJ (Hsp40) homolog, subfamily A, and member 2 DB120764 EEF1A1 eukaryotic translation elongation factor 1 alpha 1 6 4 HIST1H2AM histone cluster 1, H2am BG035867 EIF1 eukaryotic translation initiation factor 1 6 38 RALY RALY heterogeneous nuclear ribonucleoprotein AB209020 GJA1 gap junction protein, alpha 1, and 43 kDa 6 136 IFT140 intraflagellar transport 140 BG926120 PDCD4 programmed cell death 4 (neoplastic transformation inhibitor) 5 110 GAPDH glyceraldehyde-3-phosphate dehydrogenase BC001412 EEF1A1 eukaryotic translation elongation factor 1 alpha 1 5 462 LASP1 LIM and SH3 protein 1 BQ962146 E2F1 E2F TF 1 5 8 RDH11 retinol dehydrogenase 11 (all-trans/9-cis/11-cis) CK004088 NDRG2 NDRG family member 2 5 153 RPL38 ribosomal protein L38 a ChiTaRS fusion event accessions are listed along with affected genes, retained segment lengths, and tallies of UB site gain or loss. ==== Refs References Abate F. Zairis S. Ficarra E. Acquaviva A. Wiggins C.H. Frattini V. Lasorella A. Iavarone A. Inghirami G. Rabadan R. Pegasus: a comprehensive annotation and prediction tool for detection of driver gene fusions in cancer BMC Syst. Biol. 8 2014 97 25183062 Blomen V.A. Májek P. Jae L.T. Bigenzahn J.W. Nieuwenhuis J. Staring J. Sacco R. van Diemen F.R. Olk N. Stukalov A. Gene essentiality and synthetic lethality in haploid human cells Science 350 2015 1092 1096 26472760 Boisvert F.-M. Ahmad Y. Gierliński M. Charrière F. Lamont D. Scott M. Barton G. Lamond A.I. Gierlinski M. Charriere F. A quantitative spatial proteomics analysis of proteome turnover in human cells Mol. Cell. Proteomics 11 2012 M111.011429 Bondeson D.P. Mares A. Smith I.E.D. Ko E. Campos S. Miah A.H. Mulholland K.E. Routly N. Buckley D.L. Gustafson J.L. Catalytic in vivo protein knockdown by small-molecule PROTACs Nat. Chem. Biol. 11 2015 611 617 26075522 Bossi A. Lehner B. Tissue specificity and the human protein interaction network Mol. Syst. Biol. 5 2009 260 19357639 Buljan M. Chalancon G. Eustermann S. Wagner G.P. Fuxreiter M. Bateman A. Babu M.M. Tissue-specific splicing of disordered segments that embed binding motifs rewires protein interaction networks Mol. Cell 46 2012 871 883 22749400 Charloteaux B. Zhong Q. Dreze M. Cusick M.E. Hill D.E. Vidal M. Protein-protein interactions and networks: forward and reverse edgetics Methods Mol. Biol. 759 2011 197 213 21863489 Cierpicki T. Grembecka J. Targeting protein-protein interactions in hematologic malignancies: still a challenge or a great opportunity for future therapies? Immunol. Rev. 263 2015 279 301 25510283 Cui Q. Ma Y. Jaramillo M. Bari H. Awan A. Yang S. Zhang S. Liu L. Lu M. O’Connor-McCourt M. A map of human cancer signaling Mol. Syst. Biol. 3 2007 152 18091723 Davey N.E. Van Roey K. Weatheritt R.J. Toedt G. Uyar B. Altenberg B. Budd A. Diella F. Dinkel H. Gibson T.J. Attributes of short linear motifs Mol. Biosyst. 8 2012 268 281 21909575 David A. Razali R. Wass M.N. Sternberg M.J.E. Protein-protein interaction sites are hot spots for disease-associated nonsynonymous SNPs Hum. Mutat. 33 2012 359 363 22072597 Dinkel H. Van Roey K. Michael S. Davey N.E. Weatheritt R.J. Born D. Speck T. Krüger D. Grebnev G. Kuban M. The eukaryotic linear motif resource ELM: 10 years and counting Nucleic Acids Res. 42 2014 D259 D266 24214962 Dosztányi Z. Csizmok V. Tompa P. Simon I. IUPred: web server for the prediction of intrinsically unstructured regions of proteins based on estimated energy content Bioinformatics 21 2005 3433 3434 15955779 Dosztányi Z. Mészáros B. Simon I. ANCHOR: web server for predicting protein binding regions in disordered proteins Bioinformatics 25 2009 2745 2746 19717576 Frenkel-Morgenstern M. Gorohovski A. Lacroix V. Rogers M. Ibanez K. Boullosa C. Andres Leon E. Ben-Hur A. Valencia A. ChiTaRS: a database of human, mouse and fruit fly chimeric transcripts and RNA-sequencing data Nucleic Acids Res. 41 2013 D142 D151 23143107 Gianni M. Boldetti A. Guarnaccia V. Rambaldi A. Parrella E. Raska I. Jr. Rochette-Egly C. Del Sal G. Rustighi A. Terao M. Garattini E. Inhibition of the peptidyl-prolyl-isomerase Pin1 enhances the responses of acute myeloid leukemia cells to retinoic acid via stabilization of RARalpha and PML-RARalpha Cancer Res. 69 2009 1016 1026 19155306 Han H. Shim H. Shin D. Shim J.E. Ko Y. Shin J. Kim H. Cho A. Kim E. Lee T. TRRUST: a reference database of human transcriptional regulatory interactions Sci. Rep. 5 2015 11432 26066708 Hegyi H. Buday L. Tompa P. Intrinsic structural disorder confers cellular viability on oncogenic fusion proteins PLoS Comput. Biol. 5 2009 e1000552 19888473 Höglund M. Frigyesi A. Mitelman F. A gene fusion network in human neoplasia Oncogene 25 2006 2674 2678 16331252 Huang F. Chen Y.-G. Regulation of TGF-β receptor activity Cell Biosci. 2 2012 9 22420375 Huttlin E.L. Ting L. Bruckner R.J. Gebreab F. Gygi M.P. Szpyt J. Tam S. Zarraga G. Colby G. Baltier K. The BioPlex Network: A systematic exploration of the human interactome Cell 162 2015 425 440 26186194 Jin L. Wang W. Fang G. Targeting protein-protein interaction by small molecules Annu. Rev. Pharmacol. Toxicol. 54 2014 435 456 24160698 Jividen K. Li H. Chimeric RNAs generated by intergenic splicing in normal and cancer cells Genes Chromosomes Cancer 53 2014 963 971 25131334 Kuenemann M.A. Sperandio O. Labbé C.M. Lagorce D. Miteva M.A. Villoutreix B.O. In silico design of low molecular weight protein-protein interaction inhibitors: Overall concept and recent advances Prog. Biophys. Mol. Biol. 119 2015 20 32 25748546 Kumar-Sinha C. Kalyana-Sundaram S. Chinnaiyan A.M. Landscape of gene fusions in epithelial cancers: seq and ye shall find Genome Med. 7 2015 129 26684754 Latysheva N.S. Babu M.M. Discovering and understanding oncogenic gene fusions through data intensive computational approaches Nucleic Acids Res. 44 2016 4487 4503 27105842 Lobato M.N. Metzler M. Drynan L. Forster A. Pannell R. Rabbitts T.H. Modeling chromosomal translocations using conditional alleles to recapitulate initiating events in human leukemias J. Natl. Cancer Inst. Monogr. 39 2008 58 63 18648005 Lu C.-T. Huang K.-Y. Su M.-G. Lee T.-Y. Bretaña N.A. Chang W.-C. Chen Y.-J. Chen Y.-J. Huang H.-D. DbPTM 3.0: an informative resource for investigating substrate site specificity and functional association of protein post-translational modifications Nucleic Acids Res. 41 2013 D295 D305 23193290 Mertens F. Johansson B. Fioretos T. Mitelman F. The emerging complexity of gene fusions in cancer Nat. Rev. Cancer 15 2015 371 381 25998716 Mertens F. Antonescu C.R. Mitelman F. Gene fusions in soft tissue tumors: recurrent and overlapping pathogenetic themes Genes Chromosomes Cancer 55 2015 291 310 26684580 Meyer M.J. Das J. Wang X. Yu H. INstruct: a database of high-quality 3D structurally resolved protein interactome networks Bioinformatics 29 2013 1577 1579 23599502 Mi H. Muruganujan A. Casagrande J.T. Thomas P.D. Large-scale gene function analysis with the PANTHER classification system Nat. Protoc. 8 2013 1551 1566 23868073 Minguez P. Letunic I. Parca L. Garcia-Alonso L. Dopazo J. Huerta-Cepas J. Bork P. PTMcode v2: a resource for functional associations of post-translational modifications within and between proteins Nucleic Acids Res. 43 2015 D494 D502 25361965 Mitelman F. Johansson B. Mertens F. The impact of translocations and gene fusions on cancer causation Nat. Rev. Cancer 7 2007 233 245 17361217 Nam R.K. Sugar L. Yang W. Srivastava S. Klotz L.H. Yang L.-Y. Stanimirovic A. Encioiu E. Neill M. Loblaw D.A. Expression of the TMPRSS2:ERG fusion gene predicts cancer recurrence after surgery for localised prostate cancer Br. J. Cancer 97 2007 1690 1695 17971772 Oates M.E. Romero P. Ishida T. Ghalwash M. Mizianty M.J. Xue B. Dosztányi Z. Uversky V.N. Obradovic Z. Kurgan L. D2 P2 : database of disordered protein predictions Nucleic Acids Res. 41 2013 D508 D516 23203878 Pawson T. Warner N. Oncogenic re-wiring of cellular signaling pathways Oncogene 26 2007 1268 1275 17322911 Robinson D.R. Kalyana-Sundaram S. Wu Y.-M. Shankar S. Cao X. Ateeq B. Asangani I.A. Iyer M. Maher C.A. Grasso C.S. Functionally recurrent rearrangements of the MAST kinase and Notch gene families in breast cancer Nat. Med. 17 2011 1646 1651 22101766 Rolland T. Taşan M. Charloteaux B. Pevzner S.J. Zhong Q. Sahni N. Yi S. Lemmens I. Fontanillo C. Mosca R. A proteome-scale map of the human interactome network Cell 159 2014 1212 1226 25416956 Shugay M. Ortiz de Mendíbil I. Vizmanos J.L. Novo F.J. Oncofuse: a computational framework for the prediction of the oncogenic potential of gene fusions Bioinformatics 29 2013 2539 2546 23956304 Sudha G. Nussinov R. Srinivasan N. An overview of recent advances in structural bioinformatics of protein-protein interactions and a guide to their principles Prog. Biophys. Mol. Biol. 116 2014 141 150 25077409 Tognon C.E. Somasiri A.M. Evdokimova V.E. Trigo G. Uy E.E. Melnyk N. Carboni J.M. Gottardis M.M. Roskelley C.D. Pollak M. Sorensen P.H. ETV6-NTRK3-mediated breast epithelial cell transformation is blocked by targeting the IGF1R signaling pathway Cancer Res. 71 2011 1060 1070 21148487 Tomczak K. Czerwińska P. Wiznerowicz M. The Cancer Genome Atlas (TCGA): an immeasurable source of knowledge Contemp. Oncol. (Pozn.) 19 1A 2015 A68 A77 25691825 Tompa P. Davey N.E. Gibson T.J. Babu M.M. A million peptide motifs for the molecular biologist Mol. Cell 55 2014 161 169 25038412 Van Roey K. Dinkel H. Weatheritt R.J. Gibson T.J. Davey N.E. The switches.ELM resource: a compendium of conditional regulatory interaction interfaces Sci. Signal. 6 2013 rs7 23550212 Varley K.E. Gertz J. Roberts B.S. Davis N.S. Bowling K.M. Kirby M.K. Nesmith A.S. Oliver P.G. Grizzle W.E. Forero A. Recurrent read-through fusion transcripts in breast cancer Breast Cancer Res. Treat. 146 2014 287 297 24929677 Vitari A.C. Leong K.G. Newton K. Yee C. O’Rourke K. Liu J. Phu L. Vij R. Ferrando R. Couto S.S. COP1 is a tumour suppressor that causes degradation of ETS transcription factors Nature 474 2011 403 406 21572435 Vogel C. Abreu R. de S. Ko D. Le S.-Y. Shapiro B.A. Burns S.C. Sandhu D. Boutz D.R. Marcotte E.M. Penalva L.O. Sequence signatures and mRNA concentration can explain two-thirds of protein abundance variation in a human cell line Mol. Syst. Biol. 6 2010 400 20739923 Wang X. Gulbahce N. Yu H. Network-based methods for human disease gene prediction Brief. Funct. Genomics 10 2011 280 293 21764832 Wang X. Wei X. Thijssen B. Das J. Lipkin S.M. Yu H. Three-dimensional reconstruction of protein networks provides insight into human genetic disease Nat. Biotechnol. 30 2012 159 164 22252508 Wang Y. Sahni N. Vidal M. Global edgetic rewiring in cancer networks Cell Syst. 1 2015 251 253 27136053 Watson I.R. Takahashi K. Futreal P.A. Chin L. Emerging patterns of somatic mutations in cancer Nat. Rev. Genet. 14 2013 703 718 24022702 Winter G.E. Buckley D.L. Paulk J. Roberts J.M. Souza A. Dhe-Paganon S. Bradner J.E. Drug Development. Phthalimide conjugation as a strategy for in vivo target protein degradation Science 348 2015 1376 1381 25999370 Wu G. Feng X. Stein L. A human functional protein interaction network and its application to cancer data analysis Genome Biol. 11 2010 R53 20482850 Yoshihara K. Wang Q. Torres-Garcia W. Zheng S. Vegesna R. Kim H. Verhaak R.G.W. The landscape and therapeutic relevance of cancer-associated transcript fusions Oncogene 34 2015 4845 4854 25500544 Yu C.-Y. Liu H.-J. Hung L.-Y. Kuo H.-C. Chuang T.-J. Is an observed non-co-linear RNA product spliced in trans, in cis or just in vitro? Nucleic Acids Res. 42 2014 9410 9423 25053845 Zhang Y. Gong M. Yuan H. Park H.G. Frierson H.F. Li H. Chimeric transcript generated by cis-splicing of adjacent genes regulates prostate cancer cell proliferation Cancer Discov. 2 2012 598 607 22719019

PMC005xxxxxx/PMC5003814.txt

==== Front Cell SystCell SystCell Systems2405-47122405-4720Cell Press S2405-4712(16)30216-210.1016/j.cels.2016.06.011ArticleDeep Proteome Analysis Identifies Age-Related Processes in C. elegans Narayan Vikram 123Ly Tony 1Pourkarimi Ehsan 15Murillo Alejandro Brenes 1Gartner Anton 14Lamond Angus I. 14Kenyon Cynthia cynthia@calicolabs.com234∗1 Centre for Gene Regulation and Expression, School of Life Sciences, University of Dundee, Dundee DD1 5EH, UK2 Department of Biochemistry and Biophysics, University of California, San Francisco, San Francisco, CA 94158-2517, USA3 Calico Life Sciences, 1170 Veterans Boulevard, South San Francisco, CA 94080, USA∗ Corresponding author cynthia@calicolabs.com4 Co-senior author 5 Present address: Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA 24 8 2016 24 8 2016 3 2 144 159 22 1 2016 11 5 2016 21 6 2016 © 2016 The Authors2016This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).Summary Effective network analysis of protein data requires high-quality proteomic datasets. Here, we report a near doubling in coverage of the C. elegans adult proteome, identifying >11,000 proteins in total with ∼9,400 proteins reproducibly detected in three biological replicates. Using quantitative mass spectrometry, we identify proteins whose abundances vary with age, revealing a concerted downregulation of proteins involved in specific metabolic pathways and upregulation of cellular stress responses with advancing age. Among these are ∼30 peroxisomal proteins, including the PRX-5/PEX5 import protein. Functional experiments confirm that protein import into the peroxisome is compromised in vivo in old animals. We also studied the behavior of the set of age-variant proteins in chronologically age-matched, long-lived daf-2 insulin/IGF-1-pathway mutants. Unexpectedly, the levels of many of these age-variant proteins did not scale with extended lifespan. This indicates that, despite their youthful appearance and extended lifespans, not all aspects of aging are reset in these long-lived mutants. Graphical Abstract Highlights • Near doubling in C. elegans proteome coverage with age-dependent protein measurements • Putative sub-cellular localization assignments for >6,000 nematode proteins • Peroxisome protein import impaired during aging • Most age-variant proteins do not scale with biological age in insulin/IGF-1 mutants Deep proteome analysis of adult C. elegans provides a valuable resource, reveals a failure in peroxisome protein import during aging, and, surprisingly, shows that the rates of change in the levels of the majority of age-variant proteins do not scale with the reduced rate of biological aging in long-lived insulin/IGF-1 mutants. Published: July 21, 2016 ==== Body Introduction The nematode C. elegans has been used widely to study aging due to its short lifespan and hermaphroditic reproductive cycle, as a result of which large numbers of isogenic animals can be cultured easily. Much of what we know about how aging is regulated was discovered through genetic perturbations in C. elegans. For example, lifespan extension through downregulation of insulin/IGF-1 signaling, and later TOR, was first described in these animals (Kenyon et al., 1993, Vellai et al., 2003, Kimura et al., 1997). Since then, dozens of genes, primarily affecting the sensation of, or response to, stress and nutrient sensing, have been found to affect lifespan. Many of these genes also are associated with aging in other organisms, including mammals (Kapahi et al., 2010, Kenyon, 2010). In comparison with detailed studies on the genetic control of aging, relatively less is known about how the proteome changes with age. There is evidence for increased protein aggregation and a gradual failure of the ubiquitin-proteasome machinery with advancing age, leading to a disruption of proteostasis (David et al., 2010, Taylor and Dillin, 2011, Morley and Morimoto, 2004). Only a handful of proteomic-based aging studies have been reported to date, so far with limited global protein coverage and resolution. Recently, the opportunity to apply quantitative proteomic approaches in C. elegans has been aided by the development of a stable-isotope labeling with amino acids in cell culture (SILAC)-based methodology for nematodes (Larance et al., 2011, Fredens et al., 2011). A database of mRNA and protein-abundance changes during nematode development also has been compiled (Grün et al., 2014) and age-dependent changes measured (Walther et al., 2015, Liang et al., 2014, Dong et al., 2007, Zimmerman et al., 2015, Copes et al., 2015). Consistent with the measurement of transcript and protein levels during development (Grün et al., 2014), recent findings in worms have reported an unexpectedly weak correlation between mRNA-level changes and protein-level changes during aging, suggesting that a significant proportion of age-related changes in protein levels are regulated at the post-translational level (Walther et al., 2015). However, current reports quantify only about 25% of the theoretical proteome, i.e., 5,000 proteins of ∼20,000 protein-coding genes, although not every protein is likely to be expressed at all times. Together, these findings highlight the need to measure protein-level changes, and they underscore the importance of expanding the depth of protein coverage to provide a better understanding of aging. Having in-depth, high-quality proteomic datasets also will expand the number of validated protein-coding genes, and it will aid in the development of novel network analysis tools based on proteins, rather than mRNAs. In this study, we expand the coverage and quantification of the C. elegans proteome, identifying >9,300 protein groups in each of three biological replicates. We characterize how organismal aging affects the levels of a large subset of these proteins (7,380), providing a proteomic fingerprint of the aging process. This expanded analysis of the proteome identifies a large set of proteins whose abundances change with age. These data, coupled with in vivo biological experiments, also identify the concerted downregulation of specific cellular metabolic networks during aging and uncover an age-dependent dysregulation of the peroxisome. In addition, combining our new findings with a detailed analysis of previously published datasets shows that only a subset of proteins whose abundances change with age in wild-type animals scale with the rate of biological aging in long- and short-lived insulin/IGF-1 pathway mutants. Results Deep Proteome Analysis in C. elegans Reproducibly Identifies 9,398 Proteins To make global measurements of age-related changes in protein abundance, we first developed a workflow to improve proteome coverage in C. elegans, while ensuring accurate quantification. We used the recently described SILAC-for-nematodes strategy, in which worms are fed bacteria whose proteins are labeled with arginine and lysine containing either light, medium, or heavy isotopes, resulting in near-complete labeling of the worm proteome after two generations (Larance et al., 2011, Fredens et al., 2011). This technique also exploits an RNAi-via-feeding strategy (Timmons and Fire, 1998) to knock down orn-1, thus preventing ORN-1-catalyzed arginine-to-proline conversion (Larance et al., 2011). We confirmed that orn-1 RNAi had no significant effect on C. elegans lifespan (Figure S1A). Stable-isotope-labeled nematodes were harvested at days 1 (fertile young adults), 5 (early aging/post-reproductive), and 10 (aging) of adulthood (Figures 1A and S1B). Time points after day 10 of adulthood were not considered because C. elegans aging has a stochastic component that becomes particularly evident after this time (Herndon et al., 2002; Figure S1C). Importantly, although wild-type worms can live for up to 4 weeks, many phenotypes associated with aging, such as decreased rates of pharyngeal pumping, mitochondrial fission, and muscle deterioration, are visible at day 10 (Jiang et al., 2015, Croll et al., 1977, Huang et al., 2004, Garigan et al., 2002; Figure S1D). As this study includes time points at which worms are fertile, in order to eliminate progeny from our assay, labeled worms were treated with 50 μM 5-Fluoro-2′-deoxyuridine (FUdR) at the mid-late L4 larval stage. We found that 100% of the progeny from worms treated with 50 μM FUdR were unviable, and that treated worms also had fewer eggs, thus decreasing egg-protein contamination in the assay (Figure S1E). Notably, unlike animals treated with higher doses of FUdR, at the chosen dose, animals were healthy, not prone to bursting, and, importantly, had a lifespan similar to that of untreated control animals (Figure S1F). The FUdR-treated labeled worms that were harvested at days 1, 5, and 10 of adulthood were next mixed and lysed in different buffers to increase proteome coverage (Figure 1A). Following digestion with trypsin, the resulting peptides were fractionated by hydrophilic strong anion exchange (SAX) chromatography to further reduce sample complexity, as described previously (Ly et al., 2014), prior to analysis by liquid chromatography-tandem mass spectrometry (LC-MS/MS). Three independent biological replicates were analyzed. We used a differential buffer extraction methodology (QProteome Cell Compartment Kit, QIAGEN), which allowed us to sample a wider pool of proteins with different physicochemical properties (Figure S2A; Pourkarimi et al., 2012). This method also increased the number of unique peptides we identified from a single protein when the data were aggregated, thereby improving sequence coverage (Figure S2B). Additionally, we found that, although the QProteome kit was designed to work primarily with cells grown in culture, with our optimizations, the kit could be used successfully to isolate proteins from different sub-cellular compartments in day 1 adult worms, in spite of the presence of a cuticle and syncytium (Figures S2C–S2E). The validity of our method was supported by the predicted fractionation of proteins that previously have been shown to localize to specific cellular locations (Figures S2D and S2E). Using this method, we compiled a resource with putative sub-cellular localization profiles for ∼6,300 C. elegans proteins, based on our proteomic analysis of day 1 adult nematodes treated with FUdR (Table S1). The relative proportions of these proteins in four sub-cellular compartments—i.e., cytoplasm, membrane (plasma membrane and membrane-bound organelles, excluding the nucleus), nucleus, and cytoskeleton/insoluble fractions—are presented, together with lists of proteins that were predominantly localized to any of these fractions based on our proteomic analysis (Table S1). This is a significant improvement over the 4,954 protein-coding nematode genes (of ∼20,000) that currently have a manually curated gene ontology (GO) term cellular component annotation, of which only a small subset (∼2,200 genes) is based on experimental evidence (Figures S2F–S2H; GO annotations and associated data were obtained from WormBase). As worm tissue integrity is compromised during aging (Garigan et al., 2002) and we were unable to confirm the efficacy of the kit in isolating sub-cellular fractions from old nematodes, we were hesitant to make any conclusions about changes in protein localization occurring during aging based on these data. Using this combination of protein-level and peptide-level fractionation, >125,000 sequence-unique peptides were identified with a false discovery rate (FDR) of 1%. These were assembled into >9,300 protein groups that were detected in all three biological replicates (see the Supplemental Experimental Procedures; Table S2; Figure S3). Identifications were highly reproducible, with 10,361 protein groups identified in at least two of three biological replicates and 9,398 protein groups identified across all three biological replicates (Figure 1B). Relative protein quantification among the three biological replicates also was very reproducible, as shown in the principal-component analysis (PCA) plot (Figure 1C) and using Pearson’s correlation (median value for Pearson’s r across the various pairwise comparisons was 0.85; Figure S4A). The measured protein intensities spanned approximately seven orders of magnitude (Figure 1D), and mean protein sequence coverage was ∼30%. Interestingly, the abundance composition of the C. elegans proteome is dominated by a relatively small proportion of proteins (Figure 1E; 24 proteins make up 25% of the measured abundance or total protein content), with apparent saturation at ∼4,000 proteins. Protein- and peptide-level fractionation enabled access to a much deeper proportion of the proteome, i.e., ∼7,383 proteins that cumulatively represent only 10% of the measured abundance. The total of 11,020 protein groups identified in at least one of the three biological replicates (see Figure 1B) corresponds to ∼54% of the predicted C. elegans proteome as defined by WormBase (Release WS243, March 28, 2014, containing 20,480 protein-coding genes; Figure S4B). Of the 11,020 protein groups, 6,480 proteins were reproducibly identified with sequence-unique peptides (Figure S4B). Where peptides could not be assigned to a single protein unambiguously (e.g., alternatively spliced isoforms and proteins encoded by genes with high sequence similarity), the lead member of the assembled protein group, as determined by Occam’s razor rule using MaxQuant (Cox and Mann, 2008), was used for all downstream data analysis. Our dataset corresponds to ∼22% of the theoretical C. elegans MS-flyable peptide digest of the predicted proteome (Figure S4B). However, both the 54% overlap of our dataset with the theoretical proteome and 22% overlap with the theoretical peptide-digest are likely to be underestimated, as our study only used adult C. elegans. Indeed, a recent study identified only ∼3,000 proteins in young-adult nematodes (compared to ∼10,000 in our study) but ∼9,000 when developmental stages (embryo, L1, L2, L3, early L4, late L4, and young adult) were included (Grün et al., 2014). Similarly, ∼11,000 proteins were identified from a mixed-developmental stage worm population in an independent study (Schrimpf et al., 2009), although the study additionally reported a very high FDR (∼7%), a large number of single-peptide-based protein identifications (∼23% of the dataset), and an absence of biological replicates. Of the Adult C. elegans Proteome, ∼8% Changes Strikingly during Aging Of the 9,398 protein groups identified in all three biological replicates, those with at least two sequence-unique peptides in each replicate (7,380) were quantified across the day 1, 5, and 10 time points (Figure S3; Table S3). The 95% confidence intervals were defined for each SILAC pair (i.e., day 5/day 1 [M/H] ratio, day 10/day 5 [L/M] ratio, and day 10/day 1 [L/H] ratio) by calculating mean ratio ±2 SDs across the dataset. To compute a statistic for data reproducibility and significance, p values were calculated using ANOVA. Age-invariant control proteins with log2 SILAC pair ratios close to 0 were randomly sampled from the dataset for the ANOVA calculation. Additionally, as measurement error correlates inversely with intensity, the random sampling of controls was controlled for intensity (see the Supplemental Experimental Procedures and Figure S3 for details). Volcano plots show the FDR-corrected p value from ANOVA for each protein group against the respective log2(fold change) for each SILAC pair (Figure 2A). The p values less than 0.05 (95% cutoff) were considered significant, and 627 proteins were found to be significantly altered in abundance across the three time points considered (Table S4). The vast majority of these age-altered proteins had p values < 0.001. Specific Metabolic Processes and Stress Responses Are Preferentially Affected during Aging GO term analysis for biological process was performed on the 627 proteins showing the most significant change in abundance, during the three time points measured, to profile nematode aging, using the Database for Annotation, Visualization and Integrated Discovery (DAVID; Huang et al., 2009) and ReviGo (Supek et al., 2011). This analysis revealed an enrichment of GO terms including aging itself, fatty acid metabolic processes, oxidation-reduction processes, unfolded protein response, and cellular response to stress (Figure 2B). Additionally, a highly significant enrichment of the GO term lipid modification also was observed, together with lipid transport and metabolic processes such as polysaccharide metabolism and cellular nitrogen compound biosynthesis (Figure 2B). Similar results were obtained upon analyzing the same 627 proteins for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment (Kanehisa and Goto, 2000, Kanehisa et al., 2014) using DAVID. Thus, enrichment of cellular metabolic pathways, including amino acid metabolism, fatty acid metabolism, and carbohydrate metabolism, was observed (Figure 2C), along with cellular stress responses. As GO term enrichment analysis does not take into account the direction of changes (i.e., whether a network is up- or downregulated), we further probed our dataset using gene set enrichment analysis (GSEA) as described below. We generated a heatmap of the 627 proteins whose levels showed the most significant change across the three time points examined to assess whether groups of proteins show similar expression profiles during aging (Figure 2D). The heatmap groups proteins that showed similar abundance changes during adulthood, which were then further clustered using an unbiased hierarchical clustering approach. To minimize redundancy in clustering due to large differences in the magnitude of the abundance changes, the SILAC ratios were scaled to range between −1 (maximum decrease) and +1 (maximum increase), with ratios near 0 indicating no change in abundance with age. Using hierarchical clustering based on Ward’s minimum variance method (Ward, 2012), we grouped the age-variant proteins into clusters based on their abundance trend profiles. Although two clusters (proteins that increase with age and proteins that decrease with age; Figures S4C and S4D) may be used to categorize this set of 627 proteins, we found that grouping the proteins into six clusters could be more biologically meaningful, although one must be cautious in interpreting these data as only three time points were considered in this study (Figures 2E, S4C, and S4D). For example, cluster 3 in the six-cluster model, which includes 34 proteins that increase sharply in abundance between days 1 and 5 of adulthood and then decrease sharply between days 5 and 10, is enriched for the GO term lipid glycosylation (Figure 2E). Thus, it will be interesting to carry out follow-up biological experiments to examine whether lipid glycosylation or other lipid modifications decrease during aging and to study the functional consequences of this decrease. In the two-cluster model, this small group of proteins is grouped together with a larger group of proteins whose levels increase during days 1–5 and then continue to increase between days 5 and 10, and the GO term lipid glycosylation is not significantly enriched (Figure S4D). We next extended our analysis to include our entire aging dataset, rather than just the subset of 627 proteins that showed the largest fold change during the three time points, using GSEA (Subramanian et al., 2005). The advantage of this type of analysis is that, in addition to genes (or proteins) associated with gross changes in abundance that are usually defined by arbitrary fold-change cutoffs, GSEA also detects small but reproducible changes in levels of RNAs or proteins, provided a similar trend is observed in multiple constituents of a defined gene set or pathway. Because GSEA, in common with the majority of pathway analysis tools, was designed to work primarily with human genes rather than nematode genes, we first probed the dataset using two nematode gene sets that we compiled, using information from WormBook, WormBase, and the literature (Kenyon, 2010, Murphy and Hu, 2013, Lapierre and Hansen, 2012), for changes in networks known to be associated with aging in C. elegans. We could reproducibly detect and quantify known C. elegans components of the TORC1 and TORC2 complexes and 17 components of the insulin pathway involved in aging, notable exceptions being IST-1, SKN-1, and HSB-1. As shown by density plots (Figure 3A, left panel), nearly all proteins in the insulin-signaling gene set showed a consistent, albeit modest, increase in abundance between days 1 and 10 of adulthood when compared to the total quantified proteome, suggesting the hypothesis that the activity of this pathway late in life accelerates aging. The exceptions were the longevity-promoting transcription factor DAF-16 (FOXO3 in humans) and the serine/threonine protein kinase SGK-1 (SGK1), which decreased in abundance during aging. Since DAF-16 is negatively regulated by insulin/IGF-1 signaling, its downregulation makes sense; however, SGK-1, which was identified with at least ten razor/unique peptides in each of the three biological replicates (∼31% sequence coverage), seems to be a true biological exception. The increase in abundance of the proteins SIR-2.1 (SIRT1) and DAF-18 (PTEN) was more striking than other components of the insulin-signaling gene set. These appear to be biological exceptions, as both proteins are positive regulators of longevity (Kaeberlein et al., 1999, Masse et al., 2005, Schmeisser et al., 2013). At least 9 razor/unique peptides from SIR-2.1 were detected in each biological replicate, and the sequence coverage was ∼29%, while >26 razor/unique peptides from DAF-18 were detected with sequence coverage of ∼49%. A similar trend was observed when the mTOR gene set we compiled was examined, consistent with the literature (Chen et al., 2009, Perkey et al., 2013), with an increase in TORC1 and TORC2 components observed between days 1 and 10 of adulthood (Figure 3A, right panel). Having extracted information about known aging pathways from our dataset, we next performed GSEA for all pathways in the KEGG database using human homologs of the C. elegans proteins quantified in this study. Of the C. elegans protein-coding genes from our proteomic dataset, 4,394 had known human homologs (see the Supplemental Experimental Procedures), with multiple worm genes mapping to a single human homolog in many cases. Consequently, this list comprises 3,067 unique human homologs. To help exclude effects contributing to reproduction rather than organismal aging (Figure S1B), we focused on post-reproductive changes in protein abundance occurring between days 5 and 10 of adulthood. GSEA comparing changes in protein levels in day 5 versus day 10 animals confirmed an enrichment of pathways that also were enriched in the analysis of the subset of 627 proteins whose levels showed the largest change during aging (Figures 2B and 2C). As GSEA also takes into account the direction of the change in abundance, we observed that there is a marked downregulation of proteins involved in cellular metabolic pathways during aging. For example, enzymes involved in carbohydrate, amino acid, fatty acid, as well as drug and retinol metabolism are all decreased between days 5 and 10 of adulthood (Figures 3B and S5A). GSEA Uncovers an Age-Dependent Decline in Peroxisomal Protein Import In addition to age-dependent decreases in cellular metabolism, GSEA also showed a downregulation of proteins functioning in the KEGG pathway peroxisome (Figure 3B). A closer look revealed a decrease in abundance of ∼30 proteins involved in peroxisomal protein import and function. For example, PEX3 (human ortholog)/PRX-3 (C. elegans), involved in the import of peroxisome membrane proteins, decreased in abundance during aging (Figure 4A), as did levels of PEX5/PRX-5, a protein known to be involved in the import of peroxisomal-matrix enzymes (Figures 4A and 4B; Petriv et al., 2002). Additionally, peroxisomal proteins involved in fatty acid oxidation, ether phospholipid biosynthesis, sterol precursor biosynthesis, amino acid metabolism, purine metabolism, and retinol metabolism decreased in abundance between days 5 and 10 of adulthood (Figure 4A). The peroxisomal import protein PRX-5 has not been identified previously in proteomic studies on aging in wild-type nematodes, probably because of a limited depth of proteome coverage. When we mined existing microarray data of C. elegans aging (Youngman et al., 2011, Budovskaya et al., 2008), we found that prx-5 mRNA levels decreased during aging in two independent studies using wild-type nematodes as well as temperature-sensitive sterile mutants (Figure 4C, left and right panels, respectively). Interestingly, analysis of mRNA transcript levels from published data (Youngman et al., 2011) for all the peroxisome proteins identified in our proteomic study revealed that, unlike prx-5 transcript levels, the majority of mRNAs encoding peroxisome proteins were present at similar levels to the overall mRNA population, i.e., the entire dataset (Figure S5B, upper panel). This is in contrast with peroxisome protein levels measured in our proteomic study, the majority of which decreased in abundance when compared with the overall protein population (Figure S5B, lower panel). Examination of global mRNA levels (not just for the subset of peroxisome transcripts; Youngman et al., 2011) and our protein data showed a poor correlation (r = 0.25 [Pearson]; Figure S5C), consistent with previous findings (Walther et al., 2015). To validate our proteomic data and assess whether a gradual failure of the peroxisomal import machinery occurs during aging (Figure 5A), we made use of a C. elegans peroxisome reporter in which GFP is linked to the peroxisome-targeting sequence (PTS1; GFP-SKL) and expressed using an hsp-16.2 heat shock promoter (Thieringer et al., 2003). We verified that GFP-SKL localized correctly to foci indicative of peroxisomes (Thieringer et al., 2003) in the presence of functional import machinery (Figure 5B, middle panel). As predicted, when animals were treated with prx-5 RNAi during adulthood, GFP failed to localize to the peroxisome and instead showed a diffuse cytoplasmic distribution (Figure 5B, right panel; Thieringer et al., 2003). To assess whether the foci were artifacts due to heat stress, we tested and confirmed that GFP localization without the PTS1 target sequence in Phsp-16.2::gfp animals was diffuse and devoid of distinct foci (Figure S6A). Next we asked whether GFP-SKL was similarly mislocalized to the cytoplasm (see Figure 5B) during aging, as predicted from our MS analysis. Day 5 adult worms showed a mixed distribution of GFP-SKL between peroxisomes, indicated by intense spots/foci, and the cytoplasm (diffuse signal), whereas young day 1 animals showed GFP-SKL localized predominantly in peroxisomes (Figure 5C). Quantification of the cytoplasmic GFP signal confirmed an increase by day 5 relative to day 1 (Figure 5C, top left graph). However, as total GFP levels produced using the hsp-16.2 promoter increased by day 5 (Figures S6B and S6C), we normalized the cytoplasmic GFP signal to total GFP intensity. After normalization, cytoplasmic GFP levels were still found to be significantly higher in day 5 adults compared with day 1 adults (Figure 5C, top right graph). Time points beyond day 5 were not considered, as GFP-SKL expression driven by the hsp-16.2 promoter was markedly reduced by day 10 (Figures S6B and S6C). Interestingly, the number of resolved foci (peroxisomes) increased significantly between days 1 and 5 of adulthood (Figures 5C, lower graph, and S6D). A similar observation has been made in senescent human fibroblasts (Legakis et al., 2002), where the authors speculate that this could be due to an age-dependent impairment in the signals that lead to peroxisome growth and division or, alternately, a failure to remove these structures by autophagy. The cytoplasmic mislocalization of GFP-SKL in aging animals was verified using a second strain, in which expression was controlled by the intestine-specific ges-1 promoter (Figure 5D). Thus, we conclude that post-reproductive C. elegans hermaphrodites mislocalize proteins that are normally targeted to the peroxisome in young adults, likely due to a dysfunction of the peroxisome import machinery. Reducing Insulin/IGF-1 Signaling Slows the Rate of Biological Aging without Slowing the Rate of Change of Much of the Age-Dependent Proteome Long-lived insulin/IGF-1-pathway mutants exhibit reduced rates of tissue and behavioral aging. This raises the question of whether proteins that change in abundance with age in wild-type animals (Figure 2) show the expected slower rate of change when lifespan is extended. To test this, we first confirmed, in vivo, the age-dependent change in levels of several proteins whose abundance increased with age in our MS analysis, using GFP::protein-fusion worm strains (Figures 6A and S7A–S7G). Specifically, levels of GFP::fusions to either the DNA-binding protein HMG-11, the linker histone HIS-24, the histone H3 variant HIS-71, a nematode-specific protein (F55B11.4), or the p62/sequestosome homolog SQST-1, an indicator of autophagic flux, all increased during the course of aging in vivo (Figure 6A). Next we asked how these proteins behaved in lifespan mutants. When HMG-11::GFP worms were raised under life-extending conditions (daf-2 RNAi), HMG-11::GFP protein levels were significantly lower (younger pattern) on day 10 of adulthood, relative to control. Conversely, under conditions that accelerate aging and shorten lifespan (hsf-1 RNAi) (Garigan et al., 2002), HMG-11::GFP protein levels were higher (older pattern) on day 5 relative to control. Thus, HMG-11 protein accumulation scales with the overall apparent rate of aging in these cases (Figure 6B). However, surprisingly, this was not the case for any of the remaining proteins. SQST-1::GFP levels were significantly lower in daf-2(RNAi) animals, as predicted, but they were not higher in hsf-1(RNAi) animals (Figure 6C). HIS-24::GFP, HIS-71::GFP, and F55B11.4::GFP protein levels were not lowered in daf-2(RNAi) worms; in fact, the levels of F55B11.4::GFP were higher (older pattern) in daf-2(RNAi) worms when compared with wild-type (Figure S7H). In hsf-1(RNAi) animals, in some cases (day 10 for HIS-24 and day 5 for F55B11.4), protein levels were lower (younger pattern) than in wild-type animals (Figure S7H). These unexpected results prompted us to extend this analysis to include other proteins from the list of 627 proteins whose abundance changed most strikingly during aging. Of these 627 proteins, we concentrated on the 88 that also were identified in a recent study that analyzed the proteomes of wild-type (N2 Bristol), long-lived daf-2(e1370) mutants, and short-lived progeric hsf-1(sy441) mutant worms (Walther et al., 2015). Of the 88 proteins, 55 increased with age in wild-type animals in both the Walther et al. (2015) study and in our study, and 33 decreased with age. A closer look at the 55 proteins that increased with age revealed that 35% (19 proteins) accumulated to a lesser extent (more youthful pattern) in day 12 daf-2(e1370) adults relative to wild-type animals (Figure 6D). These included peptidases (F48E3.4 and T28H10.3), fatty acid-binding proteins (FAR-6 and LBP-2), the tumor necrosis factor-induced protein homolog T05E12.3, the carbohydrate-binding C-type lectin CLEC-146, transthyretin-related proteins (TTR-51, TTR-6, and TTR-30), the galectin LEC-2, which is believed to play a role in programmed cell death, the UDP glycosyltransferase UGT-31, and some nematode-specific proteins of unknown function. Interestingly, a large proportion of the 55 age-increasing proteins (40%) did not show an appreciable change in levels between age-matched N2 and daf-2 mutant animals. Further, 25% (14/55) of them were found to increase to a much greater extent (older pattern) in daf-2(e1370) mutants than in wild-type worms. In principle, these 14 proteins could function to delay aging, consistent with the findings of our lab and other labs that daf-2 mutants express cell-protective mRNA species at high levels even in young adults (Murphy et al., 2003). In support of this hypothesis, we found that this subset of proteins contained the small heat shock protein HSP-12.3. However, this subset also included the aurora-related kinase AIR-2, which could be associated with the longer period of egg laying observed in daf-2(e1370) mutants (Gems et al., 1998). Other proteins in this subset included cytoskeletal and cuticle components (TBB-6, CHT-1, and TTH-1), the fatty acid-binding protein FAR-3, the nucleotide-binding protein CATP-3, and the enzymes HEX-1 (hexosaminidase) and F13H8.11 (phospholipase B1). When the levels of the 55 proteins that increase with age in wild-type animals were measured in hsf-1(sy441) adults at day 6 (hsf-1 mutants are mostly dead by day 12; hence, proteomic measurements during this period are likely to be biased), it was found that 62% (34/55) scaled with biological age, accumulating to a greater extent in these animals when compared with day 6 wild-type control animals (Figure 6D). However, 36% (20/55) of the proteins were unchanged between hsf-1(sy441) and control animals, and one protein of unknown function, T12D8.5, accumulated to a lesser extent in hsf-1(sy441) animals. Similarly, we examined the 33 proteins from the pool of 627 age-variant proteins that were found to decrease during aging in wild-type animals in our study as well as in the Walther et al. (2015) study. Of these, 36% (12/33) scaled with biological age, decreasing to a lesser extent (i.e., the protein was more abundant, younger pattern) in long-lived daf-2(e1370) worms compared to wild-type worms at day 12 of adulthood (Figure 6E); 39% of the proteins remained unchanged between the two populations, and 24% (8/33) were present at lower levels (older pattern) in daf-2(e1370) mutants when compared to wild-type controls. In hsf-1(sy441) mutants at day 6 of adulthood, the levels of 27% (9/33) scaled with biological age, i.e., they were present at lower levels in these animals compared to age-matched controls (Figure 6E). However, the vast majority (61%) were unchanged between wild-type and hsf-1(sy441) worms, and 12% (4/33) were present at higher levels (younger pattern) in hsf-1(sy441) animals. Thus, the aging rate of much of the age-variant proteome, at least in terms of protein abundance, is unchanged under conditions where the rate of biological aging is altered by manipulating insulin/IGF signaling. Discussion In this study, we present an in-depth measurement of age-related changes in protein abundance in C. elegans, facilitated by deep MS-based quantitative proteomic analysis that provides a major increase—almost a doubling—in coverage of the adult C. elegans proteome. Specifically, we have reproducibly identified >9,300 proteins in each of three biological replicates (and >11,000 in at least one of the three experiments), and we have monitored how these vary in abundance at three time points. The depth of proteome coverage obtained, together with the high degree of reproducibility among biological replicates, allowed the definition of strict statistical thresholds for assessing the significance of age-related protein abundance changes, rather than relying simply on arbitrary fold-change cutoffs. Using these stringent thresholds, we identified, in addition to factors previously shown to vary in abundance during aging in C. elegans (Liang et al., 2014, Walther et al., 2015, Zimmerman et al., 2015; see Figures S8A–S8F for comparisons between our study and previous proteomic studies), examples of proteins whose abundance variation and function have not previously been linked to aging. Our dataset provides a valuable resource that can be used to generate hypotheses that will help to further our understanding of aging, which can then be tested in vivo. We provide two examples of this. First, our deep proteome analysis unveiled a decrease in the abundance of ∼30 peroxisomal proteins with age (Figure 4A). In particular, we observed a decrease in abundance of the peroxisomal import protein PRX-5 (PEX5 in humans; Figure 4B) during aging, and we were curious to examine whether this results in impaired peroxisomal protein import during aging. There are reports of an age-dependent mislocalization of catalase to the cytoplasm, and studies in middle- and late-passage human fibroblasts speculate that catalase mislocalization during replicative senescence may be attributed to impaired peroxisome import (Legakis et al., 2002). Whether this phenomenon also is observed during organismal aging and applies to all PTS1-containing peroxisomal proteins or only to catalase, which has a non-canonical PTS1 (Lys-Ala-Asn-Leu rather than Ser-Lys-Leu; Purdue et al., 1996), was still unclear. In this study, we used a peroxisome-targeted GFP-PTS1 reporter and showed, in vivo, that this protein was mislocalized during aging in two independent worm strains (Figures 5C and 5D). Thus, our data support a model wherein impaired peroxisome protein import caused by decreased PRX-5 levels leads to peroxisome dysfunction with age. Like mitochondria, peroxisomes generate reactive oxygen species (ROS), and, although mitochondria are responsible for the majority of the ROS produced within a cell, ROS generated in peroxisomes appear to have a profound impact on mitochondria. Peroxisomal catalase deficiency disrupts mitochondrial redox balance (Hwang et al., 2012), and the induction of ROS production in the peroxisome results in increased mitochondrial fragmentation (a hallmark of aging) and an altered redox potential (Ivashchenko et al., 2011). Peroxisomes and mitochondria therefore appear to be intimately linked, and the interplay between the two organelles could be important in the context of aging. Although prx-5 is essential for nematode larval development (Thieringer et al., 2003), surprisingly, when knocked down in adult nematodes, lifespan was not shortened but, if anything, increased slightly (Zhou et al., 2012, Curran and Ruvkun, 2007). In our hands, adult-only prx-5 RNAi treatment resulted in a significant decrease in brood size (p = 0.00019; Figure S6E), raising the possibility that lifespan is increased in nematodes via the pathway that extends lifespan in response to germline depletion or potentially by a cell-protective response triggered by, for example, an altered redox potential. Interestingly, in yeast, although Δpex5 cells show minimal peroxisome biogenesis defects, they have a dramatically shorter chronological lifespan when grown on 0.5% glucose (mean lifespan ∼7 versus ∼16 days in wild-type cells and maximum lifespan ∼10 versus ∼23 days), underscoring the importance of correctly localized peroxisome matrix proteins imported via functional Pex5p (PEX5/PRX-5) in maintaining normal lifespan (Lefevre et al., 2013). A large compendium of proteins whose levels change with age is a powerful tool to better understand how single-gene mutations, such as daf-2 insulin/IGF-1-receptor mutations, can slow aging and extend lifespan. daf-2 mutants are known to express many cell-protective proteins at elevated levels, even as young adults (e.g., proteasome subunits and molecular chaperones [Depuydt et al., 2013, Walther et al., 2015, Walker et al., 2001] as well as some autophagy components [Hashimoto et al., 2009, Hansen et al., 2008]). To test whether abundances of age-variant proteins scale with the rate of biological aging, we examined the levels of some of the age-variant proteins identified in our study in vivo in daf-2(RNAi) worms. As expected, the accumulation rates of some proteins were slower in daf-2(RNAi) animals than in wild-type. However, surprisingly, this was not the case for all the proteins we examined. When this analysis was expanded to include proteomic data from other studies (Walther et al., 2015), we confirmed that only a minor fraction of the age-variant proteins identified in wild-type animals showed reduced rates of change in long-lived daf-2 mutants. The majority of proteins accumulated at similar rates in chronologically age-matched wild-type animals versus daf-2(e1370) mutants. Stated simply, these proteins didn’t seem to know that they were in a long-lived mutant. These results agree with our lab’s previous study of mRNA changes during early adulthood (days 1–6) in daf-2(−) animals (Murphy et al., 2003), where again, the age-specific abundances of many mRNAs changed at a wild-type rate. However, the significance of this was not clear because these mRNAs might have been linked to reproduction, which takes place during this period, involves two-thirds of the animal’s cells, and was known to be fairly normal in these mutants. Our finding that this observation also holds true for protein levels, and extends well beyond the reproductive phase of adulthood, is thus significant. It suggests that the aging rate of much of the age-variant proteome, at least in terms of protein abundance, is unchanged in these long-lived animals. More fundamentally, it implies that an important biological clock still ticks at a normal rate in these animals, in spite of the change in their rates of morphological decline. These mutants probably age more slowly at the organismal level and live long because the relatively small set of cell-protective proteins and proteostasis regulators, known to be expressed constitutively at high (or low) levels even in young daf-2 adults (Murphy et al., 2003, Dong et al., 2007), alter the extent to which normal time-dependent processes affect biological aging. In summary, here we describe a highly reproducible proteomic dataset containing >11,000 proteins identified in the adult worm (>9,300 reproducibly in three biological replicates), and we show how these proteins change during aging. Network analysis of our data, together with follow-up in vivo studies to test hypotheses generated using our dataset, has identified impairments in peroxisomal protein import during aging and, surprisingly, the finding that the majority of age-variant proteins do not scale with the rate of biological aging in long-lived insulin/IGF-1-receptor mutants. Additionally, this deep proteomic analysis contributes candidate determinants for biological aging (organismal and tissue decline) and a large set of proteins whose abundances, to our knowledge, were not previously shown to change with age, including proteins that may serve as biomarkers of aging and can be tested genetically for a causal role. To maximize the value of this resource to the community, all of these data are provided in a convenient, searchable online database (https://www.peptracker.com/epd). Experimental Procedures Detailed procedures for SILAC labeling, sub-cellular fractionation, LC-MS/MS and data analysis, microscopy, RNAi treatment, and in vivo assays are included in the Supplemental Experimental Procedures. C. elegans Strains and Maintenance The N2 (Bristol) strain was used as wild-type. HZ859 was a gift from Hong Zhang and Malene Hansen, GFP-SKL was from Monica Driscoll, and TH184 and TH237 were from Mihail Sarov. All other strains were from the Caenorhabditis Genetics Center (CGC). The strains were grown and maintained at 20°C as previously described (Brenner, 1974) with sufficient food (E. coli OP50 unless otherwise indicated) for at least three generations prior to use. Data Sharing The data have been assembled into a searchable, online resource with a user-friendly graphical interface, maintained by the A.I.L. lab, to provide convenient and open access to the community (https://www.peptracker.com/epd/). Our data will also shortly be incorporated into WormBase, a resource that is used extensively by the worm community. Author Contributions V.N. conceived and performed the experiments, with input from T.L., A.G., A.I.L., and C.K. V.N., A.I.L., A.G., and C.K. secured funding. V.N. and T.L. wrote R scripts for data analysis. E.P. and V.N. optimized SILAC protocols for worms. A.B.M. integrated the proteomic data into the EPD (https://www.peptracker.com/epd/). V.N., C.K., A.I.L., T.L., and A.G. wrote the manuscript. Accession Numbers The accession numbers for all of the raw MS data, as well as MaxQuant output files (containing protein identification/matching and quantification data), reported in this paper are ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org): PXD004584 and MassIVE (http://massive.ucsd.edu/ProteoSAFe/): MSV000079263. Supplemental Information Document S1. Supplemental Experimental Procedures and Figures S1–S8 Table S1. Sub-cellular Localization Data, Related to Figure 1 Table S2. Identified Proteins, Related to Figure 1 Table S3. Quantified Proteins, Related to Figures 1 and 2 Table S4. Age-Variant Proteins, Related to Figure 2 Document S2. Article plus Supplemental Information Acknowledgments We would like to thank members of the C.K., A.G., and A.I.L. labs and WormBase (Gary Williams, Kevin Howe, and Chris Grove) for helpful discussion, advice, and experimental protocols. We are grateful to Monica Driscoll, Malene Hansen, and Mihail Sarov for strains. Some strains were provided by the CGC, which is funded by NIH Office of Research Infrastructure Programs (P40 OD010440). This research was funded by a Sir Henry Wellcome Postdoctoral Fellowship to V.N. (098646/Z/12/Z), a Wellcome Trust Strategic Award to A.I.L. (073980/Z/03/BR), a Wellcome Trust Senior Fellowship to A.G. (0909444/Z/09/Z), an NIH R01 Grant to C.K. (R01 AG11816), and by Calico Life Sciences LLC, where V.N. and C.K. are now employed. Supplemental Information includes Supplemental Experimental Procedures, eight figures, and four tables and can be found with this article online at http://dx.doi.org/10.1016/j.cels.2016.06.011. Figure 1 SILAC-Based Deep Proteome Analysis Reproducibly Identifies 9,398 Proteins in C. elegans (A) Schematic overview shows the methodology used in this study. (B) Area-proportionate Venn diagram (top) shows reproducibility in protein identifications across three biological replicates, with the number of proteins in each region of the Venn diagram indicated on a schematic (bottom). (C) Principal-component analysis (PCA) plot shows reproducibility in protein quantification across the three biological replicates. (D) Histogram shows log10-transformed protein abundance (MaxQuant intensity) for day 1 measurements. (E) A cumulative plot of protein abundance, as estimated using median protein intensity measurements based on three biological replicates. 9,398 proteins were identified in total with at least one peptide per protein. 90% of the bulk protein mass is made up of only 2,015 proteins (21% of measured proteins). The remaining protein identifications (7,383 or 79%) comprise less than 10% of the bulk protein mass. See also Figures S1–S4 and S8. Figure 2 Of the Detected C. elegans Proteome, 8.5% Changes by a Large Magnitude during Aging (A) Volcano plot of log2-transformed SILAC ratios (fold change) against the negative log10 of the FDR-corrected p value calculated using ANOVA. Shown are the graphs for day 5 versus day 1 fold changes (left), day 10 versus day 5 (center), and day 10 versus day 1 (right). The shaded dark gray box represents the mean ± 1.96 SD for each plot. Horizontal gray lines denote p value cutoffs of 0.05 (∗), 0.01 (∗∗), and 0.001 (∗∗∗). The number of proteins that increase (orange) or decrease (blue) significantly (p value < 0.05) in each graph also is indicated. (B) GO term enrichment (biological process) of the 627 age-variant proteins (blue and orange dots from the three volcano plots in A with p values < 0.05) performed using DAVID and plotted using REVIGO. The size of the bubbles is indicative of the number of proteins annotated with that GO term; bubbles are color coded according to significance. (C) KEGG pathway enrichment of the 627 age-variant proteins calculated using DAVID is shown. (D) Heatmap shows age-dependent changes in abundance of the above proteins generated using R. (E) Hierarchical clustering of the 627 proteins into six groups based on trend profiles. Results from GO term enrichment analysis (p < 0.05) of each cluster using DAVID also are depicted. Only GO terms in clusters 1, 4, 5, and 6 remained significant (p < 0.05) after multiple hypothesis correction (Benjamini). See also Figure S4. Figure 3 GSEA Uncovers a Downregulation of Cellular Metabolic Pathways with Age (A) Density plots depicting the protein abundance of self-defined gene sets (left, insulin-IGF signaling; right, mTOR) relative to the entire dataset collected in this study. C. elegans proteins and their human orthologs are indicated in the figure. Negative regulators, in the context of extended longevity, are annotated with an asterisk (∗). (B) GSEA (KEGG pathways) output of the day 10 versus day 5 changes in protein abundance depicted schematically using Enrichment Map (Cytoscape). The thickness of the lines connecting nodes (i.e., KEGG pathways) indicates the overlap of quantified proteins common to both nodes. See also Figure S5. Figure 4 Levels of the Peroxisome Import Protein PRX-5 Decrease during Aging (A) Schematic depicting an adaptation of the KEGG pathway peroxisome in humans (adapted from www.genome.jp/kegg/). C. elegans homologs of the human proteins are shaded yellow. Also indicated are the proteins quantified in this dataset, with orange, blue, or black arrows to denote whether the proteins were found to increase or decrease in abundance or to remain relatively unchanged between days 5 and 10 of adulthood, as measured in our study. (B) Graph shows the decrease in levels of the peroxisomal import protein PRX-5 during aging, at the indicated time points. (C) Graph shows the decrease in mRNA levels of prx-5 during aging observed in previously published datasets in wild-type nematodes (left; Youngman et al., 2011) and temperature-sensitive sterile mutants (right; Budovskaya et al., 2008). See also Figure S5. Figure 5 Peroxisomal Protein Import Is Impaired during Aging (A) Schematic depicting the correct peroxisomal localization of GFP targeted to the peroxisome (GFP-SKL) in young animals (left panel). If, as predicted by our MS analysis, the peroxisome import machinery is compromised during aging, GFP-SKL import into the peroxisome will be impaired (right panel). (B–D) Scale bar, 10 μm. (B) GFP-SKL is correctly localized to the peroxisome in the intestine of control animals, but mislocalized to the cytosol under conditions where prx-5 is knocked down with RNAi (n = 7). (C) GFP-SKL begins to accumulate in the cytoplasm of aging animals (day 5 adults) when compared to day 1 adults. Shown are representative images from the distal intestinal cells. Cytoplasmic (diffuse) GFP intensity was quantified, summed, and plotted from 81 individual z-sections at 0.5-μm intervals for each worm to minimize masking of the cytoplasmic signal by intense GFP foci (peroxisomes) as well as to minimize bleed through. The cytoplasmic signal was found to increase significantly with age (upper left graph; p value calculated using an unpaired t test = 7.68 × 10−5; n = 11 individuals for day 1 adults and n = 9 for day 5 animals). As the total hsp-16.2-driven GFP-SKL signal also was found to increase with age, cytoplasmic GFP measurements were corrected for total intensity in each image and age-dependent changes remained significant (upper right graph; p value from unpaired t test = 0.009). Also indicated is a count of the number of dots resolved (peroxisomes) at days 1 and 5 from a single z-section. This was found to increase significantly by day 5 (lower graph; p value from unpaired t test = 7.86 × 10−5). (D) As above, but we used worms that express GFP-SKL under the intestine-specific ges-1 promoter (strain VS15). Cytoplasmic GFP-SKL intensity increased significantly with age (p value from unpaired t test = 1.56 × 10−11 for raw cytosolic intensities [left graph] and 4.54 × 10−10 for cytosolic GFP intensity normalized to total GFP intensity [right graph]; n = 15 for day 1 adults and n = 19 for day 5 adults). As in (C), the number of dots resolved also increased significantly with age (lower graph; p value from unpaired t test = 4.46 × 10−8). See also Figure S6. Figure 6 Most Age-Variant Proteins Do Not Scale with Biological Age in Long-Lived Insulin/IGF-1 Mutants (A) Snapshots of GFP fluorescence at days 1, 5, and 10 of adulthood for the indicated GFP::protein fusions. Shown are high exposures at day 1 (upper panels) and comparable exposures at days 1, 5, and 10 (lower panels) (n = 20 per experiment). (B) HMG-11::GFP-expressing worms were grown on the indicated RNAi-bacteria (initiated at L1 stage) and GFP fluorescence was quantified on days 1, 5, and 10 of adulthood. Graphs show mean ± SD (n = 20 worms). Significance was calculated using an unpaired t test (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; n.s., not significant). The daf-2 and hsf-1 dsRNA constructs are in different vector backbones, hence the corresponding empty vector control for each was used. Snapshots of images taken at day 5 (hsf-1 RNAi) and day 10 (daf-2 RNAi) also are shown. (C) As in (B), except we used SQST-1::GFP-expressing hermaphrodites (n = 20 except for day 10 hsf-1 RNAi where n = 11). (D and E) 88 proteins from the 627 age-variant proteins identified in this study were detected in a previous study by Walther et al. (2015) in wild-type, daf-2(e1370), and hsf-1(sy441) animals, and they were found to show similar abundance trends in wild-type animals when compared to our dataset. Of these, those that increased with age in wild-type animals also were quantified in daf-2(e1370) and hsf-1(sy441) worms (D). The stacked bar plot shows how the levels of these proteins vary in the different strains relative to wild-type animals at day 6 (hsf-1) or days 6 and 12 (daf-2) of adulthood, using raw data from Walther et al. (2015). Also indicated are representative examples of proteins present at lower, comparable, or higher levels in day 12 daf-2 mutants compared to wild-type controls. (E) As in (D), except that proteins whose levels were found to decrease in wild-type animals were analyzed. As in (D), raw data from Walther et al. (2015) were used to generate the bar plots. See also Figure S7. ==== Refs References Brenner S. The genetics of Caenorhabditis elegans Genetics 77 1974 71 94 4366476 Budovskaya Y.V. Wu K. Southworth L.K. Jiang M. Tedesco P. Johnson T.E. Kim S.K. An elt-3/elt-5/elt-6 GATA transcription circuit guides aging in C. elegans Cell 134 2008 291 303 18662544 Chen C. Liu Y. Liu Y. Zheng P. mTOR regulation and therapeutic rejuvenation of aging hematopoietic stem cells Sci. Signal. 2 2009 ra75 19934433 Copes N. Edwards C. Chaput D. Saifee M. Barjuca I. Nelson D. Paraggio A. Saad P. Lipps D. Stevens S.M. Jr. Bradshaw P.C. Metabolome and proteome changes with aging in Caenorhabditis elegans Exp. Gerontol. 72 2015 67 84 26390854 Cox J. Mann M. MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification Nat. Biotechnol. 26 2008 1367 1372 19029910 Croll N.A. Smith J.M. Zuckerman B.M. The aging process of the nematode Caenorhabditis elegans in bacterial and axenic culture Exp. Aging Res. 3 1977 175 189 334555 Curran S.P. Ruvkun G. Lifespan regulation by evolutionarily conserved genes essential for viability PLoS Genet. 3 2007 e56 17411345 David D.C. Ollikainen N. Trinidad J.C. Cary M.P. Burlingame A.L. Kenyon C. Widespread protein aggregation as an inherent part of aging in C. elegans PLoS Biol. 8 2010 e1000450 20711477 Depuydt G. Xie F. Petyuk V.A. Shanmugam N. Smolders A. Dhondt I. Brewer H.M. Camp D.G. 2nd Smith R.D. Braeckman B.P. Reduced insulin/insulin-like growth factor-1 signaling and dietary restriction inhibit translation but preserve muscle mass in Caenorhabditis elegans Mol. Cell. Proteomics 12 2013 3624 3639 24002365 Dong M.-Q. Venable J.D. Au N. Xu T. Park S.K. Cociorva D. Johnson J.R. Dillin A. Yates J.R. 3rd Quantitative mass spectrometry identifies insulin signaling targets in C. elegans Science 317 2007 660 663 17673661 Fredens J. Engholm-Keller K. Giessing A. Pultz D. Larsen M.R. Højrup P. Møller-Jensen J. Færgeman N.J. Quantitative proteomics by amino acid labeling in C. elegans Nat. Methods 8 2011 845 847 21874006 Garigan D. Hsu A.L. Fraser A.G. Kamath R.S. Ahringer J. Kenyon C. Genetic analysis of tissue aging in Caenorhabditis elegans: a role for heat-shock factor and bacterial proliferation Genetics 161 2002 1101 1112 12136014 Gems D. Sutton A.J. Sundermeyer M.L. Albert P.S. King K.V. Edgley M.L. Larsen P.L. Riddle D.L. Two pleiotropic classes of daf-2 mutation affect larval arrest, adult behavior, reproduction and longevity in Caenorhabditis elegans Genetics 150 1998 129 155 9725835 Grün D. Kirchner M. Thierfelder N. Stoeckius M. Selbach M. Rajewsky N. Conservation of mRNA and protein expression during development of C. elegans Cell Rep. 6 2014 565 577 24462290 Hansen M. Chandra A. Mitic L.L. Onken B. Driscoll M. Kenyon C. A role for autophagy in the extension of lifespan by dietary restriction in C. elegans PLoS Genet. 4 2008 e24 18282106 Hashimoto Y. Ookuma S. Nishida E. Lifespan extension by suppression of autophagy genes in Caenorhabditis elegans Genes Cells 14 2009 717 726 19469880 Herndon L.A. Schmeissner P.J. Dudaronek J.M. Brown P.A. Listner K.M. Sakano Y. Paupard M.C. Hall D.H. Driscoll M. Stochastic and genetic factors influence tissue-specific decline in ageing C. elegans Nature 419 2002 808 814 12397350 Huang C. Xiong C. Kornfeld K. Measurements of age-related changes of physiological processes that predict lifespan of Caenorhabditis elegans Proc. Natl. Acad. Sci. USA 101 2004 8084 8089 15141086 Huang W. Sherman B.T. Lempicki R.A. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources Nat. Protoc. 4 2009 44 57 19131956 Hwang I. Lee J. Huh J.Y. Park J. Lee H.B. Ho Y.S. Ha H. Catalase deficiency accelerates diabetic renal injury through peroxisomal dysfunction Diabetes 61 2012 728 738 22315314 Ivashchenko O. Van Veldhoven P.P. Brees C. Ho Y.S. Terlecky S.R. Fransen M. Intraperoxisomal redox balance in mammalian cells: oxidative stress and interorganellar cross-talk Mol. Biol. Cell 22 2011 1440 1451 21372177 Jiang H.-C. Hsu J.M. Yen C.P. Chao C.C. Chen R.H. Pan C.L. Neural activity and CaMKII protect mitochondria from fragmentation in aging Caenorhabditis elegans neurons Proc. Natl. Acad. Sci. USA 112 2015 8768 8773 26124107 Kaeberlein M. McVey M. Guarente L. The SIR2/3/4 complex and SIR2 alone promote longevity in Saccharomyces cerevisiae by two different mechanisms Genes Dev. 13 1999 2570 2580 10521401 Kanehisa M. Goto S. KEGG: kyoto encyclopedia of genes and genomes Nucleic Acids Res. 28 2000 27 30 10592173 Kanehisa M. Goto S. Sato Y. Kawashima M. Furumichi M. Tanabe M. Data, information, knowledge and principle: back to metabolism in KEGG Nucleic Acids Res. 42 2014 D199 D205 24214961 Kapahi P. Chen D. Rogers A.N. Katewa S.D. Li P.W. Thomas E.L. Kockel L. With TOR, less is more: a key role for the conserved nutrient-sensing TOR pathway in aging Cell Metab. 11 2010 453 465 20519118 Kenyon C.J. The genetics of ageing Nature 464 2010 504 512 20336132 Kenyon C. Chang J. Gensch E. Rudner A. Tabtiang R. A C. elegans mutant that lives twice as long as wild type Nature 366 1993 461 464 8247153 Kimura K.D. Tissenbaum H.A. Liu Y. Ruvkun G. daf-2, an insulin receptor-like gene that regulates longevity and diapause in Caenorhabditis elegans Science 277 1997 942 946 9252323 Lapierre L.R. Hansen M. Lessons from C. elegans: signaling pathways for longevity Trends Endocrinol. Metab. 23 2012 637 644 22939742 Larance M. Bailly A.P. Pourkarimi E. Hay R.T. Buchanan G. Coulthurst S. Xirodimas D.P. Gartner A. Lamond A.I. Stable-isotope labeling with amino acids in nematodes Nat. Methods 8 2011 849 851 21874007 Lefevre S.D. van Roermund C.W. Wanders R.J. Veenhuis M. van der Klei I.J. The significance of peroxisome function in chronological aging of Saccharomyces cerevisiae Aging Cell 12 2013 784 793 23755917 Legakis J.E. Koepke J.I. Jedeszko C. Barlaskar F. Terlecky L.J. Edwards H.J. Walton P.A. Terlecky S.R. Peroxisome senescence in human fibroblasts Mol. Biol. Cell 13 2002 4243 4255 12475949 Liang V. Ullrich M. Lam H. Chew Y.L. Banister S. Song X. Zaw T. Kassiou M. Götz J. Nicholas H.R. Altered proteostasis in aging and heat shock response in C. elegans revealed by analysis of the global and de novo synthesized proteome Cell. Mol. Life Sci. 71 2014 3339 3361 24458371 Ly T. Ahmad Y. Shlien A. Soroka D. Mills A. Emanuele M.J. Stratton M.R. Lamond A.I. A proteomic chronology of gene expression through the cell cycle in human myeloid leukemia cells eLife 3 2014 e01630 24596151 Masse I. Molin L. Billaud M. Solari F. Lifespan and dauer regulation by tissue-specific activities of Caenorhabditis elegans DAF-18 Dev. Biol. 286 2005 91 101 16153634 Morley J.F. Morimoto R.I. Regulation of longevity in Caenorhabditis elegans by heat shock factor and molecular chaperones Mol. Biol. Cell 15 2004 657 664 14668486 Murphy C.T. Hu P.J. Insulin/insulin-like growth factor signaling in C. elegans WormBook 2013 1 43 24395814 Murphy C.T. McCarroll S.A. Bargmann C.I. Fraser A. Kamath R.S. Ahringer J. Li H. Kenyon C. Genes that act downstream of DAF-16 to influence the lifespan of Caenorhabditis elegans Nature 424 2003 277 283 12845331 Perkey E. Fingar D. Miller R.A. Garcia G.G. Increased mammalian target of rapamycin complex 2 signaling promotes age-related decline in CD4 T cell signaling and function J. Immunol. 191 2013 4648 4655 24078700 Petriv O.I. Pilgrim D.B. Rachubinski R.A. Titorenko V.I. RNA interference of peroxisome-related genes in C. elegans: a new model for human peroxisomal disorders Physiol. Genomics 10 2002 79 91 12181365 Pourkarimi E. Greiss S. Gartner A. Evidence that CED-9/Bcl2 and CED-4/Apaf-1 localization is not consistent with the current model for C. elegans apoptosis induction Cell Death Differ. 19 2012 406 415 21886181 Purdue P.E. Castro S.M. Protopopov V. Lazarow P.B. Targeting of human catalase to peroxisomes is dependent upon a novel C-terminal peroxisomal targeting sequence Ann. N Y Acad. Sci. 804 1996 775 776 8993621 Schmeisser K. Mansfeld J. Kuhlow D. Weimer S. Priebe S. Heiland I. Birringer M. Groth M. Segref A. Kanfi Y. Role of sirtuins in lifespan regulation is linked to methylation of nicotinamide Nat. Chem. Biol. 9 2013 693 700 24077178 Schrimpf S.P. Weiss M. Reiter L. Ahrens C.H. Jovanovic M. Malmström J. Brunner E. Mohanty S. Lercher M.J. Hunziker P.E. Comparative functional analysis of the Caenorhabditis elegans and Drosophila melanogaster proteomes PLoS Biol. 7 2009 e48 19260763 Subramanian A. Tamayo P. Mootha V.K. Mukherjee S. Ebert B.L. Gillette M.A. Paulovich A. Pomeroy S.L. Golub T.R. Lander E.S. Mesirov J.P. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles Proc. Natl. Acad. Sci. USA 102 2005 15545 15550 16199517 Supek F. Bošnjak M. Škunca N. Šmuc T. REVIGO summarizes and visualizes long lists of gene ontology terms PloS One 6 2011 e21800 21789182 Taylor R.C. Dillin A. Aging as an event of proteostasis collapse Cold Spring Harb. Perspect. Biol. 3 2011 a004440–a004440 Thieringer H. Moellers B. Dodt G. Kunau W.H. Driscoll M. Modeling human peroxisome biogenesis disorders in the nematode Caenorhabditis elegans J. Cell Sci. 116 2003 1797 1804 12665560 Timmons L. Fire A. Specific interference by ingested dsRNA Nature 395 1998 854 9804418 Vellai T. Takacs-Vellai K. Zhang Y. Kovacs A.L. Orosz L. Müller F. Genetics: influence of TOR kinase on lifespan in C. elegans Nature 426 2003 620–620 Walker G.A. White T.M. McColl G. Jenkins N.L. Babich S. Candido E.P. Johnson T.E. Lithgow G.J. Heat shock protein accumulation is upregulated in a long-lived mutant of Caenorhabditis elegans J. Gerontol. A Biol. Sci. Med. Sci. 56 2001 B281 B287 11445592 Walther D.M. Kasturi P. Zheng M. Pinkert S. Vecchi G. Ciryam P. Morimoto R.I. Dobson C.M. Vendruscolo M. Mann M. Hartl F.U. Widespread proteome remodeling and aggregation in aging C. elegans Cell 161 2015 919 932 25957690 Ward J.H. Jr. Hierarchical grouping to optimize an objective function J. Am. Stat. Assoc. 58 2012 236 244 Youngman M.J. Rogers Z.N. Kim D.H. A decline in p38 MAPK signaling underlies immunosenescence in Caenorhabditis elegans PLoS Genet. 7 2011 e1002082 21625567 Zhou B. Yang L. Li S. Huang J. Chen H. Hou L. Wang J. Green C.D. Yan Z. Huang X. Midlife gene expressions identify modulators of aging through dietary interventions Proc. Natl. Acad. Sci. USA 109 2012 E1201 E1209 22509016 Zimmerman S.M. Hinkson I.V. Elias J.E. Kim S.K. Reproductive aging drives protein accumulation in the uterus and limits lifespan in C. elegans PLoS Genet. 11 2015 e1005725 26656270

PMC005xxxxxx/PMC5003815.txt

==== Front Mol CellMol. CellMolecular Cell1097-27651097-4164Cell Press S1097-2765(16)30287-810.1016/j.molcel.2016.06.029ResourceComprehensive Identification of RNA-Binding Domains in Human Cells Castello Alfredo 125Fischer Bernd 135Frese Christian K. 1Horos Rastislav 1Alleaume Anne-Marie 1Foehr Sophia 1Curk Tomaz 14Krijgsveld Jeroen 13Hentze Matthias W. hentze@embl.de1∗1 European Molecular Biology Laboratory (EMBL), Meyerhofstrasse 1, 69117 Heidelberg, Germany2 Department of Biochemistry, University of Oxford, South Parks Road, Oxford OX1 3QU, UK3 German Cancer Research Center (DKFZ), Im Neuenheimer Feld 280, 69120 Heidelberg, Germany4 Faculty of Computer and Information Science, University of Ljubljana, 1001 Ljubljana, Slovenia∗ Corresponding author hentze@embl.de5 Co-first author 18 8 2016 18 8 2016 63 4 696 710 17 9 2015 31 5 2016 20 6 2016 © 2016 The Author(s)2016This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).Summary Mammalian cells harbor more than a thousand RNA-binding proteins (RBPs), with half of these employing unknown modes of RNA binding. We developed RBDmap to determine the RNA-binding sites of native RBPs on a proteome-wide scale. We identified 1,174 binding sites within 529 HeLa cell RBPs, discovering numerous RNA-binding domains (RBDs). Catalytic centers or protein-protein interaction domains are in close relationship with RNA-binding sites, invoking possible effector roles of RNA in the control of protein function. Nearly half of the RNA-binding sites map to intrinsically disordered regions, uncovering unstructured domains as prevalent partners in protein-RNA interactions. RNA-binding sites represent hot spots for defined posttranslational modifications such as lysine acetylation and tyrosine phosphorylation, suggesting metabolic and signal-dependent regulation of RBP function. RBDs display a high degree of evolutionary conservation and incidence of Mendelian mutations, suggestive of important functional roles. RBDmap thus yields profound insights into native protein-RNA interactions in living cells. Graphical Abstract Highlights • Experimental generation of an atlas of RNA-binding sites (RBS) in human cells • RBS overlap with enzymatic cores and protein-protein interaction sites • About half of the total RBS map to disordered protein regions • RBS are enriched for phosphorylation, acetylation, and methylation sites Many recently discovered RNA-binding proteins (RBPs) do not show architectural similarities with classical RBPs, and their modes of interaction with RNA were unclear. We developed and employed RBDmap as a method for the comprehensive determination of the RNA-interacting sites of RBPs, identifying more than a thousand such sites. These data yield unprecedented insight into RNA-protein interactions in cells with implications for numerous biological contexts. Published: July 21, 2016 ==== Body Introduction RNA metabolism relies on the dynamic interplay of RNAs with RNA-binding proteins (RBPs) forming ribonucleoprotein complexes, which control RNA fate from synthesis to decay (Glisovic et al., 2008). Due to their central role in cell biology, it is unsurprising that mutations in RBPs underlie numerous hereditary diseases (Castello et al., 2013a, Lukong et al., 2008). Many RBPs are modular, built from a limited pool of RNA-binding domains (RBDs), including the RNA recognition motif (RRM) and other canonical RBDs (Lunde et al., 2007). These domains have been characterized biochemically and structurally, furthering our understanding of protein-RNA interactions. The identification of unorthodox RBPs lacking canonical RBDs expands the scope of physiologically important protein-RNA interactions (e.g., Jia et al., 2008). System-wide approaches to identify RBPs have recently been developed, including immobilization of RNA probes (Butter et al., 2009) or proteins (Scherrer et al., 2010, Tsvetanova et al., 2010), followed by in vitro selection of their interaction partners. These experiments identified numerous proteins previously unknown to bind RNA. While informative, in vitro protein-RNA interactions may arise non-physiologically from the electrostatic properties of RNA. To address this limitation, in vivo UV crosslinking has been used to covalently stabilize native protein-RNA interactions occurring in living cells. After cell lysis, proteins covalently bound to polyadenylated [poly(A)] RNAs are isolated by oligo(dT) selection and identified by quantitative mass spectrometry (Baltz et al., 2012, Castello et al., 2012). This approach (named RNA interactome capture) identified over a thousand RBPs in HeLa and HEK293 cells, hundreds of which were previously unknown to bind RNA. Subsequently, similar data sets were obtained from mouse embryonic stem cells, Saccharomyces cerevisiae, and Caenorhabditis elegans (Beckmann et al., 2015, Kwon et al., 2013, Matia-González et al., 2015, Mitchell et al., 2013), confirming earlier findings and further uncovering the repertoire of RBPs. Several of the unorthodox RBPs identified in these studies have been characterized for their physiological roles in RNA biology. These include metabolic enzymes (Beckmann et al., 2015), regulators of alternative splicing (Papasaikas et al., 2015, Tejedor et al., 2015), the E3 ubiquitin ligase TRIM25 (Choudhury et al., 2014), or the FAST kinase domain-containing protein 2 (FASTKD2) (Popow et al., 2015). However, the RNA-binding regions of these unorthodox RBPs remain largely unknown. To identify the interaction sites of such proteins with RNA, UV crosslinking followed by extensive RNase treatment has been used to detect the peptide mass shift induced by the crosslinked RNA remnant via mass spectrometry (Schmidt et al., 2012). While conceptually simple, the mass heterogeneity of the nucleotide remnant has rendered this approach challenging in practice. Some RBDs have been characterized in vitro using this approach (reviewed in Schmidt et al., 2012), and a sophisticated algorithm allowed assignment of 257 binding sites from 124 proteins in yeast (Kramer et al., 2014). While informative, this data set is strongly enriched for interactions mediated by RRMs, because the challenging identification of peptides with aberrant mass spectra requires both abundance and high crosslinking efficiency for detection. Nonetheless, 10% of the identified interaction sites mapped to non-canonical RBDs, supporting the existence of unanticipated modes of RNA binding. Here, we develop and exploit RBDmap as a method for the in vivo identification of RBDs on a proteome-wide scale. We identified 1,174 high-confidence RNA-binding sites in 529 RBPs from HeLa cells, generating an unprecedented atlas of RNA-binding architectures in vivo. Results and Discussion Proteome-wide Mapping of RBDs by RBDmap To define how RBPs bind to RNA in living cells, we extended RNA interactome capture (Castello et al., 2013b) by addition of an analytical protease digestion step followed by a second round of oligo(dT) capture and mass spectrometry (Figure 1A). First, UV light is applied to cell monolayers to covalently stabilize native protein-RNA interactions taking place at “zero” distance (Pashev et al., 1991). While UV exposure using dosages exceeding those used here can potentially promote protein-protein crosslinking (Davidenko et al., 2016, Suchanek et al., 2005), we could not detect such crosslinks under our conditions, evidenced by the lack of UV-dependent, high molecular weight complexes in RNase-treated samples (Figures S1A and S4A; Strein et al., 2014). Proteins crosslinked to poly(A) RNA are isolated using oligo(dT) magnetic beads and purified by stringent washes that include 500 mM LiCl and chaotropic detergents (0.5% LiDS), efficiently removing non-covalent binders (Castello et al., 2012, Castello et al., 2013b). After elution, RBPs are proteolytically digested by either LysC or ArgC. These proteases were selected as best suited for RBDmap by an in silico simulation of their predicted cleavage patterns of known HeLa RBPs (Castello et al., 2012) and their compatibility with subsequent tryptic digestion (Figure S1B). Analysis by mass spectrometry (MS) of LysC- and ArgC-treated samples revealed an excellent match with the in silico predictions, as reflected by the low number of missed cleavages (Figures 1B and 1C). The extensive proteolysis of HeLa RBPs is achieved without compromising RNA integrity (Figures 1D and S1C–S1E). The average peptide length after LysC and ArgC treatment is ∼17 amino acids, which defines the resolution of RBDmap (Figure 1C). Note that the extensive protease treatment disrupts protein integrity, and thus protein-protein complexes that might have withstood the experimental conditions will be released into the supernatant. We collected an input sample aliquot after UV irradiation, oligo(dT) selection, and protease digestion, which in principle should reflect the RNA interactome (Figure 1A). When compared to a non-irradiated specificity control, the resulting high-confidence RBPs overlap 82% with the previously published human RNA interactomes (Baltz et al., 2012, Beckmann et al., 2015, Castello et al., 2012). This high concordance shows that LysC and ArgC treatments are fully compatible with the RNA interactome capture protocol. The remaining two thirds of the LysC or ArgC-treated samples were subjected to a second round of oligo(dT) purification leading to two peptide pools (Figure 1A): (1) peptides released from the RNA into the supernatant, and (2) peptides remaining covalently bound to the RNA, representing the RNA-binding sites of the respective RBPs. Importantly, subsequent tryptic digestion of the RNA-bound LysC/ArgC fragments yields two classes of peptides: the portion that still remains crosslinked to the RNA (X-link) and its neighboring peptides (N-link) (Figure 1A). While the directly crosslinked peptides (X-link) are difficult to identify due to the heterogeneous mass shift induced by the residual nucleotides (Kramer et al., 2014, Schmidt et al., 2012), the native peptides adjacent to the crosslinking site (N-link) can be identified by standard MS and peptide search algorithms. The original RNA-bound region of the RBP (i.e., RBDpep; Figure 1A), which includes both the crosslinked peptide (X-link) and its unmodified neighboring peptides (N-link), is then re-derived in silico by extending the MS-identified peptides to the two nearest LysC or ArgC cleavage sites. Analysis of the RNA-bound and released fractions by quantitative proteomics shows high correlation of the resulting peptide intensity ratios between independent biological replicates. These ratios follow a bimodal distribution with one mode representing the released peptides (gray) and the other the RNA-bound ones (red; Figures 1E and S1F). We detected 909 and 471 unique N-link peptides as significantly enriched in the RNA-bound fractions of LysC- or ArgC samples, respectively (1% false discovery rate, FDR) (Figure S1G). Notably, computed RNA-bound/released peptide intensity ratios also correlate between the LysC and ArgC data sets (Figure 1F), supporting the robustness of the workflow. Due to their different specificities, each protease also contributes unique 1% FDR RBDpeps to the complete peptide superset (Figure S1G), covering 529 RBPs that highly overlap with human RNA interactomes (Figure 1G) (Baltz et al., 2012, Beckmann et al., 2015, Castello et al., 2012). Proteins within the RBDmap data set range from low to high abundance (Figure S1H), following a similar distribution as the input fraction and the HeLa RNA interactome (Castello et al., 2012). Thus, RBDmap is not selective for highly abundant proteins. There were 154 additional RBPs that were identified here, helped by the reduction of sample complexity and of experimental noise by the additional proteolytic step and the second oligo(dT) capture. In agreement with this explanation, the relative abundance of corresponding RBDpeps is higher in the RNA-bound fractions than in the “input” samples (Figures 1H and S1I). Thus, RBDmap detects RNA-binding regions within hundreds of RBPs in one approach, even if it does not cover all RBPs identified by RNA interactome capture (Figure 1G). Proteins will be missed by RBDmap when (1) binding to non-polyadenylated RNAs, (2) displaying low crosslinking efficiency, (3) interacting with the phospho-sugar backbone, but not the nucleotide bases, or (4) lacking suitable cleavage sites for trypsin within the LysC and ArgC proteolytic fragments and hence lacking MS-identifiable N-link peptides. Thus, the distribution of arginines (R) and lysines (K) will influence whether a given RBP can be studied by RBDmap, and we used two different proteases to maximize the identification of RBDpeps. About half of the RBPs covered by RBDpeps harbor well-established RBDs and play known functions in RNA biology, reflected by a strong and significant enrichment of RNA-related protein domains and biological processes comparable to the HeLa RNA interactome (Figures 1I and S1J). Note that the reduced RBP coverage of RBDmap compared to RNA interactome capture equally affects both well-established and unorthodox RBPs (Figures 1I and S1J). RBDmap “Rediscovers” Classic RBDs Interestingly, RNA-bound and released proteolytic fragments display distinct chemical properties. Released peptides are rich in negatively charged and aliphatic residues, which are generally underrepresented in RNA-binding protein surfaces (Figures 2A, 2B, and S2A). Conversely, RBDpeps are significantly enriched in amino acids typically involved in protein-RNA interactions, including positively charged and aromatic residues. These data show that the chemical properties of the RBDpeps resemble those expected of bona fide RNA-binding surfaces. As a notable exception, glycine (G) is enriched in RBDpeps, but depleted from protein-RNA interfaces derived from available structures (Figures 2A and 2B). Flexible glycine tracks can contribute to RNA binding via shape-complementarity interactions as described for RGG boxes (Phan et al., 2011). Hence, lack of glycine at binding sites of protein-RNA co-structures reflects the technical limitations of crystallographic studies regarding disordered protein segments. Validating the RBDmap data, classical RBDs such as RRM, KH, cold shock domain (CSD), and Zinc finger CCHC, are strongly enriched in the RNA-bound fraction (Figure 2C). This enrichment can also be appreciated at the level of individual protein maps (Figures 2D and S2B–S2D). To evaluate the capacity of RBDmap to identify bona fide RBDs, we focused on RBPs that harbor at least one classical RBD (as listed in Lunde et al., 2007). MS-identified peptides from these proteins were classified as “within” or “outside” a classical RBD, according to their position within the proteins’ architecture (Figure 2E). The relative fraction of peptides within versus outside of the RBD was then plotted for each possible RNA-bound/released intensity ratio (Figure 2F). Correct re-identification of classical RBDs would lead to an ascending line (i.e., within/outside ratios should grow in parallel to the RNA-bound/release ratios; Figure 2E), while a random distribution of peptides within and outside of classical RBDs would yield a horizontal line (i.e., within/outside ratios do not vary in accordance with the RNA-bound/released ratios; Figure 2E). As shown in Figure 2F, the relative fraction of peptides mapping within classical RBDs increases in parallel with the RNA-bound/released ratios. Thus, RBDmap correctly assigns RNA-binding activity to well-established RBDs. Unexpected initially, helicase domains are underrepresented in the RNA-bound fraction (Figure 2C). However, the high number of released helicase peptides likely reflects (1) the transitory and dynamic interactions that helicases establish with RNA, (2) the large protein segments of the domain situated far from the RNA, and (3) the predominance of interactions with the phospho-sugar backbone over nucleotide bases (Figures S2C–S2E) (Bono et al., 2006). Nevertheless, high-confidence RBDpeps are found at the exit of the helicase tunnel, as discussed below (Figures S2C–S2E). High-Resolution Determination of RNA-Binding Sites For direct validation of the RBDmap data, we selected all those RBPs for which protein-RNA co-structures are available within the Protein Data Bank (PDB) repository. These were “digested” in silico with either LysC or ArgC, and the predicted proteolytic fragments were considered as “proximal” to RNA when the distance to the closest RNA molecule is 4.3 Å or less; otherwise, they were categorized as non-proximal (Figure 3A). About half of all LysC and ArgC fragments are proximal to RNA by this criterion, reflecting that many RBP structures are incomplete and focused on the RBDs (average protein coverage ∼50%). By contrast, 70.3% (LysC) and 81% (ArgC), respectively, of RBDpeps qualify as proximal, showing that RBPmap highly significantly enriches for peptides in close proximity to the RNA (Figure 3A). Several factors suggest that the pool of peptides classified as proximal in the analyzed structures even underestimates the performance of RBDmap: (1) in several structures of RBPs that harbor two or more RBDs, only one of the RBDs displays the interaction with RNA (e.g., PDB 3NNC) (Teplova et al., 2010). At least in some of these cases, structures lack RNA contacts of RBDs that likely occur in vivo. (2) Proteins are normally co-crystallized with short nucleic acids (5 to 8 nucleotides), and their physiological RNA partners likely establish additional interactions with the RBP. (3) RNA-protein co-structures usually reflect one interaction state, while protein-RNA interactions are typically more dynamic in vivo (Ozgur et al., 2015, Safaee et al., 2012). RBDmap also correctly assigns RNA-binding regions within large protein complexes such as the nuclear cap-binding complex. The small nuclear cap-binding protein (NCBP) 2 (or CBP20) directly contacts mRNA via the cap structure (m7GpppG), while the larger NCBP1 (CBP80) interacts with NCBP2 (Mazza et al., 2002). In agreement, RBDmap defines the RNA-binding region of NCBP2 within the m7GpppG-binding pocket and no RBDpep is assigned to the large NCBP1 (Figure S3A). Moreover, RBDmap defines the corresponding RNA-binding sites within NCBP2 (Mazza et al., 2002) and its cytoplasmic counterpart eIF4E (Brown et al., 2007) (Figure S3B), in spite of their low sequence identity. The glutamyl-prolyl-tRNA synthetase (EPRS) represents a large non-canonical RBP that harbors two tRNA synthase domains separated by three WHEP motifs (Figures S3C and S3D). The first and second WHEP motif bind the GAIT RNA element present in the 3′ UTRs of a number of pro-inflammatory mRNAs (Jia et al., 2008), in complete agreement with the RBDmap data. To test whether RNA-binding assignments of RBDmap can reach near single-amino acid resolution, we collected the complete set of RBDpeps and released peptides mapping to a given RBD class (e.g., RRM) and assessed their relative position within the domain (from 0 to 1) as well as its adjacent upstream (from −1 to 0) and downstream regions (from 1 to 2) (Figure 3B). The MS-identified part (N-link) of each RBDpep was then subtracted to infer the RNA-crosslinked (X-link) moiety(s), which cannot be identified by conventional MS due to their nucleotide remnant (Figures 1A and 3B). The X-link/released peptide ratio was calculated for each position in the domain, where high prevalence of X-link over released peptides will indicate RNA binding (Figure 3B). The high accuracy of this analysis is illustrated by the example profile obtained for RRMs. As shown in Figures 3C, 3D, and S3E, the highest X-link/released peptide ratio points to β strand 1, 2, and 3 as partners in the interaction with RNA, in agreement with the dozens of RNA-RRM co-structures available. Note that the LysC and ArgC proteases dissected the RRM in a differential manner: while LysC points to β strand 1 and 3, ArgC identifies β strand 2 as RNA-binding site, reflecting that the mapping capacity by these proteases depends on the distribution of lysines and arginines. Moreover, these data support the complementarity of the LysC and ArgC data sets to build accurate and comprehensive RNA-binding maps. Unexpectedly, we observed two discrete peaks of high X-link/released peptide ratio within the α helices placed at the back of the RRM. These peaks coincide with amino acids projected from the α helix to the RNA in several structures (Figure S3F) (Safaee et al., 2012, Teplova et al., 2010) and hence confirm the accuracy of RBDmap. This analysis also successfully assigned correct RNA-binding sites to KH, DEAD-box helicase, and CSD, as shown in Figures 3E–3J, S3G, and S3H. The DEAD box helicase domain establishes interactions primarily with the phospho-sugar backbone of the RNA, while nucleotide bases project away from the protein core (Figure S3I). X-link peptide coverage of RBDmap for the DEAD box domain identifies one alpha helix in the helicase tunnel exit that coincides with the only position in RNA-protein co-crystals where multiple amino acids establish direct contacts with nucleotide bases. Interestingly, different binding orientations of the double-stranded RNA-binding motif (DSRM) have been observed in structural studies (Figure S3J) (Fu and Yuan, 2013, Ramos et al., 2000). The X-link peptide coverage analysis of the DSRM domain highlights the loop separating the second and third β strands as interaction partners with the double-stranded RNA (Figures S3J and S3K). Note that this loop is shown in several RNA-protein co-structures to be projected into the minor grove of the double-stranded RNA helix, establishing numerous interactions with the Watson-Crick paired bases (Lunde et al., 2007). In summary, RBDmap faithfully re-identifies the protein surfaces of canonical RBDs that contact nucleotide bases. Identification of Non-canonical RBDs For more than half of the RBPs characterized by RBDmap, no functional or domain annotation related to RNA biology is currently available (Figures 1I and S1J). RBDpeps identify dozens of unorthodox globular RBDs associated with different molecular functions, including DNA binding, enzymatic cores, mediators of protein-protein interactions, or of protein localization (Figure 4A; Table S2). As an illustrative example, thioredoxin (TXN) catalyzes disulfide bond formation and has recently been discovered in RNA interactomes (Beckmann et al., 2015, Castello et al., 2012). RBDmap identifies an RBDpep at the N terminus of TXN (Figure 4B; Table S1) that overlaps with two solvent-exposed lysines (K3 and 8) highlighted as potential binding sites in the X-link coverage analysis for the TXN fold (Figures 4B and 4C). To evaluate this assignment functionally, we expressed TXN-eGFP fusion proteins in HeLa cells. Following in vivo UV crosslinking, oligo(dT) capture, and stringent washes, green fluorescence in eluates was measured to quantify RNA binding (Figure 4D) (Castello et al., 2013b, Strein et al., 2014). We used unfused eGFP as negative control and the well-established RNA-binding helicase MOV10 as a positive control for RNA binding (Gregersen et al., 2014). Although all the fusion proteins are expressed at similar levels in cells, only TXN-eGFP and MOV10-YFP co-purify with poly(A) RNAs significantly above background (Figure 4E). Mutation of K3 and/or K8 to glutamic acid (E) totally abrogates TXN RNA-binding activity. Conversely, conservative mutation to arginine (R) is tolerated. These results experimentally validate the accurate identification of a previously unknown RNA-binding region by RBDmap. We also noticed clusters of RBDpeps within enzymes. Peptidyl prolyl cis/trans isomerases are classified based on their domain architecture into two groups: PPI and FKBP. This protein superfamily has close links to RNA metabolism, and two members, PPIE and PPIL4, harbor classical RRMs (Mesa et al., 2008). However, RNA interactome studies found 11 additional members of this family that lack RRMs as RBPs, suggesting the existence of a still unknown mechanism of RNA binding (Castello et al., 2012). RBDmap reveals this RNA-binding activity within both the PPI and FKBP folds (Tables S1 and S2). Although lacking sufficient peptide coverage to perform an X-link peptide analysis, we noticed two clusters of RBDpeps at the N- and C-termini of the FKBP fold that are located far apart in primary sequence, but close in 3D structure (Figures S4B and S4C). The mapped candidate RBD opposes the catalytic site. Furthermore, we noticed clusters of RBDpeps in six chaperones of the heat shock protein (HSP) 90 and 70 families (Figure S4D). HSPs are induced by cellular stress and prevent protein misfolding and subsequent aggregation, which typically occur in disordered regions of RBPs in health and disease (Weber and Brangwynne, 2012). Indeed, HSPs have been functionally linked to RNA metabolism and translation (Iwasaki et al., 2010, Willmund et al., 2013). Chaperone domain binding to RNA may help to increase the local concentration of the chaperone machinery at ribonucleoprotein complexes to avoid the accumulation of pathological aggregates. Apparently, numerous enzymes of intermediary metabolism bind RNA through regions in close proximity to their substrate-binding pockets. Specifically, the di-nucleotide binding domain (or Rossmann fold) and mono-nucleotide binding folds emerge as bona fide RBDs with 12 proteins mapped by RBDmap (Table S3), extending earlier observations (Cieśla, 2006, Nagy and Rigby, 1995). RBDpeps mapping to Aldolase (ALDO) A and C delimit the fructose 1,6 bisphosphate interacting domain (Figures S4E and S4F), suggesting that RNA and metabolite may compete for this binding pocket. Overall, the RBDpeps identified within metabolic enzymes show that the few well-characterized examples such as aconitase 1 (iron regulatory protein 1, IRP1), glyceraldehyde-3-phophate dehydrogenase, and thymidylate synthase may represent the tip of the iceberg of a more general engagement of metabolic enzymes with RNA (reviewed in Castello et al., 2015). RBDmap also uncovers RNA-binding activities within PDZ, 14-3-3, ERM, and the tubulin-binding domains, which are involved in protein-protein interactions and protein localization (Figures 4F, 4G, and S4G–S4I). Due to the high peptide coverage of the PDZ domain, we could generate an X-link analysis (Figures 4F and 4G). This map shows a discrete RNA-binding site within a basic cavity formed by a short α helix and two β strands. RBDmap also identifies RNA-binding sites within domains of unknown function such as NDR and DZF. N-myc downstream-regulated genes (NDRGs) represent a family of proteins with unknown function. NDRG1 is a metastasis suppressor relevant for cancer progression and prognosis (Chang et al., 2014), its exact molecular function has remained unknown. RBDmap resolves a conserved RNA-binding region within the NDR domain of NDRG1, NDRG2, and NDRG4. RBDpeps reproducibly map to the helix-loop-β strand structure at the C terminus of the NDR fold (Figures S4J and S4K). DZF is predicted to harbor nucleotidyltransferase activity (Kuchta et al., 2009) and to promote protein dimerization (Wolkowicz and Cook, 2012). The X-link peptide coverage analysis maps the RNA-binding region to a deep, basic cleft between two symmetrical domain subunits (Figures 4H and 4I). The RNA-binding activity of the DZF domain is compatible with its proposed nucleotidyltransferase function. To independently assess RNA-binding of PDZ and DZF domains, we used the T4 polynucleotide kinase (PNK) assay as an orthogonal approach. In brief, cells are irradiated with UV light and, after lysis, RNA is trimmed with RNase I. Proteins of interest are immunoprecipitated under stringent conditions and the presence of RNA revealed by 5′ end phosphorylation with PNK and [γ-32P]-ATP, followed by SDS-PAGE and autoradiography. We generated Tet-inducible HeLa cell lines expressing the PDZ domain of β-1-syntrophin (SNTB) 1 and SNTB2, as well as the DZF domains of Zinc finger RNA-binding protein (ZFR) and interleukin enhancer-binding factor (ILF) 2 and ILF3, all fused to a FLAG-HA tag. As positive controls, we used the full-length ILF3 (FL), its DSRM domain alone, and hnRNPC, while actin (ACTB) was used as a negative control. The PNK assay shows radioactive bands of the expected molecular weight for all tagged PDZ and DFZ domains and only when UV light was applied to the cultured cells (Figures 4J and 4K). By contrast, no signal is detectable for the control ACTB. As expected, the DSRM domain of ILF3 also displays RNA-binding activity. Taken together, these data corroborate the RBDmap assignment of PDZ and DZF domains as RBDs. Even if functional studies will have to define the physiological roles of these unconventional RBDs in the future, their biological relevance warrants consideration. It is possible that these RBDs may endow RBPs with “moonlighting” activities in posttranscriptional regulation, akin to cytosolic aconitase (IRP1) (Muckenthaler et al., 2008). Alternatively, the RBDs could serve as “docking sites” for regulatory or scaffolding RNAs that inhibit, activate, or modify protein functions. In analogy, innate immune effectors such as PKR, TLR3, TLR7, TLR8, or RIG-I, can be controlled by pathogen-derived RNAs (Barbalat et al., 2011, Yu and Levine, 2011). RNA may also serve to recruit proteins to RNPs, akin to NEAT1 RNA in paraspeckle formation (Clemson et al., 2009). The identification of these RBDs and the mapping of the RNA-interaction sites for hundreds of proteins serve as a critical step toward definition of the biological functions of these RBPs in detail. Disordered Regions Emerge as Frequent RNA Interaction Sites In Vivo A high proportion of the human RBPs lack native 3D structure (Castello et al., 2012), and these disordered regions can occasionally engage in non-canonical protein-RNA interactions (45 examples reviewed in Järvelin et al., 2016). In some instances, these interactions can induce co-folding of both molecules (Phan et al., 2011). While this mode of interaction emerged recently, the scope of disordered motifs involved in RNA-binding remained unknown. Strikingly, half of the RBDpeps map to disordered regions, and RBDmap identifies a disordered RBD as the sole detectable RNA-binding site for 170 RBPs (Figures 5A 5B,, and S5A). Disordered RBDpeps largely mirror the chemical properties of the whole RBDpep superset, apart from the expected enrichment for disorder-promoting residues (proline [P], serine [S], and glycine [G]), as well as R and glutamine (Q) (Figures 5C and S5B). Detailed analysis identifies clusters of disordered RBDpeps that can be classified on the basis of sequence motifs. While a few R-rich, RGG, and SR repeats have previously been shown to bind RNA experimentally (Järvelin et al., 2016), RBDmap expands the RNA-binding role of these motifs by dozens of additional examples (Figures 5D and S5C). The superset of RNA-binding RGG boxes can be subclassified by the lengths of the glycine linkers (Thandapani et al., 2013). Because glycines can position arginines and contribute to RNA binding providing shape complementarity, G-linker length could serve in setting the motif’s specificity for RNA. In agreement, both arginine and glycine substitutions impair RGG-RNA recognition (Phan et al., 2011). Aromatic residues are typically found in hydrophobic cores. However, histidines (H), phenylalanines (F), and especially tyrosines (Y) occur within the RNA-binding disordered regions (Figures 5D and S5C). YGG repeats (also called [G/S]Y[G/S]) can promote protein aggregation in vitro, inducing hydrogel formation and amyloid-like fibers, as well as dynamic phase transitions in vivo (Han et al., 2012, Kato et al., 2012). Since YGG repeats are identified as a potential RNA-binding motif in our data set, it will be important to elucidate whether their RNA-binding capacity is affected by the aggregation state and, conversely, whether RNA-binding to such disordered linear motifs can affect phase transitions and granule formation (Zhang et al., 2015). Lysine (K) combines with negatively charged residues, G, P, or Q, to form distinctive RNA-binding motifs (Figures 5D and S5C). The stoichiometry and distances between lysines and other amino acids are similar across analogous K-rich motifs present in non-homologous proteins (Figure 5E). Several copies of a repeat combining basic and acidic residues within the neuroblast differentiation-associated protein AHNAK are identified by RBDmap (Figure S5D), suggesting that low complexity regions can contribute to modular RNA-binding architectures, similar to globular RBDs. Interestingly, the K-rich regions within RBPs display similarities with the basic tails of DNA-binding proteins. The large capture radius of these disordered regions play important roles in transcription factor activity by favoring “hopping” and “sliding” over 3D diffusion to reach their target sequences (Vuzman et al., 2010). K-rich sequences may play similar roles in RBPs. To validate the disordered regions identified by RBDmap as bona fide RNA-binding motifs, we fused the RGG-rich and the K-rich sequences from FUS and Methyl-CpG-binding protein 2 (MECP2), respectively, to eGFP and tested the fusion proteins with the same assay as in Figure 4D: both short motifs suffice to confer RNA-binding to eGFP (Figures 5F and 5G). The biological function and mode of interaction of disordered regions with RNA should be further investigated. Uncovering Biological Properties of RBDs Previously unknown RNA-binding globular and disordered regions display similar mean isoelectric points as known RBDs (Figure 6A), while their released counterparts exhibit a significantly lower isoelectric point, as expected. Thus, (1) both previously unknown and well-characterized RBDs share common chemical properties, (2) they differ from released fragments, and (3) the unorthodox RBDs do not artificially associate with RNA due to an abnormally high isoelectric point. Established RBPs and proteins harboring previously unknown globular and disordered RBDs display very similar mRNA abundance profiles, ranging from low to high levels, with a slight tendency to lower abundance for the unconventional folded and disordered RNA-binding regions (Figures 6B and 6C). Thus, proteins with unorthodox RBDs are not biased toward high abundance. Notably, RBDpeps in both globular and disordered RBDs are more highly conserved throughout evolution than their released counterparts (Figure 6D), suggesting functional relevance. Cross-referencing of the RBDpep data sets with databases of curated posttranslational modifications shows that RNA-binding sites represent hot spots for defined post-translational modifications (PTMs, p = 2.025 × 10−08), including tyrosine phosphorylation, methylation, acetylation, and malonylation (Figure 6E). This finding suggests that, reminiscent of chromatin remodeling, RBDs are posttranslationally regulated and respond to signaling and metabolic cues. The conserved amino acid contexts of these PTMs implicate sequence-selective modifying enzymes (Figure 6F). Interestingly, acetylation frequently occurs in a lysine two positions upstream of a conserved proline (Figure 6F). Proline isomerization in the basic tail of histone H3 is regulated by acetylation of adjacent lysines and has notable consequences for protein conformation (Howe et al., 2014). Our results suggest the possibility that this regulatory mechanism could also apply to RBP regulation. Our data also show that Mendelian disease mutations cluster within RBDs compared to natural variants (p = 0.0001796) (Figure 6G; Table S4). For example, one RBDpep maps to an RGG-box in FUS that is a hotspot for disease-associated mutations (Figure 6H) (Shang and Huang, 2016), and the RNA-binding activity of this region is validated here by an orthogonal approach (Figures 4D and 5G). Interestingly, a mutation in this region (R495X) causes amyotrophic lateral sclerosis (ALS) and correlates with impaired interaction of FUS with the SMN complex and reduced localization to nuclear gems (Yamazaki et al., 2012). The relationship between altered RNA-binding and disease phenotypes in this and other proteins deserves further exploration. Conclusions RBDmap provides an unprecedented identification of RNA-binding regions of RBPs in living cells. It describes 1,174 high confidence (1% FDR) RNA-binding sites within 529 proteins. These sites have been validated as a whole by stringent statistical analyses (Figure 1) and cross-correlation with well-established RBPs and domains, previously studied by biochemical and structural means (Figures 2 and 3). We also validated a small number of previously unknown RBDs (TXN, PDZ, DZF, and the disordered regions of MECP2 and FUS) individually, applying orthogonal methods (Figures 4 and 5). Against this background, we recommend similar validation experiments for any individual RBD of interest before further in depth analyses. Our data suggest that multifunctional globular domains, which combine RNA-binding with enzymatic functions or protein-protein interaction surfaces, are commonplace, not rare exceptions. These invoke additional functions for RNA, including the (allosteric or competitive) control of catalytic activities and of protein-protein interactions. Moreover, disordered regions are found to play common roles in native protein-RNA interactions, comprising half of the total RNA-binding sites identified. The RNA-binding motifs identified here share physico-chemical features of well-established RBDs, are conserved across evolution, and represent hot spots for posttranslational modifications and disease-associated mutations. Individually and in combination, these features suggest important biological roles. As a method, RBDmap can now be applied to other cell types and organisms such as S. cerevisiae, Caenorhabditis elegans, or Drosophila melanogaster to study the evolution of RBDs. It can also be applied to cells subjected to different experimental conditions to investigate the responses of RBPs to physiological cues such as e.g., stress, starvation, or differentiation. Experimental Procedures RBDmap Initial UV crosslinking and oligo(dT) purification followed the mRNA interactome capture protocol (Castello et al., 2013b). Complete proteolytic digestions were performed with LysC or ArgC for 8 hr at 37°C. Polyadenylated RNA and crosslinked peptides were diluted in 20 mM Tris-HCl, 500 mM LiCl, 1 mM DTT, and 0.5 mM EDTA and recaptured on oligo(dT) beads. The supernatant was processed for MS (released peptides). oligo(dT) beads were washed as in Castello et al. (2013b). All fractions were treated with trypsin and labeled with stable isotopes in vitro (Boersema et al., 2008). Peptides were analyzed on a liquid chromatography-tandem MS (LC-MS/MS) platform. The R-scripts used for the analyses can be found in the R/Bioconductor data-package RBDmapHeLa (http://www.bioconductor.org). RBDmap data can be accessed under http://www-huber.embl.de/users/befische/RBDmap. MS, Protein Identification, and Quantification Proteins were processed following standard protocols, and the resulting peptides were labeled with stable isotopes in vitro, fractionated, and analyzed on a nano-HPLC system (Proxeon) or nano-Acquity UPLC system (Waters) coupled directly to an LTQ Orbitrap Velos (Thermo Fisher Scientific). Data Analysis A complete description of data analysis can be found in the Supplemental Information. Fluorescence-Based Method to Measure RNA-Binding In Vivo and PNK Assay Tet-on HeLa cells expressing eGFP fusion proteins were generated as described elsewhere (Castello et al., 2012). Upon induction, cells were UV irradiated and subjected to small scale RNA interactome capture (Castello et al., 2013b). Eluates were measured in a plate reader. For PNK assays, cell monolayers were irradiated with 150 mJ/cm2 UV254 (Castello et al., 2013b). After cell lysis and RNase treatment, FLAG-HA tagged proteins were immunoprecipitated with an anti-FLAG antibody coupled to magnetic beads (M8823, Sigma Aldrich) and processed as in Beckmann et al. (2015). More detailed information can be found in the Supplemental Information. Author Contributions A.C., B.F., and M.W.H. contributed to the conception and design of the project. A.C., R.H., and A.-M.A. carried out the experimental work. C.K.F., S.F., and J.K. performed the proteomic analyses. B.F., T.C., A.C., C.K.F., J.K., and M.W.H. performed the data analyses. A.C. and M.W.H. wrote the manuscript with input from all authors. Accession Numbers The accession number for the proteomics data reported in this paper is ProteomeXchange Consortium (http://www.proteomexchange.org): PXD000883. Supplemental Information Document S1. Supplemental Experimental Procedures, Figures S1–S5, and Tables S2, S3, and S5 Table S1. List of RBDs and Their Respective Peptides, Identified by RBDmap, Related to Figures 1 and S1 Table S4. Mendelian Mutations Occurring within the RNA-Bound Fragments of RBPs and Their Associated Diseases, Related to Figure 6 Document S2. Article plus Supplemental Information Acknowledgments We thank Drs. Benedikt Beckmann and the M.W.H. group for helpful discussions. A.C. is funded by MRC Career Development Award #MR/L019434/1. M.W.H. acknowledges support by ERC Advanced Grant ERC-2011-ADG_20110310 and the Virtual Liver Network of the German Ministry for Science and Education. C.K.F. is supported by EMBO postdoctoral fellowship LTF1006-2013. Supplemental Information includes Supplemental Experimental Procedures, five figures, and five tables and can be found with this article online at http://dx.doi.org/10.1016/j.molcel.2016.06.029. Figure 1 In Vivo Identification of RBDs by RBDmap (A) Schematic representation of the RBDmap workflow. (B) LysC- and ArgC-mediated proteolysis was monitored without trypsin treatment. The protease digestion under RBDmap conditions or in buffers typically used in MS studies (optimal) were compared to in silico digestions defining 0% miscleavage. The missed cleavages were calculated and plotted. (C) Distribution of MS-identified LysC/ArgC fragments based on their number of amino acids. (D) Silver staining shows the protein pattern of purified RBPs prior to and after LysC treatment (crosslinking: CL). (E) Scatter plot comparing the peptide intensity ratios between RNA-bound and released fractions. The peptides enriched in the RNA-bound fraction at 1% (RBDpep) and 10% FDR (candidate RBDpep) are shown in red and salmon, respectively (Pearson correlation coefficient: r). (F) Peptide intensity ratios between LysC and ArgC experiments computed from three biological replicates. The dots represent released peptides (blue), RBDpeps (red), candidate RBDpeps (salmon), and background peptides (gray). (G) Venn diagram comparing the proteins within the RBDmap data set and the HeLa, HEK293, and Huh-7 RNA interactomes. (H) Comparison of the peptide intensity ratios from three biological replicates between UV-irradiated and non-irradiated inputs (x axis) and between RNA-bound and released fractions (y axis) (color code as above). (I) Number of proteins harboring recognizable or unknown RBDs in the HeLa mRNA interactome (left) and in RBDmap dataset (right). See also Table S1 and Figure S1. Figure 2 Identification of Well-Established RBDs by RBDmap (A) Amino acid enrichment within RBDpeps (salmon) over released (blue) proteolytic fragments (∗, 10% FDR and ∗∗, 1% FDR). (B) Amino acid enrichment within RNA-binding protein surfaces (≤4.3 Å to the RNA) over distant regions (>4.3 Å from the RNA) extracted from protein-RNA co-structures. (C) Bar plot showing the odds ratio of the most enriched known RBDs. (D) Distribution of RBDpeps and released fragments in a classical RBP. The x axis represents the protein sequence from N to C terminus, and the y axis shows the RNA-bound/released peptide intensity ratios. The protein domains are shown in boxes under the x axis (LysC: L and ArgC: A). (E) Schematic representation of RBDpeps mapping within or outside of classical RBDs (left). The idealized outcome of a perfect correlation between RBDpeps and classical RBDs (top right) and random distribution are shown (bottom right). (F) Computed ratio of peptides mapping within known RBDs versus outside RBDs, regarding their peptide RNA-bound to released ratios. The horizontal line represents the baseline for uncorrelated data (i.e., the proportion of peptides mapping to classical RBD in the whole validation set in absence of enrichment; see E bottom). See also Table S2 and Figure S2. Figure 3 RBDmap Identifies RNA-Binding Regions with High Accuracy (A) Schematic representation of proximal and non-proximal peptides (left). The proteins within protein-RNA co-structures were digested in silico with LysC or ArgC and predicted fragments aligned with the RBDpep supersets. The left bars represent the proportion of proximal and non-proximal LysC/ArgC fragments in the complete structure superset (random probability). The right bars show the % of aligned RBDpeps that are RNA proximal or non-proximal (∗∗∗p < 0.001). (B) Schematic representation of the X-link peptide coverage analysis. (C) x axis represents the relative position of the RRM (from 0 to 1) and their upstream (−1 to 0) and downstream (1 to 2) regions. The ratio of the X-link over released peptides at each position of the RRM and surrounding regions using the LysC data set was plotted (top). The secondary structure prediction for each position of the RRM and flanking regions is shown (bottom). (D) The ratio of X-link over released peptides was plotted in a representative RRM-RNA structural model (PDB 2FY1) using a heatmap color code. (E) As in (C), but for the DEAD-box domain. (F) As in (D), but using the PDB 2J0S as a DEAD-box helicase model. (G) As in (C), but for the KH domain. (H) As in (D), but using the PDB 4B8T as a model for a KH domain. (I) As in (C), but for the CSD. (J) As in (D), but with the PDB 3TS2 as a model for a CSD. See also Table S2 and Figure S3. Figure 4 Globular RBDs Discovered by RBDmap (A) Odds ratios for the most highly enriched RBDs. (B) RBDpep and released peptides mapping to TXN as in Figure 2D (top). The ratio of the X-link over released peptide coverage at each position of the TXN fold as in Figure 3C is shown (middle). The secondary structure prediction for each position of the TXN fold and flanking regions is shown (bottom). (C) Crystal structure of human TXN (PDB 3M9J), K3 and K8 are highlighted, and the identified RBDpep is shown in red. (D) Schematic representation of the protocol for measurement of RNA-binding using eGFP fusion proteins. (E) Relative total (input) or RNA-bound (eluate) green fluorescence signal from cells expressing different eGFP fusion proteins (∗∗∗p < 0.01, t test, and n = 9). (F) As in (B), but for PDZ domain. (G) Ratio of X-link over released peptides plotted as a heatmap in a PDZ homology model. (H) As in (B), but for DZF domain. (I) As in (G), but using a DZF homology model. (J) Autoradiography of FLAG-HA tagged proteins after PNK assay. (K) Western blotting using an antibody against the HA tag. The polypeptides of the expected molecular masses are indicated by asterisks. See also Tables S2, S3, and S5 and Figure S4. Figure 5 Disordered Protein Regions as RBDs (A) Number of RBDpeps mapping to globular and disordered domains. (B) Number of proteins mapped by at least one RBDpep solely in globular domains, in globular and disordered domains, or only in disordered motifs. (C) Amino acid enrichment between globular (violet) and disordered (pink) RBDs (∗, 10% FDR and ∗∗, 1% FDR). (D) Multiple sequence alignment of short, disordered RBDpeps with clustal omega. The sequence logos were extracted from aligned disordered fragments. (E) Examples of alignment of K-rich protein motifs. (F) Disordered RNA-binding motifs from FUS and MECP2 expressed as eGFP fusion. (G) Relative total (input) or RNA-bound (eluate) green fluorescence signal from cells expressing FUS449–518-eGFP, MECP2267–316-eGFP, or unfused eGFP as a negative control (∗∗p < 0.01, t test, and n = 6). See also Figure S5. Figure 6 Features of Known and Previously Unknown RBDs (A) Dots show the mean isoelectric point of all LysC and ArgC fragments (the bars represent SEM) (∗∗∗p < 0.01 and not statistically significant: n.s.). (B) Density plot comparing mRNA abundances of known RBPs and previously unknown globular and disordered RBPs. (C) Dots show the mean of the mRNA abundance of the protein groups described in (B) (∗p < 0.05 and not statistically significant: n.s.). (D) Bar plot showing the conservation of RBDpeps and released fragments using Homo sapiens as reference (∗p < 0.05 and ∗∗p < 0.01). (E) Odds ratios for the most enriched PTMs in RBDpeps versus released fragments. (F) Sequence logos of conserved amino acids around posttranslational modifications. A position weight matrix is computed from all 12-mer sequences around the modified residue (10% FDR amino acids are shown). (G) Bar plot showing the odds ratio of Mendelian mutations occurring in RNA-bound over released fragments. (H) RBDmap of FUS. The position of the disease-associated mutations is represented as red or blue colored circles if mapping within or outside an RBDpep, respectively. See also Table S4. ==== Refs References Baltz A.G. Munschauer M. Schwanhäusser B. Vasile A. Murakawa Y. Schueler M. Youngs N. Penfold-Brown D. Drew K. Milek M. The mRNA-bound proteome and its global occupancy profile on protein-coding transcripts Mol. Cell 46 2012 674 690 22681889 Barbalat R. Ewald S.E. Mouchess M.L. Barton G.M. Nucleic acid recognition by the innate immune system Annu. Rev. Immunol. 29 2011 185 214 21219183 Beckmann B.M. Horos R. Fischer B. Castello A. Eichelbaum K. Alleaume A.M. Schwarzl T. Curk T. Foehr S. Huber W. The RNA-binding proteomes from yeast to man harbour conserved enigmRBPs Nat. Commun. 6 2015 10127 26632259 Boersema P.J. Aye T.T. van Veen T.A. Heck A.J. Mohammed S. Triplex protein quantification based on stable isotope labeling by peptide dimethylation applied to cell and tissue lysates Proteomics 8 2008 4624 4632 18850632 Bono F. Ebert J. Lorentzen E. Conti E. The crystal structure of the exon junction complex reveals how it maintains a stable grip on mRNA Cell 126 2006 713 725 16923391 Brown C.J. McNae I. Fischer P.M. Walkinshaw M.D. Crystallographic and mass spectrometric characterisation of eIF4E with N7-alkylated cap derivatives J. Mol. Biol. 372 2007 7 15 17631896 Butter F. Scheibe M. Mörl M. Mann M. Unbiased RNA-protein interaction screen by quantitative proteomics Proc. Natl. Acad. Sci. USA 106 2009 10626 10631 19541640 Castello A. Fischer B. Eichelbaum K. Horos R. Beckmann B.M. Strein C. Davey N.E. Humphreys D.T. Preiss T. Steinmetz L.M. Insights into RNA biology from an atlas of mammalian mRNA-binding proteins Cell 149 2012 1393 1406 22658674 Castello A. Fischer B. Hentze M.W. Preiss T. RNA-binding proteins in Mendelian disease Trends Genet. 29 2013 318 327 23415593 Castello A. Horos R. Strein C. Fischer B. Eichelbaum K. Steinmetz L.M. Krijgsveld J. Hentze M.W. System-wide identification of RNA-binding proteins by interactome capture Nat. Protoc. 8 2013 491 500 23411631 Castello A. Hentze M.W. Preiss T. Metabolic enzymes enjoying new partnerships as RNA-binding proteins Trends Endocrinol. Metab. 26 2015 746 757 26520658 Chang X. Xu X. Ma J. Xue X. Li Z. Deng P. Zhang S. Zhi Y. Chen J. Dai D. NDRG1 expression is related to the progression and prognosis of gastric cancer patients through modulating proliferation, invasion and cell cycle of gastric cancer cells Mol. Biol. Rep. 41 2014 6215 6223 24985974 Choudhury N.R. Nowak J.S. Zuo J. Rappsilber J. Spoel S.H. Michlewski G. Trim25 is an RNA-specific activator of Lin28a/TuT4-mediated uridylation Cell Rep. 9 2014 1265 1272 25457611 Cieśla J. Metabolic enzymes that bind RNA: yet another level of cellular regulatory network? Acta Biochim. Pol. 53 2006 11 32 16410835 Clemson C.M. Hutchinson J.N. Sara S.A. Ensminger A.W. Fox A.H. Chess A. Lawrence J.B. An architectural role for a nuclear noncoding RNA: NEAT1 RNA is essential for the structure of paraspeckles Mol. Cell 33 2009 717 726 19217333 Davidenko N. Bax D.V. Schuster C.F. Farndale R.W. Hamaia S.W. Best S.M. Cameron R.E. Optimisation of UV irradiation as a binding site conserving method for crosslinking collagen-based scaffolds J. Mater. Sci. Mater. Med. 27 2016 14 26676860 Fu Q. Yuan Y.A. Structural insights into RISC assembly facilitated by dsRNA-binding domains of human RNA helicase A (DHX9) Nucleic Acids Res. 41 2013 3457 3470 23361462 Glisovic T. Bachorik J.L. Yong J. Dreyfuss G. RNA-binding proteins and post-transcriptional gene regulation FEBS Lett. 582 2008 1977 1986 18342629 Gregersen L.H. Schueler M. Munschauer M. Mastrobuoni G. Chen W. Kempa S. Dieterich C. Landthaler M. MOV10 Is a 5′ to 3′ RNA helicase contributing to UPF1 mRNA target degradation by translocation along 3′ UTRs Mol. Cell 54 2014 573 585 24726324 Han T.W. Kato M. Xie S. Wu L.C. Mirzaei H. Pei J. Chen M. Xie Y. Allen J. Xiao G. McKnight S.L. Cell-free formation of RNA granules: bound RNAs identify features and components of cellular assemblies Cell 149 2012 768 779 22579282 Howe F.S. Boubriak I. Sale M.J. Nair A. Clynes D. Grijzenhout A. Murray S.C. Woloszczuk R. Mellor J. Lysine acetylation controls local protein conformation by influencing proline isomerization Mol. Cell 55 2014 733 744 25127513 Iwasaki S. Kobayashi M. Yoda M. Sakaguchi Y. Katsuma S. Suzuki T. Tomari Y. Hsc70/Hsp90 chaperone machinery mediates ATP-dependent RISC loading of small RNA duplexes Mol. Cell 39 2010 292 299 20605501 Järvelin A.I. Noerenberg M. Davis I. Castello A. The new (dis)order in RNA regulation Cell Commun. Signal. 14 2016 9 27048167 Jia J. Arif A. Ray P.S. Fox P.L. WHEP domains direct noncanonical function of glutamyl-Prolyl tRNA synthetase in translational control of gene expression Mol. Cell 29 2008 679 690 18374644 Kato M. Han T.W. Xie S. Shi K. Du X. Wu L.C. Mirzaei H. Goldsmith E.J. Longgood J. Pei J. Cell-free formation of RNA granules: low complexity sequence domains form dynamic fibers within hydrogels Cell 149 2012 753 767 22579281 Kramer K. Sachsenberg T. Beckmann B.M. Qamar S. Boon K.L. Hentze M.W. Kohlbacher O. Urlaub H. Photo-cross-linking and high-resolution mass spectrometry for assignment of RNA-binding sites in RNA-binding proteins Nat. Methods 11 2014 1064 1070 25173706 Kuchta K. Knizewski L. Wyrwicz L.S. Rychlewski L. Ginalski K. Comprehensive classification of nucleotidyltransferase fold proteins: identification of novel families and their representatives in human Nucleic Acids Res. 37 2009 7701 7714 19833706 Kwon S.C. Yi H. Eichelbaum K. Föhr S. Fischer B. You K.T. Castello A. Krijgsveld J. Hentze M.W. Kim V.N. The RNA-binding protein repertoire of embryonic stem cells Nat. Struct. Mol. Biol. 20 2013 1122 1130 23912277 Lukong K.E. Chang K.W. Khandjian E.W. Richard S. RNA-binding proteins in human genetic disease Trends Genet. 24 2008 416 425 18597886 Lunde B.M. Moore C. Varani G. RNA-binding proteins: modular design for efficient function Nat. Rev. Mol. Cell Biol. 8 2007 479 490 17473849 Matia-González A.M. Laing E.E. Gerber A.P. Conserved mRNA-binding proteomes in eukaryotic organisms Nat. Struct. Mol. Biol. 22 2015 1027 1033 26595419 Mazza C. Segref A. Mattaj I.W. Cusack S. Large-scale induced fit recognition of an m(7)GpppG cap analogue by the human nuclear cap-binding complex EMBO J. 21 2002 5548 5557 12374755 Mesa A. Somarelli J.A. Herrera R.J. Spliceosomal immunophilins FEBS Lett. 582 2008 2345 2351 18544344 Mitchell S.F. Jain S. She M. Parker R. Global analysis of yeast mRNPs Nat. Struct. Mol. Biol. 20 2013 127 133 23222640 Muckenthaler M.U. Galy B. Hentze M.W. Systemic iron homeostasis and the iron-responsive element/iron-regulatory protein (IRE/IRP) regulatory network Annu. Rev. Nutr. 28 2008 197 213 18489257 Nagy E. Rigby W.F. Glyceraldehyde-3-phosphate dehydrogenase selectively binds AU-rich RNA in the NAD(+)-binding region (Rossmann fold) J. Biol. Chem. 270 1995 2755 2763 7531693 Ozgur S. Buchwald G. Falk S. Chakrabarti S. Prabu J.R. Conti E. The conformational plasticity of eukaryotic RNA-dependent ATPases FEBS J. 282 2015 850 863 25645110 Papasaikas P. Tejedor J.R. Vigevani L. Valcárcel J. Functional splicing network reveals extensive regulatory potential of the core spliceosomal machinery Mol. Cell 57 2015 7 22 25482510 Pashev I.G. Dimitrov S.I. Angelov D. Crosslinking proteins to nucleic acids by ultraviolet laser irradiation Trends Biochem. Sci. 16 1991 323 326 1835191 Phan A.T. Kuryavyi V. Darnell J.C. Serganov A. Majumdar A. Ilin S. Raslin T. Polonskaia A. Chen C. Clain D. Structure-function studies of FMRP RGG peptide recognition of an RNA duplex-quadruplex junction Nat. Struct. Mol. Biol. 18 2011 796 804 21642970 Popow J. Alleaume A.M. Curk T. Schwarzl T. Sauer S. Hentze M.W. FASTKD2 is an RNA-binding protein required for mitochondrial RNA processing and translation RNA 21 2015 1873 1884 26370583 Ramos A. Grünert S. Adams J. Micklem D.R. Proctor M.R. Freund S. Bycroft M. St Johnston D. Varani G. RNA recognition by a Staufen double-stranded RNA-binding domain EMBO J. 19 2000 997 1009 10698941 Safaee N. Kozlov G. Noronha A.M. Xie J. Wilds C.J. Gehring K. Interdomain allostery promotes assembly of the poly(A) mRNA complex with PABP and eIF4G Mol. Cell 48 2012 375 386 23041282 Scherrer T. Mittal N. Janga S.C. Gerber A.P. A screen for RNA-binding proteins in yeast indicates dual functions for many enzymes PLoS ONE 5 2010 e15499 21124907 Schmidt C. Kramer K. Urlaub H. Investigation of protein-RNA interactions by mass spectrometry--techniques and applications J. Proteomics 75 2012 3478 3494 22575267 Shang Y. Huang E.J. Mechanisms of FUS mutations in familial amyotrophic lateral sclerosis Brain Res. 2016 S0006-8993(16)30165-2 Strein C. Alleaume A.M. Rothbauer U. Hentze M.W. Castello A. A versatile assay for RNA-binding proteins in living cells RNA 20 2014 721 731 24664470 Suchanek M. Radzikowska A. Thiele C. Photo-leucine and photo-methionine allow identification of protein-protein interactions in living cells Nat. Methods 2 2005 261 267 15782218 Tejedor J.R. Papasaikas P. Valcárcel J. Genome-wide identification of Fas/CD95 alternative splicing regulators reveals links with iron homeostasis Mol. Cell 57 2015 23 38 25482508 Teplova M. Song J. Gaw H.Y. Teplov A. Patel D.J. Structural insights into RNA recognition by the alternate-splicing regulator CUG-binding protein 1 Structure 18 2010 1364 1377 20947024 Thandapani P. O’Connor T.R. Bailey T.L. Richard S. Defining the RGG/RG motif Mol. Cell 50 2013 613 623 23746349 Tsvetanova N.G. Klass D.M. Salzman J. Brown P.O. Proteome-wide search reveals unexpected RNA-binding proteins in Saccharomyces cerevisiae PLoS One 5 2010 e12671 20844764 Vuzman D. Azia A. Levy Y. Searching DNA via a “Monkey Bar” mechanism: the significance of disordered tails J. Mol. Biol. 396 2010 674 684 19958775 Weber S.C. Brangwynne C.P. Getting RNA and protein in phase Cell 149 2012 1188 1191 22682242 Willmund F. del Alamo M. Pechmann S. Chen T. Albanèse V. Dammer E.B. Peng J. Frydman J. The cotranslational function of ribosome-associated Hsp70 in eukaryotic protein homeostasis Cell 152 2013 196 209 23332755 Wolkowicz U.M. Cook A.G. NF45 dimerizes with NF90, Zfr and SPNR via a conserved domain that has a nucleotidyltransferase fold Nucleic Acids Res. 40 2012 9356 9368 22833610 Yamazaki T. Chen S. Yu Y. Yan B. Haertlein T.C. Carrasco M.A. Tapia J.C. Zhai B. Das R. Lalancette-Hebert M. FUS-SMN protein interactions link the motor neuron diseases ALS and SMA Cell Rep. 2 2012 799 806 23022481 Yu M. Levine S.J. Toll-like receptor, RIG-I-like receptors and the NLRP3 inflammasome: key modulators of innate immune responses to double-stranded RNA viruses Cytokine Growth Factor Rev. 22 2011 63 72 21466970 Zhang H. Elbaum-Garfinkle S. Langdon E.M. Taylor N. Occhipinti P. Bridges A.A. Brangwynne C.P. Gladfelter A.S. RNA controls polyQ protein phase transitions Mol. Cell 60 2015 220 230 26474065

PMC005xxxxxx/PMC5003816.txt

==== Front CellCellCell0092-86741097-4172Cell Press S0092-8674(16)30866-210.1016/j.cell.2016.07.001ArticleUBQLN2 Mediates Autophagy-Independent Protein Aggregate Clearance by the Proteasome Hjerpe Roland 127Bett John S. johnsbett@yahoo.co.uk127∗Keuss Matthew J. 2Solovyova Alexandra 3McWilliams Thomas G. 2Johnson Clare 2Sahu Indrajit 4Varghese Joby 2Wood Nicola 2Wightman Melanie 2Osborne Georgina 5Bates Gillian P. 5Glickman Michael H. 4Trost Matthias 2Knebel Axel 2Marchesi Francesco 6Kurz Thimo thimo.kurz@glasgow.ac.uk12∗∗1 Institute of Molecular, Cell and Systems Biology, College of Medical, Veterinary and Life Sciences, Davidson Building, Henry Wellcome Lab of Cell Biology, University of Glasgow, G12 8QQ Glasgow, UK2 The MRC Protein Phosphorylation and Ubiquitylation Unit, The Sir James Black Centre, College of Life Sciences, University of Dundee, Dow Street, Dundee DD1 5EH, Scotland3 Newcastle University Protein and Proteome Analysis, Devonshire Building, Devonshire Terrace, Newcastle upon Tyne NE1 7RU, UK4 Department of Biology, Technion-Israel Institute of Technology, 32000 Haifa, Israel5 Department of Medical and Molecular Genetics, King’s College London, 8th Floor Tower Wing, Guy’s Hospital, Great Maze Pond, London SE1 9RT, UK6 School of Veterinary Medicine, College of Medical, Veterinary and Life Sciences, University of Glasgow, 464 Bearsden Road, Glasgow G61 1QH, UK∗ Corresponding author johnsbett@yahoo.co.uk∗∗ Corresponding author thimo.kurz@glasgow.ac.uk7 Co-first author 11 8 2016 11 8 2016 166 4 935 949 27 11 2015 18 4 2016 2 7 2016 © 2016 The Authors2016This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).Summary Clearance of misfolded and aggregated proteins is central to cell survival. Here, we describe a new pathway for maintaining protein homeostasis mediated by the proteasome shuttle factor UBQLN2. The 26S proteasome degrades polyubiquitylated substrates by recognizing them through stoichiometrically bound ubiquitin receptors, but substrates are also delivered by reversibly bound shuttles. We aimed to determine why these parallel delivery mechanisms exist and found that UBQLN2 acts with the HSP70-HSP110 disaggregase machinery to clear protein aggregates via the 26S proteasome. UBQLN2 recognizes client-bound HSP70 and links it to the proteasome to allow for the degradation of aggregated and misfolded proteins. We further show that this process is active in the cell nucleus, where another system for aggregate clearance, autophagy, does not act. Finally, we found that mutations in UBQLN2, which lead to neurodegeneration in humans, are defective in chaperone binding, impair aggregate clearance, and cause cognitive deficits in mice. Graphical Abstract Highlights • UBQLN2 clears aggregates independent of autophagy via HSP70 and the proteasome • A disease mutation in UBQLN2 prevents its binding to HSP70 • Mutant UBQLN2 is defective in clearance of aggregates in vivo • UBQLN2 knockin mice develop cognitive impairment and brain pathology The proteasome teams up with partner proteins to clear tightly aggregated, potentially toxic species from the nucleus. Published: July 28, 2016 ==== Body Introduction The modification of proteins with ubiquitin regulates most cellular pathways. A major role for ubiquitylation is to target proteins for degradation via the 26S proteasome, forming the so-called ubiquitin-proteasome system (UPS) (Glickman and Ciechanover, 2002). Ubiquitin chains are built on substrates by E3 ubiquitin ligases, which link the first ubiquitin via its C terminus to the ε-amino group of an internal lysine residue of the substrate, followed by the conjugation of subsequent ubiquitin moieties to a lysine of the preceding ubiquitin (Thrower et al., 2000, Shabek et al., 2012, Lu et al., 2015). Specificity in the UPS is largely mediated by the ∼600 E3 ubiquitin ligases that recognize their cognate substrates, but there is also selectivity on the level of delivery to the 26S proteasome, as ubiquitylated proteins are either directly recognized by the proteasome through stoichiometric subunits (RPN10 and RPN13) or through loosely associated shuttle factors, which link polyubiquitylated proteins and the proteasome to facilitate degradation. Budding yeast has three shuttles: Dsk2, Rad23, and Ddi1 (Verma et al., 2004, Elsasser et al., 2004). These have an N-terminal ubiquitin-like (UBL) domain, which interacts with the proteasome (Elsasser et al., 2002, Saeki et al., 2002), and a C-terminal ubiquitin-associated (UBA) domain, which binds polyubiquitylated proteins. They also all contain domains between the UBL and UBA domains, whose functions are largely unexplored. An important observation is that UBL-UBA domain proteins act as inhibitors of proteasomal degradation when overexpressed (Kleijnen et al., 2000, Chen and Madura, 2002, Funakoshi et al., 2002, Raasi and Pickart, 2003). It is thus vital to study these proteins at endogenous levels, as even small increases in their abundance inhibit proteasomal degradation (Verma et al., 2004). Similarly, overexpression of Dsk2 in yeast cells causes cell-cycle arrest and cell death (Matiuhin et al., 2008), and overexpressing UBQLN in Drosophila leads to photoreceptor neurodegeneration (Ganguly et al., 2008). Most vertebrates contain four homologs of the yeast protein Dsk2, which are named ubiquilin-1–4 (UBQLN1–4). While UBQLN1, 2, and 4 are expressed widely, UBQLN3 is restricted to testis (Marín, 2014). Part of the central region of UBQLN2 contains domains with homology to a heat shock binding protein called STI1, which binds Stch (HSP13), a protein similar to HSP70 (Kaye et al., 2000). UBQLN1, 2, and 4 each contain four such STI1 domains and can all interact with Stch (Lim et al., 2006, Wang et al., 2011, Rual et al., 2005), although the physiological role for this is currently unclear. UBQLN2 is mutated in familial cases of the protein folding disorder amyotrophic lateral sclerosis (ALS) (Deng et al., 2011), and intriguingly, all familial mutations cluster to the PXXP motif, which is unique to UBQLN2 and of unknown function (Deng et al., 2011, Fahed et al., 2014, Williams et al., 2012, Vengoechea et al., 2013) (Figure 1A). The existence of shuttle factors is puzzling, and it is unclear why not all polyubiquitylated proteins are recognized by the intrinsic ubiquitin receptors of the proteasome. An attractive possibility is that shuttle factors add functionality to the proteasomal machinery to enable degradation of specialized substrates. We have explored this by studying the mammalian proteasome shuttle factor UBQLN2. Results UBQLN2 Is Required for Survival after Proteotoxic Stress To better understand the role of UBQLN2 and its relevance to neurodegenerative disease, we isolated its binding partners from mouse brain using immunoprecipitation and mass spectrometry. UBQLN2 most evidently bound to HSP70-type chaperones, UBQLN1 and UBQLN4 (Figure 1B), and to a lesser extent to proteasomal subunits (Figure 1B). Thus, UBQLN2 may be involved in the regulation of misfolded proteins. Indeed, UBQLN2 depletion by small interfering RNA (siRNA) caused hyper-sensitivity to heat shock, with a drop in cell viability comparable to the level observed after depletion of HSP70 (HSPA1A; Figure 1C). Previous work showed that UBQLN2 binds to a range of protein aggregates in patient brains (Mori et al., 2012). We established that endogenous UBQLN2 similarly co-purifies with ubiquitylated insoluble protein aggregates generated by heat shock (Figure 1D), along with HSP70 and the proteasome (Figure 1E). Under non-stressed conditions (Figure 1D) or after heat shock of pre-lysed cells (Figure S1A), endogenous UBQLN2 is soluble, suggesting that UBQLN2 is not itself heat-unstable but rather actively recruited to aggregates. Interestingly, UBQLN1 and UBQLN4 remained soluble after heat stress (Figure 1F), which was surprising given their homology to UBQLN2. Strikingly, we detected strongly increased binding of UBQLN2 to the proteasome and polyubiquitylated proteins after heat shock (Figure 1G), as well as enhanced binding to HSP70 (Figure 1H), suggesting the protein becomes activated under stress. UBQLN2 is not upregulated after heat shock (Figure S1B), indicating that it may instead be held in a repressed state under non-stressed conditions. Indeed, heat shock resulted in a loss of binding to other UBQLNs, consistent with a model where heterologous UBQLN complexes represent dormant reservoirs (Figure 1I). UBQLN2 Is a Proteasome Shuttle that Acts with the HSP70 System to Clear Aggregated Proteins Heat shock generates aggregates of polyubiquitylated proteins insoluble in up to 1% SDS (Figure S1C), which are cleared by the proteasome (Figure 2A; Figure S1D). We found that siRNA depletion of UBQLN2 resulted in a pronounced defect in the clearance of heat-induced insoluble ubiquitin conjugates (Figure 2A) but did not affect their accumulation (Figure S1E), supporting a role of UBQLN2 in protein aggregate clearance. Large aggregates are thought to be degraded by a proteolytic mechanism called autophagy. Thus, we examined autophagy-defective atg5 knockout cells and found that these were just as capable as wild-type cells in clearing heat-induced aggregates (Figure 2B). In contrast, proteasomal inhibition led to a complete abrogation of clearance for both wild-type and atg5 knockout cells (Figure 2B). Clearance also required UBQLN2, as atg5 knockout cells where UBQLN2 was downregulated also no longer efficiently cleared the aggregates (Figure 2C). These results demonstrate that UBQLN2 mediates degradation of insoluble heat-shock-induced aggregates through the proteasomal pathway, independently of autophagy. We next depleted HSP70 by siRNA and observed that HSP70 was also required to clear heat shock aggregates (Figure 2D). HSP70-mediated disaggregase activity requires the co-chaperone HSP110 (HSP105 in mice) (Nillegoda et al., 2015). To investigate whether UBQLN2 acts with the HSP70/HSP110 disaggregase pathway, we examined HSP110 (mHSP105) knockout mouse embryonic fibroblasts (MEFs) (Nakamura et al., 2008) and found that in these cells, interaction of both HSP70 and ubiquitin conjugates with UBQLN2 was increased even in the absence of heat stress (Figure 2E). This result suggested that in cells lacking HSP110, UBQLN2 becomes activated due to a higher aggregate load. In addition, heat shock induced a dramatic increase in the amount of UBQLN2, proteasome, and ubiquitin conjugates in the insoluble fraction of HSP110 knockout MEFs (Figure 2F), which also were impaired in their ability to clear heat shock aggregates (Figure 2G). These results demonstrate that UBQLN2 and the HSP70-HSP110 disaggregase act in the same pathway, and they explain how aggregates are processed by the chaperones prior to UBQLN2-mediated proteasomal degradation. We next tested if UBQLN2 also mediates the degradation of unfolded proteins independent of heat stress. The antibiotic puromycin leads to the accumulation of unfolded nascent polypeptide chains (Eggers et al., 1997), and we found that UBQLN2 depletion impaired the clearance of these faulty translation products (Figure 2H), while UBQLN2 levels remained unchanged (Figure S2C). Since many protein aggregates are found in the nucleus, where autophagy does not act, we next tested if UBQLN2 can enter the nucleus to clear protein aggregates. Using both biochemical fractionation (Figure 2I; Figure S2A) and immunofluorescence (Figure 2J), we found that UBQLN2 translocates into the nucleus upon heat stress, similar to HSP70 and other quality control components (Velazquez and Lindquist, 1984, Park et al., 2013). This did not happen using puromycin (Figure S2B), which generates unfolded proteins in the cytoplasm. To test if UBQLN2 clears nuclear substrates, we used cells stably expressing GFPu-NLS (Bennett et al., 2005), a model unfolded nuclear protein. Heat shock causes aggregation of GFPu-NLS (Figures S2D and S2E) and results in interaction of UBQLN2 with GFPu-NLS (Figure S2F), coinciding with nuclear translocation of UBQLN2. Moreover, the proteasomal degradation of GFPu-NLS after heat shock was dependent on UBQLN2 (Figure S2G), demonstrating that UBQLN2 can clear nuclear aggregates. We next examined the requirement of UBQLN2 for the clearance of a pathological Huntingtin fragment (HTTQ103), as UBQLN2 has been described to bind to aggregates in mouse models and patients with Huntington’s disease (HD) (Doi et al., 2004, Rutherford et al., 2013). We detected recruitment of endogenous UBQLN2 to HTT aggregates (Figure 2K), alongside HSP70 and the 26S proteasome (Figures S3A and S3B). We next found that the insoluble fraction from cells expressing GFP-HTTQ103, but not non-pathological GFP-HTTQ25, is retained on a filter trap alongside endogenous UBQLN2 (Figure 2L). HTTQ103 aggregates are retained in the stacking gel in SDS-PAGE, where we found that they also trap endogenous UBQLN2 (Figure S3C), and downregulation of UBQLN2 led to increased HTTQ103 aggregation (Figure 2M). Thus, UBQLN2 regulates degradation of model and disease-linked aggregation-prone proteins. Importantly, we demonstrate that the UBQLN2/HSP70/26S-proteasome pathway can clear aggregates in the nucleus. UBQLN2 Mutations Do Not Lead to UBQLN2 Aggregation We next examined the disease-linked mutations of UBQLN2 found in patients with familial ALS. Previous reports have suggested that both wild-type (WT) and mutant UBQLN2 aggregate, as exogenous expression leads to their localization to cytoplasmic foci similar in appearance to aggregates (Deng et al., 2011, Osaka et al., 2015). Indeed, exogenously expressing UBQLN2 in cells causes formation of cytosolic foci (Figure 3A), but no gross differences in size or number of foci were seen for mutant UBQLN2 (P506T, P497H) (Figure 3A). Importantly, mutating the UBA domain (L619A) to abolish ubiquitin binding (Figure 3B) leads to complete exclusion of both WT and mutant forms of UBQLN2 from the foci (Figure 3A), strongly suggesting the foci are not misfolded UBQLN2. The foci do not co-localize with as P bodies, stress granules (Figures S3D and S3E), or autophagosomes (Figure S3F). Furthermore, UBQLN2 foci formation does not render UBQLN2 insoluble, as UBQLN2 (WT) and five disease-linked mutants remained soluble when overexpressed in HEK293 cells (Figure S4E). Importantly, endogenous UBQLN2 is diffusely cytosolic (Figure 2J; Figure S4F). Next, we used purified UBQLN2 to investigate the biophysical properties of the WT and mutant proteins (Figure S4A). Small angle X-ray scattering (SAXS) experiments using WT and two mutant forms of UBQLN2 (P506T and P497H; Figure S4A) indicated that the mutations reduce the flexibility of the protein (Figure S4D). Based on circular dichroism measurements, there are no gross differences in secondary structure for any tested mutant (Figure 3C). Using analytical ultracentrifugation, we detected that both WT and mutant UBQLN2 forms dimers and trimers in a concentration-dependent manner but no higher-number oligomers or aggregates, which we also confirmed by size exclusion chromatography (Figure 3D; Figures S4B and S4C). Disease-Linked UBQLN2 Mutation Impedes Binding to HSP70 Chaperones and Sensitizes Cells to Protein Folding Stress As disease-linked mutant UBQLN2 did not aggregate, we next used stable isotope labeling with amino acids in cell culture (SILAC) proteomics to investigate changes in the interactome of cells stably expressing inducible WT or mutant UBQLN2. We found that disease-linked UBQLN2 (P506T) showed decreased binding to HSP70 chaperones and increased binding to ubiquitin (Figure 4A). We next generated a mouse knockin of the equivalent human P506T mutation (mP520T) and confirmed these changes at the endogenous level using primary MEFs from male mice (UBQLN2 is X linked) (Figures 4B–4D; Figures S6A and S6B). Strikingly, the binding of UBQLN2 to HSP70, ubiquitylated substrates and the proteasome after heat shock was strongly attenuated for mutant UBQLN2 (Figure 4E). Also, while the heat-shock-induced nuclear translocation of mutant UBQLN2 (mP520T) was unaffected (Figure S5A), it was strongly impaired in its recruitment to aggregates (Figure 4F), and cells expressing UBQLN2 (mP520T) were hypersensitive to both heat shock and puromycin stress compared to their wild-type littermate counterparts (Figures 4G and 4H). Together, these data suggest that the disease-linked forms of UBQLN2 are loss-of-function mutations. Since binding of UBQLN2 to HSP70 was unaffected by inhibiting stress inducible kinases or the ubiquitin E1 (Figures S5B and S5C) and recruitment of UBQLN2 to the insoluble fraction was also independent of ubiquitylation (Figure 5A), binding of UBQLN2 to HSP70 may in turn depend on client binding to HSP70. To test this, we used an in vitro system to examine the effect of protein aggregates on the UBQLN2-HSP70 interaction. Strikingly, the interaction between HSP70 and UBQLN2 was only induced when reactions also contained HSP70 client in the form of either mildly denatured (42°C for 30 min; Figure 5B; Figure S5D) or strongly denatured (95°C for 5 min; Figure S5E) recombinant luciferase. We next tested if the presence of HSP70 client would also result in the recruitment of purified human proteasomes (Figure 5C). Indeed, the interactions among HSP70, UBQLN2, and proteasomes in vitro were strongly induced by the addition of denatured luciferase, demonstrating that the presence of substrate leads to the formation of degradation complexes (Figure 5D). We next asked if a relevant pathological aggregate would have the same effect on HSP70/UBQLN2 complex formation. For this, we added small amounts of brain extracts from wild-type or R6/2 HD model mice (Mangiarini et al., 1996) to the in vitro interaction experiments and found that only the R6/2 extract triggered the interaction between HSP70 and UBQLN2 (Figure 5E; Figure S5F). This effect was seen with WT UBQLN2, but strikingly not with the disease-linked UBQLN2 (P506T; Figure 5E), entirely corroborating our cell-based experiments. Thus, the data strongly support a model whereby binding of clients to HSP70 triggers interaction with UBQLN2, which then bridges binding to the proteasome to mediate degradation. For disease-linked UBQLN2, mutations no longer support interaction with client-bound HSP70 and aggregate clearance is impaired. HSP70 can be roughly divided into two distinct domains, the N-terminal ATPase domain and the C-terminal substrate-binding domain, where also regulatory proteins such as the ubiquitin ligase CHIP bind (Zhang et al., 2015). We found that the C terminus of HSP70 is sufficient to bind UBQLN2, but unlike for the full-length protein, the interaction was constitutive and not regulated by heat shock (Figure 5F). We also tested if the PXXP motif is required for interaction; however, deletion of the PXXP motif had no effect on HSP70 binding, demonstrating that this region is not the direct binding site (Figure S5G). Instead, it is likely that the PXXP mutations interfere indirectly with HSP70 binding. UBQLN2 Mutation Leads to Cognitive Impairment and Inclusion Body Pathology in Mice After confirming decreased UBQLN2-HSP70 binding in knockin mouse brain (Figure 5G), we undertook a longitudinal behavioral study to determine the effect on mouse behavior. Using novel object recognition tests, where the time that a mouse spends exploring a novel versus familiar object is measured, we observed that mutant UBQLN2 (mP520T) animals were no longer able to distinguish between novel and familiar objects at 12 months of age (Figure 5H). Similarly, in novel place recognition tests (Figure 5I), mutant animals were incapable of distinguishing an object in a new location at both 9 and 12 months of age. Thus, UBQLN2 (mP520T) knockin mice develop cognitive deficits with age. As patients also have motor defects, we tested the UBQLN2 (mP520T) knockin animals using gait and rotarod analysis (Figures S6C–S6E) but observed no gross defects in either assay, although mutant mice presented with a slightly shorter stride length (Figure S6C). To assess if the cognitive deficits were accompanied by pathological changes, we performed immunohistochemical analyses on CNS tissues from 15- to 18-month-old mice. We observed regionalized UBQLN2 and p62 inclusion pathology in the hippocampus, cortex, and brainstem of mutant, but not WT, mice (Figure 5J). Interestingly, UBQLN2 is prominently present in the pellet fraction in hippocampal, but not cortical or cerebellar, tissue, despite similar expression levels (Figure 5K; Figures S7A and S7B). Importantly, our combined behavioral and histological findings demonstrate that UBQLN2 (mP520T) knockin mice recapitulate cognitive and pathological features of UBQLN2-associated neurodegeneration. UBQLN2 Mutation Impairs the Clearance of Protein Aggregates In Vivo To examine the role of UBQLN2 in handling aggregating clients in vivo, we turned to mutant Huntingtin (HTT) as a representative model. Using the R6/2 transgenic mouse (Mangiarini et al., 1996) and the HdhQ150 knockin mouse model (Lin et al., 2001), we found that immunoprecipitated UBQLN2 only associated with aggregated, but not SDS-soluble, HTT in vivo (Figures 6A and 6B). In both mouse models, binding of UBQLN2 to HTT was age- and disease-stage specific and only occurred once HTT had aggregated. HTT fragments passively diffuse into the nucleus in neurons, where they are retained upon aggregation (Cornett et al., 2005). Importantly, nuclear aggregation of HTT in both mouse models led to a translocation of UBQLN2, but not UBQLN1, to the nucleus (Figure 6C; Figure S7D). A proportion of HSP70 was present in nuclei at all ages (Figure 6C), and importantly, aggregate-associated HSP70 was trapped in the stacking gel in UBQLN2 immunoprecipitations from R6/2 brains (Figure S7E). Thus, mouse UBQLN2 behaves identically in HD mouse brains to UBQLN2 in cultured cells after heat shock. Moreover, UBQLN2 and HTT were co-captured by a ubiquitin binding resin (Hjerpe et al., 2009) (TUBE; Figure 6D), demonstrating that HTT-UBQLN2 complexes contain ubiquitin, suggesting they may be cleared by the proteasome. To directly test if UBQLN2 regulates HTT aggregation in vivo, we crossed R6/2 mice with UBQLN2 mP520T mutant knockin mice and observed a pronounced and significant increase of aggregated HTT, and a concomitant decrease of soluble HTT (Figure 6E). UBQLN2 co-localized with HTT inclusions (Figure S7D), and the number of nuclear HTT aggregates was significantly higher in the cortex of R6/2; mP520T double mutant animals compared to the R6/2 animals (Figure 6F). Moreover, a Seprion ligand assay shows significantly higher aggregate load in double-mutant brains, independently confirming our western blot and immunofluorescence analysis (Figure 6G). Thus, UBQLN2 mediates the clearance of protein aggregates in vivo, and the disease-linked forms of UBQLN2 are loss-of-function mutations, resulting in a failure to clear aggregating proteins. Discussion Proteasome Shuttle Factors as a Route for Protein Degradation Degradation through the UPS is the major cellular mechanism of selective protein turnover. We have shown that the shuttle factor UBQLN2 works with the HSP70 system for proteasomal degradation of insoluble ubiquitylated protein aggregates. UBQLN2 does this by coupling recognition of HSP70-bound clients with its proteasome shuttle properties. UBQLN2 binding to ubiquitylated proteins and the proteasome is negligible under resting conditions, suggesting it is constitutively held in an inactive state. Accumulation of clients results in an activation of UBQLN2, mediated by recognizing client-bound HSP70, where binding to ubiquitylated substrates is induced and degradation facilitated. UBQLN2 Integrates the Chaperone Network with the UPS to Clear Protein Aggregates UBQLN2 is needed both for aggregate clearance and survival after proteotoxic stress, suggesting that it is an integral component of the proteostasis network similar to HSP70 (Labbadia and Morimoto, 2015). Our finding that efficient binding of UBQLN2 to HSP70 requires the presence of HSP70 clients integrates the chaperone network with the UPS. Our conclusions are summarized in Figure 7. Briefly, under resting conditions, UBQLN2 is inactive and bound to other UBQLNs and itself. Activation of UBQLN2 occurs when HSP70 binds to client proteins, triggering exposure of a UBQLN2 binding site. A structural change in HSP70 mediated by client binding would provide efficient and fast means of activating degradation, while ensuring that complexes are only formed in the presence of unfolded client. Activation of UBQLN2 also allows binding of 26S proteasome to form a degradation-competent complex. Interestingly, initial complex formation among client-bound HSP70, UBQLN2, and proteasome does not require polyubiquitylation of the client. However, ubiquitin is an integral part of proteasomal degradation, and heat-shock-induced aggregated proteins are ubiquitylated. Ubiquitylation of an HSP70 client could thus take place with UBQLN2 already present in the complex and may enhance UBQLN2 affinity, committing the client to proteasomal degradation. This model explains why we observe the inducible binding of UBQLN2 to ubiquitylated proteins after heat shock. Moreover, it is very likely that translocation into the proteolytic chamber and degradation of the substrate by the proteasome requires polyubiquitylation of the client, even though initial complex formation does not. Whether a client is refolded by HSP70 or degraded by UBQLN2/UPS may ultimately be a question of its residence time on HSP70. The HSP70-UBQLN2-Proteasome Pathway Provides an Autophagy-Independent Means for Clearing Protein Aggregates Since proteasomes can only accommodate single unfolded polypeptide chains and not large aggregates, it has been assumed that the proteasome cannot degrade these. We demonstrate that the proteasome can clear aggregates through a UBQLN2-HSP70 pathway but suggest aggregates are first solubilized by HSP70-HSP110 disaggregase activity. Lending support to this idea, we show that the HSP70 cofactor HSP110, which is part of the HSP70-mediated disaggregase (Nillegoda et al., 2015), is also required for the efficient clearance of heat shock aggregates. UBQLN2 likely binds to HSP70 associated to both insoluble and soluble misfolded proteins as part of an ongoing disaggregation and clearance activity, which explains our observation that UBQLN2 co-purifies with insoluble ubiquitylated aggregates. This model is consistent with previous reports that demonstrate that aggregates exist in equilibrium between soluble and insoluble states (Yamamoto et al., 2000), and we propose that the soluble fraction is degraded by the proteasome, while autophagy may manage larger insoluble structures. Critically, we show that UBQLN2 can clear aggregates in the nucleus, where autophagy is absent (Lu et al., 2014). UBQLN2 Loss-of-Function Mutations Lead to Disease Due to Loss of HSP70 Binding It has been unclear whether UBQLN2 mutations cause disease through loss of function or toxic gain of function. We found that a disease-linked mutation led to a pronounced sensitivity to proteotoxic stress, effectively phenocopying the effect of UBQLN2 depletion, strongly suggesting a loss-of-function mutation. Our data demonstrate that this defect is due to impaired interaction with HSP70, ultimately leading to defective aggregate clearance (Figure 7). Interestingly, translocation of UBQLN2 into the nucleus was not affected by the disease mutation, suggesting that this aspect of the stress response is independent of HSP70 binding. This makes sense, as our model predicts that activation of UBQLN2 would rely upon association to client-bound HSP70, and it is unlikely that such a complex would be formed in the cytoplasm and then driven into the nucleus. However, the mechanism by which UBQLN2 is tranlsocated into the nucleus as inactive species is currently unclear. We also found that the mutant form of UBQLN2 binds slightly more polyubiquitin than the WT under unstressed conditions. The reason is not apparent, but it may be due to UBQLN2 occasionally dissociating from its inactive state under resting conditions, leading to binding to polyubiquitylated proteins and a possible delay of mutant UBQLN2 in returning to its inhibited state. This difference is dramatically swamped under stress conditions, where ubiquitin binding by mutant UBQLN2 is significantly decreased versus the WT protein. Together, our data provide a mechanistic understanding of UBQLN2, which in the future may allow for the design of small molecules to mediate the therapeutic activation of UBQLN2 in patients with diseases of protein aggregation. Experimental Procedures Animal Work UBQLN2 P520T constitutive knock-in mice were created and supplied by Taconic/Artemis. R6/2 mice were maintained as previously described (Bett et al., 2006). Mice were bred at the University of Dundee and Kings College London in accordance with European Union and Home Office regulations. Work was approved by the Ethical Review Committee (ERC) from the University of Dundee and was performed with a UK Home Office project license. R6/2 males were bred with heterozygous UBQLN2 P520T females at Charles River Laboratories (UK). Cell Culture and Cell Lines Cells stably expressing inducible FLAG-UBQLN2 WT, P506T, P497H, L619A, P506T/L619A, P497H/L619A, HTTQ25-GFP, and HTTQ103-GFP were created using T-Rex HEK293 (Life Technologies, R710-07). Stably expressing cells were maintained in DMEM (Life Technologies, 11995-065), 10% fetal bovine serum (FBS), 50 U/ml penicillin, 50 μg/ml streptomycin (Life Technologies, 15070-063), 2 mM L-glutamine, 100 μg/ml hygromycin (Invivogen, ant-hg-1bl), and 15 μg/ml blasticidin (Invivogen, ant-bl-1). Expression was induced with 2–5 ng/ml doxycycline. U2OS cells, HEK293 cells, and MEFs were maintained as above but without hygromycin and blasticidin. Solubility Experiments Cells were heat shocked at the indicated temperature for 2 hr followed by recovery at 37°C. Soluble and pellet fractions were generated by lysing cells in stringent lysis buffer (20 mM Tris-HCL, 2 mM EDTA, 150 mM NaCl, 1.2% deoxycholate, 1.2% Triton-X, 200 mM iodoacetamide and cOmplete protease inhibitor cocktail [Roche]), sonicating (30% power 3 × 10 s pulses), and centrifugation at 17,000 × g for 15 min. The supernatant was collected and represented the soluble fraction. The remaining pellet (insoluble fraction) was washed five times in PBS and re-suspended in Laemmli’s sample buffer. To generate the cytosolic-soluble, nuclear-soluble, and total-insoluble fractions, cells were first lysed in low-stringency buffer (10 mM HEPES [pH 7.9], 1.5 mM MgCl2, 10 mM KCL, 0.08% NP-40, and cOmplete protease inhibitor cocktail [Roche]) followed by centrifugation at 17,000 × g for 15 min. The supernatant (soluble fraction) was collected. The remaining pellet was washed five times in PBS prior to re-suspending in stringent lysis buffer, and soluble and insoluble fractions were generated as above. In this case, the supernatant represented the nuclear-soluble fraction and the pellet represented the total-insoluble fraction. Cell Viability Assays Cell viability assays were done by lysing cells in 50 mM Tris/phosphate (pH 7.8), 1.6 mM MgCl2, 2 mM DTT, 2% Triton X-100, 30% glycerol, 1% BSA, 0.250 mM D-luciferin, 8 μM sodium pyrophosphate, and 500 ng QuantiLum recombinant Luciferase (Promega). Viability was determined using Envision 2104 plate reader (Perkin Elmer). Cells were heat shocked for 2 hr followed by 24 hr recovery prior to viability assay being carried out. Antibodies Sheep antibodies to UBQLN1, UBQLN2, and UBQLN4 were produced in house, raised against the following epitopes (residues numbered): mouse UBQLN1 482-515, mouse UBQLN2 11-27, human UBQLN2 478-518, mouse UBQLN4 84-161 (Figures S7F–S7J). Additional antibodies were FLAG-M2-peroxidase (Sigma-Aldrich, A8592), HSP70 (Abcam, ab181606), GAPDH (Cell Signaling Technology), Actin (Millipore, MAB1501R), anti-ubiquitin (Dako, Z 0458), GFP (Roche), Histone H4 (Abcam), histone H3B (Abcam), HTT (Bett et al., 2006), tubulin (Sigma), RPT6 (Enzo Life Sciences, BML PW9265), puromycin 12D10, (Millipore, MABE343). For immunofluorescence, anti-UBQLN2 from Novus Biologicals (NBP2-25164SS), anti-RPT3 (Bethyl Laboratories, A303-850A), and anti-GFP (Abcam, ab13970) were used. Secondary antibodies were from Bio-Rad (anti-mouse 170-5047; anti-rabbit 170-5046) and Abcam (anti-sheep ab97130). Protein-G horseradish peroxidase (HRP) was used for secondary detection in immunoprecipitations (Abcam, ab7460). Author Contributions R.H. and J.S.B. performed all of the experiments described with the exception of the biophysical experiments (analytical ultra centrifugation [AUC], circular dichroism [CD], and SAXS), which were performed by A.S., and the Seprion Ligand assay, which was performed by G.O. and G.B. A.K. and C.J. set up conditions for purification of UBQLN2. R.H. purified UBQLN2 for AUC, CD, and SAXS experiments. M.J.K. performed immunofluorescence staining of endogenous UBQLN2 after heat shock and of HTTQ103 cells. T.G.M. performed cardiac perfusions and mouse brain sub-dissections. F.M. performed IHC on mouse brains and analyzed pathology. I.S. purified human proteasome and aided in the characterization of in vitro complexes between UBQLN2 and the 26S proteasome with advice from M.H.G. A.K. provided all other purified proteins. M.T. and J.V. performed the mass spectrometry analyses. N.W. and M.W. generated cDNA clones. R.H. initiated the work on the effect of UBQLN2 mutations in vitro and in vivo and established the UBQLN2 interactome and its role in newly synthesized protein stress. J.S.B. initiated the work on the concept of UBQLN2 as a stress-activated proteasome shuttle in the clearance of heat-induced protein aggregates and in the in vivo clearance of HTT. R.H., J.S.B., and T.K. designed, interpreted, and analyzed the experiments and wrote the paper with contributions from all the other authors. T.K. conceived the project and supervised the work. Supplemental Information Document S1. Supplemental Experimental Procedures Document S2. Article plus Supplemental Information Acknowledgments We acknowledge technical support of the MRC Protein Phosphorylation and Ubiquitylation Unit, Elaine Forsyth for assistance with mouse work, the DNA Sequencing Service (coordinated by Nicholas Helps), Thomas Macartney and other members of the cloning team (coordinated by Rachel Toth and Mark Peggie), and the Protein Purification Team and Antibody Production Team (coordinated by Hilary McLauchlan and James Hastie). We thank Prof. Mike Cheetham (UCL) for providing HTT constructs and Prof. Ron Kopito (Stanford University) for providing GFPu-NLS cells. Special thanks to Dr. Katsuaki Inoue (beamline B21, Diamond Light Source, Didcot, UK) for help with SAXS data collection and primary data treatment and the Diamond Light Source for beamtime granted (proposal SM–5025–1). We thank Prof. Jeremy H. Lakey and Dr. Helen Waller for help in carrying out CD experiments, Mrs. Lynn Stevenson and Ms. Lynn Oxford (Veterinary Diagnostic Services, School of Veterinary Medicine, University of Glasgow) for their technical assistance with immunohistochemistry, and Dr. Ian Ganley and Dr. Michael Munson for advice on autophagy-related experiments and for providing ATG5 KO cells and LC3 antibodies. We thank Professor Kazuhiro Nagata (Kyoto Sangyo University) for providing HSP105 KO cells. We thank John MacLeod for his assistance in anaesthetizing mice and Amnon Golan for aiding in purification and characterization of 26S proteasome from human cells. This work was supported by the Medical Research Council (MRC_MC_UU_12016/7), an ERC Starting Investigator Grant to T.K. (ERC_243019), a University of Glasgow Leadership Fellowship and Tenovus Scotland grant (to J.S.B.), the CDHI foundation (to G.B.), an Israel Science Foundation (ISF 909.14) grant (to M.H.G.), as well as the pharmaceutical companies supporting the Division of Signal Transduction Therapy Unit (AstraZeneca, Boehringer-Ingelheim, GlaxoSmithKline, Merck, Janssen Pharmaceutica, and Pfizer). Supplemental Information includes Supplemental Experimental Procedures and seven figures and can be found with this article online at http://dx.doi.org/10.1016/j.cell.2016.07.001. Figure 1 UBQLN2 Is Required for Cell Survival after Heat Shock (A) Schematic of the known domains of UBQLN2, their binding partners, and reported familial disease mutations shown in italics. (B) Binding partners of UBQLN2 that were identified by immunoprecipitation (IP) of UBQLN2 from mouse brain lysate followed by mass spectrometry. (C) Depletion of UBQLN2 by two independent siRNAs (72 hr) leads to cell death on heat stress. (D–F) UBQLN2, HSP70, and proteasome, but not UBQLN1 or UBQLN4, co-purify with insoluble ubiquitin-rich aggregates upon heat stress. (G–I) UBQLN2 inducibly interacts with proteasomes, ubiquitylated proteins, and HSP70 after heat shock and loses binding to UBQLN1 and UBQLN4. See also Figures S1 and S7. Figure 2 Heat Stress Activates UBQLN2 to Clear Aggregated Proteins (A) UBQLN2 depletion by siRNA leads to defective clearance of heat-shock-induced insoluble ubiquitin conjugates (left), and quantification of insoluble ubiquitin in the pellet (right) (n = 2). Error bars represent SEM. (B) Insoluble heat-shock-generated ubiquitin conjugates are cleared efficiently in ATG5 knockout (autophagy-deficient) MEFs in a proteasome-dependent manner. (C) UBQLN2 depletion in autophagy-deficient cells leads to attenuated clearance of heat-shock-induced insoluble ubiquitin conjugates. Quantification (n = 3) is shown (right). Error bars represent SD; statistical tests were two-tailed t tests. (D) HSP70 siRNA leads to a defective clearance of ubiquitylated aggregated proteins. Over time, the transcriptional heat shock response leads to increased levels of HSP70. (E) Increased interaction of UBQLN2 with HSP70 and ubiquitin was observed in HSP105 knockout (KO) MEF cells. (F) UBQLN2 and ubiquitin are more abundant in the pellet fraction after heat shock in HSP105 KO MEF cells. (G) HSP105 KO MEFs are deficient in clearing heat-shock-induced aggregates. In addition, increased binding of HSP70 and ubiquitin to UBQLN2 was detected. (H) Depletion of UBQLN2 by siRNA leads to defective clearance of puromycin-labeled truncated proteins. (I and J) UBQLN2 translocates to the nucleus after heat stress (see Figure S2A for fractionation protocol). Quantification of the normalized nuclear fluorescence intensity is shown (J, bottom) (n = 99 and 122 for 37°C and 43°C, respectively). Error bars represent SD. (K) UBQLN2 co-localizes with cellular HTT aggregates in HEK293 cells inducibly expressing pathological GFP-Huntingtin (HTTQ103). (L) UBQLN2 co-aggregates with pathological, but not non-pathological, GFP-Huntingtin, as shown by filter trap assay. (M) UBQLN2 depletion leads to increased HTT-Q103 aggregates, running in the stacking gel. Quadruplicate transfections are shown. See also Figures S2, S3, S4, and S7. Figure 3 UBQLN2 Mutations Do Not Cause Protein Aggregation (A) Inducible HEK293 cells stably overexpressing the indicated FLAG-UBQLN2 exhibit cytosolic foci for both the wild-type (WT) and P506T mutant. The L619A ubiquitin non-binding point mutation abrogates foci formation for WT and P506T mutant. (B) UBQLN2 point mutants (F594V and L619A) are defective in polyubiquitin binding. (C) Circular dichroism performed on pure wild-type and mutant protein. PONDR prediction (inset) results in a small decrease of disorder for PXXP mutant proteins (WT, P506T, and P497H shown). Experimentally, no difference is seen in the amount of disorder and secondary structure for the mutants. (D) Purified UBQLN2 was analyzed by analytical ultracentrifugation at different concentrations, showing dimer and trimer peaks for both WT and mutant protein. See also Figures S3 and S4. Figure 4 Disease Mutant UBQLN2 Loses Binding to HSP70 and Sensitizes to Protein Misfolding Stress (A) SILAC proteomics was performed on FLAG IP from cells stably expressing inducible FLAG-UBQLN2 WT or P506T. Interaction with proteasomal subunits (PSMA6 shown) is unaffected by the mutation, UBQLN2 P506T binding to HSP70 family members (HSPA1A, HSPA8) is significantly lower (p < 0.0001), and binding to ubiquitin is significantly higher (p = 0.011). Asterisks indicate a statistically significant difference from a SILAC ratio of 1 (two-tailed single-value t test). (B–D) Decreased binding to HSP70 and increased binding to ubiquitin was confirmed by UBQLN2 IP from wild-type and mP520T (equivalent to human P506T) primary male mouse embryonic fibroblasts (MEFs), derived from littermate embryos. HSP70 (B) and ubiquitin (C) were detected by western blot. (C) Quantification of mutant/wild-type signal ratio for co-immunoprecipitated HSP70 and ubiquitin. Asterisk indicates a statistically significant difference from a mean ratio of 1 (two-tailed single-value t test). (E) Stress-induced binding to HSP70, ubiquitin and proteasomes is defective for mutant UBQLN2. Asterisk indicates Rpt6 (proteasome). (F) Mutant UBQLN2 is defective in association to heat shock induced aggregates. Asterisk indicates a non-specific band. (G) mP520T MEFs are hypersensitive to heat shock as compared to WT counterparts. (H) mP520T MEFs are hypersensitive to 20-hr puromycin treatment at the indicated concentrations. Error bars represent SD. Statistical test was a two-tailed t test. See also Figures S5, S6, and S7. Figure 5 HSP70 Client Interaction Drives UBQLN2-HSP70 Binding (A) UBQLN2 association to heat-shock-induced pelleted proteins is independent of ubiquitin. Cells were treated with the ubiquitin E1 inhibitor MLN7243, heat shocked, and fractionated into supernatant and pellet. (B) Presence of HSP70-client induces UBQLN2-HSP70 interaction in vitro. Reaction components were mixed and incubated at the indicated temperature, followed by pull-down of GST-HSP70. (C) Purified human 26S proteasome. Lane 1, Coomassie staining of 2 μg purified human proteasome; lanes 2–4, in-gel LLVY-AMC (N-succinyl-leucine-leucine-valine-tyrosine-7-amino-4-methylcoumarin) chymotrypsin activity of 2 μg human proteasome, Coomassie staining, and immunoblot with anti-Rpt5 antibody in 4% native-PAGE, respectively. (D) Heat-denatured (95°C) or native recombinant luciferase was added to the other reaction components, followed by GST-HSP70 pull-down. (E) Pathological Huntingtin aggregates induce binding of GST-HSP70 to purified wild-type, but not mutant (P506T), UBQLN2 in vitro. Brain extract from wild-type or R6/2 mice was spiked into the reaction mix, followed by GST-HSP70 pull-down and analysis of UBQLN2 binding. (F) UBQLN2 binds to the C-terminal domain of HSP70. IP of HSPA8-SV5 mutants expressed in HEK293 cells and detection of endogenous UBQLN2. Cells were heat shocked as indicated. Schematic shows the HSP70 domains. (G) Mutant UBQLN2 shows reduced binding to HSP70 in knockin mouse brain. (H and I) The UBQLN2 mP520T knockin mutation leads to cognitive impairment in aged mice. Male mice (n = 11 of each genotype) were aged and tested in novel-object and novel-place recognition tests. Error bars represent SD. Statistical tests were two-tailed t tests. (J) Aged UBQLN2 mP520T knockin animals have UBQLN2- and p62-positive inclusion body pathology. Brains from aged (15- to 18-month-old) mice were subjected to immunohistochemistry (IHC) for UBQLN2 and p62 (n = 6 per genotype). Red shading in schematic shows areas of inclusion pathology. (K) Mutant UBQLN2 is specifically present in the pellet from hippocampal lysates in aged (15- to 18-month-old) knockin mice. Isolated neocortex (CTX), hippocampus (HC), and cerebellum (CB) were separated into NP40-soluble and insoluble fractions. Asterisk indicates an unspecific band. See also Figures S5, S6, and S7. Figure 6 UBQLN2 Mutation Impairs Aggregate Clearance In Vivo (A and B) UBQLN2 interacts with aggregated, but not SDS-soluble, HTT in vivo, as judged by reciprocal IP of HTT and UBQLN2 from the R6/2 transgenic (A) and HdhQ150 knockin (B) Huntington’s disease models. (C) UBQLN2, but not UBQLN1, translocates to the nucleus in the R6/2 and HdhQ150 models. (D) UBQLN2 is present in ubiquitylated Huntingtin aggregates from brains of the R6/2 and HdhQ150 mouse models. Aggregated HTT and UBQLN2 were captured with a ubiquitin binding resin (GST-TUBE). (E) The R6/2 and UBQLN2 mP520T mice were crossed to produce double-mutant animals, and 9-week-old male brains from these were assayed for aggregated HTT by western blot. Quantification of soluble HTT is shown (bottom) (n = 4 per genotype). (F) Immunofluorescence (IF) of nuclear HTT aggregates in R6/2 and R6/2;mP520T brains shows more inclusion bodies in the double mutant. Quantification is shown (right). Error bars represent SEM. Statistical test was a two-tailed t test. (G) The Seprion ligand assay independently confirms a significant increase in aggregated HTT in double mutants, compared to R6/2 littermates (n = 8 per genotype). Error bars represent SEM. See also Figure S7. Figure 7 Model of How UBQLN2 Manages Proteotoxic Stress Under non-stressed conditions, UBQLN2 is held inactive in homo- or hetero-dimers (1). In the presence of HSP70 clients, UBQLN2 binds to HSP70 and associated misfolded/aggregated proteins, which are ubiquitylated (2). HSP70/HSP110-dependent disaggregase activity pulls aggregated proteins apart, allowing for UBQLN2 to act as a proteasome shuttle connecting ubiquitylated misfolded proteins to the proteasome, after forming a HSP70-client-UBQLN2-proteasome degradation complex (3) ending in client proteolysis (4). Disease mutant UBQLN2 (star) is defective in its association to HSP70 and no longer effectively forms a degradation complex, leading to accumulation of misfolded/aggregated proteins (5). Figure S1 Heat Shock Generates Insoluble Ubiquitin-Positive Aggregates and Does Not Inactive Proteasomes, Related to Figure 1 (A) UBQLN2 is not pelleted when cells are heat shocked post lysis. Cell lysates were incubated at 37 or 42°C and then fractionated into soluble (S) and pellet (P) fraction. This indicates that UBQLN2 itself does not aggregate as a result of high temperature. (B) UBQLN2 levels are not upregulated in response to heat shock. HSP70 and GAPDH were used as a positive and negative controls, respectively. (C) Heat shock aggregates are insoluble in up to and including 1% SDS but are solubilized in 2% SDS. Blotting of soluble and pellet fractions with anti-ubiquitin and UQBLN2 antibodies confirmed dissolution of the aggregates in 2% SDS. (D) Proteasomes are active after heat shock. To confirm that proteasome activity was not affected by heat shock, we incubated U2OS and MEFs at the indicted temperatures for 2h. Cells were then harvested and cell lysates were incubated with the proteasome inhibitor MG132 or DMSO, followed by incubation with a fluorescent proteasome-activity probe, as indicated. The presence of fluorescently labeled beta-subunits at the same intensity under both heat stress and normal temperature, indicate that proteasome activity is not significantly affected by heat shock. (E) U2OS cells were treated with control or UBQLN2 siRNA and subjected to heat shock for the indicated times. Analysis of the pellet fraction revealed that insoluble ubiquitylated aggregates are generated within 5 min of heat shock, but that depletion of UBQLN2 does not noticeably alter the accumulation of these aggregates at any of the indicated time points. Figure S2 Puromycin Does Not Upregulate UBQLN2 Levels and UBQLN2 Clears Nuclear Aggregated GFP-u, Related to Figure 2 (A) Schematic representation on how the three fractions, total insoluble, nuclear soluble and total insoluble were generated. (B) Cells treated with puromycin were fractionated as indicated and treatment did not induce the nuclear localization of UBQLN2. (C) Puromycin treatment did not induce the upregulation of UBQLN2 protein. (D) Schematic showing the GFPu-NLS construct (Bennett et al., 2005). (E) HEK293 cells stably expressing GFPu-NLS were subject to heat shock for 2h at 42°C and fractionated into soluble and insoluble fractions. GFPu-NLS recruited to the insoluble fraction after heat shock indicating its heat-induced aggregation. (F) GFPu-NLS cells were subject to heat shock for 2h at 42°C and proteasome inhibition with 25 μM MG132 as indicated. UBQLN2 was immunoprecipitated and GFPu-NLS was found to co-immunoprecipitate only upon heat shock, consistent with UBQLN2 nuclear localization. Combined heat shock and proteasome inhibition increased the binding further. (G) GFPu-NLS cells were depleted of UBQLN2 or treated with a control non-targeting siRNA then treated with 50 μg/ml cycloheximide (CHX) for the indicated time to measure turnover. Turnover was quantified using data from three independent experiments. Error bars represent SE. Figure S3 Proteasomes and HSP70 Co-localize with HTTQ103 Aggregates and Overexpressed UBQLN2 Form Cytosolic Foci, Related to Figures 2 and 3 (A and B) HEK293 cells expressing inducible HttQ103-GFP were stained with antibodies to (A) HSP70 and (B) proteasome subunit RPT3. Inclusion bodies were positive for booth HSP70 and the proteasome, the latter forming a ring around the perimeter of the inclusion. (C) Endogenous UBQLN2 co-aggregates with pathological Huntingtin (HTT-Q103) but not with non-pathological HTT-Q25. GFP-HTT Q25 or Q103 expression was induced in HEK293 cells, followed by cell harvesting and fractionation into soluble (S) and pellet (P) fractions. HTT-Q103 runs as high molecular weight aggregates present in the stacking gels for the pellet fraction, and endogenous UBQLN2 is observed to also be upshifted to the stacking gel. (D) The cytosolic foci visible on UBQLN2 overexpression do not co-localize with markers for p-bodies or stress-granules. (E) FLAG-UBQLN2 was transiently transfected into U2OS cells, with either GFP-DCP1A (p-body marker) or GFP-G3BP1 (stress granule marker). UBQLN2 was detected by indirect immunofluorescence to the FLAG-tag, using mouse monoclonal anti-FLAG antibody (SIGMA ALDRICH F3165), and an anti-mouse secondary Alexa Fluor 647 (Jackons 715-605-151). (F) Cytosolic UBQLN2 foci do not co-localize with the autophagosome marker LC3. U2OS cells stably expressing inducible EGFP-UBQLN2 were induced with 2 ng/ml doxycycline for 24 hr, then treated with 50 nM Bafilomycin A1 for 1 hr prior to fixation and staining. LC3 staining was performed using a mouse monoclonal anti-LC3 (MBL M152-3). Secondary antibody was anti-mouse Alexa Fluor 647 (Jackson 715-605-151). Vehicle control was DMSO. Figure S4 UBQLN2 Does Not Form Aggregates, and the Antibody Used for UBQLN2 Immunofluorescence Is Specific for UBQLN2, Related to Figures 2 and 3 (A) Coomassie stain of bacterially expressed and purified, untagged UBQLN2 wild-type and mutant proteins. (B) Analytical ultracentrifugation was performed to investigate differences in oligomerization or aggregation for purified UBQLN2. Additional mutants shown here to support results in main Figure 3. No significant amount of aggregated protein was detected for any mutant, and no differences in dimerization or trimerisation were observed. (C) Analytical gel filtration of UBQLN2 WT, P506T and P497H show a single sharp peak migrating at an apparent molecular weight above 158 kDa, without any indication of additional UBQLN2 species or aggregated material. (D) UBQLN2 P506T and P497H mutants are more compact particles than WT – flexibility analysis based on small-angle X-ray scattering experiments. The radius of gyration based dimensionless Kratky plot (top panel) has a characteristic shape for partially disordered protein containing both ordered and disordered fragment(s) – the peak of the curve (dotted gray line) is shifted from a position characteristic for globular folded protein (solid gray line). At the same time the plot demonstrates that WT contains more disorder (the right wing on the WT curve is slightly lifted comparing with P506T and 497H). The protein concentrations were 4.67, 4.03, 4.91 mg/ml for WT, P506T and 497H respectively. The volume-of-correlation based dimensionless Kratky plot (bottom panel) gives a more in-depth analysis and reveals the increase in volume-to-surface ratio for the P506T and 497H mutants, indicating more compact particles compared to WT UBQLN2 (the maximal possible volume-to-surface ratio of 0.82 is for a sphere; see arrow). (E) Mutations in UBQLN2 do not cause the protein to become insoluble in cells. FLAG-tagged wild-type and mutant UBQLN2 were overexpressed, and cells were fractionated into 1% NP-40 soluble (S) and insoluble pellet (P) fractions, and detected with FLAG-HRP conjugated antibody (SIGMA ALDRICH A8592). No difference in distribution as compared to the wild-type was seen for any of the mutants. FUS was used as a marker for the pellet. (F and G) Validation of UBQLN2 antibody for staining of endogenous UBQLN2 in U2OS cells. U2OS cells were transfected with control siRNA or siRNA targeting UBQLN2. 72h post-transfection, cells were trypsinised, and seeded on glass slides for microscopy. Cells were either seeded as separate groups (i.e., control and UBQLN2 siRNA) or mixed 1:1 and seeded together (third panel from the left). As a separate control, cells were transfected with plasmid encoding for FLAG-tagged UBQLN2 (right-most panel only). These cells show large UBQLN2 foci not present at endogenous levels. Cells were stained using the mouse monoclonal anti-UBQLN2 6H9 (Novus NBP2-25164), at 1:250 in 2% BSA PBS for 1h. Secondary antibody was goat Anti-Mouse DyLight 488 (Abcam ab96871). Knockdown of UBQLN2 can be clearly seen to decrease the signal, indicating that the antibody is specific to UBQLN2. (G) displays zoom of the indicated areas in. For the mixed cells (third panel from the left) a white arrowhead indicates a cell transfected with control siRNA and a black arrowhead a cell transfected with UBQLN2 siRNA. Figure S5 Nuclear Translocation of UBQLN2 Is Unaffected by Disease Mutation, and HSP70 Clients Induce HSP70-UBQLN2 Interaction, Related to Figures 4 and 5 (A) Wild-type of P520T knock-in MEFs were heat shocked and fractionated as indicated and no difference was observed in the nuclear localization as a result of the disease mutation. (B) HEK293 cells were treated with the broad spectrum kinase inhibitor Staurosporine (1 μM) or the p38 (BIRB-0796 and VX-745; 1 μM) and JNK (JNKIN8; 10 μM) kinase inhibitors for 1h prior to heat shock and showed that kinase signaling is not regulating the inducible interaction of HSP70 and UBQLN2. (C) HEK293 cells were treated with the ubiquitin E1 inhibitor MLN7243 (10 μM) for 1h prior to heat shock and demonstrated that ubiquitylation or ubiquitin signaling is not involved in regulating the inducible interaction between HSP70 and UBQLN2. (D) UBQLN2 does not bind non-specifically to GST in the presence or absence of denatured luciferase. GST or GST-HSP70 and purified UBQLN2 was incubated at 42°C in the presence or absence of Luciferase, as indicated. This was followed by GST pulldown, and Western blot for associated UBQLN2. (E) Luciferase was denatured at 95°C for 5 min and found to stimulate the binding of untagged recombinant UBQLN2 to GST-HSP70 upon GST-pulldown, unlike native luciferase. (F) R6/2 brain extracts but not WT brain extracts were found to be able to stimulate the interaction of recombinant untagged UBQLN2 with GST-HSP70 in GST pulldown experiments. (G) HEK293 cells stably expressing inducible UBQLN2 WT or PXXP deletion mutants were found to both equally interact with endogenous HSP70 after heat shock, indicating that the PXXP motif does not directly mediate the interaction. Figure S6 Generation of a Constitutive Knock-in Mouse Model and Locomotor Tests of a Male Cohort, Related to Figures 4, 5, and 6 (A) Targeting strategy used to generate the UBQLN2 P520T knock-in mice. (B) Western blot of brain extracts showing that UBQLN2 levels are expressed at the same level in WT and UBQLN2 knock-in male mice (expressing one copy each of UBQLN2 due to being X-linked). (C) Gait analysis in the mP520T mouse model. Gait analysis was performed at 6, 9 and 12 months of age, showing a marginal, but significant decrease in stride length at 6 months of age for the mutant animals. At 9 and 12 months the trend persists. Habituation to handling/runway corridor was followed by assessment of gait by painting of front and hind paws. Gait parameters including stride length and width between paws was analyzed manually from the paw print records. (D) Accelerating rotarod tests showed no impairment in motor function for UBQLN2 mP520T animals at any age. The animals performed 8 trials (4 trials on day 1 and a further 4 trials on the following day). On each trial the mouse was placed on the RotaRod and the rod accelerates from a speed of 5 rpm up to 45 rpm, with a maximum trial time of 5 min. (E) Fixed speed rotarod tests showed no impairment in motor function for UBQLN2 mP520T animals at any age. Figure S7 UBQLN2 Is Aggregated in Hippocampus and Associates to HSP70 and HTT Aggregates; Characterization of UBQLN Antibodies, Related to Figures 1, 2, 4, 5, and 6 (A and B) Hippocampal or cortical extracts were examined from WT or P520T knock-in mice. UBQLN2 expression levels were quantified and found to be indistinguishable between brain regions in either genotype. (C) UBQLN2 was found in the pellet fraction of the hippocampus in P520T knock-in, but not WT mice, in three additional independent pairs of animals. (D) UBQLN2 co-localizes with HTT inclusions in R6/2 brains. Sections of 14 week R6/2 brains were stained for HTT (MW8 antibody) and UBQLN2. (E) UBQLN2 was immunoprecipitated from 14-week-old R6/2 brains and blotted for the indicated proteins. HTT and HSP70 were detected in the stacking gel, indicated UBQLN2 interacts with SDS-insoluble HTT aggregates that are positive for HSP70. (F–I) Validation of specificity for UBQLN antibodies produced in-house. All antibodies were raised in sheep. UBQLN1, UBQLN2 or UBQLN4 were knocked down using siRNA and the indicated antibody used for detection. No cross-reactivity between ubiquilins was seen. (J) Purified untagged mouse UBQLN1, 2, 3 and 4 was further used to assess specificity of the raised antibodies, which confirm that there is no cross reactions. ==== Refs References Bennett E.J. Bence N.F. Jayakumar R. Kopito R.R. Global impairment of the ubiquitin-proteasome system by nuclear or cytoplasmic protein aggregates precedes inclusion body formation Mol. Cell 17 2005 351 365 15694337 Bett J.S. Goellner G.M. Woodman B. Pratt G. Rechsteiner M. Bates G.P. Proteasome impairment does not contribute to pathogenesis in R6/2 Huntington’s disease mice: exclusion of proteasome activator REGgamma as a therapeutic target Hum. Mol. Genet. 15 2006 33 44 16311253 Chen L. Madura K. Rad23 promotes the targeting of proteolytic substrates to the proteasome Mol. Cell. Biol. 22 2002 4902 4913 12052895 Cornett J. Cao F. Wang C.-E. Ross C.A. Bates G.P. Li S.-H. Li X.-J. Polyglutamine expansion of huntingtin impairs its nuclear export Nat. Genet. 37 2005 198 204 15654337 Deng H.-X. Chen W. Hong S.-T. Boycott K.M. Gorrie G.H. Siddique N. Yang Y. Fecto F. Shi Y. Zhai H. Mutations in UBQLN2 cause dominant X-linked juvenile and adult-onset ALS and ALS/dementia Nature 477 2011 211 215 21857683 Doi H. Mitsui K. Kurosawa M. Machida Y. Kuroiwa Y. Nukina N. Identification of ubiquitin-interacting proteins in purified polyglutamine aggregates FEBS Lett. 571 2004 171 176 15280037 Eggers D.K. Welch W.J. Hansen W.J. Complexes between nascent polypeptides and their molecular chaperones in the cytosol of mammalian cells Mol. Biol. Cell 8 1997 1559 1573 9285825 Elsasser S. Gali R.R. Schwickart M. Larsen C.N. Leggett D.S. Müller B. Feng M.T. Tübing F. Dittmar G.A.G. Finley D. Proteasome subunit Rpn1 binds ubiquitin-like protein domains Nat. Cell Biol. 4 2002 725 730 12198498 Elsasser S. Chandler-Militello D. Müller B. Hanna J. Finley D. Rad23 and Rpn10 serve as alternative ubiquitin receptors for the proteasome J. Biol. Chem. 279 2004 26817 26822 15117949 Fahed A.C. McDonough B. Gouvion C.M. Newell K.L. Dure L.S. Bebin M. Bick A.G. Seidman J.G. Harter D.H. Seidman C.E. UBQLN2 mutation causing heterogeneous X-linked dominant neurodegeneration Ann. Neurol. 75 2014 793 798 24771548 Funakoshi M. Sasaki T. Nishimoto T. Kobayashi H. Budding yeast Dsk2p is a polyubiquitin-binding protein that can interact with the proteasome Proc. Natl. Acad. Sci. USA 99 2002 745 750 11805328 Ganguly A. Feldman R.M.R. Guo M. ubiquilin antagonizes presenilin and promotes neurodegeneration in Drosophila Hum. Mol. Genet. 17 2008 293 302 17947293 Glickman M.H. Ciechanover A. The ubiquitin-proteasome proteolytic pathway: destruction for the sake of construction Physiol. Rev. 82 2002 373 428 11917093 Hjerpe R. Aillet F. Lopitz-Otsoa F. Lang V. England P. Rodriguez M.S. Efficient protection and isolation of ubiquitylated proteins using tandem ubiquitin-binding entities EMBO Rep. 10 2009 1250 1258 19798103 Kaye F.J. Modi S. Ivanovska I. Koonin E.V. Thress K. Kubo A. Kornbluth S. Rose M.D. A family of ubiquitin-like proteins binds the ATPase domain of Hsp70-like Stch FEBS Lett. 467 2000 348 355 10675567 Kleijnen M.F. Shih A.H. Zhou P. Kumar S. Soccio R.E. Kedersha N.L. Gill G. Howley P.M. The hPLIC proteins may provide a link between the ubiquitination machinery and the proteasome Mol. Cell 6 2000 409 419 10983987 Labbadia J. Morimoto R.I. Repression of the heat shock response is a programmed event at the onset of reproduction Mol. Cell 59 2015 639 650 26212459 Lim J. Hao T. Shaw C. Patel A.J. Szabó G. Rual J.-F. Fisk C.J. Li N. Smolyar A. Hill D.E. A protein-protein interaction network for human inherited ataxias and disorders of Purkinje cell degeneration Cell 125 2006 801 814 16713569 Lin C.H. Tallaksen-Greene S. Chien W.M. Cearley J.A. Jackson W.S. Crouse A.B. Ren S. Li X.J. Albin R.L. Detloff P.J. Neurological abnormalities in a knock-in mouse model of Huntington’s disease Hum. Mol. Genet. 10 2001 137 144 11152661 Lu K. Psakhye I. Jentsch S. Autophagic clearance of polyQ proteins mediated by ubiquitin-Atg8 adaptors of the conserved CUET protein family Cell 158 2014 549 563 25042851 Lu Y. Lee B.-H. King R.W. Finley D. Kirschner M.W. Substrate degradation by the proteasome: a single-molecule kinetic analysis Science 348 2015 1250834 25859050 Mangiarini L. Sathasivam K. Seller M. Cozens B. Harper A. Hetherington C. Lawton M. Trottier Y. Lehrach H. Davies S.W. Bates G.P. Exon 1 of the HD gene with an expanded CAG repeat is sufficient to cause a progressive neurological phenotype in transgenic mice Cell 87 1996 493 506 8898202 Marín I. The ubiquilin gene family: evolutionary patterns and functional insights BMC Evol. Biol. 14 2014 63 24674348 Matiuhin Y. Kirkpatrick D.S. Ziv I. Kim W. Dakshinamurthy A. Kleifeld O. Gygi S.P. Reis N. Glickman M.H. Extraproteasomal Rpn10 restricts access of the polyubiquitin-binding protein Dsk2 to proteasome Mol. Cell 32 2008 415 425 18995839 Mori F. Tanji K. Odagiri S. Toyoshima Y. Yoshida M. Ikeda T. Sasaki H. Kakita A. Takahashi H. Wakabayashi K. Ubiquilin immunoreactivity in cytoplasmic and nuclear inclusions in synucleinopathies, polyglutamine diseases and intranuclear inclusion body disease Acta Neuropathol. 124 2012 149 151 22661321 Nakamura J. Fujimoto M. Yasuda K. Takeda K. Akira S. Hatayama T. Takagi Y. Nozaki K. Hosokawa N. Nagata K. Targeted disruption of Hsp110/105 gene protects against ischemic stress Stroke 39 2008 2853 2859 18658041 Nillegoda N.B. Kirstein J. Szlachcic A. Berynskyy M. Stank A. Stengel F. Arnsburg K. Gao X. Scior A. Aebersold R. Crucial HSP70 co-chaperone complex unlocks metazoan protein disaggregation Nature 524 2015 247 251 26245380 Osaka M. Ito D. Yagi T. Nihei Y. Suzuki N. Evidence of a link between ubiquilin 2 and optineurin in amyotrophic lateral sclerosis Hum. Mol. Genet. 24 2015 1617 1629 25398946 Park S.-H. Kukushkin Y. Gupta R. Chen T. Konagai A. Hipp M.S. Hayer-Hartl M. Hartl F.U. PolyQ proteins interfere with nuclear degradation of cytosolic proteins by sequestering the Sis1p chaperone Cell 154 2013 134 145 23791384 Raasi S. Pickart C.M. Rad23 ubiquitin-associated domains (UBA) inhibit 26 S proteasome-catalyzed proteolysis by sequestering lysine 48-linked polyubiquitin chains J. Biol. Chem. 278 2003 8951 8959 12643283 Rual J.-F. Venkatesan K. Hao T. Hirozane-Kishikawa T. Dricot A. Li N. Berriz G.F. Gibbons F.D. Dreze M. Ayivi-Guedehoussou N. Towards a proteome-scale map of the human protein-protein interaction network Nature 437 2005 1173 1178 16189514 Rutherford N.J. Lewis J. Clippinger A.K. Thomas M.A. Adamson J. Cruz P.E. Cannon A. Xu G. Golde T.E. Shaw G. Unbiased screen reveals ubiquilin-1 and -2 highly associated with huntingtin inclusions Brain Res. 1524 2013 62 73 23774650 Saeki Y. Sone T. Toh-e A. Yokosawa H. Identification of ubiquitin-like protein-binding subunits of the 26S proteasome Biochem. Biophys. Res. Commun. 296 2002 813 819 12200120 Shabek N. Herman-Bachinsky Y. Buchsbaum S. Lewinson O. Haj-Yahya M. Hejjaoui M. Lashuel H.A. Sommer T. Brik A. Ciechanover A. The size of the proteasomal substrate determines whether its degradation will be mediated by mono- or polyubiquitylation Mol. Cell 48 2012 87 97 22902562 Thrower J.S. Hoffman L. Rechsteiner M. Pickart C.M. Recognition of the polyubiquitin proteolytic signal EMBO J. 19 2000 94 102 10619848 Velazquez J.M. Lindquist S. hsp70: nuclear concentration during environmental stress and cytoplasmic storage during recovery Cell 36 1984 655 662 6421488 Vengoechea J. David M.P. Yaghi S.R. Carpenter L. Rudnicki S.A. Clinical variability and female penetrance in X-linked familial FTD/ALS caused by a P506S mutation in UBQLN2 Amyotroph. Lateral Scler. Frontotemporal Degener. 14 2013 615 619 23944734 Verma R. Oania R. Graumann J. Deshaies R.J. Multiubiquitin chain receptors define a layer of substrate selectivity in the ubiquitin-proteasome system Cell 118 2004 99 110 15242647 Wang J. Huo K. Ma L. Tang L. Li D. Huang X. Yuan Y. Li C. Wang W. Guan W. Toward an understanding of the protein interaction network of the human liver Mol. Syst. Biol. 7 2011 536 21988832 Williams K.L. Warraich S.T. Yang S. Solski J.A. Fernando R. Rouleau G.A. Nicholson G.A. Blair I.P. UBQLN2/ubiquilin 2 mutation and pathology in familial amyotrophic lateral sclerosis Neurobiol. Aging 33 2012 e3–310 Yamamoto A. Lucas J.J. Hen R. Reversal of neuropathology and motor dysfunction in a conditional model of Huntington’s disease Cell 101 2000 57 66 10778856 Zhang H. Amick J. Chakravarti R. Santarriaga S. Schlanger S. McGlone C. Dare M. Nix J.C. Scaglione K.M. Stuehr D.J. A bipartite interaction between Hsp70 and CHIP regulates ubiquitination of chaperoned client proteins Structure 23 2015 472 482 25684577

PMC005xxxxxx/PMC5003817.txt

==== Front Front MicrobiolFront MicrobiolFront. Microbiol.Frontiers in Microbiology1664-302XFrontiers Media S.A. 10.3389/fmicb.2016.01336MicrobiologyMini ReviewThe Role of Bacterial Secretion Systems in the Virulence of Gram-Negative Airway Pathogens Associated with Cystic Fibrosis Depluverez Sofie Devos Simon Devreese Bart *Laboratory for Protein Biochemistry and Biomolecular Engineering, Department of Biochemistry and Microbiology, Ghent UniversityGhent, BelgiumEdited by: David Wareham, Queen Mary University of London, UK Reviewed by: Burton F. Dickey, University of Texas MD Anderson Cancer Center, USA; Konstantin V. Korotkov, University of Kentucky, USA *Correspondence: Bart Devreese, bart.devreese@ugent.beThis article was submitted to Infectious Diseases, a section of the journal Frontiers in Microbiology 30 8 2016 2016 7 133612 5 2016 12 8 2016 Copyright © 2016 Depluverez, Devos and Devreese.2016Depluverez, Devos and DevreeseThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.Cystic fibrosis (CF) is the most common lethal inherited disorder in Caucasians. It is caused by mutation of the CF transmembrane conductance regulator (CFTR) gene. A defect in the CFTR ion channel causes a dramatic change in the composition of the airway surface fluid, leading to a highly viscous mucus layer. In healthy individuals, the majority of bacteria trapped in the mucus layer are removed and destroyed by mucociliary clearance. However, in the lungs of patients with CF, the mucociliary clearance is impaired due to dehydration of the airway surface fluid. As a consequence, patients with CF are highly susceptible to chronic or intermittent pulmonary infections, often causing extensive lung inflammation and damage, accompanied by a decreased life expectancy. This mini review will focus on the different secretion mechanisms used by the major bacterial CF pathogens to release virulence factors, their role in resistance and discusses the potential for therapeutically targeting secretion systems. infectioncystic fibrosispathogenesisantimicrobial resistanceprotein biosynthesisGram-negative bacteriaFederaal Wetenschapsbeleid10.13039/501100002749IAP7/44Universiteit Gent10.13039/501100004385GOA actionUniversiteit Gent10.13039/501100004385BOF research fund, phd fellowship ==== Body Bacterial Infections Involved in Cystic Fibrosis (CF) Lung Disease The combination of a highly viscous, dehydrated mucus layer, defective mucociliary clearance and a number of yet unknown factors make patients with CF extremely susceptible to infections (Lipuma, 2010). Pseudomonas aeruginosa is the most prevalent Gram-negative species, infecting about 50% of all patients. It is detected in 25% of children, but approximately 70% of patients older than 25 years tested positive (Cystic Fibrosis Foundation, 2015). Members of the Burkholderia cepacia complex (Bcc) cause chronic infections in CF patients, which results in approximately 20% of the cases in fatal ‘cepacia syndrome,’ characterized by necrotizing pneumonia, bacteremia, sepsis and eventually death (Lipuma, 2010). The prevalence of Bcc is highest in adults, affecting about 4% of the patients, with B. cenocepacia and B. multivorans accounting for 70% of the Bcc infections. Several reports indicate that the incidence of Stenotrophomonas maltophilia in CF patients has increased considerably in recent years (Denton and Kerr, 2002). This opportunistic nosocomial pathogen is mostly recovered from adolescent patients, with a prevalence of ± 15% (Razvi et al., 2009; Cystic Fibrosis Foundation, 2015). Prevalence of Haemophilus influenzae is maximal at an age of 2–5 years (32%) and decreases thereafter (Cystic Fibrosis Foundation, 2015). Achromobacter xylosoxidans is also an emerging CF pathogen with an overall prevalence around 6% (Razvi et al., 2009). Common to all these species is their dramatic intrinsic or acquired resistance against most of the currently employed antibiotics, making these infections extremely difficult to eradicate. Efflux pumps, biofilm formation, decreased outer membrane permeability, and inactivation of β-lactam antibiotics by chromosomally encoded β-lactamases are the main causes of resistance (Hoyle and Costerton, 1991; Waters, 2012). Virulence Factors Each of the abovementioned species has its own repertoire of virulence factors, specifically adapted to its needs for invasion, colonization, replication, and survival in the host (Table 1). Survival of P. aeruginosa is supported by the secretion of toxins and proteases, including pyocyanin, exotoxin A, elastase, alkaline phosphatase, and phospholipase C (Lee et al., 2005; van’t Wout et al., 2015). Similar strategies are used by B. cenocepacia to invade and colonize host cells. Two zinc metalloproteases (ZmpA and ZmpB), phospholipase C, iron-chelating siderophores, and cable pili participate in this process (Sajjan et al., 1995; Darling et al., 1998; Chung et al., 2003; Corbett et al., 2003; Uehlinger et al., 2009). Besides the production of a range of extracellular enzymes (lipase, fibrinolysin, hyaluronidase, protease, elastase, etc.), little is known about virulence factors contributing to the pathogenesis of S. maltophilia (Bottone et al., 1986). The extracellular capsule, adhesion proteins (HMW1 and HMW2, opacity-associated protein A), pili, haemocin, and the IgA1 protease play a crucial role in the onset of the patient’s inflammatory response by H. influenzae (Rosadini, 2011; Kostyanev and Sechanova, 2012). Table 1 Overview of the major virulence factors associated with the outer membrane or secreted by cystic fibrosis (CF) pathogens. Pseudomonas aeruginosa Burkholderia cenocepacia Stenotrophomonas maltophilia Haemophilus influenzae Proteases LasB2, AprA1, AprX1, Staphylolysin LasA2, aminopeptidase PaAP2, protease IV2, LepA5, elastase2 ZmpA2, ZmpB2, MprA2 StmPr12, StmPr22, elastase2 IgA1 protease5 Lipases LipA2, LipC2, phospholipase C2, PlcH2, PlcN2, ExoU3 Phospholipase C Lipase / Toxins Pyocyanin, exotoxin A2, Cif Haemolysin Zonula occludens toxin / Adhesion molecules Chitin-binding protein CbpD2, pili, ExoS3, ExoT3, alginate, fimbriae, flagellin Cable pili, flagellin, fimbriae Flagellin, fimbriae HMW15, HMW25, pili, Hap5, Hia5, Hsf5, opacity-associated protein A Hydrolytic enzymes Alkaline phosphatase, EstA5 Chitinase Fibrinolysin, hyaluronidase, DNase, chitinase Haemocin Virulence factors of Achromobacter xylosoxidans have not been characterized yet. Indices indicate a known association with a certain secretion system (number corresponds to the type of secretion system).The Role of Bacterial Secretion Systems in CF Pathogenesis and Virulence Bacterial virulence factors are delivered either in the extracellular environment or directly into host cells. Most Gram-negative CF pathogens possess one or more specialized secretion systems to accomplish this task. Eight different secretion systems have been identified (Figure 1). Type I [type I secretion system (T1SS)], type III [type III secretion system (T3SS)], type IV [type IV secretion system (T4SS)], and type VI [type VI secretion system (T6SS)] secretion pathways use a single energy-coupled step to transport proteins across both the inner and outer membranes. The outer membrane-spanning type V secretion system (T5SS) and the double membrane-spanning type II secretion system (T2SS) translocate substrates that first have been transported into the periplasm by the Sec or Tat machinery (Costa et al., 2015). Type VII secretion system (Type VII) is restricted to Gram-positive bacteria and will not be discussed here. The type VIII secretion system (type VIII) refers to the curli biogenesis pathway (Chapman et al., 2002). FIGURE 1 Schematic overview of the different secretion systems of Gram-negative airway pathogens associated with cystic fibrosis (CF). T1SS The type I secretion machinery is composed of an inner membrane associated ATP-binding cassette protein (which recognizes the secretion signal of the substrate), a membrane fusion adapter protein and a TolC-like outer membrane protein (Wandersman, 1996). Substrate proteins are often very acidic and contain distinctive glycine-rich repeats that bind Ca2+ ions (Baumann et al., 1993). Most of the transported proteins also contain repeats with a high degree of homology to adhesion molecules, suggesting a role for T1SS substrates in adherence (Hinsa et al., 2003). The heme-binding protein HasAp from P. aeruginosa, important for iron acquisition, is an example of a protein secreted by T1SS (Letoffe et al., 1998). A second T1SS in P. aeruginosa is responsible for the secretion of the alkaline proteases AprA and AprX (Guzzo et al., 1991; Duong et al., 2001). In B. pseudomallei, the major haemolysin is exported through a T1SS (Harland et al., 2007). Three T1SS clusters are present in the genome of S. maltophilia (Rocco, 2011), a potential substrate being the virulence-associated membrane protein Ax21 (Ferrer-Navarro et al., 2013). T2SS The T2SS is important for the secretion of hydrolases. It consists of an outer membrane complex, a periplasmic pseudopilus, an inner membrane platform and a cytoplasmic ATPase. Substrates are transported into the periplasm as unfolded or folded proteins by the SecYEG translocon or the Tat transporter, respectively (Costa et al., 2015). Interaction of the T2SS with its substrates presumably occurs through recognition of a structural motif, rather than a linear secretion signal (Lu and Lory, 1996; Sauvonnet and Pugsley, 1996; Francetic and Pugsley, 2005). In P. aeruginosa, the major extracellular protease LasB is secreted by the T2SS and is responsible for elastin degradation and cleavage of surfactant protein D, an important immune system protein (Olson and Ohman, 1992; Alcorn and Wright, 2004). Staphylolysin LasA, aminopeptidase PaAP, and protease IV are other examples of type II secreted proteinolytic enzymes in P. aeruginosa (Olson and Ohman, 1992; Engel et al., 1998; Cahan et al., 2001). Another important family of T2SS substrates in this pathogen are lipases, like LipA, LipC, phospholipase C, PlcH, and PlcN, which are targeting the host membrane (Diaz-Laviada et al., 1990; Ostroff et al., 1990). CbpD, a T2SS-dependent chitin-binding protein, could serve as an adhesin, mediating colonization of eukaryotic cells (Folders et al., 2000). The type II secreted exotoxin A is responsible for ADP-ribosylation of elongation factor 2, resulting in protein synthesis inhibition and cell death (Allured et al., 1986). Also the B. cenocepacia zinc-dependent metalloproteases, ZmpA and ZmpB, are T2SS substrates (Nakazawa, 1996). They cleave antimicrobial peptides involved in innate immunity, like β-defensin-1, cathelicidin LL-37, elafin, and secretory leukocyte inhibitor (Kooi and Sokol, 2009). S. maltophilia possesses two T2SS, Gsp and Xps (Karaba et al., 2013). The serine proteases StmPr1 and StmPr2 are substrates of the Xps T2SS and mediate degradation of extracellular matrix proteins (DuMont et al., 2015). H. influenzae does not contain the genes required to build a functional T2SS (Cianciotto, 2005). T3SS Bacterial T3SS are nanomachines capable of injecting effector proteins into the cytoplasm or cell membrane of eukaryotic target cells, and are therefore also called injectisomes (Cornelis, 2006). The system consists of a double-membrane-spanning base composed of stacked rings and a needle-shaped filament that extends into the extracellular space (Marlovits et al., 2004). Different translocator proteins are first transported through the needle and inserted into the eukaryotic cell membrane to form a pore of about 2.8–3.0 nm (Dacheux et al., 2001; Schoehn et al., 2003). Effectors contain a non-cleavable N-terminal secretion signal and are targeted to the secretion machinery in an unfolded state (Cornelis, 2006). Known T3SS effectors of P. aeruginosa include ExoS and ExoT, both containing a GTPase-activating function and an ADP-ribosyltransferase activity. By acting on the actin cytoskeleton, they are able to protect P. aeruginosa from phagocytosis (Barbieri and Sun, 2004). Accumulation of cyclic AMP in host cells is caused by the action of ExoY, an adenylate cyclase (Yahr et al., 1998). ExoU is responsible for acute cytotoxicity and lung tissue damage by its phospholipase A2 activity. Together with ExoS, it prevents interleukin production by alveolar macrophages and modulates the early inflammatory response (Sato and Frank, 2004). A T3SS mutant of B. cenocepacia was attenuated in virulence in a murine model of infection, which indicates a role for the T3SS in evasion of the host immune system (Tomich et al., 2003). Currently, no effectors have been identified for this species. T3SS genes are not present in S. maltophilia (Crossman et al., 2008) or H. influenzae (Harrison et al., 2005). T4SS Like the T3SS, the T4SS is composed of a core complex spanning the inner and outer membrane and a pilus that protrudes into the extracellular environment (Christie et al., 2014). The secretion signals needed for translocation of effector proteins are generally localized at the C-terminus and consist of clusters of hydrophobic or positively charged residues (Alvarez-Martinez and Christie, 2009). Two T4SSs with different functions are present in B. cenocepacia. The first is located on a 92 kb plasmid and is responsible for secretion of plant cytotoxic proteins. It also plays a role in the intracellular survival of B. cenocepacia in phagocytes. The second T4SS is chromosomally encoded and might be involved in plasmid mobilization, although the exact function is still unknown (Zhang et al., 2009). T4SS effectors of Xanthomonas citri, a close relative of S. maltophilia, have the capability of killing other bacterial species, thereby conferring a selective growth advantage in mixed bacterial communities (Souza et al., 2015). Whether the T4SS of S. maltophilia has a similar function, remains unknown. H. influenzae and P. aeruginosa do not contain a conventional T4SS. A unique feature of the T4SS is that it can also transport nucleic acids. P. aeruginosa and H. influenzae possess one or more genomic island-associated T4SSs (GI-T4SS) that play a crucial role in horizontal gene transfer (HGT) of integrative and conjugative elements (ICEs; Juhas et al., 2007a). ICEs not only contain genes required for excision/integration and various accessory genes, but they often also harbor a T4SS, which completes the machinery for efficient transfer from donor to recipient cell (Juhas et al., 2008; Wozniak and Waldor, 2010; Guglielmini et al., 2011). A considerable part of the accessory genes are involved in antibiotic resistance or virulence. ICEHin1056 of H. influenzae carries ampicillin, tetracycline and chloramphenicol resistance genes (Juhas et al., 2007b), while PAPI-1 of P. aeruginosa encodes CupD type fimbriae essential for attachment and the PvrSR/RcsCB regulatory system involved in biofilm formation and antibiotic resistance (Mikkelsen et al., 2013). The chromosomally encoded T4SS of B. cenocepacia was also linked to plasmid mobilization (Zhang et al., 2009). Taken together, these mechanisms of HGT pose a major threat to our ability to combat infections occurring in CF patients by potentially transforming the lung microbiota into an antibiotic resistant community. T5SS The T5SS is a single-membrane-spanning system that secretes virulence factors and mediates cell-to-cell adhesion and biofilm formation. The substrates are fused to their secretion pore to form a single polypeptide, also known as autotransporter. Unfolded autotransporters are delivered to the periplasm via the SecYEG translocon. The exoproteins either remain associated with the outer membrane or are released in the extracellular environment after proteolytic cleavage (Leo et al., 2012). In a second type of T5SS, two-partner secretion (TPS), the substrate or passenger domain and the pore-forming domain are two separate proteins. There is only one known autotransporter in P. aeruginosa, i.e., EstA. It can hydrolyze glycerol esters through its esterase activity and is involved in the production of rhamnolipids, cell motility and biofilm formation (Wilhelm et al., 2007). Three TPS systems have been characterized in P. aeruginosa: the LepA/LepB system, in which LepA is a protease activating NF-κB through digestion of PAR receptors (Kida et al., 2008), the CupB system, involved in the assembly of CupB fimbriae (Ruer et al., 2008) and the PdtA/PdtB system, where PdtA is related to High Molecular Weight (MWH) adhesins (Faure et al., 2014). The genome of B. cenocepacia J2315 contains four T5SS, two of them contain pertactin domains involved in adhesion, and the other two contain haemagglutinin repeats (Holden et al., 2009). Haemagglutinin autotransporters are also present in S. maltophilia (Ryan et al., 2009). The HMW1 and HMW2 from H. influenzae are also TPS systems. The H. influenzae Hap, Hia, and Hsf autotransporters mediate bacterial aggregation and microcolony formation and promote adherence to epithelial cells and extracellular matrix proteins (Fink et al., 2003; Spahich and St Geme, 2011). Another T5SS substrate is the IgA protease, responsible for degradation of the major mucosal immunoglobulin (Fernaays, 2008). T6SS The type VI secretion machinery consists of a membrane complex and a tail complex, composed of structural elements that are equivalent to contractile phage tails (Basler et al., 2012). Although the T6SS plays a major role in the pathogenesis toward eukaryotic cells, it can also be used to target other bacteria in polymicrobial infections (Ho et al., 2014). Three T6SS are present in P. aeruginosa, but only two major substrates have been identified so far, Hcp and VgrGs. Hcp is believed to form nanotubes on the bacterial surface, which may allow transport of other T6SS effectors (Ballister et al., 2008). VgrGs could form trimeric complexes puncturing membranes allowing the passage of other proteins (Leiman et al., 2009). The B. cenocepacia T6SS modulates actin cytoskeleton dynamics and NADPH oxidase complex assembly, also through the action of Hcp and VgrGs (Pukatzki et al., 2007). S. maltophilia and H. influenzae do not contain T6SS genes. Membrane Vesicles Secretion of membrane vesicles (MVs) by both Gram-negative and Gram-positive bacteria is now considered as a true secretion system. The membranous nanoparticles are pinched off from the cell surface and carry membrane-associated and soluble proteins, nucleotides, and other molecules into the extracellular environment. MVs are involved in a series of biological functions, including nutrient acquisition, iron scavenging, antibiotic resistance and biofilm formation (Haurat et al., 2015). Membrane vesicles contribute to pathogenesis by delivering virulence factors and/or through modulation of the host immune system (Schwechheimer and Kuehn, 2015). P. aeruginosa MVs enable long-distance delivery of multiple virulence factors including alkaline phosphatase, hemolytic phospholipase C and Cif, a toxin that inhibits CFTR-mediated chloride secretion in the airways (Bomberger et al., 2009). Cif also enhances ubiquitination and subsequent degradation of the transporter associated with antigen processing (TAP), reducing MHC class I activation (Bomberger et al., 2014). Secretion of MV-associated hydrolases like (metallo)proteases, (phospho)lipases and peptidoglycan-degrading enzymes was also shown in B. cenocepacia (Allan et al., 2003). H. influenzae MVs activate B-cells in a T-cell independent manner, possibly creating a diversion on the adaptive immune system and promoting survival within the host (Deknuydt et al., 2014). Several studies highlighted the importance of MVs in antibiotic resistance. Exposure of S. maltophilia cells to β-lactam antibiotics led to a significant increase in MVs that are packed with β-lactamases (Devos et al., 2015). These MVs are capable of degrading β-lactams extracellularly, and even increase the β-lactam tolerance of the species P. aeruginosa and B. cenocepacia (Devos et al., 2016). Furthermore, β-lactamases were found in MVs of P. aeruginosa and H. influenzae, indicative for a general mechanism to respond to β-lactam stress (Ciofu et al., 2000; Schaar et al., 2011). MVs can also mediate export of antibiotics or extracellular capturing of antibiotics. When P. aeruginosa is treated with the aminoglycoside gentamycin, it secretes gentamycin-containing MVs. These MVs also contain peptidoglycan hydrolase and were shown to be bactericidal against B. cenocepacia (Allan and Beveridge, 2003). Finally, MVs can aid in the inter- and intraspecies spread of resistance genes (Schwechheimer and Kuehn, 2015). Secretion Systems As Targets For Anti-Infective Drugs Development of novel therapies is crucial to manage the spread and impact of these pathogens on CF patients. Classical antibiotics mostly exert their function by inhibiting the growth of bacteria through interference with cell wall biogenesis, DNA replication, transcription, and protein synthesis (Baron and Coombes, 2007). Unfortunately, the rate at which resistance against these traditional antibiotics emerges is alarming, partly due to the rise of mutations in the genes coding for antibiotic targets. Secretion system inhibitors are a novel class of anti-infectives that do not inhibit bacterial growth per se and therefore do not provoke selection for mutations causing resistance. Another advantage is the fairly high degree of conservation of these systems between a whole range of Gram-negative pathogens. Since secreted effectors often play a major role in immune evasion, targeting these important bacterial virulence mechanisms may restore pathogen clearance by the host’s own immune system. Kauppi et al. (2003) found that a family of acylated hydrazones of different salicylaldehydes can inhibit the T3SS at the level of substrate secretion/translocation. The related halogenated salicylaldehydes are capable of inhibiting the transcription of genes encoding T3SS components (Kenny et al., 1997). Thiazolidinones were found to target the formation or assembly of the T3SS needle apparatus. These compounds could also inhibit the T2SS in Pseudomonas and the type IV pili secretion system of Francisella, therefore it is hypothesized that they might act on the conserved outer membrane secretin (Felise et al., 2008; Kline et al., 2009). Other promising targets are the energy-generating ATPases of T2SS and T4SS (Sayer et al., 2014), the accessory lytic transglycosylases of T2SS, T3SS, and T4SS (Koraimann, 2003) and the translocated effector proteins (Coburn et al., 2007; Figueira et al., 2013; Kidwai et al., 2013). By inhibiting T4SS-dependent secretion, horizontal transfer of antibiotic resistance genes could be reduced. Concluding Remarks With as many as 90% of CF patients dying of fatal lung infections every year, it is crucial to find means to eradicate or at least control the growth and spread of these major CF pathogens. Secretion systems provide a useful target, since their effector proteins are responsible for a wealth of host cell compromising actions. Due to the fairly high degree of conservation in the composition of these secretion systems, an inhibitor has the potential to target a whole array of Gram-negative pathogens. Because the growth of the pathogens is unaffected by such compounds, the risk for resistance development is highly reduced. It is therefore essential to keep investing in the identification of novel effector proteins and structural elements of secretion systems, as well as in ways to block secretion of virulence factors and MVs. Author Contributions SoD wrote the chapters on secretion systems. SiD wrote the chapter on outer membrane vesicles. BD edited the manuscript and is the supervisor of the two other authors. Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Funding. This work was supported by funding of the Belgian Science Policy via IAP grant 7/44. BOF research Fund of Ghent University supported the fellowship of SoD and provided funds for research on protein secretion in CF pathogens through a GOA grant. ==== Refs References Alcorn J. F. Wright J. R. (2004 ). Degradation of pulmonary surfactant protein D by Pseudomonas aeruginosa elastase abrogates innate immune function. J. Biol. Chem. 279 30871 –30879 . 10.1074/jbc.M400796200 15123664 Allan N. D. Beveridge T. J. (2003 ). Gentamicin delivery to Burkholderia cepacia group IIIa strains via membrane vesicles from Pseudomonas aeruginosa PAO1. Antimicrob. Agents Chemother. 47 2962 –2965 . 10.1128/AAC.47.9.2962-2965.2003 12937002 Allan N. D. Kooi C. Sokol P. A. Beveridge T. J. (2003 ). Putative virulence factors are released in association with membrane vesicles from Burkholderia cepacia. Can. J. Microbiol. 49 613 –624 . 10.1139/w03-078 14663495 Allured V. S. Collier R. J. Carroll S. F. McKay D. B. (1986 ). Structure of exotoxin A of Pseudomonas aeruginosa at 3.0-Angstrom resolution. Proc. Natl. Acad. Sci. U.S.A. 83 1320 –1324 . 10.1073/pnas.83.5.1320 3006045 Alvarez-Martinez C. E. Christie P. J. (2009 ). Biological diversity of prokaryotic type IV secretion systems. Microbiol. Mol. Biol. Rev. 73 775 –808 . 10.1128/MMBR.00023-09 19946141 Ballister E. R. Lai A. H. Zuckermann R. N. Cheng Y. Mougous J. D. (2008 ). In vitro self-assembly of tailorable nanotubes from a simple protein building block. Proc. Natl. Acad. Sci. U.S.A. 105 3733 –3738 . 10.1073/pnas.0712247105 18310321 Barbieri J. T. Sun J. (2004 ). Pseudomonas aeruginosa ExoS and ExoT. Rev. Physiol. Biochem. Pharmacol. 152 79 –92 . 10.1007/s10254-004-0031-7 15375697 Baron C. Coombes B. (2007 ). Targeting bacterial secretion systems: benefits of disarmament in the microcosm. Infect. Disord. Drug Targets 7 19 –27 . 10.2174/187152607780090685 17346208 Basler M. Pilhofer M. Henderson G. P. Jensen G. J. Mekalanos J. J. (2012 ). Type VI secretion requires a dynamic contractile phage tail-like structure. Nature 483 182 –186 . 10.1038/nature10846 22367545 Baumann U. Wu S. Flaherty K. M. McKay D. B. (1993 ). Three-dimensional structure of the alkaline protease of Pseudomonas aeruginosa: a two-domain protein with a calcium binding parallel beta roll motif. EMBO J. 12 3357 –3364 .8253063 Bomberger J. M. Ely K. H. Bangia N. Ye S. Green K. A. Green W. R. (2014 ). Pseudomonas aeruginosa Cif protein enhances the ubiquitination and proteasomal degradation of the transporter associated with antigen processing (TAP) and reduces major histocompatibility complex (MHC) class I antigen presentation. J. Biol. Chem. 289 152 –162 . 10.1074/jbc.M113.459271 24247241 Bomberger J. M. Maceachran D. P. Coutermarsh B. A. Ye S. O’Toole G. A. Stanton B. A. (2009 ). Long-distance delivery of bacterial virulence factors by Pseudomonas aeruginosa outer membrane vesicles. PLoS Pathog. 5 :e1000382 10.1371/journal.ppat.1000382 Bottone E. J. Reitano M. Janda J. M. Troy K. Cuttner J. (1986 ). Pseudomonas maltophilia exoenzyme activity as correlate in pathogenesis of ecthyma gangrenosum. J. Clin. Microbiol. 24 995 –997 .3537006 Cahan R. Axelrad I. Safrin M. Ohman D. E. Kessler E. (2001 ). A secreted aminopeptidase of Pseudomonas aeruginosa. Identification, primary structure, and relationship to other aminopeptidases. J. Biol. Chem. 276 43645 –43652 . 10.1074/jbc.M106950200 11533066 Chapman M. R. Robinson L. S. Pinkner J. S. Roth R. Heuser J. Hammar M. (2002 ). Role of Escherichia coli curli operons in directing amyloid fiber formation. Science 295 851 –855 . 10.1126/science.1067484 11823641 Christie P. J. Whitaker N. Gonzalez-Rivera C. (2014 ). Mechanism and structure of the bacterial type IV secretion systems. Biochim. Biophys. Acta 1843 1578 –1591 . 10.1016/j.bbamcr.2013.12.019 24389247 Chung J. W. Altman E. Beveridge T. J. Speert D. P. (2003 ). Colonial morphology of Burkholderia cepacia complex genomovar III: implications in exopolysaccharide production, pilus expression, and persistence in the mouse. Infect. Immun. 71 904 –909 . 10.1128/IAI.71.2.904-909.2003 12540572 Cianciotto N. P. (2005 ). Type II secretion: a protein secretion system for all seasons. Trends Microbiol. 13 581 –588 . 10.1016/j.tim.2005.09.005 16216510 Ciofu O. Beveridge T. J. Kadurugamuwa J. Walther-Rasmussen J. Hoiby N. (2000 ). Chromosomal beta-lactamase is packaged into membrane vesicles and secreted from Pseudomonas aeruginosa. J. Antimicrob. Chemother. 45 9 –13 . 10.1093/jac/45.1.9 10629007 Coburn B. Sekirov I. Finlay B. B. (2007 ). Type III secretion systems and disease. Clin. Microbiol. Rev. 20 535 –549 . 10.1128/CMR.00013-07 17934073 Corbett C. R. Burtnick M. N. Kooi C. Woods D. E. Sokol P. A. (2003 ). An extracellular zinc metalloprotease gene of Burkholderia cepacia. Microbiology 149(Pt 8) , 2263 –2271 . 10.1099/mic.0.26243-0 12904566 Cornelis G. R. (2006 ). The type III secretion injectisome. Nat. Rev. Microbiol. 4 811 –825 . 10.1038/nrmicro1526 17041629 Costa T. R. Felisberto-Rodrigues C. Meir A. Prevost M. S. Redzej A. Trokter M. (2015 ). Secretion systems in Gram-negative bacteria: structural and mechanistic insights. Nat. Rev. Microbiol. 13 343 –359 . 10.1038/nrmicro3456 25978706 Crossman L. C. Gould V. C. Dow J. M. Vernikos G. S. Okazaki A. Sebaihia M. (2008 ). The complete genome, comparative and functional analysis of Stenotrophomonas maltophilia reveals an organism heavily shielded by drug resistance determinants. Genome Biol. 9 :R74 10.1186/gb-2008-9-4-r74 Cystic Fibrosis Foundation (2015 ). Patient Registry. 2014 Annual Data Report. Bethesda, MD : Cystic Fibrosis Foundation . Dacheux D. Goure J. Chabert J. Usson Y. Attree I. (2001 ). Pore-forming activity of type III system-secreted proteins leads to oncosis of Pseudomonas aeruginosa-infected macrophages. Mol. Microbiol. 40 76 –85 . 10.1046/j.1365-2958.2001.02368.x 11298277 Darling P. Chan M. Cox A. D. Sokol P. A. (1998 ). Siderophore production by cystic fibrosis isolates of Burkholderia cepacia. Infect. Immun. 66 874 –877 .9453660 Deknuydt F. Nordstrom T. Riesbeck K. (2014 ). Diversion of the host humoral response: a novel virulence mechanism of Haemophilus influenzae mediated via outer membrane vesicles. J. Leukoc. Biol. 95 983 –991 . 10.1189/jlb.1013527 24550522 Denton M. Kerr K. G. (2002 ). Molecular epidemiology of Stenotrophomonas maltophilia isolated from cystic fibrosis patients. J. Clin. Microbiol. 40 :1884 10.1128/JCM.40.5.1884.2002 Devos S. Stremersch S. Raemdonck K. Braeckmans K. Devreese B. (2016 ). Intra- and interspecies effects of outer membrane vesicles from Stenotrophomonas maltophilia on beta-Lactam resistance. Antimicrob. Agents Chemother. 60 2516 –2518 . 10.1128/AAC.02171-15 26787686 Devos S. Van Oudenhove L. Stremersch S. Van Putte W. De Rycke R. Van Driessche G. (2015 ). The effect of imipenem and diffusible signaling factors on the secretion of outer membrane vesicles and associated Ax21 proteins in Stenotrophomonas maltophilia. Front. Microbiol. 6 :298 10.3389/fmicb.2015.00298 Diaz-Laviada I. Larrodera P. Diaz-Meco M. T. Cornet M. E. Guddal P. H. Johansen T. (1990 ). Evidence for a role of phosphatidylcholine-hydrolysing phospholipase C in the regulation of protein kinase C by ras and src oncogenes. EMBO J. 9 3907 –3912 .2123453 DuMont A. L. Karaba S. M. Cianciotto N. P. (2015 ). Type II secretion-dependent degradative and cytotoxic activities mediated by Stenotrophomonas maltophilia serine proteases StmPr1 and StmPr2. Infect. Immun. 83 3825 –3837 . 10.1128/IAI.00672-15 26169274 Duong F. Bonnet E. Geli V. Lazdunski A. Murgier M. Filloux A. (2001 ). The AprX protein of Pseudomonas aeruginosa: a new substrate for the Apr type I secretion system. Gene 262 147 –153 . 10.1016/S0378-1119(00)00541-2 11179678 Engel L. S. Hill J. M. Caballero A. R. Green L. C. O’Callaghan R. J. (1998 ). Protease IV, a unique extracellular protease and virulence factor from Pseudomonas aeruginosa. J. Biol. Chem. 273 16792 –16797 . 10.1074/jbc.273.27.16792 9642237 Faure L. M. Garvis S. de Bentzmann S. Bigot S. (2014 ). Characterization of a novel two-partner secretion system implicated in the virulence of Pseudomonas aeruginosa. Microbiology 160(Pt 9) , 1940 –1952 . 10.1099/mic.0.079616-0 25009238 Felise H. B. Nguyen H. V. Pfuetzner R. A. Barry K. C. Jackson S. R. Blanc M. P. (2008 ). An inhibitor of gram-negative bacterial virulence protein secretion. Cell Host Microbe 4 325 –336 . 10.1016/j.chom.2008.08.001 18854237 Fernaays M. M. (2008 ). Virulence Determinants of Pathogenic Nontypeable Haemophilus Influenzae . Buffalo, NY : State University of New York . Ferrer-Navarro M. Planell R. Yero D. Mongiardini E. Torrent G. Huedo P. (2013 ). Abundance of the quorum-sensing factor Ax21 in four strains of Stenotrophomonas maltophilia correlates with mortality rate in a new zebrafish model of infection. PLoS ONE 8 :e67207 10.1371/journal.pone.0067207 Figueira R. Watson K. G. Holden D. W. Helaine S. (2013 ). Identification of Salmonella pathogenicity island-2 type III secretion system effectors involved in intramacrophage replication of S. enterica serovar typhimurium: implications for rational vaccine design. MBio 4 :e00065 10.1128/mBio.00065-13 Fink D. L. Buscher A. Z. Green B. Fernsten P. (2003 ). The Haemophilus influenzae Hap autotransporter mediates microcolony formation and adherence to epithelial cells and extracellular matrix via binding regions in the C-terminal end of the passenger domain. Cell. Microbiol. 5 175 –186 . 10.1046/j.1462-5822.2003.00266.x 12614461 Folders J. Tommassen J. van Loon L. C. Bitter W. (2000 ). Identification of a chitin-binding protein secreted by Pseudomonas aeruginosa. J. Bacteriol. 182 1257 –1263 . 10.1128/JB.182.5.1257-1263.2000 10671445 Francetic O. Pugsley A. P. (2005 ). Towards the identification of type II secretion signals in a nonacylated variant of pullulanase from Klebsiella oxytoca. J. Bacteriol. 187 7045 –7055 . 10.1128/JB.187.20.7045-7055.2005 16199575 Guglielmini J. Quintais L. Garcillan-Barcia M. P. de la Cruz F. Rocha E. P. (2011 ). The repertoire of ICE in prokaryotes underscores the unity, diversity, and ubiquity of conjugation. PLoS Genet. 7 :e1002222 10.1371/journal.pgen.1002222 Guzzo J. Pages J. M. Duong F. Lazdunski A. Murgier M. (1991 ). Pseudomonas aeruginosa alkaline protease: evidence for secretion genes and study of secretion mechanism. J. Bacteriol. 173 5290 –5297 .1832151 Harland D. N. Dassa E. Titball R. W. Brown K. A. Atkins H. S. (2007 ). ATP-binding cassette systems in Burkholderia pseudomallei and Burkholderia mallei. BMC Genomics 8 :83 10.1186/1471-2164-8-83 Harrison A. Dyer D. W. Gillaspy A. Ray W. C. Mungur R. Carson M. B. (2005 ). Genomic sequence of an otitis media isolate of nontypeable Haemophilus influenzae: comparative study with H. influenzae serotype d, strain KW20. J. Bacteriol. 187 4627 –4636 . 10.1128/JB.187.13.4627-4636.2005 15968074 Haurat M. F. Elhenawy W. Feldman M. F. (2015 ). Prokaryotic membrane vesicles: new insights on biogenesis and biological roles. Biol. Chem. 396 95 –109 . 10.1515/hsz-2014-0183 25178905 Hinsa S. M. Espinosa-Urgel M. Ramos J. L. O’Toole G. A. (2003 ). Transition from reversible to irreversible attachment during biofilm formation by Pseudomonas fluorescens WCS365 requires an ABC transporter and a large secreted protein. Mol. Microbiol. 49 905 –918 . 10.1046/j.1365-2958.2003.03615.x 12890017 Ho B. T. Dong T. G. Mekalanos J. J. (2014 ). A view to a kill: the bacterial type VI secretion system. Cell Host Microbe 15 9 –21 . 10.1016/j.chom.2013.11.008 24332978 Holden M. T. Seth-Smith H. M. Crossman L. C. Sebaihia M. Bentley S. D. Cerdeno-Tarraga A. M. (2009 ). The genome of Burkholderia cenocepacia J2315 an epidemic pathogen of cystic fibrosis patients. J. Bacteriol. 191 261 –277 . 10.1128/JB.01230-08 18931103 Hoyle B. D. Costerton J. W. (1991 ). Bacterial resistance to antibiotics: the role of biofilms. Prog. Drug Res. 37 91 –105 . 10.1007/978-3-0348-7139-6_2 1763187 Juhas M. Crook D. W. Dimopoulou I. D. Lunter G. Harding R. M. Ferguson D. J. (2007a ). Novel type IV secretion system involved in propagation of genomic islands. J. Bacteriol. 189 761 –771 . 10.1128/JB.01327-06 17122343 Juhas M. Crook D. W. Hood D. W. (2008 ). Type IV secretion systems: tools of bacterial horizontal gene transfer and virulence. Cell. Microbiol. 10 2377 –2386 . 10.1111/j.1462-5822.2008.01187.x 18549454 Juhas M. Power P. M. Harding R. M. Ferguson D. J. Dimopoulou I. D. Elamin A. R. (2007b ). Sequence and functional analyses of Haemophilus spp. genomic islands. Genome Biol. 8 :R237 10.1186/gb-2007-8-11-r237 Karaba S. M. White R. C. Cianciotto N. P. (2013 ). Stenotrophomonas maltophilia encodes a type II protein secretion system that promotes detrimental effects on lung epithelial cells. Infect. Immun. 81 3210 –3219 . 10.1128/IAI.00546-13 23774603 Kauppi A. M. Nordfelth R. Uvell H. Wolf-Watz H. Elofsson M. (2003 ). Targeting bacterial virulence: inhibitors of type III secretion in Yersinia. Chem. Biol. 10 241 –249 . 10.1016/S1074-5521(03)00046-2 12670538 Kenny B. DeVinney R. Stein M. Reinscheid D. J. Frey E. A. Finlay B. B. (1997 ). Enteropathogenic E. coli (EPEC) transfers its receptor for intimate adherence into mammalian cells. Cell 91 511 –520 . 10.1016/S0092-8674(00)80437-7 9390560 Kida Y. Higashimoto Y. Inoue H. Shimizu T. Kuwano K. (2008 ). A novel secreted protease from Pseudomonas aeruginosa activates NF-kappaB through protease-activated receptors. Cell. Microbiol. 10 1491 –1504 . 10.1111/j.1462-5822.2008.01142.x 18331590 Kidwai A. S. Mushamiri I. Niemann G. S. Brown R. N. Adkins J. N. Heffron F. (2013 ). Diverse secreted effectors are required for Salmonella persistence in a mouse infection model. PLoS ONE 8 :e70753 10.1371/journal.pone.0070753 Kline T. Barry K. C. Jackson S. R. Felise H. B. Nguyen H. V. Miller S. I. (2009 ). Tethered thiazolidinone dimers as inhibitors of the bacterial type III secretion system. Bioorg. Med. Chem. Lett. 19 1340 –1343 . 10.1016/j.bmcl.2009.01.047 19195888 Kooi C. Sokol P. A. (2009 ). Burkholderia cenocepacia zinc metalloproteases influence resistance to antimicrobial peptides. Microbiology 155(Pt 9) , 2818 –2825 . 10.1099/mic.0.028969-0 19542010 Koraimann G. (2003 ). Lytic transglycosylases in macromolecular transport systems of Gram-negative bacteria. Cell. Mol. Life Sci. 60 2371 –2388 . 10.1007/s00018-003-3056-1 14625683 Kostyanev T. S. Sechanova L. P. (2012 ). Virulence factors and mechanisms of antibiotic resistance of Haemophilus influenzae. Folia Med. (Plovdiv) 54 19 –23 . 10.2478/v10153-011-0073-y 22908826 Lee V. T. Smith R. S. Tummler B. Lory S. (2005 ). Activities of Pseudomonas aeruginosa effectors secreted by the Type III secretion system in vitro and during infection. Infect. Immun. 73 1695 –1705 . 10.1128/IAI.73.3.1695-1705.2005 15731070 Leiman P. G. Basler M. Ramagopal U. A. Bonanno J. B. Sauder J. M. Pukatzki S. (2009 ). Type VI secretion apparatus and phage tail-associated protein complexes share a common evolutionary origin. Proc. Natl. Acad. Sci. U.S.A. 106 4154 –4159 . 10.1073/pnas.0813360106 19251641 Leo J. C. Grin I. Linke D. (2012 ). Type V secretion: mechanism(s) of autotransport through the bacterial outer membrane. Philos. Trans. R. Soc. Lond. B Biol. Sci. 367 1088 –1101 . 10.1098/rstb.2011.0208 22411980 Letoffe S. Redeker V. Wandersman C. (1998 ). Isolation and characterization of an extracellular haem-binding protein from Pseudomonas aeruginosa that shares function and sequence similarities with the Serratia marcescens HasA haemophore. Mol. Microbiol. 28 1223 –1234 . 10.1046/j.1365-2958.1998.00885.x 9680211 Lipuma J. J. (2010 ). The changing microbial epidemiology in cystic fibrosis. Clin. Microbiol. Rev 23 299 –323 . 10.1128/CMR.00068-09 20375354 Lu H. M. Lory S. (1996 ). A specific targeting domain in mature exotoxin A is required for its extracellular secretion from Pseudomonas aeruginosa. EMBO J. 15 429 –436 .8617218 Marlovits T. C. Kubori T. Sukhan A. Thomas D. R. Galan J. E. Unger V. M. (2004 ). Structural insights into the assembly of the type III secretion needle complex. Science 306 :1040 10.1126/science.1102610 Mikkelsen H. Hui K. Barraud N. Filloux A. (2013 ). The pathogenicity island encoded PvrSR/RcsCB regulatory network controls biofilm formation and dispersal in Pseudomonas aeruginosa PA14. Mol. Microbiol. 89 450 –463 . 10.1111/mmi.12287 23750818 Nakazawa T. (1996 ). Molecular Biology of Pseudomonads. Washington, DC : ASM Press . Olson J. C. Ohman D. E. (1992 ). Efficient production and processing of elastase and LasA by Pseudomonas aeruginosa require zinc and calcium ions. J. Bacteriol. 174 4140 –4147 .1597429 Ostroff R. M. Vasil A. I. Vasil M. L. (1990 ). Molecular comparison of a nonhemolytic and a hemolytic phospholipase C from Pseudomonas aeruginosa. J. Bacteriol. 172 5915 –5923 .2120196 Pukatzki S. Ma A. T. Revel A. T. Sturtevant D. Mekalanos J. J. (2007 ). Type VI secretion system translocates a phage tail spike-like protein into target cells where it cross-links actin. Proc. Natl. Acad. Sci. U.S.A. 104 15508 –15513 . 10.1073/pnas.0706532104 17873062 Razvi S. Quittell L. Sewall A. Quinton H. Marshall B. Saiman L. (2009 ). Respiratory microbiology of patients with cystic fibrosis in the United States, 1995 to 2005. Chest 136 1554 –1560 . 10.1378/chest.09-0132 19505987 Rocco F. (2011 ). Core and Variable Components in Prokaryotic Genomes. Naples : University of Napoli Federico II . Rosadini C. V. (2011 ). Roles of Secreted Virulence Factors in Pathogenecity of Haemophilus influenzae. Doctor of Philosophy, thesis , University of Massachussetts Medical School, Worcester, MA . Ruer S. Ball G. Filloux A. de Bentzmann S. (2008 ). The ‘P-usher,’ a novel protein transporter involved in fimbrial assembly and TpsA secretion. EMBO J. 27 2669 –2680 . 10.1038/emboj.2008.197 18833195 Ryan R. P. Monchy S. Cardinale M. Taghavi S. Crossman L. Avison M. B. (2009 ). The versatility and adaptation of bacteria from the genus Stenotrophomonas. Nat. Rev. Microbiol. 7 514 –525 . 10.1038/nrmicro2163 19528958 Sajjan U. S. Sun L. Goldstein R. Forstner J. F. (1995 ). Cable (cbl) type II pili of cystic fibrosis-associated Burkholderia (Pseudomonas) cepacia: nucleotide sequence of the cblA major subunit pilin gene and novel morphology of the assembled appendage fibers. J. Bacteriol. 177 1030 –1038 .7532166 Sato H. Frank D. W. (2004 ). ExoU is a potent intracellular phospholipase. Mol. Microbiol. 53 1279 –1290 . 10.1111/j.1365-2958.2004.04194.x 15387809 Sauvonnet N. Pugsley A. P. (1996 ). Identification of two regions of Klebsiella oxytoca pullulanase that together are capable of promoting beta-lactamase secretion by the general secretory pathway. Mol. Microbiol. 22 1 –7 . 10.1111/j.1365-2958.1996.tb02650.x 8899703 Sayer J. R. Walldén K. Pesnot T. Campbell F. Gane P. J. Simone M. (2014 ). 2- and 3-substituted imidazo(12-a)pyrazines as inhibitors of bacterial type IV secretion. Bioorg. Med. Chem. 22 6459 –6470 . 10.1016/j.bmc.2014.09.036 25438770 Schaar V. Nordstrom T. Morgelin M. Riesbeck K. (2011 ). Moraxella catarrhalis outer membrane vesicles carry beta-lactamase and promote survival of Streptococcus pneumoniae and Haemophilus influenzae by inactivating amoxicillin. Antimicrob. Agents Chemother. 55 3845 –3853 . 10.1128/AAC.01772-10 21576428 Schoehn G. Di Guilmi A. M. Lemaire D. Attree I. Weissenhorn W. Dessen A. (2003 ). Oligomerization of type III secretion proteins PopB and PopD precedes pore formation in Pseudomonas. EMBO J. 22 4957 –4967 . 10.1093/emboj/cdg499 14517235 Schwechheimer C. Kuehn M. J. (2015 ). Outer-membrane vesicles from Gram-negative bacteria: biogenesis and functions. Nat. Rev. Microbiol. 13 605 –619 . 10.1038/nrmicro3525 26373371 Souza D. P. Oka G. U. Alvarez-Martinez C. E. Bisson-Filho A. W. Dunger G. Hobeika L. (2015 ). Bacterial killing via a type IV secretion system. Nat. Commun. 6 :6453 10.1038/ncomms7453 Spahich N. A. St Geme J. W. III. (2011 ). Structure and function of the Haemophilus influenzae autotransporters. Front. Cell. Infect. Microbiol. 1 :5 10.3389/fcimb.2011.00005 Tomich M. Griffith A. Herfst C. A. Burns J. L. Mohr C. D. (2003 ). Attenuated virulence of a Burkholderia cepacia type III secretion mutant in a murine model of infection. Infect. Immun. 71 1405 –1415 . 10.1128/IAI.71.3.1405-1415.2003 12595458 Uehlinger S. Schwager S. Bernier S. P. Riedel K. Nguyen D. T. Sokol P. A. (2009 ). Identification of specific and universal virulence factors in Burkholderia cenocepacia strains by using multiple infection hosts. Infect. Immun. 77 4102 –4110 . 10.1128/IAI.00398-09 19528212 van’t Wout E. F. van Schadewijk A. van Boxtel R. Dalton L. E. Clarke H. J. Tommassen J. (2015 ). Virulence factors of Pseudomonas aeruginosa induce both the unfolded protein and integrated stress responses in airway epithelial cells. PLoS Pathog. 11 :e1004946 10.1371/journal.ppat.1004946 Wandersman C. (1996 ). Escherichia coli and Salmonella tiphymurium Cellular and Molecular Biology. Washington, DC : ASM Press . Waters V. (2012 ). New treatments for emerging cystic fibrosis pathogens other than Pseudomonas. Curr. Pharm. Des. 18 696 –725 . 10.2174/138161212799315939 22229574 Wilhelm S. Gdynia A. Tielen P. Rosenau F. Jaeger K. E. (2007 ). The autotransporter esterase EstA of Pseudomonas aeruginosa is required for rhamnolipid production, cell motility, and biofilm formation. J. Bacteriol. 189 6695 –6703 . 10.1128/JB.00023-07 17631636 Wozniak R. A. Waldor M. K. (2010 ). Integrative and conjugative elements: mosaic mobile genetic elements enabling dynamic lateral gene flow. Nat. Rev. Microbiol. 8 552 –563 . 10.1038/nrmicro2382 20601965 Yahr T. L. Vallis A. J. Hancock M. K. Barbieri J. T. Frank D. W. (1998 ). ExoY, an adenylate cyclase secreted by the Pseudomonas aeruginosa type III system. Proc. Natl. Acad. Sci. U.S.A. 95 13899 –13904 . 10.1073/pnas.95.23.13899 9811898 Zhang R. LiPuma J. J. Gonzalez C. F. (2009 ). Two type IV secretion systems with different functions in Burkholderia cenocepacia K56-2. Microbiology 155(Pt 12) , 4005 –4013 . 10.1099/mic.0.033043-0 19744991

PMC005xxxxxx/PMC5003818.txt

==== Front Front MicrobiolFront MicrobiolFront. Microbiol.Frontiers in Microbiology1664-302XFrontiers Media S.A. 10.3389/fmicb.2016.01376MicrobiologyOriginal ResearchDetection of Thermal Sublethal Injury in Escherichia coli via the Selective Medium Plating Technique: Mechanisms and Improvements Espina Laura García-Gonzalo Diego Pagán Rafael *Departamento de Producción Animal y Ciencia de los Alimentos, Facultad de Veterinaria, Instituto Agroalimentario de Aragón – IA2, CITA-Universidad de ZaragozaZaragoza, SpainEdited by: Avelino Alvarez-Ordóñez, Teagasc Food Research Centre, Ireland Reviewed by: Gonzalo García De Fernando, Complutense University of Madrid, Spain; Hélène Simonin, Agrosup Dijon, France; Paula María Periago, Universidad Politécnica de Cartagena, Spain *Correspondence: Rafael Pagán, pagan@unizar.esThis article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology 30 8 2016 2016 7 137603 5 2016 19 8 2016 Copyright © 2016 Espina, García-Gonzalo and Pagán.2016Espina, García-Gonzalo and PagánThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.In food preservation, the synergistic combination of different technologies aims to maximize the total lethality of the process and minimize the intensity of each hurdle. This is especially the case when at least one of the treatments can cause sublethal (reparable) injury in a great proportion of the population, so that sublethally injured cells can end up being entirely inactivated by the other hurdle(s). The selective medium plating technique (SMPT) is extensively used to enumerate bacterial sublethal injury after inimical treatments, being sodium chloride added to the recovery medium to detect damaged bacterial envelopes. However, little work has been done to explain the reasons for the inability of sublethally injured cells to outgrow in selective agar media, whereas they are able to grow in non-selective agar. In the present paper, the performance of SMPT on Escherichia coli cells after heat treatments is explored by applying different selective agents in the recovery media, using mutants lacking factors involved in osmoregulation, and also by examining the integrity of the cytoplasmic membrane. In view of the results, the possibility of a specific toxic effect of Na+ as the main mechanism under SMPT was discarded, since the same level of sublethal injury was detected using KCl instead of NaCl. The synthesis of the osmoprotectant trehalose determined the maximum osmotolerance of intact cells to the selective agents, but was not crucial in the quantification of sublethal injury. Moreover, for the first time, the extent of sublethal injury detected via SMPT was directly correlated with the physical loss of integrity of the cell membrane in 99.999% of the initial population. This was achieved through statistical analysis of flow cytometry data using propidium iodide-exclusion technique when that dye was added before thermal treatments. The present work confirms the adequacy of SMPT as a tool for detecting the occurrence and quantity of sublethally injured cells after thermal treatments and thus, for efficiently designing the combination of heat with other preservation techniques. We also propose the study of statistical analysis from flow cytometry data for a more rapid quantification of bacterial sublethal injury in a broad detection range. sublethal injuryosmoregulationselective mediaEscherichia coliflow cytometryMinisterio de EconomÍa y Competitividad10.13039/501100003329AGL2012–32165, AGL2015–69565 ==== Body Introduction In bacteriology, viability has been traditionally defined and measured as the ability of organisms to self-replicate in culture media (Bogosian and Bourneuf, 2001; Nyström, 2001). However, it has long been known that the failure of a bacterial cell to produce a colony on a standard nutrient plate may not necessarily mean that the cell was dead at the time of sampling (Nyström, 2001). For instance, microorganisms that are metabolically active despite their inability to grow in laboratory culture media are said to be in a “viable but non-culturable” (VBNC) state, which, under harsh environmental conditions, can be triggered as a survival mechanism (Bogosian and Bourneuf, 2001). On other occasions, exposure to chemical or physical processes can lead to the sublethal injury of bacterial cells: this state is considered to be transient, since cells are able to repair their damages and resume growth if suitable environmental conditions emerge (Mackey, 2000). In food preservation it has been demonstrated that, once one has applied preservation treatments to control bacterial food contamination, a considerable proportion of the population may become sublethally injured in addition to both the surviving (non-injured) and the inactivated populations (Wesche et al., 2009). The adequate identification and quantification of the sublethally injured population plays a key role in food safety. Since damaged cells are not generally able to grow on the conventional selective enrichment media used in the food industry (Restaino et al., 2001), they can remain undetected, subsequently repair their damages and reach infective concentrations (Mackey, 2000). On the other hand, according to the “hurdle effect” (Leistner and Gorris, 1995), repair of sublethally injured cells after a preservation treatment can be adequately prevented by the combination of additional preservation agents (hurdles) that interfere with cellular homeostasis maintenance, thereby synergistically increasing the combined process’s global lethality (Mackey, 2000). In this regard, although new methods are being developed for the detection of sublethally injured bacteria (Back et al., 2012; Gelaw et al., 2014), the most widely used strategy among microbiologists is still differential enumeration on non-selective and selective agar, following the so-called selective medium plating technique (SMPT; Mackey, 2000). For this purpose, out of all possible selective agents, subinhibitory concentrations of sodium chloride have been consistently incorporated in the recovery medium (Chilton et al., 2001; Ulmer et al., 2002; Miller et al., 2006). It is believed that the increase in osmotic pressure caused by the addition of sodium chloride explains the selective outgrowth of only those cells whose cytoplasmic membrane remains intact (Mackey, 2000). Despite the observed selective effect of the osmolyte NaCl on bacterial growth, little research has been done to study the osmoregulatory mechanisms of sublethally injured cells and, therefore, to find out more about their ability to maintain selective permeability after different stresses. In intact bacteria, an osmotic upshock unleashes a cascade of events intended to maintain turgor pressure within limits by regulating the total osmotic solute pool in the cytoplasm (and in the periplasm in Gram-negative bacteria; Wood, 2011). As the osmolality of the surrounding environment increases, turgor pressure drops and growth slows or halts (Wood, 2011). The most rapid response to this osmotic upshock is an increase in potassium ion influx that increases cytosolic osmolality (Wood, 2011). Since high intracellular concentrations of K+ interfere with many important cellular functions, the cell starts to accumulate large quantities of so-called compatible solutes, which are more congruous with its physiology (Wood et al., 2001). The compatible solute trehalose is synthesized (via Ots system) and accumulated up to levels that may comprise as much as 20% of cytoplasmic osmolality under conditions of high osmolality (Wood, 1999). Other compounds, when present externally (such as glycine betaine), can be incorporated via transporters such as BetT or ProP (Lucht and Bremer, 1994; Wood, 2011), leading to a decrease in trehalose levels and stimulating bacterial growth rates under hyperosmotic conditions. It has been estimated that after 1 h of osmotic stress, a cell’s physiology and structure are largely restored via these osmoregulatory systems (Wood, 1999). A better knowledge of the interaction between cellular osmoregulatory mechanisms and the permeabilization of the cytoplasmic membrane and how they influence bacterial ability to outgrow in selective media could facilitate the estimation of sublethal injury and, therefore, help us improve the design of food preservation processes. In the present study, SMPT is applied as the primary technique to detect and quantify the proportion of sublethally injured cells in their cytoplasmic membrane after exposure to a lethal stress. Thermal treatment was selected as the lethal stress, since it is the most studied and best understood treatment known to sublethally injure microorganisms (Wesche et al., 2009). It should be noted that mild thermal treatments applied in fluid environments have been demonstrated to disturb the permeability of the outer membrane earlier and more intensely than the permeability of the cytoplasmic membrane (Mackey, 2000; Shigapova, 2004); thus, the outer membrane does not interfere with the detection of sublethal injury in the cytoplasmic membrane. The microorganism Escherichia coli was also selected, since it is the model microorganism for studying bacterial osmoregulation (Shabala et al., 2009). Besides, the availability of a great variety of E. coli mutants lacking factors involved in the osmoregulatory system (Baba et al., 2006) can be used to determine those factors’ role in SMPT. The primary objective of this study was (i) to gain a better understanding of the mechanisms underlying SMPT by trying to identify which bacterial osmoregulatory mechanisms or physical structures are modified by heat and are thus responsible for the prevention of bacterial growth in selective media. Additionally, we aimed (ii) to improve traditional SMPT by testing the effect of different variations in the composition of the recovery media, and also (iii) to explore the possible use of flow cytometry as a complementary technique to assess sublethal injury. Materials and Methods Preparation of Media Minimal medium M9 was chosen as the broth and treatment medium, since it is commonly used for the culture of E. coli (Neidhardt et al., 1974), and because its minimal composition reduces the presence of osmolytes or osmoprotectants influencing the osmoregulation processes. M9 minimal broth was prepared following the steps indicated in Maniatis et al. (1982): its composition is of 38 mM Na2HPO4, 20 mM KH2PO4, 7.7 mM NaCl, 17 mM NH4Cl, 1 mM MgSO4, 0.1 mM CaCl2, and 0.2% glucose. Regarding the recovery media, both minimal and rich agar plates were prepared to cover a whole range of culture conditions, as both types are commonly used in the study of sublethal injury (Wesche et al., 2009). In addition to the ingredients in M9 minimal broth, the M9 minimal agar medium contained 15 g/L of Agar Technical No. 3 (Oxoid, Basingstoke, UK). Tryptic soy agar (Biolife, Milan, Italy) plus 0.6% of yeast extract (Biolife; TSAYE) was selected as the rich recovery medium, given its widespread use in the enumeration of bacterial injury (Miller et al., 2006; Noriega et al., 2013). Preliminary experiments showed that recovery in M9 minimal agar medium after different thermal treatments yielded similar counts than in TSAYE (data not shown). Although, NaCl is the solute most commonly used to inhibit growth in selective agar media when evaluating sublethal injury in the cytoplasmic membrane, we also tested the osmolytes KCl and saccharose. With the objective of determining the influence of the type of osmolyte in the detection of sublethal injury, each solute was added in the concentration required to achieve the same osmolality values in the agar medium. For this purpose, the osmolality values of the agar (Os/kg of M9 agar medium) were chosen to correspond with those created by the addition of 1–6% of NaCl, and resulted in a range of 0.34–2.05 Os/kg of agar medium. The KCl and saccharose concentrations required to achieve such osmolality values were 1.27–7.68% KCl or 11.63–70.17% saccharose. Betaine was added as osmoprotectant at 1 mM, following the lines of previous research (Le Rudulier et al., 1984; McLaggan et al., 2002). Higher concentrations were not proven more effective to osmotically protect cells (data not shown). Micro-Organisms and Growth Conditions The strains used were E. coli BW25113 and its deleterious mutants E. coli ΔotsA, ΔproP, ΔnhaA, ΔnhaB, and ΔnhaR. While factors OtsA and ProP are involved in the synthesis of trehalose and the uptake of betaine respectively, the different subunits of the factor Nha are involved in the excretion of Na+. All strains were obtained from the KEIO collection (Baba et al., 2006). The cultures were maintained in cryovials at -80°C prior to use. Broth subcultures were prepared by inoculating one single colony from a plate in a 50-mL flask containing 10 mL of sterile M9 minimal medium. After inoculation, the flasks were incubated overnight at 37°C. With these subcultures, 250-mL Erlenmeyer flasks containing 50 mL of M9 medium were inoculated into a final concentration of 3 × 106 CFU/mL. These flasks were incubated with agitation (130 rpm; Selecta, mod. Rotabit, Barcelona, Spain) at 37°C until stationary growth phase was reached (24 h). Thermal Treatments Before inoculation, cultures were centrifuged at 6000 × g for 5 min and resuspended in the treatment medium (M9 medium). For the preparation of heat-treated samples, 0.1 mL of culture at 109 CFU/mL was added to a tube containing 0.9 mL of M9 medium tempered at 55 ± 0.2°C or at 53, 57, or 59 ± 0.2°C (FX Incubator, A. F. Ingeniería S. L., Valencia, Spain). The actual temperature was controlled with a thermocouple wire introduced in a 0.9 mL M9 broth test tube inside the incubator. After each individual treatment interval, samples were taken, immediately placed on ice, and adequately diluted in 0.1% w/v peptone water (Biolife). Survivors were evaluated as explained below. Exceptionally for an experiment aimed to compare inactivation kinetics in the absence and presence of the osmoprotectant betaine, survival curves to heat treatments were obtained in a specially designed thermo-resistometer, as previously described (Condón et al., 1993). This device has a thermocouple (Pt 100) to monitor the temperature during heat treatment and for the injection of inoculum. Once the temperature had stabilized (at 58, 61, 64, 67, or 70°C), 0.2 mL of culture was injected via a solenoid-valve-operated automatic syringe into the 400-mL treatment chamber containing the treatment medium under constant agitation. Samples were taken at regular intervals and survivors were evaluated as explained below. Collection of Samples, Counts of Culturable Cells and Quantification of Sublethally Injured Cells In order to quantify bacterial cell injury, in a first step the maximum non-inhibitory concentration (MNIC) of each osmolyte in M9 agar medium was determined. To achieve this, untreated cells were spread plated onto M9 agar media with different concentrations of each solute (NaCl, KCl, or saccharose), and plates were incubated at 37°C for 48 h. According to previous work (Cebrián et al., 2014), the MNIC was defined as the highest concentration which inhibited less than 20% of the initial untreated bacterial population. After treatments, 0.02 mL volumes of adequately diluted samples (using M9 broth as the dilution medium) were spread on the surface of prepared M9 agar and/or TSA plates, in both non-selective and selective plates. Exceptionally, samples treated with the thermo-resistometer were poured either directly on plates (for treatment temperatures of 58, 61, and 64°C) or were pour-plated after having been collected in agar-medium-containing tubes placed on a rotating carousel (for experiments performed at 67 and 70°C). This sample-collection device allowed for the characterization of survival curves despite the high inactivation rates at these treatment temperatures. In all cases, plates were incubated at 37°C for 48 h. Previous experiments showed that longer incubation times did not influence the amount of surviving cells regardless of the added osmolyte. For each dilution, 10–200 colonies were counted on the surface of the agar medium in spread-plated samples. For pour-plated samples, colonies were counted with an improved Image Analyzer Automatic Counter (Protos; Analytical Measuring Systems, Cambridge, UK) as described in earlier work (Condón et al., 1993). Taking into account the initial cell concentration in the thermoresistance experiments (108 CFU/mL), the detection limit was of 5 log10 cycles. Inactivation was expressed in terms of the extent of reduction in log10 counts (CFU) after any treatment. Survival curves were obtained by plotting the decimal log10 fraction of survivors versus the treatment time for each independent experiment. The extent of sublethal injury was expressed as the difference between the log10 count (CFU) on non-selective medium (M9) and the log10 count on selective media. Likewise, the percentage of injured cells at each treatment time corresponded to the following equation (Busch and Donnelly, 1992): (1) %Injuredcells = 1 − (CFU/mLselectiveCFU/mLnonselective × 100) According to this representation, “2 log10 cycles of injured cells” means a 2-log10 difference in the count on selective and non-selective media, or that 99% of survivors were sublethally injured. Experimental data were obtained from at least three independent experiments performed on different days. Thermotolerance Parameters When appropriate, survival curves were fitted by a model based on a Weibull-like distribution, which was chosen based on their linear and concave upward profiles. For this investigation we used the equation proposed by Mafart et al. (2002) (Eq. 1): (2) Log10NtN0 = −(tδ)p where t is the treatment time (min); Nt and N0 are the population densities (CFU/mL) at time t and time 0, respectively; and δ and ρ are two characteristic parameters of the equation. The δ value is called the time to the first decimal reduction (time necessary to inactivate the first 1 log10 CFU of the microbial population). The ρ value is the shape parameter. Determination of the State of Cells Grown in Agar Media Containing NaCl To determine the state of cells (viable, inhibited, or inactivated) when grown in agar media with different concentrations of NaCl, wild type (WT) or ΔproP untreated or heat-treated cells (10 min at 55°C) were carefully sampled onto plates with M9 agar medium added with 0–10% NaCl. The initial sampled concentration of cells was 5 × 106 CFU/plate, and plates were incubated for 48 h. After that first incubation, a method was developed to recover colonies from colony-lacking plates in a highly reproducible way. For this, 4 g of agar of each plate from the first incubation were carefully extracted, placed in sterile plastic bags with peptone water 0.1%, and homogenized for 20 s at 230 rpm in a stomacher laboratory blender (model 400, Tekmar, Co., Cincinnati, OH, USA). Next, 1 mL-aliquots were spread plated onto non-selective M9 agar plates and incubated once more for 48 h. After that second incubation, the surface of the plates was visually inspected and classified into positive growth (presenting a high enumerable concentration of CFU/plate) or negative growth (with less than 5 CFU/plate). For each degree of NaCl concentration in the first agar medium, the state of intact or heat-treated cells was classified as viable (when colonies were observed after the first incubation at the expected concentration of 4 × 106–5 × 106 CFU/plate), inhibited (colony-lacking plates after the first incubation but with positive growth after the second incubation) or inactivated (colony-lacking plates after the first incubation and with negative growth after the second incubation). Data shown are results from a representative experiment repeated twice with similar results. Measurement of Cell Permeabilization via Propidium Iodide (PI) Uptake For the evaluation of cell permeabilization, PI at a concentration of 0.08 mM (Pagán and Mackey, 2000) was added to the treatment medium prior to the thermal treatment. Alternatively, PI was not added before treatments and was incorporated immediately after each treatment in order to obtain additional information. Cell permeabilization was analyzed by fluorescence microscopy and by flow cytometry. For the analysis under the fluorescence microscope, treatments were applied at 55°C for 0–5 min. For the flow cytometry analysis, the treatment temperature was 53°C in order to achieve longer intervals between samples. After each treatment, samples were immediately placed on ice, subsequently incubated for 15 min at 20°C, centrifuged at 6000 × g for 5 min, and washed three times. For the flow cytometry analysis, samples were also immediately fixated with a preparation of 4% paraformaldehyde in PBS, washed three times and diluted to a concentration of 105 CFU/mL in PBS. The measurement of cell permeabilization with the fluorescence microscope (Nikon, Mod. L-Kc, Nippon Kogaku KK, Japan) was performed by direct counting of non-fluorescent and fluorescent bacteria at 1000× magnification. About 200 bacteria were visible in a field of vision, and bacteria from five fields of vision were counted per sample and replicate. For each sample analyzed by flow cytometry, 10,000 events were counted using a MACSQuant Analyzer (Miltenyi Biotec, Cologne, Germany) flow cytometer. Fluorescence data were collected using the 488 nm excitation laser and the 614–650 nm filter, corresponding to the B2 channel in the MACSQuant Analyzer. The evaluation of PI uptake by each of those two techniques was run in triplicate on separate days. Statistical Analyses and Management of Flow Cytometry Data For kinetics analysis of the data from survival curves, the least-squares criterion of the GraphPad PRISM program (GraphPad Software, San Diego, CA, USA) was used. This program was also used to perform ANOVA and t-test; differences were considered significant if p ≤ 0.05. Data from flow cytometry was analyzed with FCS Express 5 (De Novo Software, Los Angeles, CA, USA). For the measurement of fluorescence intensity, the parameter “area under the curve” was chosen over “pulse height” in order to consider not only the maximum fluorescence of each event, but also the time required to collect data. No gates were created to obtain histograms or statistical data thereof. Before running the actual samples, unstained and stained cells were analyzed in the flow cytometer to establish the adequate threshold levels for the identification of “events” as “cells” and for the sensitivity of the fluorescence signals. Results and Discussion The SMPT allows for the estimation of the occurrence of sublethal injury after each treatment by measuring the difference between the inactivation level achieved in a selective medium and the inactivation level achieved in a non-selective medium (Mackey, 2000). In order to assess the damage in the cytoplasmic membrane, sodium chloride is added at its MNIC, so that only non-damaged cells are able to multiply. In the present study we primarily intended to offer a simple example of the performance of SMPT after thermal treatments on E. coli. Cells were recovered in M9 agar with 1, 2, o 3% NaCl. Concentrations over 3% NaCl (MNIC) in the agar inhibited the growth of untreated cells. The results, depicted in Figure 1, show that after 10 min of treatment less than 0.2 log10 cycles of the initial population failed to grow in non-selective agar medium. However, when recovered in agar medium containing 1, 2, or 3% NaCl, the population of cells unable to grow increased in 0.2, 1.6, or 4.8 log10 cycles respectively. This graph demonstrates that even a very mild thermal treatment can result in an increased sensitivity to NaCl in the agar media in the majority of the initial bacterial population, corresponding to sublethally injured cells. FIGURE 1 Survival curves of Escherichia coli BW25113 (initial concentration: 108 CFU/mL) to a heat treatment at 55°C in M9 broth for different treatment times. E. coli cells were cultured in M9 agar (non-selective agar) (●) or M9 agar supplemented with 1 (■), 2 (▲), or 3% (▼) NaCl. Error bars represent standard deviation of the mean from three replicates. On the other hand, there was a gradual inverse relationship between the osmolality of the recovery medium and the proportion of growing cells. Therefore, the more severely injured cells are, the lower the NaCl concentration required to prevent their growth – which fits perfectly with the previously stated hypothesis of the coexistence of different levels of damage, from minor to eventually lethal (Wesche et al., 2009; Noriega et al., 2013). Insights into the Failure of Sublethally Injured Cells to Grow on Osmotically Selective Media The increased sensitivity of cells to NaCl after thermal treatments does not have a clear origin, although it has been traditionally ascribed to the loss of permeability control, leading to their irreversible inactivation (Mackey, 2000). However, little research has been done to identify mechanisms or structures that are damaged by heat and therefore prevent bacterial growth in the presence of osmotically selective agents. As different factors could be involved, in the present study we decided to investigate the mechanisms underlying SMPT by individually considering (i) the osmoregulatory mechanisms aimed to upregulate the solute pool, (ii) the possible toxicity of the selective agent in the agar media, and (iii) the selective permeability of the cytoplasmic membrane. Role of the Upregulation of the Solute Pool in SMPT The synthesis and accumulation of trehalose, or the influx of other osmoprotectants when present in the media, are the result of a cascade of osmoregulatory events triggered in living bacterial cells by osmotic upshocks and intended to maintain their correct turgor pressure (Wood, 2011). In the present work we explored the osmoregulatory response of WT and mutant cells impaired in trehalose synthesis or in the influx of osmoprotectants, with the objective of determining the involvement of those osmoregulatory mechanisms in SMPT. Upregulation of the solute pool through the accumulation of trehalose High osmolarity stimulates the transcription of Ots system to synthesize trehalose in media devoid of osmoprotectants, and mutants impaired in otsA are osmotically sensitive due to their inability to synthetize trehalose (Lucht and Bremer, 1994). For the present work, we decided to compare the state of untreated or thermally treated WT cells with that of ΔotsA cells when plated onto agar with different NaCl concentrations. Also, the proportions of sublethally injured cells were calculated, for each strain and treatment time, by calculating the difference between the survival level in the presence of its MNIC of NaCl and in the absence of NaCl. Table 1 shows that, as expected, E. coli ΔotsA presented a lower NaCl MNIC value (2%) than the WT; the fact that untreated ΔotsA cells are unable to grow in agar medium with 3% NaCl is probably due to the absence of the osmoprotectant effect of accumulated trehalose. Furthermore, the reduced osmotolerance of ΔotsA cells was also detected in the finding that NaCl concentrations above 8% were capable of inactivating untreated cells (instead of only inhibiting their growth, as observed for the WT cells). Table 1 State of untreated or thermally treated cells after the incubation in M9 agar medium added with each NaCl concentration. % NaCl in M9 agar State of WT untreated cells State of ΔotsA untreated cells State of WT thermally treated cells State of ΔotsA thermally treated cells 0 Viable Viable Viable Viable 1 Viable Viable Viable Viable 2 Viable Viable Viable Inhibited 3 Viable Inhibited Inhibited Inhibited 4 Inhibited Inhibited Inhibited Inhibited 5 Inhibited Inhibited Inhibited Inhibited 6 Inhibited Inhibited Non-viable Non-viable 7 Inhibited Inhibited Non-viable Non-viable 8 Inhibited Inhibited Non-viable Non-viable 9 Inhibited Non viable Non-viable Non-viable 10 Inhibited Non viable Non-viable Non-viable The application of a prior thermal treatment resulted in the inactivation of otherwise inhibited cells when plated with 6–10% NaCl (Table 1). Therefore, we were able to confirm that thermally treated cells of both strains lost their ability to survive in media containing high NaCl concentrations. This observation could be related to the increase in the intracellular accumulation of Na+ in cells when plated onto agar with an external osmolality of 2 Os/kg, corresponding to 6% NaCl (Shabala et al., 2009). When considering the proportion of sublethal injury at their respective MNICs, both strains behaved similarly (2,5 ± 0,5 and more than 5 log cycles of sublethal injury after 5 and 20 min of heat treatment respectively, data not shown). The higher osmosensitivity of the mutant lacking the complete trehalose synthesis pathway in comparison with the WT exposes the relevance of trehalose synthesis in SMPT. This finding also agrees with a previously observed reduction in MNIC values of several osmolytes in E. coli mutants in the Ots-controlling sigma factor RpoS (Cebrián et al., 2015). Regarding the specific role of trehalose synthesis or accumulation in the detection of sublethal injury by SMPT, the similar proportions of sublethal injury detected in both strains seem to suggest that, once cells have been thermally damaged, other mechanisms or cellular structures are responsible for their difficulty to outgrow in selective agar media. Further research should be done on the thermosensitivity of Ots as a key factor in the way trehalose and its synthesis pathway are involved in the inhibition and inactivation of sublethally injured cells. Additionally, an unexpected discovery was made in the results in Table 1. Whereas untreated WT cells were inhibited when grown in the presence of concentrations above the MNIC, thermally treated cells remained inhibited when recovered onto agar medium with 3% NaCl, which corresponds to their MNIC and therefore is commonly used to determine the degree of sublethal injury (García et al., 2005). These results contradict, for the first time, the previously accepted hypothesis that sublethally injured cells are inactivated when plated at the MNIC determined for untreated cells (Mackey, 2000): the explanation is that the cells are not being actually inactivated but inhibited in hyperosmotic agar media. For simplicity, throughout the present study we continue to use the term “inactivation” to describe the lack of growth in the recovery medium. On the other hand, this discovery can turn out to be of great relevance, from an applicative point of view, in helping us correctly interpret the lethality of each treatment in combined preservation processes. This is especially true when low water activity is considered as one of the hurdles: since inhibited cells can resume growth under favorable conditions, the error of considering them as inactivated cells would imply that one would underestimate the bacterial content in food and thereby incur in possible health risks for the consumers. Upregulation of the solute pool through the influx of external osmoprotectants Nutritionally rich agars containing osmoprotectants are commonly used for the detection of sublethal injury in food preservation (Wu, 2008; Wesche et al., 2009). Bacteria take up osmoprotectants from surrounding media via membrane transporters such as ProP or BetT (Haardt et al., 1995; Wood et al., 2001), and their stability could be impaired after thermal treatments and therefore influence the outcome of SMPT. Among osmoprotectants, betaine has been demonstrated to increase growth of E. coli cells in hyperosmotic media (Le Rudulier et al., 1984), so E. coli mutants lacking the betaine transporter ProP (ΔproP) were selected to help determine the role of osmoprotectant transporters in SMPT. The addition of betaine to the recovery agar medium resulted in an increase in the NaCl MNIC value from 3 to 5% in untreated WT cells, as represented in Figure 2A by comparing the black bars (showing inactivation of more than 80% of the initial cell population when plated in the presence of 4% NaCl) with blue bars (showing that inactivation of more than 80% of the initial cell population is only achieved when plated in the presence of 6% NaCl). When WT cells were treated at 55°C for 10 min, a significantly higher proportion of cells were recovered at each % NaCl when recovered in media with betaine, than when betaine was absent (Figure 2B; p < 0.05). In contrast, Figure 2B shows that thermally treated mutants lacking ProP were unable to incorporate betaine: the proportion of growing cells was the same (p > 0.05) regardless of the osmoprotectant. Therefore, ProP was still active after thermal treatment (as indicated by the difference between treated WT and ΔproP cells when recovered in the presence of betaine), while other transporter(s) possibly responsible for the uptake of betaine in untreated ΔproP cells were inactive after the thermal treatment. FIGURE 2 Log10 cycles of survival fractions of untreated (A) or treated for 10 min at 55°C in M9 broth (B) of E. coli BW25113 WT (black, blue bars) or ΔproP (gray, green bars) after incubation in the absence (black, gray bars) or presence (blue, green bars) of betaine 1 mM in the M9 agar of different NaCl concentrations. Error bars represent standard deviation of the mean from three replicates. The observation of the remaining activity of ProP after heat prompted us to attempt to ascertain whether more intense thermal treatments could impair ProP and therefore interfere with the detection of sublethal injury. For this purpose, the thermosensitivity of ProP was analyzed by culturing thermally treated WT cells in agar medium containing 3% NaCl, with and without betaine added. Survival curves were modelized so that the time required to inactivate 1 log10 cycle of the initial cell population could be compared between the two treatments. According to the results (Figure 3), no statistically significant differences were observed between the slopes of the two TDT curves (p > 0.05). This implies that the osmoprotectant effect of betaine was maintained throughout the whole range of assayed temperatures (57–70°C), demonstrating the functionality of ProP in the assayed conditions. As a consequence, the possibility that cells might be unable to incorporate osmoprotectants in order to repair their sublethal injury was generally discarded for thermal treatments at temperatures up to 70°C. FIGURE 3 Log10 times for the decimal reduction of E. coli BW25113 at different treatment temperatures in M9 broth and recovered in M9 agar added with 3% NaCl in the absence (●) or presence (○) of 1 mM betaine. Error bars represent standard deviation of the mean from three replicates. On the other hand, the evident influence of added betaine on the osmoregulatory response of E. coli and in the results obtained with SMPT using M9 agar medium showed the relevance of the composition of the recovery medium in the interpretation of the sublethally injured fraction via SMPT. Moreover, previous results have demonstrated that the presence of betaine in the recovery medium can compensate for defective phenotypes in their osmoregulatory systems (Cebrián et al., 2015). However, the evaluation of the occurrence of sublethal injury incurred in E. coli after inimical treatments via SMPT usually employs complete and nutritionally rich agar media (such as tryptic soy yeast extract agar [TSAYE] or plate count agar) as the non-selective medium (Wuytack et al., 2002; Noriega et al., 2013). In contrast with the controlled and osmoprotectant-free composition of the M9 agar medium, TSAYE contains the osmoprotectant betaine (Dulaney et al., 1968). In our experiments, the presence of betaine in TSAYE was demonstrated by the determination of a MNIC of NaCl at 5% (as in M9 agar medium with betaine), and by the repetition of the treatments applied to obtain Figure 2 but with TSAYE as the recovery medium: both E. coli WT and ΔproP behaved in TSAYE similarly as in M9 agar medium with betaine (p > 0.05; data not shown). In order to control the adequacy of the SMPT using TSAYE as the recovery medium, several thermal treatments of different durations and at different temperatures causing less than 0.5 log10 cycles of inactivation in non-selective M9 agar medium were applied to E. coli WT cells. The number of log10 cycles of inactivation was measured in M9 agar medium, M9 agar medium with betaine and TSAYE, having added their respective MNIC of NaCl (3, 5, and 5%; Figure 4). The good correlations obtained between the measurements in the M9 + 3% NaCl agar medium with those in the M9 + 5% NaCl + betaine agar (R2> 0.95) or with those in the TSAYE + 5% NaCl agar (R2> 0.92) suggest that, despite the presence of osmoprotectants like betaine, the estimation of the amount of sublethal injury remains constant because of the corresponding increase in the MNIC. Therefore, these results lead to the conclusion that the selection of TSAYE, a recovery agar with osmoprotectants, does not lead to an underestimation of the proportion of sublethally injured cells. FIGURE 4 Log10 cycles of survival fractions detected for E. coli BW25113 in: M9 minimal agar with its MNIC of NaCl (3%) (●), M9 minimal agar with betaine 1 mM and its MNIC of NaCl (5%) () or TSAYE with its MNIC of NaCl (5%) (). Thermal treatments of 53–59°C of different durations (0–20 min) were applied in M9 broth to obtain a variety of samples, being each of them depicted in a different column. Figures 3 and 4, when read together, can lead to a further conclusion. It is easily understandable that only those cells which take up betaine are able to outgrow in agar medium with 5% NaCl, since that osmoprotectant prevents them from being inhibited or inactivated at concentrations above 3% NaCl. However, considering that the applied thermal treatments do not affect ProP (Figure 3), the ability to introduce betaine in the cytoplasm is not a limiting factor for heat-treated cells to grow. This would mean that only cells with functional osmoregulatory mechanisms (without considering osmoprotectants) continue to grow after thermal treatments. Given the good correlation between the inactivation detected in treated cells growing in the presence of 5% NaCl with betaine and those growing in the presence of 3% NaCl, it was demonstrated that the latter selective medium is correctly preventing the growth of those cells whose osmoregulation is not completely functional, as previously assumed (Mackey, 2000). Possible Toxicity of Na+ in the Cell Sublethal injury to microbial cell membranes caused by inimical treatments has been linked to the cell’s ability to exclude toxic materials (Gilbert, 1984). Sodium can be considered one of those toxic materials, since E. coli cells have to maintain an intracellular Na+ concentration lower than the extracellular concentration via the active extrusion systems NhaA and NhaB, regulated by NhaR (Padan and Krulwich, 2000). Moreover, Na+ stress is enhanced under conditions in which membrane integrity is compromised, and it has been suggested that in E. coli high osmolarity may lead to the induction of specific Na+ efflux pathways (Padan and Krulwich, 2000). In order to investigate the possible toxic effect of the presence of Na+ in the selective recovery medium, we studied the MNICs of different solutes on untreated cells, as well as their effect on the survival kinetics of thermally treated cells. This way, equivalent osmotic values were achieved in the agar media by incorporation of different concentrations of the ionic osmolytes Na+ or K+ (as NaCl or KCl), or the non-ionic osmolyte saccharose. In order to facilitate the comparison among osmolytes, the level of inactivation achieved after a very mild thermal treatment (which caused no inactivation in non-selective agar) was obtained in the presence of 25, 50, 75, and 100% MNIC of each osmolyte. The MNIC values of NaCl and KCl were obtained at the same osmolality value (1.02 Os/kg), while the MNIC of saccharose was determined at a greater osmolality value (1.70 Os/kg). In this regard, it has been observed that ionic and non-ionic osmotica trigger different osmoregulatory responses (Shabala et al., 2009). However, this distinction did not seem relevant in SMPT, since similar levels of thermal inactivation (p > 0.05) were detected in the presence of the MNIC or lower concentrations of either NaCl, KCl or saccharose (Figure 5). Besides, mutants lacking the Na+ extrusion systems NhaA, NhaB, or NhaR showed the same MNIC of NaCl than E. coli WT cells (3%; data not shown). All these observations suggest that thermal treatments impairing the Na+ efflux systems could be dismissed as one of the factors intervening in the detection of the sublethal injury under the conditions assayed. This hypothesis agrees with previous observations by Cebrián et al. (2014), who concluded that no specific inhibition mechanisms could be attributed to the ionic osmolytes NaCl or KCl other than the same hyperosmotic stress as imposed by saccharose. FIGURE 5 Log10 cycles of survival fractions of E. coli BW25113 in M9 broth after 5 min of thermal treatment at 55°C and after incubation in M9 agar added with NaCl (black bars), KCl (gray bars), or saccharose (white bars) at different proportions of their respective MNICs (3.00% NaCl, 3.88% KCl, or 58.53% saccharose). Error bars represent standard deviation of the mean from three replicates. In our attempt to update SMPT, we also noted that not only the MNIC of NaCl and KCl were obtained at the same osmolality value, but also similar E. coli survival curves after thermal treatments of 5, 10, or 20 min were obtained for each osmolality value (data not shown). Therefore, although NaCl is commonly used as the selective agent at its MNIC in SMPT (García et al., 2006; Miller et al., 2006; Arroyo et al., 2009), its substitution with KCl could be an alternative possibility. Impairment of Cytoplasmic Membrane Integrity After the exploration of specific osmoregulatory mechanisms triggered by osmotic upshocks, research into the reason for the inability of sublethally injured cells to outgrow in selective agar medium in SMPT should further explore the physical integrity of the cytoplasmic membrane. Not only is membrane integrity considered to be a key for the maintenance of osmoregulation (Wood et al., 2001), but the inability of cells to overcome the action of the selective agent is considered to reveal structural damage in the cytoplasmic membrane (Wesche et al., 2009). However, little research has been carried out to prove the relationship between the extent of sublethal injury and the physical integrity of the cytoplasmic membrane. For the study of membrane integrity, measurement of its degree of permeabilization with the membrane-impermeant dye propidium iodide (PI) has been extensively used to quantify cell damage by penetrating membranes with pores larger than 660 Da (Pagán and Mackey, 2000; García et al., 2005; Kennedy et al., 2011). Figure 6 shows the correlation between the percentage of permeabilized cells and the level of inactivation measured when thermally treated cells were recovered in agar media with NaCl or KCl at their MNIC. The percentage of permeabilization corresponds to the fraction of cells in which PI had entered through membrane pores during treatment, while the “level of inactivation” factor refers to the percentage of cells – out of the total initial sample population – which were unable to outgrow in the selective agar media (comprising both dead and sublethally injured cells). The total percentage of inactivation – determined by the proportion of cells unable to outgrow in non-selective agar medium – was under 5% even after the longer treatment times of 3 and 5 min (data not shown). FIGURE 6 Correlation and linear regressions between percentage of permeabilization of E. coli BW25113 cells after PI staining (measured in fluorescence microscope) and the proportion of inactivated cells (measured by plate count in M9 agar containing 3% NaCl). Thermal treatments at 55°C of different durations (0–3 min) were applied in M9 broth to obtain a variety of samples. Error bars represent standard deviation of the mean from three replicates. The good correlation between both factors (no significant differences between their slopes, p > 0.05), confirms the hypothesis that damage is due to the impairment of membrane permeability (Mackey, 2000; Wesche et al., 2009). Figure 6 also demonstrates that, independently from the functionality of the osmoregulatory mechanisms, a direct relationship exists between the extent of sublethal injury detected via SMPT and the physical state of the cytoplasmic membrane. It is noteworthy that this good correlation was obtained when PI had been added before the treatment, corresponding to the creation of pores throughout the whole treatment (Pagán and Mackey, 2000). In contrast, the incorporation of PI immediately after each thermal treatment required more than 5 min of thermal treatment in order to lead to the permeabilization of the majority of cells as evidenced by microscopy, and the resulting staining intensity measured by flow cytometry was much lower at any treatment time than when PI was added beforehand (data not shown). This would agree with previous observations which determined that 20-min treatments at 60°C were unable to permeabilize more than 80% of the E. coli population via staining with post-treatment PI (Shigapova, 2004; Kennedy et al., 2011). Furthermore, these results agree with the view that PI is a sensitive marker of cell damage, but a poor indicator of cell death (Amor et al., 2002). Exploration of the Possible Use of Flow Cytometry as a Complementary Technique to Assess Sublethal Injury Counting the number of PI-positive cells in terms of percentage only allows for the accurate evaluation of ca. 1 log10 cycle of the initial population, as opposed to the 5-log10 scale obtained in the previous results by viable plate count. In an attempt to overcome this major limitation, we decided to use flow cytometry as a more appropriate methodology to assess membrane permeabilization through PI uptake. Not only is flow cytometry highly convenient in view of this goal (Amor et al., 2002; Kennedy et al., 2011), but there is also increasing interest in its possible use to complement or even substitute plate count techniques (Nebe-von-Caron et al., 2000; Aronsson et al., 2005). In view of this objective, flow cytometer data of PI-stained samples were subjected to statistical analysis and complemented with the measurement of sublethal injury via SMPT. Each overlay in the histogram of Figure 7 depicts the frequency distribution of the fluorescence intensity’s logarithmic value for a specific treatment time. The position of the overlays alongside the x-axis shows a clear tendency toward increasing fluorescence with longer treatments. Given this observation and the Gaussian aspect of the histogram overlays, we decided to obtain, for each treatment time, a simple parameter characterizing the average fluorescence intensity. For this purpose, the median value of the total of fluorescence-area values of all events was calculated and divided by the maximum average fluorescence achieved by any sample for that assay (Figure 8). The inactivation levels detected in M9 agar medium containing 3% NaCl for different treatment times were also plotted (Figure 8). FIGURE 7 Fluorescence histogram overlays obtained by flow cytometry measurement of PI-stained E. coli BW25113 cells of untreated cells and after thermal treatments at 53°C. Each overlay represents the frequency distribution of the fluorescence intensity (fluorescence-area values) of a total of 10000 events per sample, corresponding to the untreated sample (lighter overlay), or samples treated for 2, 4, 6, 10, 16, or 20 (darker overlay) min. Results shown from a representative experiment repeated three times with similar results. FIGURE 8 Median values (●) of the fluorescence-area values obtained from single cell-flow cytometry from PI-stained cells and their linear regression lines, plotted against each duration of the thermal treatment at 53°C. The graph also shows the log10 cycles of inactivation (-log10 cycles of the survival fractions) as measured in M9 agar with 3% NaCl (red bars, corresponding to right Y-ax). Error bars represent standard deviation of the mean from three replicates. Firstly, we wanted to distinguish between the proportion of stained cells and the total fluorescence value for each sample. In order to calculate the proportion of stained cells, in the data grid obtained through flow cytometry we selected and counted only those events that surpassed the fluorescence threshold. Results showed that the proportion of stained cells increased in parallel with treatment duration for the first 5 min (data not shown). At this point, about 90% of the cells were already stained independently of flourescence intensity (data not shown). Therefore, we confirmed the previously observed correlation (assessed by optical microscopy) between the proportion of permeabilized cells and the proportion of cells unable to grow on the selective medium, since after 4 min of treatment around 85% of cells (nearly 1 log10 cycle) had been unable to outgrow in the selective agar medium. In contrast with the proportion of fluorescent cells, the median value of fluorescence intensities refers to the total fluorescence emitted by the entire bacterial population of each sample, i.e., it assesses the number of PI-stained cells and the fluorescence emitted by each of them. At this point, it should be noted that different staining intensities often occur (Shi et al., 2007), as we have observed from previous experiments using the fluorescence microscope. As can be seen in Figure 8, the median values followed a linear evolution throughout treatment time until reaching a maximum value when treatments were 20 min or longer. Therefore, although nearly all the cells had been stained after 5 min of treatment, they were mostly weakly stained, and the fluorescence intensity of the whole population went on increasing throughout treatment at a constant rate. The linearity in the average fluorescence intensity of different samples is a promising concept that has been barely approached. In this regard, Berney et al. (2007) correlated the geometric mean of fluorescence intensity with the amount of nucleic acids, but research could be expanded to different fluorescent probes in order to reveal different grades of a high variety of metabolic processes. A new methodology for the determination of the occurrence of sublethal injury at a broad detection range (at least 5 log10 cycles, depending on sample size) could be developed following the results depicted in Figure 8. The meticulous determination of cell plate counts and fluorescence measurements after inimical treatments, as well as calculations of the correlation between both factors, should be performed in order to establish a reference data matrix for further studies. In addition, simultaneous staining with other fluorochromes could provide a better description of the composition of each bacterial sample (Nebe-von-Caron et al., 2000), and therefore help us understand the evolution of treated cells from the viable to dead conditions. From a practical point of view, the rapid detection of the extent of sublethal injury via flow cytometry (and not only the extent of inactivation, as commonly performed) could significantly help in the design of food preservation processes by determining which treatment conditions could be more favorable in the synergistic combination of different hurdles. Conclusion on the Evidence of Sublethal Injury through SMPT According to the results, in SMPT only cells with intact osmoregulatory properties can overcome the osmotic pressure in the selective agar medium (Figures 3 and 4). In contrast, those which are considered sublethally injured remain inhibited at the MNIC of the selective agent (Table 1). Therefore, cells whose osmoregulatory mechanisms or physical structures become non-functional after thermal treatments are unable to outgrow in osmotically challenging agar media, although they can outgrow in the absence of the selective agent. The identification of such mechanisms or structures, as well as their relationship with the extent of sublethal injury detected, are a key in understanding the performance of SMPT. In the present study, two of the hypothesized osmoregulatory mechanisms have been discarded as key factors in the performance of SMPT in detecting sublethal injury after heat in E. coli: the exclusion of Na+ from the cytoplasm and the uptake of osmoprotectants from the agar media. The toxicity of Na+ as a cause of sublethal injury had been previously proposed (Gilbert, 1984; Padan and Krulwich, 2000), but we have found evidence neither of Na+ toxicity, nor of thermal treatments affecting the Na+ extrusion systems. Regarding osmoprotectants, their uptake is absolutely necessary for cells to outgrow in rich media added with NaCl at its MNIC (Figure 4). Since the transporter ProP remained active after intense thermal treatments (Figure 3), the inability to uptake betaine should not be hypothesized as the reason for the non-survival of sublethally injured cells in selective agar media. In the absence of osmoprotectants, the main osmoregulatory mechanisms accumulate trehalose. Its absence leads to an increased osmosensitivity and thermosensitivity in untreated and treated cells, and impairs the correct quantification of sublethal injury via SMPT (Table 1). However, no direct relationship between the impairment of trehalose synthesis and accumulation systems and the extent of sublethal injury could be established. In contrast, we found a direct relationship between the structural damage of the cell membrane and SMPT via the PI-exclusion technique when PI was added before thermal treatments. In this way, the extent of sublethal injury detected via SMPT could be ascribed to the physical loss of integrity of the cell membrane independently of specific functional osmoregulatory processes. The detection of sublethal injury of E. coli after thermal stress has been previously ascribed to the physical loss of integrity of the cell membrane (Mackey, 2000; Ukuku et al., 2008; Wesche et al., 2009). However, to the best of our knowledge, this is the first time that a direct correlation between both factors has been demonstrated, especially at such a high proportion of the bacterial population. Furthermore, some of the results of the present study can result in the improvement of SMPT. For instance, variations in the composition of the selective media without affecting the outcome of the technique are being proposed: M9 agar + 3% NaCl, M9 agar + 3.88% KCl, M9 agar + betaine + 5% NaCl, and TSAYE + 5% NaCl yielded similar levels of sublethal injury. On the other hand, the possibility of complementing SMPT with flow cytometry to detect bacterial inactivation and injury at a detection range of 5 log10 cycles is presented here, since the extent of cell permeabilization (measured simply and rapidly thanks to flow cytometry) was found to be an indicator of the extent of sublethal injury detected with SMPT. Conclusion This work demonstrates, for the first time, that the incorporation in the recovery agar of selective agents which increase its osmotic pressure (such as sodium chloride or potassium chloride) inhibits the growth of E. coli cells whose envelopes are physically impaired by mild thermal treatments. Previous hypotheses regarding the implication of two different factors on the performance of SMPT (the toxicity of Na+ in the agar and the destruction of transporters of osmoprotectans) were discarded. Moreover, the extent of this physical damage was found to be correlated with the proportion of treated cells unable to grow in selective agar, confirming the adequacy of the SMPT to assess thermal sublethal injury. Further investigation aimed to improve the performance of the SMPT or its combination with flow cytometry could help to maximize its usefulness in food preservation. Author Contributions Conceived and designed the experiments: LE, DG-G, and RP. Performed the experiments: LE. Analyzed the data: LE, DG-G, and RP. Wrote the paper: LE, DG-G, and RP. Funding. This work was supported by the Spanish Ministerio de Economía y Competitividad (CICYT Projects AGL2012–32165 and AGL2015–69565–P). Spanish Ministerio de Educación, Cultura y Deporte that provided LE with a grant to carry out this investigation. The authors would like to thank Dr. Santiago Condón for his assistance during the research. The authors also wish to thank Stanley Hanks (translator) for having revised and proofread the final version of this manuscript. This manuscript is dedicated to the loving memory of Bernard Mackey. ==== Refs References Amor K. B. Breeuwer P. Verbaarschot P. Rombouts F. M. Akkermans A. D. De Vos W. M. (2002 ). Multiparametric flow cytometry and cell sorting for the assessment of viable, injured, and dead Bifidobacterium cells during bile salt stress. Appl. Environ. Microbiol. 68 5209 –5216 . 10.1128/AEM.68.11.5209-5216.2002 12406706 Aronsson K. Rönner U. Borch E. (2005 ). Inactivation of Escherichia coli, Listeria innocua and Saccharomyces cerevisiae in relation to membrane permeabilization and subsequent leakage of intracellular compounds due to pulsed electric field processing. Int. J. Food Microbiol. 99 19 –32 .15718026 Arroyo C. Condón S. Pagán R. (2009 ). Thermobacteriological characterization of Enterobacter sakazakii. Int. J. Food Microbiol. 136 110 –118 . 10.1016/j.ijfoodmicro.2009.09.013 19811846 Baba T. Ara T. Hasegawa M. Takai Y. Okumura Y. Baba M. (2006 ). Construction of Escherichia coli K–12 in-frame, single-gene knockout mutants: the Keio collection. Mol. Syst. Biol. 2 :2006.0008 10.1038/msb4100050 Back K.-H. Kim S.-O. Park K.-H. Chung M.-S. Kang D.-H. (2012 ). Spray method for recovery of heat-Injured Salmonella Typhimurium and Listeria monocytogenes. J. Food Prot. 75 1867 –1872 . 10.4315/0362-028X.JFP-11-512 23043840 Berney M. Hammes F. Bosshard F. Weilenmann H.-U. Egli T. (2007 ). Assessment and interpretation of bacterial viability by usng the LIVE/DEAD BacLight Kit in combination with flow cytometry. Appl. Environ. Microbiol. 73 3283 –3290 . 10.1128/AEM.02750-06 17384309 Bogosian G. Bourneuf E. V. (2001 ). A matter of bacterial life and death. EMBO Rep. 2 770 –774 . 10.1093/embo-reports/kve182 11559589 Busch S. V. Donnelly C. W. (1992 ). Development of a repair-enrichment broth for resuscitation of heat-injured Listeria monocytogenes and Listeria innocua. Appl. Environ. Microbiol. 58 14 –20 .1531746 Cebrián G. Arroyo C. Condón S. Mañas P. (2015 ). Osmotolerance provided by the alternative sigma factors σB and rpoS to Staphylococcus aureus and Escherichia coli is solute dependent and does not result in an increased growth fitness in NaCl containing media. Int. J. Food Microbiol. 214 83 –90 . 10.1016/j.ijfoodmicro.2015.07.011 26256716 Cebrián G. Arroyo C. Mañas P. Condón S. (2014 ). Bacterial maximum non-inhibitory and minimum inhibitory concentrations of different water activity depressing solutes. Int. J. Food Microbiol. 188 67 –74 . 10.1016/j.ijfoodmicro.2014.07.011 25090605 Chilton P. Isaacs N. S. Mañas P. Mackey B. M. (2001 ). Biosynthetic requirements for the repair of membrane damage in pressure-treated Escherichia coli. Int. J. Food Microbiol. 71 101 –104 . 10.1016/S0168-1605(01)00566-9 11764887 Condón S. Arrizubieta M. J. Sala F. J. (1993 ). Microbial heat-resistance determinations by the multipoint system with the Thermoresistometer TR-SC – Improvement of this methodology. J. Microbiol. Methods 18 357 –366 . 10.1016/0167-7012(93)90017-C Dulaney E. Dulaney D. Rickes E. (1968 ). Factors in yeast extract which relieve growth inhibition of bacteria in defined medium of high osmolarity. Dev. Ind. Microbiol. 9 260 –269 . García D. Gómez N. Mañas P. Condón S. Raso J. Pagán R. (2005 ). Occurrence of sublethal injury after pulsed electric fields depending on the micro-organism, the treatment medium ph and the intensity of the treatment investigated. J. Appl. Microbiol. 99 94 –104 . 10.1111/j.1365-2672.2005.02611.x 15960669 García D. Mañas P. Gómez N. Raso J. Pagán R. (2006 ). Biosynthetic requirements for the repair of sublethal membrane damage in Escherichia coli cells after pulsed electric fields. J. Appl. Microbiol. 100 428 –435 . 10.1111/j.1365-2672.2005.02795.x 16478482 Gelaw T. K. Espina L. Pagán R. García-Gonzalo D. De Lamo-Castellví S. (2014 ). Prediction of injured and dead inactivated Escherichia coli O157: H7 cells after heat and pulsed electric field treatment with attenuated total reflectance infrared microspectroscopy combined with multivariate analysis technique. Food Bioproc. Tech. 7 2084 –2092 . Gilbert P. (1984 ). The revival of micro-organisms sublethally injured by chemical inhibitors. Soc. Appl. Bacteriol. Symp. Ser. 12 175 –197 .6208618 Haardt M. Kempf B. Faatz E. Bremer E. (1995 ). The osmoprotectant proline betaine is a major substrate for the binding-protein-dependent transport system ProU of Escherichia coli K-12. Mol. Gen. Genet. 246 783 –796 . 10.1007/BF00290728 7898450 Kennedy D. Cronin U. P. Wilkinson M. G. (2011 ). Responses of Escherichia coli, Listeria monocytogenes, and Staphylococcus aureus to simulated food processing treatments, determined using fluorescence-activated cell sorting and plate counting. Appl. Environ. Microbiol. 77 4657 –4668 .21602370 Le Rudulier D. Strom A. Dandekar A. Smith L. Valentine R. (1984 ). Molecular biology of osmoregulation. Science 224 1064 –1068 . 10.1126/science.224.4653.1064 16827211 Leistner L. Gorris L. G. (1995 ). Food preservation by hurdle technology. Trends Food Sci. Tech. 6 41 –46 . 10.1016/S0924-2244(00)88941-4 Lucht J. M. Bremer E. (1994 ). Adaptation of Escherichia coli to high osmolarity environments: osmoregulation of the high-affinity glycine betaine transport system ProU. FEMS Microbiol. Rev. 14 3 –20 . 10.1111/j.1574-6976.1994.tb00067.x 8011357 Mackey B. M. (2000 ). “Injured bacteria,” in The Microbiological Safety and Quality of Food eds Lund B. M. Baird-Parker T. C. Gould G. W. (Gaithersburg, MD : Aspen Publisher, Inc. ). Mafart P. Couvert O. Gaillard S. Leguérinel I. (2002 ). On calculating sterility in thermal preservation methods: application of the Weibull frequency distribution model. Int. J. Food Microbiol. 72 107 –113 . 10.1016/S0168-1605(01)00624-9 11843401 Maniatis T. Fritsch E. F. Sambrook J. (1982 ). Molecular Cloning: A Laboratory Manual. New York, NY : Cold Spring Harbor Laboratory . McLaggan D. Jones M. A. Gouesbet G. Levina N. Lindey S. Epstein W. (2002 ). Analysis of the kefA2 mutation suggests that KefA is a cation-specific channel involved in osmotic adaptation in Escherichia coli. Mol. Microbiol. 43 521 –536 . 10.1046/j.1365-2958.2002.02764.x 11985727 Miller F. A. Brandão T. R. S. Teixeira P. Silva C. L. M. (2006 ). Recovery of heat-injured Listeria innocua. Int. J. Food Microbiol. 112 261 –265 . 10.1016/j.ijfoodmicro.2006.04.013 16784792 Nebe-von-Caron G. Stephens P. Hewitt C. Powell J. Badley R. (2000 ). Analysis of bacterial function by multi-colour fluorescence flow cytometry and single cell sorting. J. Microbiol. Methods. 42 97 –114 . 10.1016/S0167-7012(00)00181-0 11000436 Neidhardt F. C. Bloch P. L. Smith D. F. (1974 ). Culture medium for enterobacteria. J. Bacteriol. 119 736 –747 .4604283 Noriega E. Velliou E. Van Derlinden E. Mertens L. Van Impe J. F. M. (2013 ). Effect of cell immobilization on heat-induced sublethal injury of Escherichia coli, Salmonella Typhimurium and Listeria innocua. Food Microbiol. 36 355 –364 . 10.1016/j.fm.2013.06.015 24010617 Nyström T. (2001 ). Not quite dead enough: on bacterial life, culturability, senescence, and death. Arch. Microbiol. 176 159 –164 . 10.1007/s002030100314 11511862 Padan E. Krulwich T. A. (2000 ). “Sodium stress,” in Bacterial Stress Responses eds Storz G. Hengge-Aronis R. (Washington, DC : ASM Press ) 117 –130 . Pagán R. Mackey B. (2000 ). Relationship between membrane damage and cell death in pressure-treated Escherichia coli cells: differences between exponential–and stationary–phase cells and variation among strains. Appl. Environ. Microbiol. 66 2829 –2834 . 10.1128/AEM.66.7.2829-2834.2000 10877775 Restaino L. Frampton E. W. Spitz H. (2001 ). Repair and growth of heat-and freeze-injured Escherichia coli O157: H7 in selective enrichment broths. Food Microbiol. 18 617 –629 . 10.1006/fmic.2001.0427 Shabala L. Bowman J. Brown J. Ross T. Mcmeekin T. Shabala S. (2009 ). Ion transport and osmotic adjustment in Escherichia coli in response to ionic and non-ionic osmotica. Environ. Microbiol. 11 137 –148 . 10.1111/j.1462-2920.2008.01748.x 18793315 Shi L. Günther S. Hübschmann T. Wick L. Y. Harms H. Müller S. (2007 ). Limits of propidium iodide as a cell viability indicator for environmental bacteria. Cytometry. 71 592 –598 . 10.1002/cyto.a.20402 17421025 Shigapova N. (2004 ). Alternations of Membrane Physical State Regulate the E. coli Heat Shock Response. Ph.D. thesis, Biological Research Centre Szeged . Ukuku D. O. Jin T. Zhang H. (2008 ). Membrane damage and viability loss of Escherichia coli K-12 and Salmonella enteritidis in liquid egg by thermal death time disk treatment. J. Food Prot. 71 1988 –1995 .18939742 Ulmer H. M. Heinz V. Gänzle M. G. Knorr D. Vogel R. F. (2002 ). Effects of pulsed electric fields on inactivation and metabolic activity of Lactobacillus plantarum in model beer. J. Appl. Microbiol. 93 326 –335 . 10.1046/j.1365-2672.2002.01699.x 12147082 Wesche A. M. Gurtler J. B. Marks B. P. Ryser E. T. (2009 ). Stress, sublethal injury, resuscitation, and virulence of bacterial foodborne pathogens. J. Food Protect. 72 1121 –1138 . Wood J. M. (1999 ). Osmosensing by bacteria: signals and membrane-based sensors. Microbiol. Mol. Biol. 63 230 –262 . Wood J. M. (2011 ). Bacterial osmoregulation: a paradigm for the study of cellular homeostasis. Annu. Rev. Microbiol. 65 215 –238 . 10.1146/annurev-micro-090110-102815 21663439 Wood J. M. Bremer E. Csonka L. N. Kraemer R. Poolman B. Van Der Heide T. (2001 ). Osmosensing and osmoregulatory compatible solute accumulation by bacteria. Comp. Biochem. Physiol. A Mol. Integr. Physiol. 130 437 –460 . 10.1016/S1095-6433(01)00442-1 11913457 Wu V. C. H. (2008 ). A review of microbial injury and recovery methods in food. Food Microbiol. 25 735 –744 . 10.1016/j.fm.2008.04.011 18620965 Wuytack E. Y. Diels A. M. J. Michiels C. W. (2002 ). Bacterial inactivation by high-pressure homogenisation and high hydrostatic pressure. Int. J. Food Microbiol. 77 205 –212 . 10.1016/S0168-1605(02)00054-5 12160080

PMC005xxxxxx/PMC5003819.txt

==== Front Front Plant SciFront Plant SciFront. Plant Sci.Frontiers in Plant Science1664-462XFrontiers Media S.A. 10.3389/fpls.2016.01282Plant ScienceHypothesis and TheoryPotential Global-Local Inconsistency in Species-Area Relationships Fitting Pan Xubin 1*Zhang Xiuling 2Wang Feng 3*Zhu Shuifang 1*1Institute of Plant Quarantine, Chinese Academy of Inspection and QuarantineBeijing, China2School of Mathematics, Tsinghua UniversityBeijing, China3Institute of Desertification Studies, Chinese Academy of ForestryBeijing, ChinaEdited by: Boris Rewald, University of Natural Resources and Life Sciences, Vienna, Austria Reviewed by: Martin Karl-Friedrich Bader, Scion, New Zealand; Zhenzhu Xu, Institute of Botany, China *Correspondence: Xubin Pan xubin.hu.pan@gmail.comFeng Wang wangfeng@caf.ac.cnShuifang Zhu zhusf@caiq.gov.cnThis article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science 30 8 2016 2016 7 128208 5 2016 11 8 2016 Copyright © 2016 Pan, Zhang, Wang and Zhu.2016Pan, Zhang, Wang and ZhuThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.The Species-Area Relationship (SAR) has been widely employed to assess species diversity and predict species extinction. Thus far, although many functions were proposed to fit SAR based on field observations or simulation results, the shape of SAR curve has been debated extensively over decades. Here we uncover a potential global-local inconsistency in SARs fitting simulation blocked by the limitation of large area sampling before. The results indicated that power and logarithm SAR formulas were good for the fitting if the sampling area range is not large which is also the practical sampling interval in the field. However, for the logarithm SAR fitting, a sigmoid curve occurred in the log10 Area−Number of Species plane, and for the power SAR fitting, the curve is convex instead of a straight line as assumed when linear regression was applied. In conclusion, neither the power SAR nor the logarithm SAR fitted to simulated data is linear at large sampling range as commonly assumed in previous studies, no matter the distribution of species abundance is log-normal or negative-binomial, which unmasks the global-local inconsistency in SARs fitting. Thus, misestimates of total number of species or other derivation parameters can occur if the fitted relationship is extrapolated beyond the range of the small and intermediate sampling size. log-normal distributionnegative-binomial distributionpower SARlogarithm SARspecies-abundance distributionextrapolation ==== Body Introduction Species-Area Relationship (SAR) is one of the most studied patterns in ecology, and has been widely employed to assess species diversity and predict species extinction (Tjørve and Tjørve, 2008). Thus far, although many functions were proposed for fitting SAR based on field observations or simulation results, the shape of SAR curve has been debated extensively over decades (Tjørve, 2003; Tjørve et al., 2008). Among various functions of SAR, two are best known and most commonly applied, the power format proposed by Arrhenius (1921), SA=cAz, where SA is the number of species in area A, and c and z are fitted constants (Arrhenius, 1921), and the logarithm format proposed by Gleason (1922), SA=a+b*lnA, where a and b are fitted constants (Gleason, 1922). Compared to logarithm SAR, power SAR has parameters corresponding to ecological meanings (Tjørve, 2003), where c is the number of species per area (analogous to α diversity) and z is the self-similarity index (analogous to β diversity) (Tjørve and Tjørve, 2008). The power SAR was even proposed as a universal model (Dengler, 2008). The application of the power SAR, however, is still in controversy due to potential risks in the process of sampling and parameters estimating, which often leads to underestimate or overestimate of species diversity and extinction rate (Collins et al., 2002; He and Hubbell, 2011; Pan, 2013, 2015). One reason is that an important global factor of the identification of total area and corresponding total number of species has been overlooked for years (Pan, 2013, 2015). Moreover, the shape of SAR curve can be affected by species-abundance distribution (SAD), and several studies attempted to address potential links between the two (He and Legendre, 2002; Green and Ostling, 2003; Tjørve and Tjørve, 2008). For example, for the community with species distributed randomly and independently, SAR can be calculated from SAD (the formula is shown in Methods) (Coleman, 1981). Obviously, the way of sampling is crucial for bridging the SAR and SAD, and accurate fitting is possible only if complete and detail sampling is carried out in accordance with statistic requirement. However, since detail sampling at a large scale is not practical, the fitting (i.e., parameterization) of SAR is usually based on the sampling at a small scale. However, high goodness of fit at the local range does not necessarily expect the same goodness of fit at the global range, partly because local sampling is more likely to misestimate or overlook the existence of rare species (Preston, 1948; Verberk, 2011). Compared to field sampling subject to incomplete surveying, computer simulation sampling can provide a more feasible approach to fitting SAR (Tjørve and Tjørve, 2008). Moreover, computer simulation enables us to scrutinize the patterns of SAR at any level, and thereby can help explore whether the inconsistency of SAR may occur between global and local levels. As abovementioned, the range of sampling is crucial for fitting SAR, and therefore this study will try to reveal potential misguidance and risks of extrapolation. In this study, we tested whether the patterns of the two SARs were consistent at the global and local levels through numerical analysis. The power and logarithm SARs were used to simulate data from two types of species-abundant distributions (negative-binomial (NB) and log-normal (LN) distributions) at the global level. We also evaluated parameter variation and potential misguidance of extrapolation. Methods Data simulation A simulation program in the R platform (R version 3.2.0, R Core Team, 2015) was used to generate sampling data. The total area was set as 1,000,000 points, and each individual of every species occupied one point. The occurrence of plant species was simulated following two distribution patterns, negative-binomial (NB) and log-normal (LN) distributions (selected from dozens of SADs, McGill et al., 2007). Individuals of 100 and 500 species were generated randomly at initial status of simulation species distribution. Data transformation As former studies proposed (Coleman, 1981), for a community where resident species is distributed randomly and independently, the SAR curve can be formulated as R Core Team (2015) (1) SA=STA-∑i=1STA(1-ATA)Ni=∑i=1STA(1-(1-ATA)Ni)  where STA is the total number of species in the total area (TA), and Ni is the number of individuals of per species i. This formula was used to calculate SAR based on simulated data. Two functions, which are SA=a+b*lnA (i.e., the logarithm SAR) and logSA = logc + zlogA (i.e., the logarithm format of the power SAR) are used to fit SARs based on simulated data. Thus, the area is log-transformed in both fittings. The number of species was not transformed in the logarithm SAR fitting but log-transformed in the power SAR fitting. Results and discussion The simulated SADs are shown in Figure 1. The range of number of individuals of each species of negative-binomial distribution is smaller than that of log-normal distribution, which means the latter has more rare species compared to the former. In addition, the average number of individuals of each species for 100 species is more (five times) than that for 500 species, meaning that the latter SAD has more rare species compared to the former SAD. Figure 1 (A) Is the diagram for Species Rank (ranked from most abundant to least abundant)−Abundance (number of individuals). (B) Is the schematic diagram for log10 (Area)−log10 (Number of Species). Different SAD (negative-binomial, NB; log-normal, LN) and number of total species (100 and 500). For the sampling data, the (log-transformed) area was plotted against the number of species (log-transformed or not) in Figures 2, 3. For the log10 (Area)−Number of Species, the curves showed downward trend (concave) when the sampling area was small, while the curves shifted to upward trend (convex) when the sampling area reached an inflection point (Figures 2A,B). And it is faster for 100 species to reach the total number of species than that of 500 species. This situation is the same as the NB compared to the LN. For the log10 (Area)−log10 (Number of Species), the curves showed an upward trend (convex) (Figures 3A,B). Similarly, it is faster for 100 species reach the total number of species than that of 500 species. This situation is the same as the NB compared to the LN. Moreover, the shape of the curves was not largely different for both NB and LN distributions and the total number of species (100 and 500), while the detailed shape of the curves was affected. In summary, the shape of curves was steeper for the NB distribution than for the LN distribution, and the shape of curves was steeper for 100 species than for 500 species. Figure 2 (A) Is the diagram for log10 (Area)—Number of Species for the number of total species equals 100. (B) Is the diagram for log10 (Area)—Number of Species for the number of total species equals 100. Different SAD (NB, negative-binomial; LN, log-normal). Figure 3 (A) Is the diagram for log10 (Area)—log10 (Number of Species) for the number of total species equals 100. (B) Is the diagram for log10 (Area)—log10 (Number of Species) for the number of total species equals 100. Different SAD (NB, negative-binomial; LN, log-normal). Back to the SAR calculated from SAD, the sampling size, the number of individuals per species and their correspondence are important, besides the evenness of SAD. In Figure 4, it showed the 1-(1-A/TA)Ni = 0.01, 0.05, 0.95, and 0.99 for the number of individual of a single species. If the sampling area in the space between the 0.01–0.99 lines cyan and blue (or 0.05–0.95 lines pink and green, Section 2 in Figure 4), the number of individuals per species will play a numerical function in the SAR function. These situations also included the transition from the area (Section 1 in Figure 4) below the line cyan (or pink) to the area (Section 2 in Figure 4) above the line cyan (or pink), and the transition from the area (Section 2 in Figure 4) below the line blue (or green) to the area (Section 3 in Figure 4) above the line blue (or green). Obviously, different number of individuals will lead to different additional function in different sampling areas (Tjørve and Tjørve, 2008). If the sampling area is small, only common species have influence on the SAR curve, and rare species are rarely present in the samples due to their low abundance (Preston, 1948; Verberk, 2011); if the sampling area is intermediate, only rare species have influence on the SAR curve, because the value calculated from common species (almost) equals 1; if the sampling area is large, no species has influence on the SAR curve because all species (almost) equals 1. Obviously the parameters of curves (Figures 2, 3) in this study are affected by the SAD of simulated data (Figure 1) and the total number of species. Figure 4 The diagram for the A/TA−Number of Individuals (Ni). 1-(A/TA)Ni = 0.01, 0.05, 0.95 and 0.99. As showed in a generalized schematic diagram of the SARs (Figures 5A,B), species abundance distributions largely affect the shape of SAR curves, which is in accordance with the findings in previous studies (Allen and White, 2003; Green and Ostling, 2003; Šizling and Storch, 2004; Dengler, 2008; Tjørve and Tjørve, 2008; Tjørve et al., 2008; Mokany et al., 2013; Rybicki and Hanski, 2013; Guo, 2015; Harte and Kitzes, 2015). The curve of the logarithm SAR sampling [i.e., log10 (Area)–Number of Species] showed a sigmoid shape that can be divided into two sections, concave section when the sampling area is small until the inflection point, and convex section. However, the curve of the power SAR sampling (i.e., log10 (Area)–log10 (Number of Species)) only has convex section. As shown in Figure 5, the power and logarithm SAR relationship can be linearly well-fitted if the sampling size is not large. And the total number of species is the determinant factor on how height of the plateau will be. The classical SARs were usually fitted to field observations when the sampling size is small or intermediate. It is, however, not practical to scrutinize all the species with accurate numbers in a large area (Pan and Zhu, 2015), and that is the way of extrapolation often used in the literature. Figure 5 (A) Is the schematic diagram for log10 (Area)−Number of Species. (B) Is the schematic diagram for log10 (Area)−log10 (Number of Species). In the convex section of a curve, the slope decreases as the area increases, therefore it can lead to overestimate of parameters if one assumes the slope is constant (Figure 6A). For example, the SAR fitting in the small sampling area (a1 and a2) and in the large sampling area (a3 and a4) causes an overestimate of the left intercept (LI) and the right intercept (RI), respectively. Meanwhile, the SAR fitting in the small sampling area has lower LI and higher RI than that in the large sampling area. Overestimate of the number of species would be even higher at the end of a curve. However, in the concave section of an SAR curve, underestimate would occur for the left and right intercepts (Figure 5A), respectively. Therefore, the linear extrapolation of the SAR fitting would be problematic, since the range of the sampling area greatly affects the linearity of the SAR. If the sampling range covers both the concave and convex sections [e.g., in the log10 (Area)−Number of Species], misestimate can also occur and would be a little complicated, with one possibility that the left intercept would be underestimated while the right intercept would be overestimated. Thus, the total number of species, derived from the extrapolation from power and logarithm SARs, is not accurate, although this is a very global important parameter for other parameter estimates such as extinction rate (Pan, 2013). Figure 6 (A) Is the schematic diagram of one Species-Area Relationship for log10 (Area)−Number of Species or log10 (Number of Species). (B) Is the schematic diagram of two Species—Area Relationships for log10 Area−Number of Species or log10 (Number of Species). In the convex section of the two SAR curves (Figures 5A,B), the fittings can also be different. Moreover, the estimated parameters (i.e., the slope and intercept) would vary if the sampling areas vary, and even a curve will not exist when the sampling area reached a certain value (Figure 6B). A pattern similar to that in Figure 6B was found in the log-log SAR of Highlands Hammock State Park, Florida, thus there is not necessarily proportionately fewer species loss at broader spatial scales (Powell et al., 2013). This implied that the linear fitting and the comparison of two or more power or logarithm SARs is less problematic only when the sampling area is within the appropriate range. Considering the impact of incomplete surveying and Preston and Pan's effect on the SAD, this SAR comparison will not make any ecological meaning without mathematical endorsement. In conclusion, neither the power SAR nor the logarithm SAR fitted to simulated data is linear at large sampling range as commonly assumed in previous studies, no matter the distribution of species abundance is log-normal or negative-binomial. Therefore, misestimates can occur if the fitted relationship is extrapolated beyond the range of the small and intermediate sampling sizes. However, if we know the full spatial distribution of all species, we can calculate the SAR curve from SAD, and the sampling and fitting is not useful anymore. Here the dilemma of SAR fitting emerges: you will get the SAR but make mistakes using the sampling and fitting if you do the extrapolation; you can avoid the mistakes using more information, but you do not need sampling and fitting anymore. Obviously, the SAR should be used with caution, as the extrapolation or prediction should not be made if one does not know the whole picture, because the global-local inconsistency exists in SAR (Elith and Leathwick, 2009). In the future, detailed sampling of SAD with full spatial information is the direction, instead of counting the number of species in the area, which also has the Preston and Pan effects in the practice (Pan and Zhu, 2015). For different types of SAR, fitting functions and SAD, such as the island SAR with areas of varying size, whether the linear regression displays global-local consistency deserves more research (Scheiner, 2003). In addition, the global-local consistency and inconsistency should be given more concerns in ecology. In this study, the community with species distributed randomly and independently is a simplified case, which still has this inconsistency. For the complex ecosystem with inaccurate sampling, spatial-temporal heterogeneity and scale effect, this inconsistency may be more obvious or more unpredictable. For example, the effect of global climate change on different places is different, while one will be hotter, and the other will be drier. In this situation, how to sample to infer the whole picture from limited samples is a challenge for us. A potential method is mass complete survey of one area conducted by an integrated research group/program, rather than three or nine repeated samples per sites. Author contributions XP, FW, and SZ designed the research. XP and XZ carried out the model simulation. XP did the data analysis. XP and FW drafted the manuscript. XP and SZ revised the manuscript. Conflict of interest statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The work is supported by the National Basic Research Program of China (973 program, 2013CB429905-04), Beijing Nova Program (Z1511000003150107) and National Key Projects for Research and Development (2016YFC1200802). ==== Refs References Allen A. P. White E. P. (2003 ). Effects of range size on species-area relationships . Evol. Ecol. Res. 5 , 493 –499 . Available online at: http://www.evolutionary-ecology.com/abstracts/v05/1544.html Arrhenius O. (1921 ). Species and area . J. Ecol. 9 , 95 –99 . 10.2307/2255763 Coleman B. D. (1981 ). On random placement and species-area relations . Math. Biosci. 54 , 191 –215 . 10.1016/0025-5564(81)90086-9 Collins M. D. Vázquez D. P. Sanders N. J. (2002 ). Species-area curves, homogenization and the loss of global diversity . Evol. Ecol. Res. 4 , 457 –464 . Available online at: http://www.evolutionary-ecology.com/abstracts/v04/1338.html Dengler J. (2008 ). Pitfalls in small-scale species-area sampling and analysis . Folia Geobot. 43 , 269 –287 . 10.1007/s12224-008-9014-9 Elith J. Leathwick J. R. (2009 ). Species distribution models: ecological explanation and prediction across space and time . Ann. Rev. Ecol. Evol. Syst. 40 , 677 –697 . 10.1146/annurev.ecolsys.110308.120159 Gleason H. A. (1922 ). On the relation between species and area . Ecology 3 , 158 –162 . 10.2307/1929150 Green J. L. Ostling A. (2003 ). Endemics-area relationships: the influence of species dominance and spatila aggregation . Eoclogy 84 , 3090 –3097 . 10.1890/02-3096 Guo Q. (2015 ). Island biogeography theory: emerging patterns and human effects , in Reference Module in Earth Systems and Environmental Sciences (Elsevier ). Available online at: http://www.forestthreats.org/products/publications/Island_Biogeography_Theory.pdf/view; http://www.sciencedirect.com/science/module/topic/9780124095489/Concept-000009?_si=1&_ct=25 Harte J. Kitzes J. (2015 ). Inferring regional-scale species diversity from small-plot censuses . PLoS ONE 10 :e0117527 10.1371/journal.pone.0117527 25706536 He F. Hubbell S. P. (2011 ). Species-area relationships always overestimate extinction rates from habitat loss . Nature 473 , 368 –371 . 10.1038/nature09985 21593870 He F. Legendre P. (2002 ). Species diversity patterns derived from species-area models . Ecology 83 , 1185 –1198 . 10.2307/3071933 Available online at: https://www.jstor.org/stable/3071933?seq=1#page_scan_tab_contents McGill B. J. Etienne R. S. Gray J. S. Alonso D. Anderson M. J. Benecha H. K. (2007 ). Species abundance distributions: moving beyond single prediction theories to integration within an ecological framework . Ecol. Lett. 10 , 995 –1015 . 10.1111/j.1461-0248.2007.01094.x 17845298 Mokany K. Jones M. M. Harwood T. D. (2013 ). Scaling pairwise β-diversity and α-diversity with area . J. Biogeogr. 40 , 2299 –2309 . 10.1111/jbi.12175 Pan X. (2013 ). Fundamental equations for species-area theory . Sci. Rep. 3 , 1334 10.1038/srep01334 23434841 Pan X. (2015 ). Reconstruct species-area theory using set theory . Natl. Acad. Sci. Lett. 38 , 173 –177 . 10.1007/s40009-014-0319-3 Pan X. Zhu S. (2015 ). Matthew effect in counting the number of species . Biodivers. Conserv. 24 , 2865 –2868 . 10.1007/s10531-015-0956-y Powell K. I. Chase J. M. Knight T. M. (2013 ). Invasive plants have scale-dependent effects on diversity by altering species-area relationships . Science 339 , 316 –318 . 10.1126/science.1226817 23329045 Preston F. W. (1948 ). The Commonness, and rarity, of species . Ecology 29 , 254 –283 . 10.2307/1930989 R Core Team (2015 ). R: A Language and Environment for Statistical Computing . Vienna : R Foundation for Statistical Computing Available online at: http://www.R-project.org. Rybicki J. Hanski I. (2013 ). Species-area relationships and extinctions caused by habitat loss and fragmentation . Ecol. Lett. 16 , 27 –38 . 10.1111/ele.12065 23452159 Scheiner S. M. (2003 ). Six types of species-area curves . Global Ecol. Biogeogr. 12 , 441 –447 . 10.1046/j.1466-822X.2003.00061.x Šizling A. L. Storch D. (2004 ). Power-law species-area relationships and self-similar species distributions within finite areas . Ecol. Lett. 7 , 60 –68 . 10.1046/j.1461-0248.2003.00549.x Tjørve E. (2003 ). Shapes and functions of species-area curves: a review of possible models . J. Biogeogr. 30 , 827 –835 . 10.1046/j.1365-2699.2003.00877.x Tjørve E. Kunin W. E. Polce C. Tjørve K. M. C. (2008 ). The species-area relationship: separating the effects of species abundance and spatial distribution . J. Ecol. 96 , 1141 –1151 . 10.1111/j.1365-2745.2008.01433.x Tjørve E. Tjørve K. M. C. (2008 ). The species-area relationship, self-similarity, and the true meaning of the z-value . Ecology 89 , 3528 –3533 . 10.1890/07-1685.1 19137957 Verberk W. C. E. P. (2011 ). Explaining general patterns in species abundance and distributions . Nat. Edu. Knowl. 3 , 38 Available online at: http://www.nature.com/scitable/knowledge/library/explaining-general-patterns-in-species-abundance-and-23162842

PMC005xxxxxx/PMC5003820.txt

==== Front Front Plant SciFront Plant SciFront. Plant Sci.Frontiers in Plant Science1664-462XFrontiers Media S.A. 10.3389/fpls.2016.01279Plant ScienceOriginal ResearchThe SOD Gene Family in Tomato: Identification, Phylogenetic Relationships, and Expression Patterns Feng Kun 12Yu Jiahong 23Cheng Yuan 2Ruan Meiying 2Wang Rongqing 2Ye Qingjing 2Zhou Guozhi 2Li Zhimiao 2Yao Zhuping 2Yang Yuejian 2Zheng Qingsong 1*Wan Hongjian 2*1Key Laboratory of Marine Biology, College of Resources and Environmental Science, Nanjing Agricultural UniversityNanjing, China2State key Laboratory Breeding Base for Zhejiang Sustainable Pest and Disease Control, Institute of Vegetables, Zhejiang Academy of Agricultural SciencesHangzhou, China3College of Chemistry and Life Science, Zhejiang Normal UniversityJinhua, ChinaEdited by: Hinanit Koltai, Agricultural Research Organization, Israel Reviewed by: Natalie Brezinova Belcredi, Mendel University, Czech Republic; Paola Leonetti, National Research Council, Italy *Correspondence: Qingsong Zheng, qszheng@njau.edu.cn Hongjian Wan, wanhongjian@sina.comThis article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science 30 8 2016 2016 7 127930 6 2016 11 8 2016 Copyright © 2016 Feng, Yu, Cheng, Ruan, Wang, Ye, Zhou, Li, Yao, Yang, Zheng and Wan.2016Feng, Yu, Cheng, Ruan, Wang, Ye, Zhou, Li, Yao, Yang, Zheng and WanThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.Superoxide dismutases (SODs) are critical antioxidant enzymes that protect organisms from reactive oxygen species (ROS) caused by adverse conditions, and have been widely found in the cytoplasm, chloroplasts, and mitochondria of eukaryotic and prokaryotic cells. Tomato (Solanum lycopersicum L.) is an important economic crop and is cultivated worldwide. However, abiotic and biotic stresses severely hinder growth and development of the plant, which affects the production and quality of the crop. To reveal the potential roles of SOD genes under various stresses, we performed a systematic analysis of the tomato SOD gene family and analyzed the expression patterns of SlSOD genes in response to abiotic stresses at the whole-genome level. The characteristics of the SlSOD gene family were determined by analyzing gene structure, conserved motifs, chromosomal distribution, phylogenetic relationships, and expression patterns. We determined that there are at least nine SOD genes in tomato, including four Cu/ZnSODs, three FeSODs, and one MnSOD, and they are unevenly distributed on 12 chromosomes. Phylogenetic analyses of SOD genes from tomato and other plant species were separated into two groups with a high bootstrap value, indicating that these SOD genes were present before the monocot-dicot split. Additionally, many cis-elements that respond to different stresses were found in the promoters of nine SlSOD genes. Gene expression analysis based on RNA-seq data showed that most genes were expressed in all tested tissues, with the exception of SlSOD6 and SlSOD8, which were only expressed in young fruits. Microarray data analysis showed that most members of the SlSOD gene family were altered under salt- and drought-stress conditions. This genome-wide analysis of SlSOD genes helps to clarify the function of SlSOD genes under different stress conditions and provides information to aid in further understanding the evolutionary relationships of SOD genes in plants. tomatosuperoxide dismutaseSOD gene familypromoterabiotic stressgene expressionNational Natural Science Foundation of China10.13039/50110000180931301774312721563150110311Natural Science Foundation of Zhejiang Province10.13039/501100004731Q15C150010 ==== Body Introduction It is well known that toxic free radicals caused by environmental stresses such as cold, drought, and water-logging are great challenges for crop production (Mittler and Blumwald, 2010). Among them are reactive oxygen species (ROS), toxic free radicals produced in plant cells in response to stress, which can damage membranes, oxidize proteins, and cause DNA lesions (Sun et al., 2007; Wang and Prabakaran, 2011; Feng et al., 2015). However, in the process of evolution, plants have developed defense mechanisms to alleviate the damage caused by adverse environmental conditions. For example, some well-known ROS-scavenging enzymes can defend plants against environmental stress by controlling the expression of enzyme responsive family genes, including superoxide dismutase (SOD), peroxidase (POD), catalase (CAT), glutathione peroxidase (GPX), and peroxiredoxin (PrxR) (Mittler et al., 2004; Filiz and Tombuloğlu, 2014). Superoxide dismutases, a group of metalloenzymes, were first found in bovine erythrocytes in Mann and Keilin (1938). Subsequently, they were also described in bacteria, higher plants, and vertebrates (Rabinowitch and Sklan, 1980; Tepperman and Dunsmuir, 1990; Kim et al., 1996; Zelko et al., 2002). McCord and Fridovich (1969), researchers found that SODs can catalyze the dismutation of the superoxide O2– to O2 and H2O2 (Tepperman and Dunsmuir, 1990). In plants, SODs have been detected in roots, leaves, fruits, and seeds (Giannopolitis and Ries, 1977; Tepperman and Dunsmuir, 1990), where they provide basic protection to cells against oxidative stress. Based on their metal cofactors, protein folds, and subcellular distribution, SODs are mainly categorized as Cu/ZnSODs, FeSODs, and MnSODs (Alscher and Erturk, 2002; Molina-Rueda et al., 2013; Filiz and Tombuloğlu, 2014; Feng et al., 2015). Cu/ZnSODs can be found in prokaryotic and eukaryotic organisms, were first isolated from Photobacterium leiognathi in Puget and Michelson (1974; Deshazer et al., 1994), and are present in the cytoplasm, chloroplasts, and/or the extracellular space in plant cells (Pilon et al., 2011). In mammalian cells, the molecular weight of Cu/ZnSODs is about 32 kDa, and they can be found in cytoplasm, nuclear compartments, and lysosomes (Chang et al., 1988; Keller et al., 1991; Crapo et al., 1992; Liou et al., 1993). FeSODs have been found in plant chloroplasts and cytoplasm (Moran et al., 2003; Miller, 2012). MnSODs, widely present in all major kingdoms, have been observed in eukaryotic mitochondria (Lynch and Kuramitsu, 2000) and can protect mitochondria by scavenging ROS (Moller, 2001). MnSODs also play an important role in promoting cellular differentiation (Wispé et al., 1992; Clair et al., 1994). In addition, a new type of SOD, NiSOD, has been reported in Streptomyces (Youn et al., 1996). However, no evidence for NiSOD has been found in plants (Felisa et al., 2005). In recent years, some studies have reported that SODs can protect plants against abiotic and biotic stresses, such as heat, cold, drought, salinity, abscisic acid and ethylene (Wang et al., 2004; Pilon et al., 2011; Asensio et al., 2012; Feng et al., 2015). Under various environmental stress conditions, researchers have found that different types of SOD genes have different expression patterns. For example, under drought stress, expression patterns of banana genes MaMSD1A and MaCSD1B were completely opposite to one another (Feng et al., 2015). Moreover, SODs with the same metal cofactor did not always play the same role in different species. For example, while we found that expression of MnSODs was not altered under oxidative stress conditions in Arabidopsis, researchers found that MnSOD expressions were changed significantly under salt stress in pea and cold and drought stresses in wheat (Gómez et al., 1999; Wu et al., 1999; Baek and Skinner, 2003). These results show that different SOD genes have different expression patterns in plants in response to diverse environmental stresses. Additionally, researchers have also found that alternative splicing and miRNAs may participate in the regulation of SOD expression (Srivastava et al., 2009; Lu et al., 2011). To date, the SOD gene family has been described in many plant species, including Arabidopsis thaliana, Musa acuminata, Sorghum bicolor, and Populus trichocarpa (Kliebenstein and Last, 1998; Molina-Rueda et al., 2013; Filiz and Tombuloğlu, 2014; Feng et al., 2015). Tomato is not only an important staple and economic crop, but also plays a vital role as an experimental model plant (Mueller et al., 2005). In previous studies, two SOD-coding cDNA sequences isolated from tomato leaves were identified as Cu/ZnSOD genes and were found to be distributed on different chromosomes (Perl-Treves et al., 1990, 1988). Subsequently, the expression levels of two Cu/ZnSOD genes in tomato organs during leaf growth and fruit ripening were reported (Perl-Treves and Galun, 1991). Further, Perl et al. (1993) found that over-expression of Cu/ZnSODs in potato showed that transgenic plants exhibited increased tolerance to oxidative stress. Recently, great efforts have been made to explore the roles of SOD genes in improving tomato tolerance to various environmental stresses, such as heat, salt, drought, cold, and bacteria (Mazorra et al., 2002; Li, 2009; Sasidharan et al., 2013; Soydam et al., 2013; Aydin et al., 2014). In addition, several recent studies have indicated that SOD played an important role of hormone and insecticide stresses. For example, Bernal et al. (2009) detected that the expression level of cytosolic Cu/Zn-SOD was increased after 1 day of auxin treatment for tomato. Chahid et al. (2015) studied the activity of SOD in tomato in response to xenobiotics stresses such as alpha-cypermethrin, chlorpyriphos, and pirimicarb. As described above, SOD gene families have been widely implicated in responses to abiotic and biotic stresses in tomato. However, these studies have mainly concentrated on the expression of a single form of SOD enzyme and on changes in the enzymatic antioxidant system under various environmental stresses, and little has been reported on the SOD gene family in tomato. To comprehensively understand the putative roles of SOD genes in tomato, a systematic analysis of the SOD gene family was necessary at the whole-genome level. Recently, the whole-genome sequence of tomato was made available, which provided opportunities for analyzing the expression patterns and regulation mechanisms of the tomato SOD (SlSOD) gene family in response to environment stresses. Hence, the objectives of this study were (i) to identify the SOD gene family in tomato; (ii) to analyze gene structure, duplication and expression patterns in different tomato tissues; (iii) to illustrate chromosomal locations and phylogenetic relationships with SODs from other plants; and (iv) to reveal the regulating mechanisms of the SlSOD gene family under abiotic (salt and drought) stress by using real-time fluorescence quantitative PCR (qRT-PCR). Materials and Methods Identification of SOD Genes in Tomato and Other Plant Species In this study, two methods were used to identify potential SOD genes in tomato. First, the whole tomato genome was downloaded from Sol Genomics Network (SGN1) and a local database was constructed using the software Bioedit 7.0 (Pan and Jiang, 2014). Four SOD amino acid sequences from Solanum lycopersicum (AF527880-CuSOD), S. tuberosum (EU545469-FeSOD), Arabidopsis thaliana (AAM62550.1-MnSOD) and Musa acuminata AAA Group (AEZ56248.2-FeSOD) were used as a query against the local tomato amino acid database. Second, Hidden Markov Model (HMM) profiles of Cu/ZnSOD (PF00080), Fe-MnSOD (PF00081, PF02777) were downloaded from Pfam2. A BlastP search was performed to retrieve candidate tomato SOD genes. For BlastP, e-value was set at 1e-5. All redundant putative SOD sequences were excluded. The remaining SOD sequences were examined for copper/zinc and iron/manganese SOD domains by the Pfam server2. Physicochemical characteristics of SOD amino acid sequences were predicted by the Protparam tool3, including molecular weight (MW), and theoretical isoelectric point (pI) (Gasteiger et al., 2005). In addition, to reveal the evolutionary relationships between SOD genes in different plant species, potential SOD genes from eight plant species were selected for phylogenetic analysis. Among them, SOD genes from four plant species (Vitis vinifera, Solanum tuberosum, Zea mays, and Panicum miliaceum) were identified using the same method above. The SOD genes of the remaining plant species (Poplar, Arabidopsis thaliana, Oryza sativa, Sorghum bicolor) were derived from previously published studies (Kliebenstein and Last, 1998; Dehury et al., 2012; Molina-Rueda et al., 2013; Filiz and Tombuloğlu, 2014). Subcellular Localization, Conserved Motifs, and Gene Structure Analysis of SlSOD Proteins Subcellular localization of SOD proteins from different plant species was obtained from the ProtComp9.0 server4 (Feng et al., 2015). Conserved motif analysis of SlSOD genes was performed by the Multiple Em for Motif Elicitation (MEME Suite 4.11.1) server5 (Bailey et al., 2009). We used the method described by Feng et al. (2015), except that the number of motifs was set to 8. Intron/exon configurations of SlSOD genes were determined via the Gene Structure Display Server6, for both coding sequences and genomic sequences (Hu et al., 2015). Chromosomal Location and Gene Duplication Information about chromosomal location of SlSOD genes was obtained from the SGN database and gene duplications were identified by the Plant Genome Duplication Database (PGDD)7 (Singh and Jain, 2015). Tandemly duplicated SlSOD genes were identified according to methods reported by previous researchers (Yang and Tuskan, 2008; Tuskan et al., 2006). Chromosomal locations of the SlSOD genes were performed with the MapDraw V2.1 tool based on information from the SGN database (Huang et al., 2012). Sequence similarity of SODs was calculated using the program DNAMAN. Phylogenetic Tree Construction of SODs To investigate the phylogenetic relationships of SlSOD genes, a total of 108 SOD protein sequences were identified from nine plant species. Among them, 24 SOD protein sequences were excluded owing to the SOD domains being incomplete. Multiple sequence alignments of the remaining 84 SOD amino acid sequences were performed with ClustalW (Higgins et al., 1996) using default parameters. A phylogenetic tree was constructed using the software MEGA5.04 via neighbor-joining method (Tamura et al., 2011). In the phylogenetic tree, the degree of support for a particular grouping pattern was evaluated using bootstrap (1000 replicates) value (Filiz et al., 2014). The tree was viewed with FigTree (v1.3.1). Promoter Sequence Analysis Regions 1,000 bp upstream from the start codons of each SlSOD gene were downloaded from SGN. Then, cis-elements in promoters of each SlSOD gene were predicted using the PlantCARE server (Postel et al., 2002). Expression Patterns of SlSOD Genes Based on RNA-seq and Microarray Data For RNA-seq analysis, transcription data of the genome-wide gene expression of the tomato cultivar Helnz were downloaded from the tomato functional genomics database (TFGD)8 (Fei et al., 2011). Ten different tissues were selected: root, leaf, bud, flower, 1-cm fruit, 2-cm fruit, 3-cm fruit, mature green fruit (MG), fruit at the fruit breaking stage (B), and fruit 10 days after the fruit breaking stage (B10). RPKM (Reads Per Kilo bases per Million mapped Reads) values of SlSOD genes were log2-transformed (Wei et al., 2012). Heat maps of SlSOD genes in different tissues were generated using MeV4.9 software (Singh and Jain, 2015). In addition, microarray data for salt and drought stresses were downloaded from the TOM2 cDNA array and Affymetrix Tomato Genome Array platform from TFGD (Fei et al., 2011). Probe sets corresponding to SlSOD genes were identified using the Probe Match Tool in NetAffx Analysis Center9 (Altschul et al., 1997). A BlastN search was performed based on sequence alignment between probe sequences and SlSOD sequences. Expression patterns of SlSOD genes under salt and drought stresses were viewed using MeV4.9 software (Singh and Jain, 2015). Plant Materials and Stress Treatment Seeds of a tomato breeding strain, Zhe fen 202, were germinated on water-saturated filter paper. Before germinating, a 10% hypochlorous acid solution was used to sterilize seeds for 5 min. Then, seeds were washed three times with distilled water. Seedlings were grown on Hoagland nutrient solution, in a controlled chamber (25°C/20°C, day/night, 16 h/12 h light/dark cycle). Upon development of the fourth true leaf, seedlings were cultivated in Hoagland nutrient solution with 150 mM sodium chloride (NaCl) and 18% polyethylene glycol (PEG) for treatment (3 and 12 h), respectively. Leaves were collected and frozen in liquid nitrogen immediately and stored at -80°C. Three biological replications were carried out. RNA Extraction and qRT-PCR Data Analysis RNA extraction was performed using the RNAsimple Total RNA Kit (TIANGEN, Beijing, China) according to manufacturer instructions. Before reverse transcription, the quality of RNA samples was checked by agarose gel electrophoresis. All RNAs were reverse transcribed into cDNA using the FastQuant RT Kit with gDNase (TIANGEN, Beijing, China) according to manufacturer instructions. Specific primers for qRT-PCR analysis were designed using the GenScript server10 (Supplementary Table S1) and were synthesized by Sangon Biotech (Shanghai). The GAPDH gene was used as an internal control (Expósito-Rodríguez et al., 2008). An ABI StepOne real time fluorescence quantitative PCR instrument (Applied Biosystems, American) was used for qRT-PCR analysis. The quality and specificity of primers was determined by the melt curve (Feng et al., 2015). Three independent technical replicates were performed for each of the SlSOD genes. The PCR program consisted of an initial denaturation at 95°C for 15 min, followed by 40 cycles of 95°C for 10 s, 55°C for 20 s, and 72°C for 30 s. The relative expression levels were calculated using the 2-ΔΔCt method, and were presented by histogram (Livak and Schmittgen, 2001). Results Genome-Wide Identification of SOD Genes Family in Tomato A total of nine SOD genes, classified into two major groups (Cu/ZnSODs and Fe-MnSODs), were identified in tomato. The former group included four members with a copper-zinc domain (SlSOD1, 2, 3, and 4); the latter was composed of five members with an iron/manganese SOD alpha-hairpin domain and an iron/manganese SOD, C-terminal domain (SlSOD5, 6, 7, 8 and 9) (Table 1). The physicochemical analysis showed that the length of amino acid sequences, MW, and pI values varied among these SlSOD proteins. The length ranged from 152 to 311AA, MW ranged from 15.3 to 34.6 kDa, and pI ranged from 5.38 to 7.13 (Table 1). No significant difference in the acid-base properties of SlSOD proteins was observed, except for SlSOD9, which was slightly basic. Using the ProtComp9.0 program, subcellular localizations of SlSOD proteins were determined. Among them, two Cu/ZnSODs (SlSOD1 and 2) and one Fe-MnSOD (SlSOD8) were predicted to localize in the cytoplasm. One Fe-MnSOD (SlSOD9) was localized in mitochondrion. The remaining members were localized in the chloroplast. Table 1 The characteristics of SOD genes from Solanum lycopersicum. Gene name Sequence ID Chromosome ORF Length (bp) Intron number Length (aa) MW (KDA) pI Predicted Pfam domain Subcellular prediction by PC SlSOD1 Solyc01g067740.2 01 459 6 152 15.3 5.47 CZ Cytoplasm SlSOD2 Solyc03g062890.2 03 471 6 156 15.9 6.53 CZ Cytoplasm SlSOD3 Solyc11g066390.1 11 654 7 217 22.3 6.01 CZ Chloroplast SlSOD4 Solyc08g079830.2 08 936 5 311 32.9 6.45 HMA,CZ Chloroplast SlSOD5 Solyc06g048410.2 06 750 8 249 27.9 6.6 IMA,IMC Chloroplast SlSOD6 Solyc03g095180.2 03 912 8 303 34.6 5.38 IMA,IMC Chloroplast SlSOD7 Solyc02g021140.2 02 759 7 252 29.1 6.65 IMA,IMC Chloroplast SlSOD8 Solyc06g048420.1 06 483 4 160 17.9 6.41 IMA,IMC Cytoplasm SlSOD9 Solyc06g049080.2 06 687 5 228 25.3 7.13 IMA,IMC Mitochondrion MW, Molecular weight; pI, isoelectric points; CZ, copper/zinc superoxide dismutase (SODC); HMA, heavy-metal-associated domain; IMA, Iron/manganese superoxide dismutases alpha-hairpin domain; IMC, Iron/manganese superoxide dismutases, C-terminal domain; PC, ProtComp9.0 server.Conserved Motif Analysis of SlSOD Proteins Four motifs in SlSOD proteins (motif 1 to motif 4) were identified by MEME. Among them, three motifs (motifs 1, 3, and 4) were related to iron/manganese SOD domains, while motif 2 was related to copper/zinc SOD domains (Table 2). As shown in Figure 1, motif 2 was located in Cu/ZnSODs (SlSOD1, 2 and 3), except SlSOD4; motif 1 and motif 3 were shared in Fe-MnSODs (SlSOD5, 6, 7, 8, and 9). Motif 4 was widely present in Fe-MnSODs, except for SlSOD8. Table 2 Eight different motifs commonly observed in tomato protein sequences by MEME server. Motif number Width Protein sequences Pfam domain 1 65 ESMKPGGGGEPSGELLQLINRDFGSYDTFVKEFKAAAATQFGSGWAWLAYKPEDKRLAIVKTPNA IMC (shorten) KAVAVLNGNDNVQGTIQFTQDDDGPTTVNGRITGLAPGLHGFHIHALGDTTNGCMSTGPH CZ (SODC) 2 149 FNPNKKDHGAPMDEVRHAGDLGNIVAGPDGVAEITITDMQIPLTGPHSIIGRAVVVHADPDDLGKGGHE LSKTTGNAGGRIACGVIGLQ 3 28 WEHAYYLDFQNRRPDYISIFMEKLVSWE IMC (shorten) 4 42 KFDLPPPPYPMDALEPHMSRRTFEFHWGKHHRAYVDNLNKQI IMA (shorten) 5 28 IILVTYNNGNPLPPFNNAAQAWNHQFFW 6 22 FLPPQGFNESCRSLQWRTQKKQ 7 23 CCCQRCVSAVKSDLWRQFGIPNV 8 20 METHSIFHQTSSDNGFVYPE Among them, Pfam domain of four mitifs (motif5, 6, 7 and 8) were not identify by Pfam server. CZ, copper/zinc superoxide dismutase (SODC); IMA, Iron/manganese superoxide dismutases alpha-hairpin domain; IMC, Iron/manganese superoxide dismutases, C-terminal domain.FIGURE 1 Conserved motif analysis of SlSOD proteins. Different color boxes represent different types of motifs. Chromosomal Distribution and Intron/Exon Configurations of SlSOD Genes The chromosomal distributions of nine SlSOD genes were determined. As shown in Figure 2, six out of the twelve chromosomes harbored SlSOD genes. Chromosome 6 and 3 possessed three and two SlSOD genes, respectively, while each of the remaining four chromosomes (chromosome 1, 2, 8, and 11) contained only one SlSOD gene. Notably, chromosome 6 had one gene cluster (SlSOD5 and 8), which was identified as tandem duplication event. Segmental duplication was identified between SlSOD5 (chromosome 6) and SlSOD6 (chromosome 3) by PGDD database. However, despite the sequence similarity between SlSOD5 and SlSOD6, no tandem-duplicated paralogous genes were found in the region surrounding SlSOD6. FIGURE 2 Chromosomal locations of 9 SlSOD genes on 12 chromosomes of tomato. Red lines represent the position of SlSOD genes on chromosomes. The chromosome numbers are indicated at the top of chromosomes. The segment duplication event occurred between SlSOD5 (chromosome 6) and SlSOD6 (chromosome 3) and one cluster including tandemly duplicated genes (SlSOD5 and SlSOD8) on chromosome 6. Intron/exon configurations of SlSOD genes were constructed using the Gene Structure Display Server (GSDS 2.011) by aligning the cDNA sequences with the corresponding genomic DNA sequences (Figure 3). We found that intron numbers among these SlSOD genes ranged from 4 to 8. Two SlSOD genes (SlSOD5 and 6) exhibited the highest intron number (8), whereas SlSOD8 only had four introns (Table 1). In addition, two groups of SlSOD genes (SlSOD1 and 2, SlSOD5 and 6) exhibited similar intron/exon organization patterns, respectively. The rest of the SlSOD genes exhibited diverse intron/exon organization patterns. FIGURE 3 Intron/exon configurations of SlSOD genes. Exons and introns are shown as yellow boxes and thin lines, respectively. UTRs are shown with blue boxes. 0 = intron phase 0; 1 = intron phase 1; 2 = intron phase 2. Phylogenetic Analysis of SOD Genes in Plants To investigate the phylogenetic relationships of SOD proteins between tomato and other plant species, a total of 108 SOD proteins were identified from S. lycopersicum, Populus trichocarpa, Arabidopsis thaliana, Vitis vinifera, S. tuberosum, O. sativa, Zea mays, Sorghum bicolor, and Panicum miliaceum. Among them, 24 SOD proteins were excluded due to incomplete SOD domains. An unrooted phylogenetic tree was constructed based on the remaining 84 SOD proteins using the program MEGA5.04 (Tamura et al., 2011). As shown in Figure 4, the SOD proteins from different plant species were divided into two major groups, Cu/ZnSODs and Fe-MnSODs. The former group was subdivided into three subgroups (a, b, and c), and the latter was separated into two subgroups (d and e), which was strongly supported by high bootstrap values. FIGURE 4 Phylogenetic tree of 84 SOD proteins from tomato and other plants including Populus trichocarpa, Arabidopsis thaliana, Vitis vinifera, S. tuberosum, O. sativa, Zea mays, Sorghum bicolor, and Panicum miliaceum. Two groups (Cu/ZnSODs and Fe-MnSOD) were identified and the tree was divided into five groups (a, b, c, d and e) based on high bootstrap values. SlSOD proteins are marked in red. Three colors (coral, light blue and light green) represent the subcellular locations of SOD proteins: coral represents proteins localized in the cytoplasm, light blue represents proteins localized in chloroplasts, and light green represents proteins localized in mitochondria. In our study, phylogenetic analysis showed that SlSOD1 grouped with OsSOD1 (Loc-Os03g22810), OsSOD2 (Loc-Os07g46990) and other plants’ cytosolic Cu/ZnSODs clustered in subgroup a. SlSOD4 grouped with OsSOD3 and other plants’ chloroplastic Cu/ZnSODs clustered in subgroup c, while two SlSOD genes (SlSOD2 and SlSOD3) were clustered with other plants’ cytosolic Cu/ZnSODs and chloroplastic Cu/ZnSODs in subgroup b, respectively. These results indicated that diversity in the Cu/ZnSODs gene family occurred before the splitting of mono- and dicot plants. Interestingly, Cu/ZnSODs genes in subgroup a were separated into mono- and dicot-specific branches, which suggested that they evolved independently after the splitting of mono- and dicot plants. In addition, FeSODs and MnSODs from different plant species were separated by a high bootstrap value (95%). Four genes (SlSOD5, 6, 7, and 8) were clustered with other plants’ chloroplastic FeSODs in subgroup d and SlSOD9 was clustered with other plants’ mitochondrial MnSODs in subgroup e. Analysis of Cis-Elements in Putative SlSOD Gene Promoters To further understand gene function and regulation patterns, cis-elements in SlSOD gene promoter sequences were researched. Regions of 1,000 bp upstream from the start codons of each SlSOD gene were determined using PlantCARE. The results showed that the cis-elements could be divided into three major classes: stress-responsive, hormone-responsive, and light-responsive. Six stress-responsive cis-elements were identified, including HSE, MBS, LTR, TC-rich, ARE and Box-W1, which reflected plant responses to heat, drought, low-temperature, defense stresses, anaerobic induction and fungal elicitors, respectively. Ten kinds of hormone-responsive cis-elements were identified (e.g., salicylic acid-SA, methyl jasmonate-MeJA, gibberellins-GA, auxin-IAA, and ethylene) (Figure 5). A relatively large number of light-responsive cis-elements in SlSOD promoters was observed (Supplementary Table S2). FIGURE 5 Cis-elements in the promoters of putative SlSOD genes that are related to stress responses. Different cis-elements with the same or similar functions are present with the same color. Expression Analysis of SlSOD Genes in Different Tissues To explore the expression patterns of SOD genes during tomato growth and development, expression profiles were analyzed for 10 different tissues (root, leaf, bud, flower, 1-cm fruit, 2-cm fruit, 3-cm fruit, MG, B, and B10) of the tomato cultivar Helnz using the RNA-seq atlas. As shown in Figure 6A, five SlSOD genes (SlSOD1, 2, 3, 4 and 9) had similar expression patterns in all the tested tissues, while two genes (SlSOD6 and 8) displayed distinct tissue-specific expression patterns. Interestingly, SlSOD1 demonstrated a consistently high expression in all ten tissues, whereas SlSOD6 and 8 were mainly expressed in young fruit. In addition, SlSOD7 was expressed strongly in young fruit, weakly in root and moderately in the other tissues. FIGURE 6 Expression profiles of SlSOD genes in different tomato tissues and under various biotic and abiotic stresses. (A) The expression patterns of SlSOD genes in different tissues. The tested tissues are: root, leaf, bud, flower, 1 cm- fruit, 2 cm- fruit, 3 cm-fruit, mature green fruit (MG), fruit breaking (B), 10 days after fruit breaking (B10). (B) Expression profiles of SlSOD genes under salt stress. (C) Expression profiles of SlSOD genes under drought stress. Expression Patterns of SlSOD Genes in Response to Abiotic Treatments To gain further insight into the role of SlSOD genes under abiotic stress, we analyzed the expression profiles of SlSOD genes in response to salt and drought stresses using microarray data. A total of 17 independent tomato microarray probes were identified by means of a BlastN search. As shown in Figure 6B, expression of SlSOD genes was significantly altered under different abiotic stress treatments. For the salt treatment, eight probes corresponding to SlSOD genes were found, with the exception being SlSOD7. Expression of four SlSOD genes (SlSOD2, 5, 6, and 8) were down-regulated, and expression of three SlSOD genes (SlSOD3, 4, and 9) remained constant. Notably, SlSOD1 was significantly up-regulated. For the drought treatment, microarray probes for each SlSOD gene were identified. Four SlSOD genes (SlSOD2, 5, 6, and 8) were up-regulated and three (SlSOD3, 4, and 9) were down-regulated (Figure 6C). Notably, SlSOD6 expression increased twofold. SlSOD1 and SlSOD7 were unchanged. We found that the expression levels of most of the SlSOD genes changed significantly under salt and drought stresses. Verification of SlSOD Gene Expression Patterns with qRT-PCR To further verify the expression profiles of SlSOD genes determined by microarray data analysis, qRT-PCR was used to analyze expression patterns of the nine SlSOD genes under salt and drought stresses. As shown in Figure 7, most of the SlSOD gene results were consistent with the microarray patterns. In response to salt treatment, expression levels of most SlSOD genes were down-regulated, in accordance with the microarray profiles. However, expression patterns of three SlSOD genes (SlSOD1, 6 and 9) determined by qRT-PCR analysis were different than those determined using the microarray profiles. We found that in response to salt treatment, there was enhanced expression of SlSOD6 and SlSOD9 and a decreased transcript level of SlSOD1. In response to drought treatment, expression levels of five SlSOD genes (SlSOD1, 3, 4, 6, and 7), were consistent with the microarray data. However, in contrast to our microarray results, qRT-PCR analysis showed down-regulation of three SlSOD genes (SlSOD2, 5, and 8) and up-regulation of SlSOD9. FIGURE 7 Gene expression profiles of SlSOD genes in response to salt and drought treatments using qRT-PCR. (A) Expression patterns of SlSOD genes under salt stress conditions. (B) Expression patterns of SlSOD genes under drought stress condition. Error bars represent standard deviations from three independent technical replicates. Discussion Environmental stresses pose considerable challenges for crop production. Gene expression and SOD enzyme activities are influenced by environmental stresses such as high salinity, drought and metal toxicity (Schützendübel and Polle, 2002; Atkinson et al., 2013). However, plants have evolved defense mechanisms to alleviate the damage caused by adverse environmental conditions. Tomato, an important staple and economic crop, is affected by various abiotic stresses (Mueller et al., 2005). SODs are key enzymes in many oxidation processes, and provide basic protection against ROS in plants (Alscher and Erturk, 2002). Therefore, a systematic analysis of the SlSOD gene family was performed and gene expression patterns were determined for plants under various abiotic stresses. In this study, nine SlSOD genes (four Cu/ZnSODs, four FeSODs, and one MnSOD) were identified in the tomato genome, including all three major types of plant SOD genes. Chromosome location analysis revealed that SlSOD5 and SlSOD8 formed a gene cluster on chromosome 6 and identified as tandem duplicated event. Segmental duplication was identified between SlSOD5 (chromosome 6) and SlSOD6 (chromosome 3) by PGDD. Although SlSOD5 and SlSOD6 have similar sequence, no tandem duplicated events were occurred in the region surrounding SlSOD6. Considering the lower sequence similarity of SlSOD5 to SlSOD6 than that of SlSOD5 to SlSOD8, it appears that segmental duplication events predate the tandem duplication in the SlSOD5 and SlSOD8 gene cluster (Supplementary Table S3). Therefore, we concluded that segmental duplication and tandem duplication played key roles in the expansion of SOD genes in the tomato genome. Gene structure analysis revealed that the intron number of SlSOD genes ranges from 4 to 8. Previous researchers reported that intron patterns of plant SOD genes were highly conserved, and that all cytosolic and chloroplastic SOD genes included seven introns except for one member (Fink and Scandalios, 2002; Filiz and Tombuloğlu, 2014). However, in this study, we observed that the intron numbers of six out of eight SlSOD genes (excepting SlSOD9) were varied, and only two SlSOD genes (SlSOD3, chloroplastic Cu/ZnSOD and SlSOD7, chloroplastic FeSOD) included seven introns. Thus, our data did not support the previous reports of plant SOD gene intron patterns. Variation in exon-intron structures was accomplished by three main mechanisms (exon/intron gain/loss, exonization/pseudoexonization and insertion/deletion), each of which contributed to structural divergence (Xu et al., 2012; Filiz and Tombuloğlu, 2014). Moreover, two groups of SlSOD genes (SlSOD1 and 2, SlSOD5 and 6) exhibited similar intron/exon organization patterns, respectively, which suggested high conservation in the evolutionary process. Phylogenetic analysis showed that FeSODs and MnSOD from different plant species were separated by a high bootstrap value (95%). This result was in agreement with previous report (Fink and Scandalios, 2002). In plants, MnSODs had 70% homology but were different from FeSODs, which suggested that they originated from different ancestral genes (Miller, 2012). These data were well support our results that FeSODs and MnSOD of tomato were diverged with 95% bootstrap value. To better understand the role of SlSOD genes under various environmental stresses, cis-elements in the promoter sequences were predicted by PlantCARE server. The results showed that three major classes of cis-elements were identified, including stress-responsive, hormone-responsive, and light-responsive. Many identified cis-elements in the promotes of SlSOD genes were related to heat, drought, low-temperature, defense stresses, anaerobic induction, fungal elicitors, SA, MeJA, GA, IAA and ethylene. As previously stated, cis-elements play an important role in plant stress responses; cis-elements such as ABRE, DRE, CRT, SARE and SURE respond to abscisic acid (ABA), dehydration, cold, SA, and sulfur, respectively (Sakuma et al., 2002; Maruyama-Nakashita et al., 2005; Shi et al., 2010; Osakabe et al., 2014). Thus, these results will contribute to further understand the various function role of SlSOD genes under complex abiotic stress conditions. To further clarify the potential functions of SlSOD genes, expression profiles of all the SlSOD genes in different tissues were analyzed. Based on RNA-seq, nine SlSOD genes exhibited two disparate expression patterns: constitutive and tissue-specific expression patterns of SlSOD genes. During tomato growth and development, two genes (SlSOD1 and 9) sustained high expression in all the tested tissues. Two genes (SlSOD6 and 8) demonstrated a tissue-specific expression pattern, being mainly expressed in young fruits. SlSOD7 was expressed strongly in young fruit tissue, weakly in root tissue, and moderately in the rest of the tissues tested. Under various natural conditions, plant growth and development are frequently affected by high salinity, drought, cold, bacteria, and insecticides (Xia et al., 2012). Previous researchers reported that three types of SODs (Cu/ZnSODs, FeSODs and MnSODs) have been exploited to eliminate ROS caused by abiotic stress (Kliebenstein and Last, 1998; Wu et al., 1999). To further clarify the putative roles of SOD genes in tomato response to abiotic stresses, we examined the expression patterns of nine SlSOD genes under salt and drought treatment conditions using microarray and qRT-PCR. The results showed that the expression patterns of most SlSOD genes obtained by qRT-PCR were in conformity with those obtained from the microarray analysis. Expression analysis revealed that most SlSOD gene expression levels were changed in response to two abiotic stresses (salt and drought). For salt treatment, SlSOD1 was the only gene that showed significant up-regulation among the nine SlSOD genes, demonstrating that the function of SlSOD1 relates to salt stress. Compared to the other SlSOD genes, a greater variety and quantity of cis-elements were found in the promoter for SlSOD1, including two TC-rich motifs, an HSEs motif, and an MBS motif, which had been demonstrated to responsible for abiotic stress in banana (Feng et al., 2015). This could explain why SlSOD1 expression changed significantly under salt treatment. Moreover, although many cis-elements related to abiotic stresses were also found in the SlSOD8 promoter, SlSOD8 expression was significantly down-regulated under salt treatment. This suggested that some unidentified cis-element could play a vital role in regulating the expression of the SOD gene family in tomato under abiotic stress. Similar results have also been found in other plant species (Singh and Jain, 2015). Additionally, we observed different expression patterns of SlSOD genes under salt and drought stress conditions (Figure 6). For example, the expression of four SlSOD genes (SlSOD2, 5, 6, and 8) decreased during the salt treatment, while increased expression levels were observed for these four genes in response to drought treatment. This suggested that different SOD genes in tomato could play different roles in eliminating ROS caused by different environment stresses. Taken together, the data we obtained provides more information about the SOD gene family in tomato including sequence information, gene duplications, conserved motifs, gene structures, and phylogenetic relationships. Promoter analysis and complex regulation patterns of the SlSOD genes under abiotic stress contribute to our understanding of the expression patterns of SOD genes in plants and provide clues for further studies of the roles of SOD genes under different stress conditions. Author Contributions Conceived and designed the experiments: HW, QZ, and KF. Performed the experiments: KF, JY, YC, MR, RW, QY, GZ, ZY, and YY. Analyzed the data: KF, JY, and ZL. Wrote the paper: KF and HW. All authors have read and approved the manuscript. Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Research is supported by National Natural Science Foundation of China (31301774, 31272156, and 3150110311), Zhejiang Provincial Natural Science Foundation of China (Q15C150010), and Young Talent Cultivation Project of Zhejiang Academy of Agricultural Sciences (2015R23R08E07, 2015R23R08E09), Public Agricultural Technology Research in Zhejiang (2016C32101, 2015C32049), and Technological System of Ordinary Vegetable Industry (CARS-25-G-16). 1 https://solgenomics.net/ 2 http://pfam.xfam.org/ 3 http://web.expasy.org/ 4 http://linux1.softberry.com/ 5 http://meme-suite.org/tools/meme 6 http://gsds.cbi.pku.edu.cn/ 7 http://chibba.agtec.uga.edu/duplication/ 8 http://ted.bti.cornell.edu/ 9 http://www.affymetrix.com 10 https://www.genscript.com/ssl-bin/app/primer 11 http://gsds.cbi.pku.edu.cn/index.php Supplementary Material The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.01279 Click here for additional data file. Click here for additional data file. Click here for additional data file. ==== Refs References Alscher R. G. Erturk N. (2002 ). Role of superoxide dismutases (SODs) in controlling oxidative stress in plants. J. Exp. Bot. 53 1331 –1341 . 10.1093/jexbot/53.372.1331 11997379 Altschul S. F. Madden T. L. Schäffer A. A. Zhang J. Zhang Z. Miller W. (1997 ). Gapped blast and psi-blast: a new generation of protein database search programs. Nucleic Acids Res. 25 3389 –3402 . 10.1093/nar/25.17.3389 9254694 Asensio A. C. Gil-Monreal M. Pires L. Gogorcena Y. Aparicio-Tejo P. M. Moran J. F. (2012 ). Two Fe-superoxide dismutase families respond differently to stress and senescence in legumes. J. Plant Physiol. 169 1253 –1260 . 10.1016/j.jplph.2012.04.019 22840995 Atkinson N. J. Lilley C. J. Urwin P. E. (2013 ). Identification of genes involved in the response of Arabidopsis to simultaneous biotic and abiotic stresses. Plant Physiol. 162 2028 –2041 . 10.1104/pp.113.222372 23800991 Aydin S. Büyük İ. Aras E. S. (2014 ). Expression of SOD gene and evaluating its role in stress tolerance in NaCl and PEG stressed Lycopersicum esculentum. Turk. J. Bot. 38 89 –98 . 10.3906/bot-1305-1 Baek K. H. Skinner D. Z. (2003 ). Alteration of antioxidant enzyme gene expression during cold acclimation of near-isogenic wheat lines. Plant Sci. 165 1221 –1227 . 10.1016/S0168-9452(03)00329-7 Bailey T. L. Boden M. Buske F. A. Frith M. Grant C. E. Clementi L. (2009 ). MEME SUITE: tools for motif discovery and searching. Nucleic Acids Res. 37 W202 –W208 . 10.1093/nar/gkp335 19458158 Bernal A. Torres J. Reyes A. Rosado A. (2009 ). Exogenous auxin regulates H2O2 metabolism in roots of tomato (Lycopersicon esculentum mill.) seedlings affecting the expression and activity of CuZn-superoxide dismutase, catalase, and peroxidase. Acta Physiol. Plant. 31 249 –260 . 10.1007/s11738-008-0225-8 Chahid K. Laglaoui A. Zantar S. Ennabili A. (2015 ). Antioxidant-enzyme reaction to the oxidative stress due to alpha-cypermethrin, chlorpyriphos, and pirimicarb in tomato (Lycopersicon esculentum, Mill.). Environ. Sci. Pollut. Res. Int. 22 18115 –18126 . 10.1007/s11356-015-5024-3 26178835 Chang L. Y. Slot J. W. Geuze H. J. James D. C. (1988 ). Molecular immunocytochemistry of the CuZn superoxide dismutase in rat hepatocytes. J. Cell Biol. 107 2169 –2179 . 10.1083/jcb.107.6.2169 3058718 Clair D. K. S. Oberley T. D. Muse K. E. Clair W. H. S. (1994 ). Expression of manganese superoxide dismutase promotes cellular differentiation. Free Radic. Biol. Med. 16 275 –282 . 10.1016/0891-5849(94)90153-8 7516302 Crapo J. D. Oury T. Rabouille C. Slot J. W. Chang L. Y. (1992 ). Copper, zinc superoxide dismutase is primarily a cytosolic protein in human cells. Proc. Natl. Acad. Sci. U.S.A. 89 10405 –10409 . 10.1073/pnas.89.21.10405 1332049 Dehury B. Sarma K. Sarmah R. Sahu J. Sahoo S. Sahu M. (2012 ). In silico analyses of superoxide dismutases (SODs) of rice (Oryza sativa L.). J. Plant Biochem. Biotechnol. 22 150 –156 . 10.1007/s13562-012-0121-6 Deshazer D. Barnnan J. D. Moran M. J. Friedman R. L. (1994 ). Characterization of the gene encoding superoxide dismutase of Bordetella pertussis and construction of a SOD-deficient mutant. Gene 142 85 –89 . 10.1016/0378-1119(94)90359-X 8181762 Expósito-Rodríguez M. Borges A. A. Borges-Pérez A. Pérez J. A. (2008 ). Selection of internal control genes for quantitative real-time RT-PCR studies during tomato development process. BMC Plant Biol. 8 :131 10.1186/1471-2229-8-131 Fei Z. Joung J. G. Tang X. Zheng Y. Huang M. Lee J. M. (2011 ). Tomato functional genomics database: a comprehensive resource and analysis package for tomato functional genomics. Nucleic Acids Res. 39 1156 –1163 . 10.1093/nar/gkq991 Felisa W. Daniel G. Oscar S. Falkowski P. G. (2005 ). The role and evolution of superoxide dismutases in algae. J. Phycol. 41 453 –465 . 10.1111/j.1529-8817.2005.00086.x Feng X. Lai Z. Lin Y. Lai G. Lian C. (2015 ). Genome-wide identification and characterization of the superoxide dismutase gene family in Musa acuminata cv. Tianbaojiao (AAA group). BMC Genomics 16 :823 10.1186/s12864-015-2046-7 Filiz E. Koç İ. Tombuloğlu H. (2014 ). Genome-wide identification and analysis of growth regulating factor genes in Brachypodium distachyon: in silico approaches. Turk. J. Biol. 38 296 –306 . 10.3906/biy-1308-57 Filiz E. Tombuloğlu H. (2014 ). Genome-wide distribution of superoxide dismutase (SOD) gene families in Sorghum bicolor. Turk. J. Biol. 39 49 –59 . 10.3906/biy-1403-9 Fink R. C. Scandalios J. G. (2002 ). Molecular evolution and structure–function relationships of the superoxide dismutase gene families in angiosperms and their relationship to other eukaryotic and prokaryotic superoxide dismutases. Arch. Biochem. Biophys. 399 19 –36 . 10.1006/abbi.2001.2739 11883900 Gasteiger E. Hoogland C. Gattiker A. Duvaud S. Wilkins M. R. Appel R. D. (2005 ). “Protein identification and analysis tools on the expasy server ,” in Proteomics Protocols Handbook, ed. Walker J. M. (Totowa, NJ : Humana Press ), 571 –607 . Giannopolitis C. N. Ries S. K. (1977 ). Superoxide dismutase: I. Occurrence in higher plants. Plant Physiol. 59 309 –314 .16659839 Gómez J. M. Hernández J. A. Jiménez A. del Río L. A. Sevilla F. (1999 ). Differential response of antioxidative enzymes of chloroplasts and mitochondria to long-term NaCl stress of pea plants. Free Radic. Res. 31(Suppl. 1) , 11 –18 . 10.1080/10715769900301261 Higgins D. G. Thompson J. D. Gibson T. J. (1996 ). Using CLUSTAL for multiple sequence alignments. Methods Enzymol. 266 383 –402 . 10.1016/S0076-6879(96)66024-8 8743695 Hu B. Jin J. Guo A. Y. Zhang H. Luo J. Gao G. (2015 ). GSDS 2.0: an upgraded gene feature visualization server. Bioinformatics 31 1296 –1297 . 10.1093/bioinformatics/btu817 25504850 Huang S. Gao Y. Liu J. Peng X. Niu X. Fei Z. (2012 ). Genome-wide analysis of WRKY transcription factors in Solanum lycopersicum. Mol. Genet. Genomics 287 495 –513 . 10.1007/s00438-012-0696-6 22570076 Keller G. A. Warner T. G. Steimer K. S. Hallewell R. A. (1991 ). Cu, Zn superoxide dismutase is a peroxisomal enzyme in human fibroblasts and hepatoma cells. Proc. Natl. Acad. Sci. U.S.A. 88 7381 –7385 . 10.1073/pnas.88.16.7381 1651504 Kim E. J. Kim H. P. Hah Y. C. Roe J. H. (1996 ). Differential expression of superoxide dismutases containing Ni and Fe/Zn in Streptomyces coelicolor. Eur. J. Biochem. 241 178 –185 . 10.1111/j.1432-1033.1996.0178t.x 8898904 Kliebenstein D. J. Last R. L. (1998 ). Superoxide dismutase in Arabidopsis: an eclectic enzyme family with disparate regulation and protein localization. Plant Physiol. 118 637 –650 . 10.1104/pp.118.2.637 9765550 Li Y. (2009 ). Physiological responses of tomato seedlings (lycopersicon esculentum) to salt stress. Mod. Appl. Sci. 3 171 –176 . 10.5539/mas.v3n3p171 Liou W. Chang L. Y. Geuze H. J. Strous G. J. Crapo J. D. Slot J. W. (1993 ). Distribution of CuZn superoxide dismutase in rat liver. Free Radic. Biol. Med. 14 201 –207 . 10.1016/0891-5849(93)90011-I 8425722 Livak K. J. Schmittgen T. D. (2001 ). Analysis of relative gene expression data using real-time quantitative PCR and the 2 -ΔΔct method. Methods 25 402 –408 . 10.1006/meth.2001.1262 11846609 Lu Y. Feng Z. Bian L. Xie H. Liang J. (2011 ). miR398 regulation in rice of the responses to abiotic and biotic stresses depends on CSD1 and CSD2 expression. Funct. Plant Biol. 38 44 –53 . 10.1071/FP10178 Lynch M. Kuramitsu H. (2000 ). Expression and role of superoxide dismutases (SOD) in pathogenic bacteria 1. Microbes Infect. 2 1245 –1255 . 10.1016/S1286-4579(00)01278-8 11008114 Mann T. Keilin D. (1938 ). Haemocuprein and hepatocuprein, copper-protein compounds of blood and liver in mammals. Proc. R. Soc. B 126 303 –315 . 10.1098/rspb.1938.0058 Maruyama-Nakashita A. Nakamura Y. Watanabe-Takahashi A. Inoue E. Yamaya T. Takahashi H. (2005 ). Identification of a novel cis -acting element conferring sulfur deficiency response in Arabidopsis roots. Plant J. 42 305 –314 . 10.1111/j.1365-313X.2005.02363.x 15842617 Mazorra L. M. Núñez M. Hechavarria M. Coll F. Sánchez-Blanco M. J. (2002 ). Influence of brassinosteroids on antioxidant enzymes activity in tomato under different temperatures. Biol. Plant 45 593 –596 . 10.1023/A:1022390917656 McCord J. M. Fridovich I. (1969 ). Superoxide dismutase. An enzymic function for erythrocuprein (hemocuprein). J. Biol. Chem. 244 6049 –6055 .5389100 Miller A. F. (2012 ). Superoxide dismutases: ancient enzymes and new insights. FEBS Lett. 586 585 –595 . 10.1016/j.febslet.2011.10.048 22079668 Mittler R. Blumwald E. (2010 ). Genetic engineering for modern agriculture: challenges and perspectives. Annu. Rev. Plant Biol. 61 443 –462 . 10.1146/annurev-arplant-042809-112116 20192746 Mittler R. Vanderauwera S. Gollery M. Breusegem F. V. (2004 ). Reactive oxygen gene network of plants. Trends Plant Sci. 9 490 –498 . 10.1016/j.tplants.2004.08.009 15465684 Molina-Rueda J. J. Tsai C. J. Kirby E. G. (2013 ). The Populus superoxide dismutase gene family and its responses to drought stress in transgenic poplar overexpressing a pine cytosolic glutamine synthetase (GS1a). PLoS ONE 8 :e56421 10.1371/journal.pone.0056421 Moller I. M. (2001 ). Plant mitochondria and oxidative stress: electron transport, NADPH turnover, and metabolism of reactive oxygen species. Annu. Rev. Plant Physiol. Biol. 52 561 –591 . 10.1146/annurev.arplant.52.1.561 Moran J. F. James E. K. Rubio M. C. Sarath G. Klucas R. V. Becana M. (2003 ). Functional characterization and expression of a cytosolic iron-superoxide dismutase from cowpea root nodules. Plant Physiol. 133 773 –782 . 10.1104/pp.103.023010 14512518 Mueller L. A. Solow T. H. Taylor N. Skwarecki B. Buels R. Binns J. (2005 ). The SOL genomics network: a comparative resource for Solanaceae biology and beyond. Plant Physiol. 138 1310 –1317 . 10.1104/pp.105.060707 16010005 Osakabe Y. Yamaguchi-Shinozaki K. Shinozaki K. Tran L. S. P. (2014 ). ABA control of plant macroelement membrane transport systems in response to water deficit and high salinity. New Phytol. 202 35 –49 . 10.1111/nph.12613 24283512 Pan L. J. Jiang L. (2014 ). Identification and expression of the WRKY transcription factors of Carica papaya in response to abiotic and biotic stresses. Mol. Biol. Rep. 41 1215 –1225 . 10.1007/s11033-013-2966-8 24390238 Perl A. Perl-Treves R. Galili S. Aviv D. Shalgi E. Malkin S. (1993 ). Enhanced oxidative-stress defense in transgenic potato expressing tomato Cu, Zn superoxide dismutases. Theor. Appl. Genet. 85 568 –576 . 10.1007/BF00220915 24195931 Perl-Treves R. Abu-Abied M. Magal N. Galun E. Zamir D. (1990 ). Genetic mapping of tomato cDNA clones encoding the chloroplastic and the cytosolic isozymes of superoxide dismutase. Biochem. Genet. 28 543 –552 . 10.1007/BF00554381 2085316 Perl-Treves R. Galun E. (1991 ). The tomato Cu,Zn superoxide dismutase genes are developmentally regulated and respond to light and stress. Plant Mol. Biol. 17 745 –760 . 10.1007/BF00037058 1912497 Perl-Treves R. Nacmias B. Aviv D. Zeelon E. P. Galun E. (1988 ). Isolation of two cDNA clones from tomato containing two different superoxide dismutase sequences. Plant Mol. Biol. 11 609 –623 . 10.1007/BF00017461 24272495 Pilon M. Ravet K. Tapken W. (2011 ). The biogenesis and physiological function of chloroplast superoxide dismutases. Biochim. Biophys. Acta 1807 989 –998 . 10.1016/j.bbabio.2010.11.002 21078292 Postel D. Vanlemmens P. Gode P. Ronco G. Villa P. (2002 ). Plantcare, a database of plant cis-acting regulatory elements and a portal to tools for in silico analysis of promoter sequences. Nucleic Acids Res. 30 325 –327 . 10.1093/nar/30.1.325 11752327 Puget K. Michelson A. M. (1974 ). Isolation of a new copper-containing superoxide dismutase bacteriocuprein. Biochem. Biophys. Res. Commun. 58 830 –838 . 10.1016/S0006-291X(74)80492-4 4836277 Rabinowitch H. D. Sklan D. (1980 ). Superoxide dismutase: a possible protective agent against sunscald in tomatoes (Lycopersicon esculentum Mill.). Planta 148 162 –167 . 10.1007/BF00386417 24309704 Sakuma Y. Qiang L. Dubouzet J. G. Abe H. Shinozaki K. Yamaguchi-Shinozaki K. (2002 ). DNA-Binding specificity of the ERF/AP2 domain of Arabidopsis, DREBs, transcription factors involved in dehydration- and cold-inducible gene expression. Biochem. Biophys. Res. Commun. 290 998 –1009 . 10.1006/bbrc.2001.6299 11798174 Sasidharan S. Kulangara N. R. Sailas B. (2013 ). Biotic stress induced biochemical and isozyme variations in ginger and tomato by Ralstonia solanacearum. Am. J. Plant Sci. 4 1601 –1610 . 10.4236/ajps.2013.48194 Schützendübel A. Polle A. (2002 ). Plant responses to abiotic stresses: heavy metal-induced oxidative stress and protection by mycorrhization. J. Exp. Bot. 53 1351 –1365 . 10.1093/jexbot/53.372.1351 11997381 Shi Z. Maximova S. N. Liu Y. Verica J. Guiltinan M. J. (2010 ). Functional analysis of the Theobroma cacao NPR1 gene in Arabidopsis. BMC Plant Biol. 10 :248 10.1186/1471-2229-10-248 Singh V. K. Jain M. (2015 ). Genome-wide survey and comprehensive expression profiling of Aux/IAA gene family in chickpea and soybean. Front. Plant Sci. 6 :918 10.3389/fpls.2015.00918 Soydam A. S. Büyük I. Aras S. (2013 ). Relationships among lipid peroxidation, SOD enzyme activity, and SOD gene expression profile in Lycopersicum esculentum L. exposed to cold stress. Genet. Mol. Res. 12 3220 –3229 . 10.4238/2013.August.29.6 24065665 Srivastava V. Srivastava M. K. Chibani K. Nilsson R. Rouhier N. Melzer M. (2009 ). Alternative splicing studies of the reactive oxygen species gene network in Populus reveal two isoforms of high-isoelectric-point superoxide dismutase. Plant Physiol. 149 1848 –1859 . 10.1104/pp.108.133371 19176719 Sun L. H. Shen L. T. Ye S. (2007 ). Simultaneous overexpression of both Cu/Zn superoxide dismutase and ascorbate peroxidase in transgenic tall fescue plants confers increased tolerance to a wide range of abiotic stresses. J. Plant Physiol. 164 1626 –1638 . 10.1016/j.jplph.2007.01.003 17360071 Tamura K. Peterson D. Peterson N. Stecher G. Nei M. Kumar S. (2011 ). MEGA5 : molecular, evolutionary, genetics, analysis, using maximum, likelihood, evolutionary, distance, and maximum, parsimony, methods. Mol. Biol. Evol. 28 2731 –2739 . 10.1093/molbev/msr121 21546353 Tepperman J. M. Dunsmuir P. (1990 ). Transformed plants with elevated levels of chloroplastic SOD are not more resistant to superoxide toxicity. Plant Mol. Biol. 14 501 –511 . 10.1007/BF00027496 1966384 Tuskan G. A. Difazio S. Jansson S. Bohlmann J. Grigoriev I. Hellsten U. (2006 ). The genome of black cottonwood, populus trichocarpa (Torr.&Gray). Science 313 1596 –1604 . 10.1126/science.1128691 16973872 Wang B. Lüttge U. Ratajczak R. (2004 ). Specific regulation of SOD isoforms by NaCl and osmotic stress in leaves of the C3, halophyte Suaeda salsa L. J. Plant Physiol. 161 285 –293 . 10.1078/0176-1617-01123 15077627 Wang H. W. Prabakaran N. (2011 ). 2,4-dichlorophenoxyacetic acid-induced leaf senescence in mung bean (Vigna radiata L. Wilczek) and senescence inhibition by co-treatment with silver nanoparticles. Plant Physiol. Biochem. 49 168 –177 . 10.1016/j.plaphy.2010.11.007 21144762 Wei K. F. Chen J. Chen Y. F. Wu L. J. Xie D. X. (2012 ). Molecular phylogenetic and expression analysis of the complete WRKY transcription factor family in maize. DNA Res. 19 153 –164 . 10.1093/dnares/dsr048 22279089 Wispé J. R. Warner B. B. Clark J. C. Dey C. R. Neuman J. Glasser S. W. (1992 ). Human Mn-superoxide dismutase in pulmonary epithelial cells of transgenic mice confers protection from oxygen injury. J. Biol. Chem. 267 23937 –23941 .1385428 Wu G. Wilen R. W. Robertson A. J. Gusta L. V. (1999 ). Isolation, chromosomal localization, and differential expression of mitochondrial manganese superoxide dismutase and chloroplastic copper/zinc superoxide dismutase genes in wheat. Plant Physiol. 120 513 –520 . 10.1104/pp.120.2.513 10364402 Xia Z. Liu Q. Wu J. Ding J. (2012 ). ZmRFP1, the putative ortholog of SDIR1, encodes a RING-H2 E3 ubiquitin ligase and responds to drought stress in an ABA-dependent manner in maize. Gene 495 146 –153 . 10.1016/j.gene.2011.12.028 22245611 Xu G. Guo C. Shan H. Kong H. (2012 ). Divergence of duplicate genes in exon-intron structure. Proc. Natl. Acad. Sci. U.S.A. 109 1187 –1192 . 10.1073/pnas.1109047109 22232673 Yang X. Tuskan G. A. (2008 ). The F-box gene family is expanded in herbaceous annual plants relative to woody perennial plants. Plant Physiol. 148 1189 –1200 . 10.1104/pp.108.121921 18775973 Youn H. D. Kim E. J. Roe J. H. Hah Y. C. Kang S. O. (1996 ). A novel nickel-containing superoxide dismutase from streptomyces spp. Biochem. J. 318 889 –896 . 10.1042/bj3180889 8836134 Zelko I. N. Mariani T. J. Folz R. J. (2002 ). Superoxide dismutase multigene family: a comparison of the CuZn-SOD (SOD1), Mn-SOD (SOD2), and EC-SOD (SOD3) gene structures, evolution, and expression. Free Radic. Biol. Med. 33 337 –349 . 10.1016/S0891-5849(02)00905-X 12126755

PMC005xxxxxx/PMC5003821.txt

==== Front Front Plant SciFront Plant SciFront. Plant Sci.Frontiers in Plant Science1664-462XFrontiers Media S.A. 10.3389/fpls.2016.01308Plant ScienceOriginal ResearchAnalysis of Network Topologies Underlying Ethylene Growth Response Kinetics Prescott Aaron M. 1McCollough Forest W. 2Eldreth Bryan L. 1Binder Brad M. 2*Abel Steven M. 13*1Department of Chemical and Biomolecular Engineering, University of TennesseeKnoxville, TN, USA2Department of Biochemistry and Cellular and Molecular Biology, University of TennesseeKnoxville, TN, USA3National Institute for Mathematical and Biological Synthesis, University of TennesseeKnoxville, TN, USAEdited by: Daniel H. Chitwood, Donald Danforth Plant Science Center, USA Reviewed by: Tsu-Wei Chen, Leibniz University of Hanover, Germany; Ashok Prasad, Colorado State University, USA *Correspondence: Brad M. Binder bbinder@utk.eduSteven M. Abel abel@utk.eduThis article was submitted to Plant Biophysics and Modeling, a section of the journal Frontiers in Plant Science 30 8 2016 2016 7 130825 6 2016 16 8 2016 Copyright © 2016 Prescott, McCollough, Eldreth, Binder and Abel.2016Prescott, McCollough, Eldreth, Binder and AbelThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.Most models for ethylene signaling involve a linear pathway. However, measurements of seedling growth kinetics when ethylene is applied and removed have resulted in more complex network models that include coherent feedforward, negative feedback, and positive feedback motifs. The dynamical responses of the proposed networks have not been explored in a quantitative manner. Here, we explore (i) whether any of the proposed models are capable of producing growth-response behaviors consistent with experimental observations and (ii) what mechanistic roles various parts of the network topologies play in ethylene signaling. To address this, we used computational methods to explore two general network topologies: The first contains a coherent feedforward loop that inhibits growth and a negative feedback from growth onto itself (CFF/NFB). In the second, ethylene promotes the cleavage of EIN2, with the product of the cleavage inhibiting growth and promoting the production of EIN2 through a positive feedback loop (PFB). Since few network parameters for ethylene signaling are known in detail, we used an evolutionary algorithm to explore sets of parameters that produce behaviors similar to experimental growth response kinetics of both wildtype and mutant seedlings. We generated a library of parameter sets by independently running the evolutionary algorithm many times. Both network topologies produce behavior consistent with experimental observations, and analysis of the parameter sets allows us to identify important network interactions and parameter constraints. We additionally screened these parameter sets for growth recovery in the presence of sub-saturating ethylene doses, which is an experimentally-observed property that emerges in some of the evolved parameter sets. Finally, we probed simplified networks maintaining key features of the CFF/NFB and PFB topologies. From this, we verified observations drawn from the larger networks about mechanisms underlying ethylene signaling. Analysis of each network topology results in predictions about changes that occur in network components that can be experimentally tested to give insights into which, if either, network underlies ethylene responses. ethylenesignal transductionnetwork topologiescomputational modelingevolutionary algorithmNational Science Foundation10.13039/100000001IOS-1254423MCB-1517032 ==== Body 1. Introduction Ethylene is the simplest of olefin gases and functions as a plant hormone, affecting many processes throughout the lifetime of a plant including seed germination, growth, formation of the apical hook, senescence, fruit ripening, abscission, and responses to various stresses (Mattoo and Suttle, 1991; Abeles et al., 1992). Ethylene inhibits the growth of dark-grown eudicot seedlings (Abeles et al., 1992), and sustained exposure to ethylene leads to a growth-inhibition response that has been used to screen for mutants and to provide information about the ethylene signaling network (Bleecker et al., 1988; Guzman and Ecker, 1990). Most proposed models of ethylene signaling consist of a linear pathway (Figure 1), where in air, ethylene receptors signal to the CONSTITUTIVE RESPONSE1 (CTR1) protein kinase which functions as a negative regulator of ethylene signaling (Kieber et al., 1993). CTR1 prevents ethylene signaling by phosphorylating the ETHYLENE INSENSITIVE2 (EIN2) protein, leading to its ubiquitination and proteolysis (Chen et al., 2011; Ju et al., 2012; Qiao et al., 2012). The binding of ethylene to ethylene receptors reduces the activity of the receptors, leading to reduced activity of CTR1 kinase and reduced phosphorylation of EIN2 protein (Chen et al., 2011; Ju et al., 2012; Qiao et al., 2012). The reduction in EIN2 phosphorylation leads to a decrease in ubiquitination of EIN2, causing a rise in EIN2 protein levels and allowing for proteolytic release of the C-terminal portion (EIN2-C) of the protein (Qiao et al., 2009; Ju et al., 2012; Qiao et al., 2012; Wen et al., 2012). EIN2-C affects the levels of two transcription factors, EIN3 and EIN3-Like1 (EIL1), in part by regulating their ubiquitination via S-PHASE KINASE-ASSOCIATED1-CULLIN-F-BOX (SCF) ubiquitin ligase complexes containing EIN3-BINDING F-BOX1 and 2 (EBF1 and 2) F-box proteins (Guo and Ecker, 2003; Potuschak et al., 2003; Yanagisawa et al., 2003; Gagne et al., 2004; Binder et al., 2007; An et al., 2010). In the presence of ethylene, ubiquitination of EIN3 and EIL1 is reduced, leading to accumulation of these transcription factors causing most ethylene responses (Guo and Ecker, 2003; Potuschak et al., 2003; Yanagisawa et al., 2003; Binder et al., 2004a, 2007; Gagne et al., 2004; An et al., 2010). Figure 1 Basic linear ethylene signaling network. The above model was developed based on end-point analyses. Even though end-point analysis of ethylene responses continues to be an instructive bioassay, it is limited because transient events are overlooked. Time-lapse imaging has provided information about the kinetics of ethylene growth responses. The kinetics of ethylene responses have been studied for several plant species (Laan, 1934; Warner and Leopold, 1971; Burg, 1973; Goeschl and Kays, 1975; Rauser and Horton, 1975; Jackson, 1983) and most extensively studied in the model flowering plant, Arabidopsis thaliana (Binder et al., 2004a,b; Potuschak et al., 2006; Binder et al., 2007; Gao et al., 2008; Christians et al., 2009; Vandenbussche et al., 2010; van Zanten et al., 2010; Žádníková et al., 2010; Kim et al., 2011, 2012; McDaniel and Binder, 2012; Bakshi et al., 2015; Merchante et al., 2015; Rai et al., 2015). Studies of ethylene growth kinetics in Arabidopsis have revealed two phases of growth inhibition at saturating ethylene levels (1 ppm or above, see Figure 2A) (Binder et al., 2004b). The first phase is rapid, with a decrease in growth rate beginning approximately 10 min after the application of ethylene and lasting approximately 10 min, at which point the growth rate reaches a plateau. This plateau lasts approximately 30 min when a second, slower phase of growth inhibition is observed. Approximately 90 min after the application of ethylene, the growth rate reaches a minimum that lasts for as long as saturating levels of ethylene are present. If ethylene is removed after 2 h, seedlings recover to pre-treatment growth rates in approximately 90 min (Figure 2A). Figure 2 Growth response kinetics of dark-grown Arabidopsis seedling hypocotyls. (A) The normalized growth rate of wildtype Arabidopsis. Seedlings were grown in air for 1 h at which time 10 ppm ethylene was applied for 2 h at which time ethylene-free air was used to replace the ethylene. (B) The normalized growth rate of wildtype Arabidopsis compared to ein3;eil1 and ein2 mutants. Seedlings were grown in air for 1 h at which time 10 ppm ethylene was applied. (C) The normalized growth rate of wildtype Arabidopsis treated with varying concentrations of ethylene as indicated. Seedlings were grown in air for 1 h prior to application of ethylene. In all panels, the growth rate was normalized to the growth rate in air prior to treatment with ethylene. Based on data from Binder et al. (2004a,b). The two phases of growth inhibition are genetically separable (Binder et al., 2004a). Mutants lacking EIN3 and EIL1 (ein3;eil1) have a normal first phase of growth inhibition but fail to have a second phase response and over time return to pre-treatment growth rates in the continued presence of ethylene (Figure 2B). This demonstrates that the first phase of growth inhibition is EIN3/EIL1-independent. Mutants lacking EIN2 (ein2) have no response to ethylene. Additionally, in wild type plants, adaptation is observed at sub-saturating levels of ethylene. At intermediate to high sub-saturating levels (e.g., 100 ppb), seedlings initially show both phases of growth inhibition but then have a partial recovery to an intermediate growth rate. At lower levels of ethylene (e.g., 2 and 10 ppb), only the first phase of growth inhibition occurs and is followed by recovery of the growth rate (Figure 2C) (Binder et al., 2004a). The experiments described above and other observations that indicate possible alternative pathways and feedback control (Kieber et al., 1993; Roman et al., 1995; Larsen and Chang, 2001; Hall and Bleecker, 2003; Qiu et al., 2012; Rai et al., 2015) suggest that ethylene signal transduction is not simply a linear pathway. Several network models have been proposed that involve more complicated topologies (Binder et al., 2004a; Gao et al., 2008; Kim et al., 2012). These and related networks are the focus of this paper. A study comparing the time-dependent growth responses of several plant species led to a proposed network that included both coherent feedforward and negative feedback (CFF/NFB) signaling motifs (Figure 3A; Kim et al., 2012). In the CFF/NFB model, in air (i.e., without ethylene) the receptors signal to CTR1, which in turn inhibits downstream signaling. This leads to fast growth with feedback on growth occurring via the modulation of gibberellin (GA), a hormone known to stimulate growth. Application of ethylene inhibits the receptors, leading to a reduction of CTR1 activity and hence an increase in EIN2 levels. EIN2 is predicted to cause an initial growth inhibition response independently of EIN3 and EIL1. EIN2 also inhibits the EBF1 and EBF2 F-box proteins, leading to increases in EIN3 and EIL1. In the CFF/NFB network model, EIN3 and EIL1 inhibit growth via a GA-dependent and GA-independent pathway. The indirect inhibition of growth from EIN2 through EIN3 and EIL1 is hypothesized to be responsible for the second phase of growth inhibition. Qualitatively, this network topology provides a framework to understand the molecular basis for phase 1 and phase 2 growth inhibition and its regulation by the coherent feedforward signal. It also provides a mechanism for transient growth inhibition in the absence of EIN3 and EIL1 that is regulated by negative feedback components. Figure 3 Diagrams illustrating proposed network topologies. We were curious to determine whether other network topologies would yield response kinetics similar to experiments. We previously proposed a model where one function of EIN3 and EIL1 is to provide feedback to regulate growth (Binder et al., 2004a). More recent data has shown that ethylene signaling requires the accumulation of EIN2 followed by the proteolytic cleavage of the EIN2 C-terminal tail (Alonso et al., 1999; Ju et al., 2012; Qiao et al., 2012; Wen et al., 2012). This accumulation of EIN2 protein is reduced in ein3;eil1 double mutants (Qiao et al., 2009) suggesting a mechanism for feedback by EIN3 and EIL1. We therefore developed a network topology where the function of EIN3 and EIL1 is to provide feedback stimulation in the form of increased synthesis of EIN2 (Figure 3B). EIN2-C feeds back through EBF1/2 and EIN3 to promote EIN2. Without accumulation of EIN2, the prediction is that the levels of EIN2-C would decrease, resulting in growth reversal. We refer to this network as the cleavage with positive feedback (PFB) network. Few computational models of ethylene signaling have been published (Díaz and Álvarez-Buylla, 2006; Díaz and Alvarez-Buylla, 2009), and none take into account the dynamical information obtained from kinetic studies of ethylene response. Considering experimental results of this nature provides an opportunity to identify network features underlying the observed growth responses. Therefore, we developed computational models that account for proposed interactions within the CFF/NFB and PFB networks described above. These networks are modeled as sets of coupled ordinary differential equations (ODEs) describing the time evolution of network components. We are interested in (i) whether the proposed networks are capable of producing dynamical growth-response behavior consistent with experiments, and (ii) mechanisms underlying the network response when it does recapitulate experimental results. The CFF/NFB and PFB networks we consider have relatively high-dimensional parameter spaces (35 and 26 parameters, respectively). The parameters regulate numerous coupled, nonlinear ODEs describing the dynamics of the network, and changing the value of one parameter can have unexpected effects on network responses. Few in vivo measurements of network parameters are available. As such, we used an evolutionary algorithm (EA) to search for sets of parameters that produce network behavior consistent with experimental growth response kinetics. EAs are a class of numerical optimization techniques and have been used to investigate networks in a variety of biochemical applications (Bäck and Schwefel, 1993; Bray and Lay, 1994; François and Hakim, 2004; Patil et al., 2005; Auliac et al., 2008; Sun et al., 2012; Spirov and Holloway, 2013; Feng et al., 2015). For example, time-course data has been used to determine parameters of small genetic networks (Kikuchi et al., 2003) and parameters associated with signal transduction in neurons (Arisi et al., 2006). We used an EA to evolve parameter sets that produce ethylene growth responses similar to those observed in experiments. In particular, we focused on evolving two-phase growth inhibition (2-PGI) and ein3;eil1 mutant partial growth recovery (MPGR). By repeatedly performing independent runs of our EA, we created libraries of evolved parameter sets. We gain insight into mechanisms underlying network responses by analyzing the dynamics of individual network components and the distributions of parameters governing the network. We additionally screen the parameter sets for partial growth recovery in the presence of sub-saturating ethylene doses, which is a property that emerges in some of the evolved parameter sets. We further explore each network by identifying simplified networks producing both 2-PGI and MPGR. 2. Methods 2.1. Kinetics of the ethylene response network We model the ethylene growth-response networks proposed above using systems of coupled ordinary differential equations (ODEs). Ethylene concentration is treated as an input variable that is varied to mimic experimental conditions. We treat the growth rate, denoted by [Growth], as a concentration-like variable that measures the fraction of maximal growth rate. The unbound ethylene receptor concentration is described by a production term and a mass-action binding term representing ethylene binding. The time-dependence of all other components is described by production and degradation terms. As an example, the ODE describing CTR1 dynamics is written d[CTR1]dt= (kprod[R]NKprodN+[R]N)R(1−[CTR1])               −kdegrCTR1[CTR1] Concentrations are denoted by square brackets, k denotes a reaction rate, K denotes an activation coefficient in a Hill equation, and N is the associated Hill coefficient. All concentrations are restricted to the range of 0 to 1 and can be interpreted as the fraction of the maximum concentration for each species. The concentration range is constrained by describing the production as a logistic production term, with production vanishing when the concentration approaches 1. Interactions between network components are described with Hill-like kinetics. For example, in the above equation, CTR1 is promoted by receptors (R), which is captured by the Hill equation in the production term on the right-hand side. Inhibitory interactions promote the rate of degradation (e.g., EBF increases the degradation rate of EIN3 and EIL1). Stimulatory interactions promote the rate of production. The complete sets of ODEs for the networks studied are included in the Supplementary Material. We initially equilibrate the system with no ethylene present, allowing it to reach steady state. We then introduce a step-change in ethylene to mimic experimental conditions. We treat EIN3 and EIL1 as a single entity, and therefore, to model the ein3;eil1 mutant, the production rate for EIN3 is set equal to zero. 2.2. Evolutionary algorithm Given the system of ODEs describing network dynamics, we use an evolutionary algorithm (EA) to identify sets of parameters that produce growth-response behavior similar to that observed experimentally. EAs are a class of optimization techniques that utilize the principle of inherited fitness to optimize parameters. A population of parameter sets is evolved over multiple generations. At each generation, each parameter set is evaluated by a fitness function and ranked by its fitness. Parameter sets with better rankings are modified in order to produce a new population of parameter sets for evaluation. The modifications consist of mutation and crossover operations. Mutations change a parameter value within a set to a new, randomly sampled value. Crossovers are events in which subsets from two high-performing parameter sets are recombined to form a new parameter set. Details of the mutations and crossovers depend on the specific implementation of the EA. Each iteration in which the population of parameter sets is updated is termed a generation. We used the following fitness function for all proposed ethylene signaling networks: fitness=∑i=1Nwtαi([Growth]calc(ti)−[Growth]targ(ti))wt2          +∑i=1Nmtβi([Growth]calc(ti)−[Growth]targ(ti))mt2 Calculated growth values ([Growth]calc(ti)]) are determined by numerically solving the system of ODEs in MATLAB using the ode45 numerical solver. Target values ([Growth]targ(ti)]) were obtained using experimentally-determined growth rates at select times (Figures 2A,B). Target values were selected from wildtype (wt) and ein3;eil1 mutant (mt) experiments, with the index i in each sum indexing the target values (there are Nwt target values for the wildtype response and Nmt target values for the mutant response). For each time point, the squared deviation of the calculated growth rate from the target value is multiplied by a weighting factor (αi, βi) that emphasizes important regimes of the growth response. Specifically, we emphasize pre-ethylene steady state values, the wildtype two-phase growth inhibition response, minimal growth rate following ethylene introduction, and maximum growth recovery levels. Target values and weighting factors are provided in Supplemental Tables 2.1.1, 2.1.2. Target growth rates were scaled by dividing all experimental growth rates by the maximum observed growth rate under both wildtype and mutant conditions. This gives a pre-ethylene growth rate target value that is less than one, in contrast with Figure 2 in which growth rates were scaled so that pre-ethylene growth rates were unity. The evolutionary algorithm was designed to minimize the fitness function. In our EA, we use a population size of 200 parameters sets at each generation. The initial parameter sets are generated by selecting uniformly distributed random values for each parameter from allowed parameter ranges (see Supplemental Table 2.1.3). Hill coefficients (N) are restricted to integer values. After evaluating each parameter set, the 50 parameter sets with the lowest fitness scores are selected as source parameter sets. These source sets are used to generate the population of parameter sets for the next generation. New parameter sets are produced by performing a two-point crossover followed by mutations. The crossover events and mutations allow a balance of global and local exploration of parameter space, and the algorithm converged to local minima of the fitness function for the signaling networks studied. Two-point crossovers are performed by randomly selecting two source parameter sets with replacement (i.e., the same set can be chosen twice). Two crossover points are chosen at random from the list of parameters. Two blocks of parameters are taken from the first source and the other block is taken from the second source, leading to a newly constructed set of parameters. Additionally, the probability of mutation for each parameter is chosen such that on average three parameters within the crossover product are mutated (the probability is 3/35 for the CFF/NFB network and 3/26 for the PFB network). When a parameter is selected for mutation, the decision to increase or decrease the value is made with equal probability. The parameter is then multiplied or divided, respectively, by a uniformly distributed value between 1 and 2. When this decision would result in a parameter exceeding its upper bound, a uniform random value between the current parameter value and its upper limit is used instead. The source parameter sets are updated each generation by replacing 25 randomly selected source sets with parameter sets having the lowest fitness scores from the current population. We ran the EA for 400 generations and recorded the parameter set with the lowest fitness score for additional analysis. This evolutionary process was repeated independently 500–4000 times depending on the network topology. Each parameter set was screened for the targeted network behavior and the resulting data was used to characterize ethylene response kinetics and identify features of the evolved parameters. 2.3. Screening for targeted responses After running the EA, we check whether the resulting growth responses exhibit wildtype two-phase growth inhibition and/or ein3;eil1 mutant partial growth recovery. Specifically, network responses are checked to ensure that the wildtype response meets the following conditions: With no ethylene, steady state growth is sufficiently high. After applying ethylene, the minimal growth rate is sufficiently low. A plateau-like region separates the first and second phases of growth inhibition. Following removal of ethylene, growth recovers to a sufficiently high level. Logic diagrams for discriminant functions are included in Supplemental Figures S1, S2. For mutant behavior, traits 1 and 2 were used to check for proper behavior. The discriminant functions were designed to make the inclusion of false positives unlikely. 2.4. Ethylene dose response We additionally screen evolved parameters sets to identify whether they exhibit sub-saturating ethylene dose response kinetics similar to experimental observations (see Figure 2C). The level of ethylene that leads to a sub-saturating response depends on the parameters of the network. Thus, we first identify the range of ethylene concentrations over which the network is responsive to concentration variations. For each evolved parameter set, we use a binary search method to identify the ethylene concentration range in which (i) the maximum dose produces long-time growth rate between 0.5 and 1.0% above minimum growth observed in the saturated response and (ii) the minimum dose produces a minimum growth rate between 0.5 and 1.0% below pre-ethylene steady state growth. We consider 20 evenly distributed ethylene concentrations between these bounds to test for partial growth recovery in the presence of sustained ethylene exposure. A parameter set is considered to exhibit sub-saturating growth recovery if there exists at least one ethylene concentration at which the growth maximum that occurs 1 h or longer after the introduction of ethylene exceeds the minimum growth observed within the first hour of ethylene exposure by at least 0.1 (maximum possible growth is unity). 3. Results 3.1. Coherent feedforward/negative feedback network The CFF/NFB network (Figure 3A) was proposed by Kim et al. (2012) based on growth kinetics in response to the addition and removal of ethylene. It can be broken down into three distinct regions: (i) the initial linear signaling cascade consisting of ethylene receptors, CTR1, and EIN2; (ii) a coherent feedforward loop with EIN2 as the initial node that inhibits growth both directly and indirectly (via EBF and EIN3); (iii) a negative feedback loop consisting of growth and GA. The coherent feedforward cascade interacts with the negative feedback loop as a result of the inhibitory effect of EIN3 on GA. As indicated in Figure 3A, we treat EBF1 and EBF2 as well as EIN3 and EIL1 as single entities. We refer to these nodes as EBF and EIN3, respectively. This reduces the complexity of the model and the dimensionality of the parameter space while keeping key topological features of the network. Additionally, with existing experimental data, it is difficult to elucidate differences between these individual components, which could be included in a more detailed computational model. We conducted multiple independent trials of the EA, obtaining 3774 sets of optimized parameters. Using the discriminant functions described previously, each evolved parameter set was screened for wildtype two-phase growth inhibition (2-PGI) and ein3;eil1 mutant partial growth recovery (MPGR). Figure 4 shows examples of results that exhibit both 2-PGI and MPGR, as well as those that exhibit only one of the responses. Approximately 36% of the evolved parameter sets exhibit both 2-PGI and MPGR, 20% exhibit only 2-PGI, and 23% exhibit only MPGR. The large number of parameter sets yielding one or both of the targeted growth responses provides a large data set for analysis. Figure 4 Characteristic growth responses at saturating ethylene doses. Columns show examples of time-dependent growth responses passing different combinations of screening criteria. Each column corresponds to a single set of evolved parameters. Rows show different simulated conditions (wildtype and ein3;eil1 mutant). Targeted growth rates are denoted by x and the evolved response is shown by solid lines. Dashed vertical lines indicate time points at which ethylene was introduced (green) and removed (red). 3.1.1. Dynamical response of network components In Figure 5, we plot the time dependence of each network component. This provides insight into how the feedforward and feedback loops shape growth response dynamics. For each component, we plot the mean response of the 1344 parameter sets exhibiting both 2-PGI and MPGR behavior. We also display the standard deviation about the mean (shaded regions) to characterize the heterogeneity of the response. Analogous results for parameter sets exhibiting only 2-PGI or MPGR behavior are provided in Supplemental Figures S3, S4. Figure 5 Time evolution of CFF/NFB network components. Figures show the mean ±1 SD for each component of the CFF/NFB network for cases exhibiting both wildtype and ein3;eil1 mutant growth responses (1344 parameters sets). Components include: (A) early signaling components (wildtype conditions), (B) components downstream of EIN2 (wildtype conditions), and (C) negative feedback components affecting growth response (ein3;eil1 mutant conditions). Black regions in (B,C) indicate the mean ±1 standard deviation of growth. Dashed vertical lines indicate time points at which ethylene was introduced (green) and removed (red). The response of the linear portion of the signaling cascade is identical for the wildtype (shown in Figure 5A) and ein3;eil1 mutant (not shown) topologies upon addition of ethylene. In response to the addition of ethylene at 1 h, the concentration of unbound receptors rapidly declines to levels near zero. This results in a decrease of CTR1 from a relatively high pre-ethylene concentration to a much lower concentration. Following this, EIN2 is no longer inhibited by active CTR1 and rapidly increases in concentration. There is a slight delay in the EIN2 response to ethylene due to the time required for the signal to propagate through the upstream components of the linear signaling cascade. Upon removal of ethylene at 3 h in the wildtype response, components return to their pre-ethylene levels. The remaining network connections differentiate the wildtype response from the ein3;eil1 mutant response. Components of the wildtype network are shown in Figure 5B. Increasing EIN2 concentration acts to inhibit both growth and EBF, which is part of the indirect feedforward loop. The direct inhibitory effect of EIN2 on growth is responsible for the first phase of growth inhibition. In response to decreasing EBF concentration, EIN3 concentration increases from an initially low value approximately 30 min after ethylene is introduced. Once EIN3 reaches a sufficiently high concentration, a pronounced second phase of growth inhibition begins. Thus, the coherent feedforward loop leads to the desired 2-PGI. The effect of the negative feedback loop can be understood by examining the dynamics of growth and GA. For the wildtype topology, GA is inhibited by both EIN3 and growth. A limited increase in GA levels accompanies the first phase of growth inhibition, and is driven by decreasing growth rates. During the second phase of growth inhibition, increasing EIN3 concentration inhibits GA, with EIN3 inhibition outcompeting the effect of decreasing growth rate. This drives GA to a low concentration, minimizing the effect of the negative feedback loop on growth. Thus, in the presence of increased EIN3, the effects of the negative feedback loop are suppressed. In the ein3;eil1 mutant, however, the inhibitory action of EIN3 on the negative feedback loop is lost. Thus, the negative feedback loop plays a more prominent role since GA is not inhibited by EIN3 (Figure 5C). Additionally, the indirect path of growth inhibition is removed, eliminating the second phase of growth inhibition. Figure 5C shows the response of key network components under these conditions. When increasing EIN2 levels cause a decrease in growth, GA levels increase in response, promoting growth and leading to partial growth recovery. This illustrates the importance of the negative feedback loop for partial growth recovery in the ein3;eil1 mutant. After the removal of ethylene at 3 h, it is interesting to note that the average growth rate does not exhibit a large overshoot compared with pre-ethylene levels (Figure 5B). When analyzing individual parameter sets, none of the responses exhibit an overshoot that exceeds pre-ethylene levels by more than 10%, only 5 of 1344 exhibit >5% overshoot, and only 44 of 1344 exhibit >1% overshoot. This is in contrast with experimental results and suggests that modifications of the network or additional components might be needed to adequately capture the overshoot behavior. However, as discussed above, our results show that the core CFF/NFB topology generates key features of the 2-PGI and MPGR responses. 3.1.2. Analysis of parameter sets The roles of the feedforward and feedback loops can be further understood by examining evolved parameters. In particular, it is instructive to characterize the distributions of evolved parameter values for the CFF/NFB network, as certain parameters are constrained to small ranges or excluded from certain parameter regimes. In Figure 6, we compare the distributions of select parameters when the evolved parameter sets are categorized by their behavior (both 2-PGI and MPGR, only 2-PGI, or only MPGR). The distributions of all parameters are shown in Supplemental Figure S5. Parameter values are scaled by normalizing the maximum value to one, and the width of each distribution is scaled such that the maximum width is equal in each distribution. Comparing the distributions for specific parameters highlights key network features leading to each response. Figure 6 Distributions of parameters from evolved sets of CFF/NFB network parameters. Parameter sets exhibiting different combinations of responses are shown. Parameter labels K and k indicate activation coefficients and rate constants, respectively, that are associated with the ODEs governing the species labeled above each figure. Subscripts indicate if the parameter regulates degradation (degr) or production (prod) and superscripts indicate the network component regulating the reaction. Basal rates indicate that the parameter is not regulated by another network component. All parameters were unit normalized using range rescaling. The parameter distributions in Figure 6 provide additional evidence that the feedforward loop plays a key role in generating 2-PGI. The activation coefficient for EIN2-regulated growth degradation of EBF (KdegrEIN2) is excluded from low values in evolved parameter sets exhibiting 2-PGI. As a consequence, EIN2 concentration must reach high levels to significantly inhibit EBF, which contributes to a delay before the second phase of growth inhibition. Interestingly, this parameter is most significantly constrained in evolved parameter sets exhibiting both 2-PGI and MPGR. This is in contrast with the broader distributions seen for the cases exhibiting only one of the targeted responses. Additionally, in parameter sets exhibiting 2-PGI, the rate constant associated with basal production of EIN3 (kprodbasal) occurs at low values. This also contributes to a time delay in the feedforward loop, which is needed for a second phase of growth inhibition. It is also informative to consider the parameters governing the negative feedback loop (Figure 6). Parameter sets exhibiting MPGR have highly restricted ranges associated with the rate constant for basal production of GA and the parameters for GA inhibition by growth. These restrictions lead to low pre-ethylene levels of GA and a reasonable response of GA as growth declines following ethylene exposure. In cases exhibiting MPGR, the rate constant governing promotion of growth by GA is also restricted to high values and the activation coefficient for the promotion of growth by GA is excluded from the lowest values. Thus, a moderate increase in GA concentration will result in a significant increase in growth. However, excluding the activation coefficient from low values prevents increases from occurring with small changes in GA. Thus, tight regulation of parameters of the negative feedback loop is most readily apparent in parameter sets exhibiting MPGR. This further suggests the importance of the negative feedback loop for MPGR. 3.1.3. Ethylene dose response Given that our network parameters were evolved to target only 2-PGI and MPGR behavior, we were interested in whether other experimentally observed behavior emerged as well. As such, we examined the sub-saturating ethylene dose-response behavior of evolved parameter sets exhibiting both 2-PGI and MPGR. Figure 7A shows an example of sub-saturating ethylene growth recovery (SSGR) that passes our screening criteria. Figure 7B shows an example of a typical growth response failing to exhibit SSGR behavior. Here, there is no growth recovery observed at any ethylene concentration. Of the evolved parameter sets exhibiting both 2-PGI and MPGR, 26% also exhibit SSGR. Partial growth recovery at large sub-saturating ethylene concentrations was observed experimentally (e.g., at 100 ppb in Figure 2C) but was not observed in the CFF/NFB network model. However, the observed adaptive behavior occurring at lower ethylene concentrations is qualitatively consistent with experiments. This emergent property provides additional support for the proposed CFF/NFB network topology. Figure 7 Characteristic growth responses at sub-saturating ethylene doses. Parameter sets exhibiting both 2-PGI and MPGR were screened for partial growth recovery to sustained sub-saturating ethylene doses (SSGR). Figures show typical growth responses for parameter sets: (A) passing SSGR screening and (B) failing SSGR screening. To gain insight into features that lead to SSGR, we compared the parameter distributions that passed SSGR screening to those that failed SSGR screening. Surprisingly, this revealed nearly identical parameter distributions except for the activation coefficient for GA inhibition by EIN3 and the rate constant associated with inhibition of growth by EIN3 (Figure 8 and Supplemental Figure S6). The activation coefficient regulating GA inhibition by EIN3 occurs at higher values in sets producing SSGR. The rate constant for growth inhibition by EIN3 occurs at lower values more frequently in cases giving SSGR. The distributions of these parameters across all evolved sets exhibiting 2-PGI and/or MPGR are shown in Figure 6. Higher values of the activation coefficient for GA inhibition by EIN3 are found primarily in parameter sets exhibiting 2-PGI, while lower values of the rate constants for inhibition of growth by EIN3 occur primarily in parameter sets exhibiting MPGR. These restrictions apply to the regulation of EIN3 on components of the negative feedback loop. This again suggests that inhibition of the negative feedback loop by the coherent feedforward loop may play a key role in ethylene signaling. Figure 8 Distributions of parameters from the CFF/NFB network screened for SSGR behavior. A comparison of parameter distributions passing and failing SSGR screening (screened parameter sets exhibit both 2-PGI and MPGR). 3.1.4. Simplified CFF/NFB networks We have shown that the CFF/NFB network can produce multiple experimentally-observed features of Arabidopsis growth responses to ethylene. Using this network topology as a guide, we probed simplified networks containing coherent feedforward and negative feedback motifs. The networks explored are shown in Figure 9. Ethylene (E) acts as either the first node of the coherent feedforward loop (Figures 9A–C) or as a direct input into the first node of the loop (Figures 9D,E). Additionally, EBF and EIN3 are combined into a single node (Y) which acts to inhibit growth. In the simplest network (Figure 9A), we remove the GA node and allow growth to directly inhibit its own production. For the remaining networks, the role of GA in the negative feedback loop is performed by node Z. To probe the inhibition of the negative feedback loop by the coherent feedforward loop, we tested network topologies with and without the inhibition of Z by Y. Approximately 500 independent optimization runs were performed for each simplified network topology. Ein3;eil1 mutants were simulated by eliminating node Y. Evolved parameter sets were screened for 2-PGI and MPGR responses and a summary of results are shown in Table 1. Figure 9 Simplified CFF/NFB networks. Simplified networks tested. (A–E) Only networks (D,E) exhibit both 2-PGI and MPGR. Table 1 Screening results for ethylene growth responses of simplified CFF/NFB networks. Network Total 2-PGI & MPGR 2-PGI only MPGR only A 504 0 0 0 B 500 0 2 0 C 500 0 80 18 D 499 5 1 3 E 500 92 73 37 The simplest network (Figure 9A) failed to produce any parameter sets passing 2-PGI or MPGR screening procedures. Examining the dynamical response of evolved parameter sets revealed two phases of growth inhibition that occurred too early and above the desired growth range. Additionally, no growth recovery was observed upon removal of Y. The addition of node Z to the negative feedback loop (Figure 9B) produced two parameter sets exhibiting 2-PGI but none showing MPGR. When the inhibition of Z by Y is included (Figure 9C), we begin to observe substantial numbers of parameter sets exhibiting either 2-PGI or MPGR. However, no parameter sets simultaneously produced both responses. 2-PGI and MPGR were observed together only when ethylene promoted the first node of the coherent feedforward cascade, which more closely mimics the initial linear signaling cascade. In networks in which Z is not directly inhibited by Y (Figure 9D), 1.0% of parameter sets exhibit both 2-PGI and MPGR responses. When Y regulates Z (Figure 9E), 18.4% of parameter sets exhibit both targeted growth responses. These results suggest the importance of (i) the initial linear cascade in achieving proper timing of growth inhibition and (ii) the inhibition of negative feedback by the coherent feedforward loop in expanding the parameter space in which 2-PGI and MPGR are observed. 3.2. EIN2 cleavage with positive feedback In this section, we consider the second proposed network topology with EIN2 cleavage and a positive feedback loop (PFB network, Figure 3B). The network was evolved with the same target responses as before. We obtained evolved parameter sets that produced both 2-PGI and MPGR behavior, but the number was substantially lower than in the CFF/NFB network. Out of 1247 independent runs of the EA, only 5 evolved parameter sets produce both 2-PGI and MPGR. Parameter sets exhibiting only 2-PGI were also uncommon (3 sets). However, a significant proportion of parameter sets exhibited only MPGR (726 sets). Two of the 5 parameter sets exhibiting both 2-PGI and MPGR also display sub-saturating ethylene growth response (SSGR). The limited number of parameter sets exhibiting both targeted responses precludes analysis of parameter distributions. However, studying the dynamic response of the best-performing parameter set exhibiting 2-PGI, MPGR, and SSGR provides valuable insight (Figure 10). Results are representative of the other parameter sets exhibiting 2-PGI and MPGR. Complete results of evolved sets exhibiting 2-PGI, MPGR, and SSGR are presented in the Supplemental Figure S7. Figure 10 Time evolution of PFB network components. Figures show the behavior of the best-performing evolved parameter set that passed 2-PGI, MPGR, and SSGR screening. (A) Response of growth, EIN2, and EIN2-C (wildtype conditions). (B) Response of components in the positive feedback loop (wildtype conditions). (C) Response of growth, EIN2, and EIN2-C (ein3;eil1 mutant conditions). 3.2.1. Dynamical response of network components As in the CFF/NFB network, the introduction of ethylene decreases CTR1 levels (Figure 10A). In the PFB network, the cleavage of EIN2 is no longer inhibited and EIN2-C is produced (Figure 10A). As EIN2-C increases in concentration it inhibits both growth and EBF (Figures 10A,B). EIN2 levels drop during this phase of network response as basal production of EIN2 cannot compensate for the rapid conversion of EIN2 to EIN2-C. EIN2 reaches low concentrations, limiting the resources available for production of EIN2-C. This leads to a transient decline in EIN2-C, which causes the plateau-like region of growth inhibition. Within the feedback loop, lower EBF levels decrease the inhibition of EIN3, which rises and promotes production of EIN2. The rapid rise of EIN2 provides more resources for EIN2-C production. This leads to the second phase of growth inhibition. Within the ein3;eil1 mutant, the positive feedback loop is absent. The addition of ethylene leads to EIN2 being converted to EIN2-C, resulting in a decline of EIN2. Without the positive feedback loop, there is no mechanism to further increase EIN2 production and its concentration monotonically decreases. EIN2-C initially increases but then declines as basal degradation eventually dominates the low rates of EIN2-C production associated with low levels of EIN2. As EIN2-C concentration decreases, partial growth recovery is observed (Figure 10C). 3.2.2. Simplified PFB network We again explored a simplified network topology that keeps key features of the PFB network. We found that a four component network in which ethylene directly promotes the conversion of X (EIN2) to Y (EIN2-C) can produce both 2-PGI and MPGR (Figure 11). In the network, Y directly inhibits growth and promotes the production of X. We performed 500 optimizations of this network and obtained 142 evolved parameter sets exhibiting both 2-PGI and MPGR. The marked increase in the fraction of parameter sets exhibiting both 2-PGI and MPGR suggests parameter evolution in the full PFB network is hindered by interactions in the positive feedback loop. Three parameters regulate the positive feedback from Y to X in the simplified network, while 9 parameters govern the positive feedback loop in the full network (associated with interactions between EIN2-C, EBF, EIN3, and EIN2). This apparently makes it difficult for our EA to evolve large numbers of parameter sets producing both 2-PGI and MPGR. The large fraction of simplified PFB networks exhibiting both 2-PGI and MPGR again provides support for EIN2 cleavage with positive feedback as a viable network topology. Figure 11 Simplified PFB network. Network diagram of the minimal PFB network. The table enumerates results of growth-response screening for evolved parameter sets. 4. Conclusions We used computational methods to explore hypothesized network topologies underlying ethylene signaling responses in Arabidopsis. We focused on two core networks that are topologically distinct. Using an evolutionary algorithm to explore parameter space, we showed that both network topologies can produce dynamical responses consistent with experimental time-dependent growth data. The core topologies are (i) a coherent feedforward loop that inhibits growth and a negative feedback from growth onto itself (CFF/NFB), and (ii) a network in which ethylene promotes the cleavage of EIN2, with the product of the cleavage inhibiting growth and promoting the production of EIN2 through a positive feedback loop (PFB). For the CFF/NFB network, high-throughput use of the evolutionary algorithm led to a large number of parameter sets producing responses consistent with experimental growth kinetics under various conditions and genotypes. The results emphasize the importance of various network features for regulating dynamic responses. For example, the two branches of the coherent feedforward loop collectively produce two-phase growth inhibition (2-PGI), and the negative feedback loop is critical for mutant partial growth recovery (MPGR). Our study additionally suggests that 2-PGI and MPGR coexist in a broader parameter regime when the negative feedback loop is suppressed by an intermediate component of the coherent feedforward cascade. The large number of parameter sets producing 2-PGI and MPGR behavior provide insight into important regimes of parameter space. Additionally, a large fraction of these parameter sets also exhibit sub-saturating ethylene growth response (SSGR), even though this was not a targeted response by the evolutionary algorithm. Taken together, these results provide support for the CFF/NFB network as a viable network topology underlying ethylene signaling. For the PFB network, the evolutionary algorithm led to far fewer parameter sets producing both 2-PGI and MPGR behavior, yet the dynamics of their responses provided insight into the mechanisms underlying the network topology. A key feature of the network is that EIN2 is converted to EIN2-C and its transient depletion upon the addition of ethylene is responsible for the plateau phase of growth inhibition. Two of the evolved parameter sets also exhibited SSGR, indicating that this emergent behavior is also possible in the PFB network. Although we generated far more parameter sets producing 2-PGI and MPGR for the CFF/NFB network, this does not necessarily imply that it is biologically more likely. For example, the region of parameter space for the PFB network that gives the desired behavior may be smaller or more difficult to identify with our EA, but this does not exclude the PFB network as biologically feasible. It is interesting to note that different plant species have qualitatively different ethylene response kinetics (Kim et al., 2012). For example, some plant species (millet) have only a transient first phase response and some (rice) have only a prolonged second phase response. The paper by Kim et al. first proposed the CFF/NFB network studied here. For millet, Kim et al. proposed that the circuit controlled by EIN3/EIL1 was missing to give the transient response; for rice, it was proposed that the rapid, EIN3/EIL1-independent output of EIN2 is missing. The first case was analyzed in this paper when we analyzed the MPGR response. An interesting feature of the CFF/NFB model is that there is a simple conceptual way to modify the network to generate responses consistent with other species. It is less clear how the PFB network could be modified in an analogous manner to generate growth response kinetics consistent with the rice and millet studies. Further exploration of the network topology across species is an interesting area for future exploration. Even though both models exhibited SSGR behavior that was similar to what has been observed experimentally, the kinetics of the computational responses are subtly different from experimental observations. In particular, there was no long-time recovery at high sub-saturating ethylene concentrations and incomplete recovery at low concentrations. It has been suggested that responses to low levels of ethylene are in large part a result of receptor clustering, where ligand occupancy of one receptor affects the signaling state of surrounding receptors through direct interactions and results in signal amplification at low ethylene levels (Gamble et al., 2002; Binder and Bleecker, 2003; Binder et al., 2004b). Computational models invoking receptor clustering indicate this element can affect both sensitivity and adaptation (Bray et al., 1998). Our models did not incorporate this feature, which would likely affect features of the SSGR. Additionally, our models do not incorporate spatial information. For instance, it is now known that EIN2-C translocates to the nucleus to affect ethylene signaling (Ju et al., 2012; Qiao et al., 2012; Wen et al., 2012). Cleavage of EIN2 was not incorporated into the CFF/NFB network and translocation of EIN2-C was not explicitly incorporated into either model. This translocation also may have diverse functions since it has recently been found that EIN2-C in the cytosol also has a role in ethylene signaling (Li et al., 2015; Merchante et al., 2015). It is likely that spatial changes in important components such as EIN2-C have a role in adaptation. Despite these differences, our calculations show that several simple networks can recapitulate the ethylene growth responses observed experimentally. The dynamic responses observed provide opportunities for experimental exploration. A comparison of the dynamical response of individual components for each network is shown in Supplemental Figure S7. For example, the PFB network shows that when ethylene is added there is a transient decrease in EIN2 levels followed by accumulation of EIN2. By contrast, the CFF/NFB model predicts qualitatively different accumulation kinetics for EIN2, with no transient decrease. Thus, one avenue of experimentation can be to obtain more detailed spatio-temporal information about the accumulation of EIN2 (and EIN2-C) to determine if the details predicted by the calculations in either model occur when saturating levels of ethylene are added. For example, a detailed time-course of EIN2-C accumulation or EIN2 full-length protein is lacking. Such information would help determine which, if either, model correctly predicts the accumulation pattern for EIN2. Additionally, removing EIN3 from the CFF/NFB model has minimal effect on the time-course of EIN2 accumulation when ethylene is added, but has a profound effect on both EIN2 and EIN2-C levels in the PFB model. Thus, experiments examining EIN2 and EIN2-C levels in ein3;eil1 double mutants would also be informative. Another example is the involvement of GA in the CFF/NFB network where it plays a larger role in the growth kinetics observed in the ein3;eil1 mutants. Detailed information about changes in GA levels would provide a test of this model and whether the negative feedback loop needs to be incorporated into the PFB network. Such experimental details will help determine which network topology, if either, could serve as the ethylene signaling transduction network of Arabidopsis. It is also possible that a combination of the two models or different network topologies will yield emergent properties that are closer to experimental observations. Additional experimental details about the spatio-temporal changes that occur in each component of the pathway will allow us to refine the above models or develop additional network topologies. In summary, these calculations show that a basic mechanistic understanding of ethylene growth response and recovery kinetics is possible without detailed knowledge of the molecular mechanisms or enzymatic kinetic parameters. Given that ethylene signal transduction has been highly studied for several decades, we anticipate that major advances in our understanding about this pathway will be to provide details about network interactions, reaction kinetics, and changes in the spatial distribution of proteins in the pathway. Our hope is that with more refined experimental input, we can refine the network models to provide insights into how plants respond to ethylene. Author contributions AP, BB, and SA designed research. AP, FM, and BE performed research. AP, BB, and SA analyzed data and wrote the paper. Funding This work was funded by NSF Grants (IOS-1254423, MCB-1517032) to BB. Conflict of interest statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Supplementary material The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.01308 Click here for additional data file. ==== Refs References Abeles F. B. Morgan P. W. Saltveit M. E. Jr. (1992 ). Ethylene in Plant Biology . San Diego, CA : Academic Press . Alonso J. M. Hirayama T. Roman G. Nourizadeh S. Ecker J. R. (1999 ). EIN2, a bifunctional transducer of ethylene and stress responses in arabidopsis . Science 284 , 2148 –2152 . 10.1126/science.284.5423.2148 10381874 An F. Zhao Q. Ji Y. Li W. Jiang Z. Yu X. . (2010 ). Ethylene-induced stabilization of ETHYLENE INSENSITIVE3 and EIN3-LIKE1 is mediated by proteasomal degradation of EIN3 binding F-box 1 and 2 that requires EIN2 in arabidopsis . Plant Cell 22 , 2384 –2401 . 10.1105/tpc.110.076588 20647342 Arisi I. Cattaneo A. Rosato V. (2006 ). Parameter estimate of signal transduction pathways . BMC Neurosci. 7 (Suppl. 1 ):S6 . 10.1186/1471-2202-7-S1-S6 17118160 Auliac C. Frouin V. Gidrol X. d'Alché Buc F. (2008 ). Evolutionary approaches for the reverse-engineering of gene regulatory networks: a study on a biologically realistic dataset . BMC Bioinformatics 9 :91 . 10.1186/1471-2105-9-91 18261218 Bäck T. Schwefel H.-P. (1993 ). An overview of evolutionary algorithms for parameter optimization . Evol. Comp. 1 , 1 –23 . 10.1162/evco.1993.1.1.1 Bakshi A. Wilson R. L. Lacey R. F. Kim H. Wupalapapati S. K. Binder B. (2015 ). Identification of regions in the receiver domain of the ETHYLENE RESPONSE1 ethylene receptor of arabidopsis important for functional divergence . Plant Physiol. 169 , 219 –232 . 10.1104/pp.15.00626 26160962 Binder B. M. Bleecker A. B. (2003 ). A model for ethylene receptor function and 1-methylcyclopropene action . Acta Hortic. 628 , 177 –187 . 10.17660/ActaHortic.2003.628.21 Binder B. M. Mortimore L. A. Stepanova A. N. Ecker J. R. Bleecker A. B. (2004a ). Short-term growth responses to ethylene in arabidopsis seedlings are EIN3/EIL1 independent . Plant Physiol. 136 , 2921 –2927 . 10.1104/pp.104.050393 15466219 Binder B. M. O'Malley R. C. Wang W. Moore J. M. Parks B. M. Spalding E. P. . (2004b ). Arabidopsis seedling growth response and recovery to ethylene. A kinetic analysis . Plant Physiol. 136 , 2913 –2920 . 10.1104/pp.104.050369 15466220 Binder B. M. Walker J. M. Gagne J. M. Emborg T. J. Hemmann G. Bleecker A. B. . (2007 ). The arabidopsis EIN3 binding F-box proteins EBF1 and EBF2 have distinct but overlapping roles in ethylene signaling . Plant Cell 19 , 509 –523 . 10.1105/tpc.106.048140 17307926 Bleecker A. B. Estelle M. A. Somerville C. Kende H. (1988 ). Insensitivity to ethylene conferred by a dominant mutation in Arabidopsis thaliana . Science 241 , 1086 –1089 . 10.1126/science.241.4869.1086 17747490 Bray D. Lay S. (1994 ). Computer simulated evolution of a network of cell-signaling molecules . Biophys. J. 66 , 972 –977 . 10.1016/S0006-3495(94)80878-1 8038401 Bray D. Levin M. D. Morton-Firth C. J. (1998 ). Receptor clustering as a cellular mechanism to control sensitivity . Nature 393 , 85 –88 . 10.1038/30018 9590695 Burg S. P. (1973 ). Ethylene in plant growth . Proc. Natl. Acad. Sci. U.S.A. 70 , 591 –597 . 10.1073/pnas.70.2.591 16592065 Chen R. Binder B. M. Garrett W. M. Tucker M. L. Chang C. Cooper B. (2011 ). Proteomic responses in Arabidopsis thaliana seedlings treated with ethylene . Mol. Biosyst. 7 , 2637 –2650 . 10.1039/c1mb05159h 21713283 Christians M. J. Gingerich D. J. Hansen M. Binder B. M. Kieber J. J. Vierstra R. D. (2009 ). The BTB ubiquitin ligases ETO1, EOL1 and EOL2 act collectively to regulate ethylene biosynthesis in arabidopsis by controlling type2 ACC synthase levels . Plant J. 57 , 332 –345 . 10.1111/j.1365-313X.2008.03693.x 18808454 Díaz J. Álvarez-Buylla E. R. (2006 ). A model of the ethylene signaling pathway and its gene response in Arabidopsis thaliana: pathway cross-talk and noise-filtering properties . Chaos 16 :023112 . 10.1063/1.2189974 16822015 Díaz J. Alvarez-Buylla E. R. (2009 ). Information flow during gene activation by signaling molecules: ethylene transduction in arabidopsis cells as a study system . BMC Syst. Biol. 3 :48 . 10.1186/1752-0509-3-48 19416539 Feng S. Ollivier J. F. Swain P. S. Soyer O. S. (2015 ). Biojazz: in silico evolution of cellular networks with unbounded complexity using rule-based modeling . Nucleic Acids Res. 43 :e213 . 10.1093/nar/gkv595 26101250 François P. Hakim V. (2004 ). Design of genetic networks with specified functions by evolution in silico . Proc. Natl. Acad. Sci. U.S.A. 101 , 580 –585 . 10.1073/pnas.0304532101 14704282 Gagne J. M. Smalle J. Gingerich D. J. Walker J. M. Yoo S.-D. Yanagisawa S. . (2004 ). Arabidopsis EIN3-binding F-box 1 and 2 form ubiquitin-protein ligases that repress ethylene action and promote growth by directing EIN3 degradation . Proc. Natl. Acad. Sci. U.S.A. 101 , 6803 –6808 . 10.1073/pnas.0401698101 15090654 Gamble R. L. Qu X. Schaller G. E. (2002 ). Mutational analysis of the ethylene receptor ETR1. Role of the histidine kinase domain in dominant ethylene insensitivity . Plant Physiol. 128 , 1428 –1438 . 10.1104/pp.010777 11950991 Gao Z. Wen C.-K. Binder B. M. Chen Y.-F. Chang J. Chiang Y.-H. . (2008 ). Heteromeric interactions among ethylene receptors mediate signaling in arabidopsis . J. Biol. Chem. 283 , 23801 –23810 . 10.1074/jbc.M800641200 18577522 Goeschl J. D. Kays S. J. (1975 ). Concentration dependencies of some effects of ethylene on etiolated pea, peanut, bean, and cotton seedlings . Plant Physiol. 55 , 670 –677 . 10.1104/pp.55.4.670 16659145 Guo H. Ecker J. R. (2003 ). Plant responses to ethylene gas are mediated by SCF EBF1/EBF2-dependent proteolysis of EIN3 transcription factor . Cell 115 , 667 –677 . 10.1104/pp.55.4.670 14675532 Guzman P. Ecker J. R. (1990 ). Exploiting the triple response of arabidopsis to identify ethylene-related mutants . Plant Cell 2 , 513 –523 . 10.1105/tpc.2.6.513 2152173 Hall A. E. Bleecker A. B. (2003 ). Analysis of combinatorial loss-of-function mutants in the arabidopsis ethylene receptors reveals that the ERS1 ETR1 double mutant has severe developmental defects that are EIN2 dependent . Plant Cell 15 , 2032 –2041 . 10.1105/tpc.013060 12953109 Jackson M. B. (1983 ). Regulation of root growth and morphology by ethylene and other externally applied growth substances , in Growth Regulators in Root Development , Monograph No. 10, eds Jackson M. B. Stead A. D. (London : British Plant Growth Regulator Group ), 103 –116 . Ju C. Yoon G. M. Shemansky J. M. Lin D. Y. Ying Z. I. Chang J. . (2012 ). CTR1 phosphorylates the central regulator EIN2 to control ethylene hormone signaling from the er membrane to the nucleus in arabidopsis . Proc. Natl. Acad. Sci. U.S.A. 109 , 19486 –19491 . 10.1073/pnas.1214848109 23132950 Kieber J. J. Rothenberg M. Roman G. Feldmann K. A. Ecker J. R. (1993 ). CTR1, a negative regulator of the ethylene response pathway in arabidopsis, encodes a member of the raf family of protein kinases . Cell 72 , 427 –441 . 10.1016/0092-8674(93)90119-B 8431946 Kikuchi S. Tominaga D. Arita M. Takahashi K. Tomita M. (2003 ). Dynamic modeling of genetic networks using genetic algorithm and s-system . Bioinformatics 19 , 643 –650 . 10.1093/bioinformatics/btg027 12651723 Kim H. Helmbrecht E. E. Stalans M. B. Schmitt C. Patel N. Wen C.-K. . (2011 ). Ethylene receptor ethylene receptor1 domain requirements for ethylene responses in arabidopsis seedlings . Plant Physiol. 156 , 417 –429 . 10.1104/pp.110.170621 21386032 Kim J. Wilson R. L. Case J. B. Binder B. M. (2012 ). A comparative study of ethylene growth response kinetics in eudicots and monocots reveals a role for gibberellin in growth inhibition and recovery . Plant Physiol. 160 , 1567 –1580 . 10.1104/pp.112.205799 22977279 Laan P. (1934 ). Der Einfluss von Aethylen auf die Wuchsstoffbildung bei Avena und Vicia . Ph.D. thesis, Utrecht : Van der Laan . Larsen P. B. Chang C. (2001 ). The arabidopsis EER1 mutant has enhanced ethylene responses in the hypocotyl and stem . Plant Physiol. 125 , 1061 –1073 . 10.1104/pp.125.2.1061 11161061 Li W. Ma M. Feng Y. Li H. Wang Y. Ma Y. . (2015 ). EIN2-directed translational regulation of ethylene signaling in arabidopsis . Cell 163 , 670 –683 . 10.1016/j.cell.2015.09.037 26496607 Mattoo A. K. Suttle J. C. (1991 ). The Plant Hormone Ethylene. Boca Raton, FL : CRC Press . McDaniel B. K. Binder B. M. (2012 ). Ethylene receptor 1 (ETR1) is sufficient and has the predominant role in mediating inhibition of ethylene responses by silver in Arabidopsis thaliana . J. Biochem. 287 , 26094 –26103 . 10.1074/jbc.M112.383034 22692214 Merchante C. Brumos J. Yun J. Hu Q. Spencer K. R. Enríquez P. . (2015 ). Gene-specific translation regulation mediated by the hormone-signaling molecule EIN2 . Cell 163 , 684 –697 . 10.1016/j.cell.2015.09.036 26496608 Patil K. R. Rocha I. Förster J. Nielsen J. (2005 ). Evolutionary programming as a platform for in silico metabolic engineering . BMC Bioinformatics 6 :308 . 10.1186/1471-2105-6-308 16375763 Potuschak T. Lechner E. Parmentier Y. Yanagisawa S. Grava S. Koncz C. . (2003 ). EIN3-dependent regulation of plant ethylene hormone signaling by two arabidopsis F box proteins: EBF1 and EBF2 . Cell 115 , 679 –689 . 10.1016/S0092-8674(03)00968-1 14675533 Potuschak T. Vansiri A. Binder B. M. Lechner E. Vierstra R. D. Genschik P. (2006 ). The exoribonuclease XRN4 is a component of the ethylene response pathway in arabidopsis . Plant Cell 18 , 3047 –3057 . 10.1105/tpc.106.046508 17085683 Qiao H. Chang K. N. Yazaki J. Ecker J. R. (2009 ). Interplay between ethylene, ETP1/ETP2 F-box proteins, and degradation of EIN2 triggers ethylene responses in arabidopsis . Gene Dev. 23 , 512 –521 . 10.1101/gad.1765709 19196655 Qiao H. Shen Z. Huang S.-S. C. Schmitz R. J. Urich M. A. Briggs S. P. . (2012 ). Processing and subcellular trafficking of er-tethered ein2 control response to ethylene gas . Science 338 , 390 –393 . 10.1126/science.1225974 22936567 Qiu L. Xie F. Yu J. Wen C.-K. (2012 ). Arabidopsis RTE1 is essential to ethylene receptor ETR1 amino-terminal signaling independent of CTR1 . Plant Physiol. 159 , 1263 –1276 . 10.1104/pp.112.193979 22566492 Rai M. I. Wang X. Thibault D. M. Kim H. J. Bombyk M. M. Binder B. M. . (2015 ). The argos gene family functions in a negative feedback loop to desensitize plants to ethylene . BMC Plant Biol. 15 :157 . 10.1186/s12870-015-0554-x 26105742 Rauser W. E. Horton R. F. (1975 ). Rapid effects of indoleacetic acid and ethylene on the growth of intact pea roots . Plant Physiol. 55 , 443 –447 . 10.1104/pp.55.3.443 16659098 Roman G. Lubarsky B. Kieber J. J. Rothenberg M. Ecker J. R. (1995 ). Genetic analysis of ethylene signal transduction in Arabidopsis thaliana: five novel mutant loci integrated into a stress response pathway . Genetics 139 , 1393 –1409 . 7768447 Spirov A. Holloway D. (2013 ). Using evolutionary computations to understand the design and evolution of gene and cell regulatory networks . Methods 62 , 39 –55 . 10.1016/j.ymeth.2013.05.013 23726941 Sun J. Garibaldi J. M. Hodgman C. (2012 ). Parameter estimation using metaheuristics in systems biology: a comprehensive review . IEEE ACM Trans. Comput. Biol. Bioinf. 9 , 185 –202 . 10.1109/TCBB.2011.63 van Zanten M. Basten Snoek L. van Eck-Stouten E. Proveniers M. C. Torii K. U. Voesenek L. A. . (2010 ). Ethylene-induced hyponastic growth in Arabidopsis thaliana is controlled by erecta . Plant J. 61 , 83 –95 . 10.1111/j.1365-313X.2009.04035.x 19796369 Vandenbussche F. Petrášek J. Žádníková P. Hoyerová K. Pešek B. Raz V. . (2010 ). The auxin influx carriers AUX1 and LAX3 are involved in auxin-ethylene interactions during apical hook development in arabidopsis thaliana seedlings . Development 137 , 597 –606 . 10.1242/dev.040790 20110325 Warner H. Leopold A. (1971 ). Timing of growth regulator responses in peas . Biochem. Biophys. Res. Commun. 44 , 989 –994 . 10.1016/0006-291X(71)90809-6 5125238 Wen X. Zhang C. Ji Y. Zhao Q. He W. An F. . (2012 ). Activation of ethylene signaling is mediated by nuclear translocation of the cleaved EIN2 carboxyl terminus . Cell Res. 22 , 1613 –1616 . 10.1038/cr.2012.145 23070300 Yanagisawa S. Yoo S.-D. Sheen J. (2003 ). Differential regulation of ein3 stability by glucose and ethylene signalling in plants . Nature 425 , 521 –525 . 10.1038/nature01984 14523448 Žádníková P. Petrášek J. Marhavý P. Raz V. Vandenbussche F. Ding Z. . (2010 ). Role of pin-mediated auxin efflux in apical hook development of Arabidopsis thaliana . Development 137 , 607 –617 . 10.1242/dev.041277 20110326

PMC005xxxxxx/PMC5003822.txt

==== Front Front MicrobiolFront MicrobiolFront. Microbiol.Frontiers in Microbiology1664-302XFrontiers Media S.A. 10.3389/fmicb.2016.01372MicrobiologyOriginal ResearchRapid and Sensitive Detection of Didymella bryoniae by Visual Loop-Mediated Isothermal Amplification Assay Yao Xiefeng Li Pingfang Xu Jinghua Zhang Man Ren Runsheng Liu Guang Yang Xingping *Institute of Vegetable Crops, Jiangsu Academy of Agricultural Sciences/Jiangsu Key Laboratory for Horticultural Crop Genetic ImprovementNanjing, ChinaEdited by: Vijai Kumar Gupta, National University of Ireland, Galway, Ireland Reviewed by: Yun Chen, Zhejiang University, China; Janmeajy Pandey, Central University of Rajasthan, India *Correspondence: Xingping Yang, xingping@jaas.ac.cnThis article was submitted to Fungi and Their Interactions, a section of the journal Frontiers in Microbiology 30 8 2016 2016 7 137215 6 2016 18 8 2016 Copyright © 2016 Yao, Li, Xu, Zhang, Ren, Liu and Yang.2016Yao, Li, Xu, Zhang, Ren, Liu and YangThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.Didymella bryoniae is a pathogenic fungus that causes gummy stem blight (GSB) in Cucurbitaceae crops (e.g., cantaloupe, muskmelon, cucumber, and watermelon). GSB produces lesions on the stems and leaves, and can also be spread by seeds. Here, we developed a rapid, visual, and sensitive loop-mediated amplification (LAMP) assay for D. bryoniae detection based on sequence-characterized amplified regions (GenBank accession nos GQ872461 and GQ872462) common to the two random amplification of polymorphic DNA group genotypes (RGI and RGII) of D. bryoniae; ideal conditions for detection were optimized for completion in 45 min at 63°C. The sensitivity and specificity of the LAMP assay were further analyzed in comparison with those of a conventional polymerase chain reaction (PCR). The sensitivity of the LAMP assay was 1000-fold higher than that of conventional PCR with a detection limit of 0.1 fg μL-1 of targeted DNA. The LAMP assay could be accomplished in about 45 min, with the results visible to the naked eye. The assay showed high specificity in discriminating all D. bryoniae isolates from seven other fungal pathogens that occur in Cucurbitaceae crops. The LAMP assay also detected D. bryoniae infection in young muskmelon leaves with suspected early symptoms of GSB disease. Hence, the technique has great potential for developing rapid and sensitive visual detection methods for the D. bryoniae pathogen in crops and seeds. This method has potential application in early prediction of disease and reducing the risk of epidemics. gummy stem blightDidymella bryoniaemuskmelonloop-mediated isothermal amplificationprimer design ==== Body Introduction Gummy stem blight (GSB) is a highly prevalent disease on cucurbit crops (e.g., cantaloupe, muskmelon, watermelon, and cucumber) worldwide (Keinath, 2011; Li et al., 2015). The disease is caused by the ascomycete Didymella bryoniae (Fuckel) Rehm [anamorph Phoma cucurbitacearum (Fr.) Sacc.], synonym Stagonosporopsis cucurbitacearum (Fr.) Aveskamp et al. (2010), which produces necrotic spots on leaves and lesions on the vines and stems (Keinath et al., 1995; Somai et al., 2002; Keinath, 2014). D. bryoniae can also be spread by a low level of contaminated seeds (Somai et al., 2002; Sudisha et al., 2006; Keinath, 2011). GSB is one of the most important biotic constraints for muskmelon and watermelon production in moist environments such as plastic tunnels and greenhouses, leading to significant yield losses and a reduction in fruit quality in cucurbit crops in China (Wolukau et al., 2007; Li et al., 2015). D. bryoniae is well adapted to infect cucurbits, and can rapidly colonize the host tissue and reproduce abundantly in humid greenhouses. As reported previously, the pathogen has been found to move from infected source seedlings to adjacent seedlings in transplant greenhouses (Keinath, 1996, 2011). Thus, a low level of latently infected seedlings could potentially result in a major disease epidemic under favorable environmental conditions (Ling et al., 2010). Although cultural practices and fungicides play an important role in GSB management (Finger et al., 2014), they require diagnosis of the pathogen during early stages of disease development in cucurbit crop production. Thus, early diagnosis, sensitive and rapid detection of D. bryoniae is very important to limit the spread of GSB in plants being moved in greenhouses or in the field. The most common methods currently used for the rapid detection of D. bryoniae are based on several molecular detection tools. These include conventional polymerase chain reaction (PCR), PCR-enzyme-linked immunosorbent assay (PCR-ELISA; Keinath et al., 2001; Somai et al., 2002), and magnetic-capture hybridization multiplex real-time PCR, but these are only specific for detection of isolates of D. bryoniae in the random amplified polymorphic DNA (RAPD) group (RG) I genotype (Somai et al., 2002; Ha et al., 2009). Recently, to obtain a reliable, sensitive, and broad-spectrum diagnostic method for D. bryoniae isolates of both genotypes (RGI and RGII), an improved real-time PCR assay that is capable of detecting D. bryoniae isolates regardless of their genotype was developed (Ling et al., 2010), but the primer set has not been used in a direct seedling healthy assay. Among the currently used PCR assays, a high background in PCR-ELISA limits its adaptability and usefulness in general disease diagnosis (Ling et al., 2010), and PCR-based diagnostic procedures still have several intrinsic disadvantages, including the requirements for expensive laboratory instrumentation and reagents, and for appropriate training and technical expertise, which are often not available in poorly resourced laboratories and in rural areas of developing countries (Francois et al., 2011; Duan et al., 2014a; Moradi et al., 2014; Shen et al., 2016). Thus, there is a need to develop a straightforward, sensitive, rapid, and cost-effective method for the early diagnosis and in situ testing of D. bryoniae. Notomi et al. (2000) developed a novel DNA amplification technique named loop-mediated isothermal amplification (LAMP) that can be an effective method to address deficiencies of PCR-based methods, overcoming common limitations of current diagnostic methods (Notomi et al., 2000; Mori et al., 2013; Niessen, 2015). LAMP can rapidly amplify nucleic acids under isothermal conditions with a set of four to six specially designed primers and the large fragment of Bst DNA polymerase, which undergoes strand displacement activity to amplify target DNA in less than 60 min (Notomi et al., 2000; Nagamine et al., 2002b). The entire procedure is not difficult to perform and requires only an isothermal instrument, such as a water bath or heating block (Nagamine et al., 2002b). Additionally, the LAMP reaction is thought to have a higher tolerance to inhibitory substances than many PCR-based assays (Francois et al., 2011). LAMP products can easily be visualized by gel electrophoresis or by measuring turbidity caused by a white precipitate of magnesium pyrophosphate (Notomi et al., 2000; Nagamine et al., 2002b). Moreover, it can be monitored with the naked eye by adding colorimetric indicators, such as calcein, which produces a green fluorescent signal if the LAMP reaction is positive (Zhang et al., 2014). With these advantages, the LAMP method has been widely used for detection of plant pathogens, such as viruses (Hadersdorfer et al., 2011; Peng et al., 2012), bacteria (Hodgetts et al., 2015), nematodes (Kang et al., 2015), oomycetes (Hansen et al., 2016), and fungi (Shen et al., 2016). In recent years, the LAMP-based assay has grown in popularity for the detection of many plant-pathogenic fungi (Niessen and Vogel, 2010; Denschlag et al., 2012; Niessen, 2015). Taking these advantages into account, in this study the LAMP method was adapted to investigate latent infection and the early stages of disease in field samples infected with D. bryoniae. Although a very recent report has described a LAMP assay targeting the conserved RNA polymerase II RPB140 (RPB2) gene to detect D. bryoniae in cucurbit seeds (Tian et al., 2016), the detection sensitivity and efficiency were lower. In addition, the study focused only on detection of D. bryoniae infection in seed samples and not on infected samples in the field. The present study was undertaken to develop a LAMP assay for the detection of D. bryoniae based on targeting a sequence-characterized amplified region (SCAR; GenBank accession nos GQ872461 and GQ872462) of genomic DNA from the two genotypes (RGI and RGII) of D. bryoniae. The method was applied to detection of D. bryoniae from young muskmelon leaves with suspected early symptoms of GSB disease. The results demonstrated that this method is specific and efficient. This new LAMP assay will provide important reference data for monitoring and controlling GSB caused by D. bryoniae. Early and accurate detection of the causal agent (D. bryoniae) of GSB in cucurbit crops also could lead to reduced use of fungicides, thus benefiting the environment, and may reduce the risk of disease epidemics. Materials and Methods Ethics Statement Our study does not involve human specimens or tissue samples, or vertebrate animals, embryos or tissues. In our study, D. bryoniae isolates were collected from its host muskmelon or watermelon. The field is not protected in any way. The field study did not involve endangered or protected species. Fungal and Culture Conditions Five D. bryoniae isolates were obtained from muskmelon and watermelon plants that we collected from Jiangsu, Anhui, and Zhejiang provinces in east China. Seven other fungal pathogens of Cucurbitaceae crops were collected from muskmelon, watermelon, and gourd from Jiangsu province, and an Ascochyta pinodes isolate was obtained from a pea plant at Zhejiang province. The plant-pathogenic fungi used in this study, as well as their host, geographical origin, and RAPD group, are listed in Table 1. All isolates of fungal species were maintained in the collection of Jiangsu Academy of Agricultural Sciences. Fungal isolates were stored as monoconidial cultures grown on potato dextrose agar (PDA) plates (200 g potato, 20 g glucose, 15 g agar in 1 L water) at 25°C and were stored on PDA at 4°C. The sporulation and inoculum density of spores were prepared based on our previous study (Li et al., 2015). Table 1 Fungi isolates used in the conventional PCR and LAMP assay. Speciesa RAPD groupb Origin PCR productc Host Geographical Conventional PCR LAMP Didymella bryoniae (DBJSJY2) RGI Muskmelon Jiangsu + + Didymella bryoniae (DBAHHF2) RGI Muskmelon Anhui + + Didymella bryoniae (DBZJNB5) RGI Muskmelon Zhejiang + + Didymella bryoniae (DBJSNJ60) RGI Watermelon Jiangsu + + Didymella bryoniae (DBZJNB7) RGII Watermelon Jiangsu + + Ascochyta pinodes ZJ-1 – Pea Zhejiang - - Colletotrichum orbiculare NJ-1 – Watermelon Jiangsu - - Pythium paroecandrum Drechsler – Gourd Jiangsu - - Alternaria alternata LH1401 – Muskmelon Jiangsu - - Fusarium verticillioide – Gourd Jiangsu - - Fusarium oxysporum f.sp. niveum Race 0 – Watermelon Jiangsu - - Fusarium oxysporum f.sp. niveum Race 1 – Watermelon Jiangsu - - Fusarium oxysporum f.sp. niveum Race 2 – Watermelon Jiangsu - - aAll isolates of fungi species were maintained in the collection of Jiangsu Academy of Agricultural Sciences.bBased on the presence of a PCR product of the expected size for RAPD group assignments (RGI and RGII) using RG-specific SCAR primers (Supplementary Figure S1; Keinath et al., 2001; Babu et al., 2015).cNote that presence (+) or absence (-) are based on the presence of a PCR or LAMP product of the expected size.Reagents Bst DNA polymerase was purchased from New England Biolabs (Beijing) Ltd (Beijing, China). Calcein was bought from Sigma-Aldrich Co. LLC (Sigma, USA). Taq DNA polymerase, MnCl2, and dNTPs were purchased from TaKaRa Biotechnology (Dalian) Co., Ltd (Dalian, China). All other reagents were analytical grade and were purchased from Sinopharm Chemical Reagent Co., Ltd (Suzhou, China). Plant Materials Muskmelon (Cucumis melo L. var. ‘Japanese Sweet Babe’) plants containing three true leaves were inoculated by spraying with the spore suspension until the leaves were completely wet. The plants were then incubated at 25°C in a misted plastic tunnel at 90–100% relative humidity, as described previously (Li et al., 2015). DNA Extraction For DNA extraction, isolates were grown on dried filter paper disks on PDA plates for 7 days. Mycelia were then harvested as described previously (Ling et al., 2010). Muskmelon genomic DNA was extracted from the leaves. All genomic DNA was extracted using a DNeasy Plant Mini Kit (Qiagen, Valencia, CA, USA) according to the manufacturer’s instructions. The DNA was quantified using 1% (w/v) agarose gel electrophoresis and stored at -20°C for further use. LAMP Primer Design In a previous study, based on random amplification of polymorphic DNA (RAPD) markers, Ling et al. (2010) developed sequence characterized amplified regions (SCAR) primers (DB17 primer set) with broad-spectrum specificity that amplified a conserved sequence region common to both genotypes (RGI and RGII) of D. bryoniae. This SCAR, common to both genotypes of D. bryoniae, was identified in PCR products generated using the DB17 primer set (Table 2) (Ling et al., 2010). According to this conserved SCAR (GenBank accession nos GQ872461 and GQ872462), a set of LAMP primers, comprising two outer primers (DB17RG-F3 and DB17RG-B3), two inner primers (DB17RG-FIP and DB17RG-BIP), and one loop-backward primer (LDB17RG-LB) was designed using the LAMP primer software PrimerExplorer v.41 (Eiken Chemical Co., Japan) (Figure 1). All of the oligomers were synthesized and purified by Beijing Genomics Institute (Shanghai, China). The designed primer sequences for D. bryoniae are shown in Table 2. Table 2 Details of conventional PCR and LAMP primers used in this study. Primer Sequences (5′–3′) Reference Conventional PCR primers RG I-F TGTCGTTGACATCATTCCAGC Keinath et al., 2001; Somai et al., 2002 RG I-R ACCACTCTGCTTAGTATCTGC RG II-F GCTAAGCCTTAATCTAGCTGC Keinath et al., 2001; Somai et al., 2002 RG II-R GAGAGTAAGCTAACCTAAAGG DB17F GCAGTCAATCCTTATCC Ling et al., 2010 DB17R CGAAAGATTGTGTGACC LAMP primers This study DB17RG-F3 AGACCGCACTTTCGAGCT DB17RG-B3 GCGAACTGGCCAATGTGT DB17RG-LB TCCACAAGGTCCCGCAAT DB17RG-FIP GTGAGGGCCCTGAGATGTTTGA (F1c plus F2) ATTATTCGCCTACAAGCCGC DB17RG-BIP CCGCATCCGACATCACCCTT (B1c plus B2) GCTTCGCCTTCCTCATCG FIGURE 1 Design of LAMP primers for detection of D. bryoniae (DB). Schematic diagram of LAMP and conventional PCR primer binding sites within the alignment D. bryoniae RAPD marker sequence from RGI and RGII (GenBank accession nos GQ872461 and GQ872462; Ling et al., 2010) were used for this study. The sequences used for LAMP primers are indicated by different colors and arrows. FIP and BIP primers contain two distinct sequences: F1c plus F2 and B1c plus B2, respectively. Optimization of LAMP Reaction Conditions The LAMP reaction was performed in a total volume of 25 μL. A visual fluorescent metal indicator (calcein) was added to the reaction mixture before amplification. For optimization of reagents, a range of concentrations of Bst DNA polymerase large fragment (2–8 U), dNTPs (2–10 mM), Mg2+ (2–8 mM), primers (2–4 μM), MnCl2 (0.1–1 mM), and calcein (2–8 μM) were evaluated. The LAMP reaction was performed in 0.2 mL microcentrifuge tubes in a water bath. Genomic DNA of D. bryoniae (strain DBJSJY2) as a template and ddH2O as a negative control were included in each assay. To identify the optimal reaction temperature and time for the visual detection of the LAMP reaction, the LAMP mixtures were incubated for 45 min at 61, 62, 63, 64, 65, 66, or 68°C in a water bath for 15, 30, 45, or 60 min. The reactions were terminated by heat inactivation at 80°C for 10 min. The reaction mixtures in the microcentrifuge tubes were visually inspected by the naked eye to determine the color change. Each product was confirmed by 2.0% agarose gel electrophoresis following staining with ethidium bromide (0.5 μg mL-1), and photographed under a UV transilluminator. There were three replications for each treatment, and the experiment was repeated twice. PCR Reaction Assignment to the RGI and RGII genotypes of D. bryoniae strains in this study was performed by an RG-specific PCR according to previously published protocols (Keinath et al., 2001; Somai et al., 2002; Babu et al., 2015). In brief, PCRs were performed in a 25 μL reaction mixture containing 40 ng DNA, 2.0 μL 10x reaction buffer [50 mM Tris/HCl (pH 8.8), 1.5 mM MgCl2, 15 mM (NH4)2SO4 and 0.1% Triton X-100], 200 μM dNTPs, 2.5 U Taq DNA polymerase (5 U μL-1), and 0.25 μM each primer. The thermal cycling program was: 94°C for 2 min, and 35 cycles of 94°C for 1 min, 62°C for 1 min, and 72°C for 2 min (Babu et al., 2015). The DB17 primer set was used to amplify the conserved SCAR common to both genotypes of D. bryoniae (Table 2) (Ling et al., 2010). The DB17 primer set-based PCR was carried out with the following cycling conditions: 95°C for 2 min, and 35 cycles of 95°C for 1 min, 50°C for 45 s, and 72°C for 1 min (Ling et al., 2010). PCR amplifications were performed in a 25 μL reaction volume using Platinum® Pfx DNA polymerase (Thermo Fisher Scientific Inc., Waltham, MA, USA). All of the primers were synthesized and purified by Beijing Genomics Institute (Shanghai, China). PCR products were detected by electrophoresis on a 2.0% (w/v) agarose gel containing ethidium bromide (0.5 μg mL-1) in 1x TAE buffer (pH 8.0) at 90 V for 1 h and were visualized under UV light. Specificity of the LAMP Assay The specificity of the LAMP assay was verified with genomic DNA of both genotypes (RGI and RGII) of the D. bryoniae strains, seven other fungal pathogens found in Cucurbitaceae crops, and one fungal pea pathogen (A. pinodes anamorph: D. pinodes; Table 1), which is in the same genus as D. bryoniae. Optimal LAMP reaction components and conditions were used as described earlier. The LAMP assay was performed and assessed twice. Sensitivity of Detection between LAMP and Conventional PCR Assays Template DNA from D. bryoniae (strain DBJSJY2) was prepared as described above and was 10-fold serially diluted (from 105 to 10-2 fg μL-1). The samples were then subjected to LAMP and PCR assays using the conditions described above. When the reactions were completed, the LAMP products were visualized as described above, while the PCR products were observed by 2.0% agarose gel electrophoresis. The LAMP and PCR assays were performed and assessed twice. Evaluation of the LAMP Assay Using Infected Plants Muskmelon seedlings were inoculated with the two genotypes of D. bryoniae (strains DBJSJY2 and DBZJNB7; Table 1; Supplementary Figure S1). After 3 days, leaves showing suspected early symptoms were collected in greenhouse. The samples were prepared as described above. The LAMP assay was used to detect the presence or absence of D. bryoniae in muskmelon leaves using optimal assay temperatures and times. Healthy muskmelon plant leaves and ddH2O were used as negative controls, while purified DNA from D. bryoniae (strains DBJSJY2 and DBZJNB7) was used as a positive control. The reaction mixtures in the microcentrifuge tubes were visually inspected by the naked eye to determine the color change. The LAMP assay was performed and assessed twice. Results Optimization of the LAMP Assay To optimize the efficiency of the LAMP reaction, the concentration of LAMP components was optimized using genomic DNA of D. bryoniae (strain DBJSJY2) as the template. The best results according to a fluorescence metal indicator (calcein; Figure 2A) and the typical ladder-like pattern on 2% agarose gel electrophoresis (Figure 2B) were obtained in a 25 μL volume containing 8 U Bst DNA polymerase, 2.5 μL 10x ThermoPol Buffer [New England Biolabs (Beijing) Ltd. Beijing, China], 8 mM MgSO4, 10 mM dNTPs, 4 μM each of DB17RG-FIP and DB17RG-BIP, 0.5 μM each of DB17RG-F3 and DB17RG-B3, 2 μM DB17RG-LB, 0.3 mM MnCl2, 8 μM calcein, and 1 μL target DNA. Based on the optimized reaction reagents, LAMP was performed using genomic DNA of D. bryoniae (strain DBJSJY2) as a template to determine the optimal temperature and reaction time. Positive results were obtained with temperatures of 61–65°C based on color change (Figure 3A) and the ladder-like pattern of the LAMP products (Figure 3B); however, the yellowish-green color intensity and ladder-like pattern of the LAMP products were strongest at 63°C. The positive reaction time in terms of color change (Figure 4B) and the ladder-like pattern of the LAMP products (Figure 4A) was as early as 30 min, but the yellowish-green color was more intense at 45 min than at 30 min. Thus, the optimal reaction conditions for the LAMP assay were 63°C for 45 min. FIGURE 2 LAMP detection of D. bryoniae (DBJSJY2). Assessment is based on (A) LAMP for detection of D. bryoniae was using a fluorescence metal indicator (calcein) as a visual indicator. The positive reaction becomes yellowish-green, and the negative is still brown; (B) LAMP product was manifested as a ladder-like pattern on a 2.0% agarose gel. M: Trans DNA Marker II (Transgen Biotech, Beijing). In (A,B), 1: Negative reaction (without target DNA), 2: Positive reaction (with target DNA). The same results were obtained in all three replicates. FIGURE 3 Optimal reaction temperatures of LAMP. (A) Detecting LAMP products by adding fluorescence metal indicator (calcein); the assessment was based on visualization of a color change from brown to yellowish-green. (B) Agarose gel electrophoresis analysis of the LAMP products. In (A,B), lane 1: 61°C, lane 2: 62°C, lane 3: 63°C, lane 4: 64°C, lane 5: 65°C, lane 6: 66°C, lane 7: 68°C. M: Trans DNA Marker II (Transgen Biotech, Beijing). The same results were obtained in all three replicates. FIGURE 4 Optimal reaction time of LAMP. (A) Agarose gel electrophoresis analysis of the LAMP products. (B) Detecting LAMP products by adding a fluorescence metal indicators (calcein). In (A,B), lane 1: 60 min, lane 2: 45 min, lane 3: 30 min, lane 4: 15 min, M: Trans DNA Marker II (Transgen Biotech, Beijing). The same results were obtained in two repeat assessments. Specificity of the LAMP Assay Using the optimal reaction conditions described above, and based on the presence of a PCR product of the expected size, the D. bryoniae strains used within this study were confirmed as RGI and RGII genotypes as shown in Supplementary Figure S1 and Table 1. The LAMP assay was performed using DNA from fungal isolates (Table 1). As expected, the LAMP assay showed a yellowish-green color change and visible turbidity only for D. bryoniae strains, whether RGI or RGII genotype (Figures 5A,B). The assay showed high specificity in discriminating all D. bryoniae isolates from seven other fungal pathogens of Cucurbitaceae crops and A. pinodes (teleomorph: Didymella pinodes), a fungal pea pathogen, which is in the same genus as D. bryoniae. The typical ladder-like pattern of the LAMP products was obtained by agarose gel electrophoresis and confirmed the specificity of the LAMP assay (Figure 5C). FIGURE 5 Specificity of LAMP detection of D. bryoniae. Assessment was based on (A) fluorescence metal indicator calcein visualization of color change, (B) the turbidity analysis of the LAMP products or (C) agarose gel electrophoresis analysis of the LAMP products. Lane 1, Didymella bryoniae (strain DBJSJY2) RGI; lane 2, Didymella bryoniae (strain DBAHHF2,) RGI; lane 3, Didymella bryoniae (strain DBZJNB5) RGI; lane 4, Didymella bryoniae (strain DBJSNJ60) RGI; lane 5, Didymella bryoniae (strain DBZJNB7) RGII; lane 6, Ascochyta pinodes ZJ-1; lane 7, Colletotrichum orbiculare NJ-1; lane 8, Pythium paroecandrum Drechsler; lane 9, Alternaria alternata LH1401; lane 10, Fusarium verticillioide; lane 11, Fusarium oxysporum f.sp. niveum Race 0; lane 12, Fusarium oxysporum f.sp. niveum Race 1; lane 13, Fusarium oxysporum f.sp. niveum Race 2; lane 14, positive control; lane 15, negative control. M, Trans DNA Marker II (Transgen Biotech, Beijing). The same results were obtained in two repeat assessments. Sensitivity of Detection among LAMP and Conventional PCR Assays To determine and compare the detection limit, PCR and LAMP assays were performed using 10-fold serial dilutions of D. bryoniae genomic DNA. PCR products were detected by 2% agarose gel electrophoresis, and a 556 bp band specific for both genotypes of D. bryoniae could be seen (Figure 6C). As shown in Figures 6A,B, the limit of detection for the LAMP assay of genomic DNA of D. bryoniae was 0.1 fg μL-1 (Figures 6A,B), whereas the detection limit for conventional PCR was 100 fg μL-1 (Figure 6C). Thus, the LAMP assay was 1000-fold more sensitive than the conventional PCR. The same detection limits for the LAMP assay and for conventional PCR were obtained with four other D. bryoniae isolates (data not shown). FIGURE 6 Sensitivity of the LAMP and conventional PCR. LAMP and conventional PCR assays using 10-fold serial dilutions of purified target DNA from D. bryoniae genomic DNA (strain DBJSJY2). (A) Detecting LAMP products by adding a fluorescence metal indicator (calcein). (B) Agarose gel electrophoresis analysis of the LAMP products. (C) Conventional PCR. Concentrations of template DNA (fg μL-1) per reaction in (A,B) were: lane 1 = 105, lane 2 = 104, lane 3 = 103, lane 4 = 102, lane 5 = 10, lane 6 = 1, lane 7 = 10-1 and lane 8 = 10-2. Concentrations of template DNA (fg μL-1) per reaction in (C) were: lane 1 = 105, lane 2 = 104, lane 3 = 103, lane 4 = 102, lane 5 = 10, lane 6 = 1 and lane 7 = 10-1. In (B,C), M indicates a Trans DNA Marker II (Transgen Biotech, Beijing). The same results were obtained in two repeat assessments. LAMP Assay Using Infected Plants Application of the LAMP assay for detection of D. bryoniae in infected muskmelon leaves was tested. Total DNA extracted from muskmelon samples was used as the template for the LAMP assay. All samples from seedling leaves with suspected primary symptoms reacted positively, but the samples from uninoculated leaves were negative (Figure 7). These results verified the potential field use of the LAMP assay. FIGURE 7 LAMP detection of D. bryoniae from infected muskmelon leaves. Lanes 1 and 2, purified DNA from D. bryoniae strains DBJSJY2 and DBZJNB7, respectively (positive controls); lanes 3 and 4, DNA from leaves infected with D. bryoniae strains DBJSJY2 and DBZJNB7, respectively; lane 5, control healthy muskmelon plant leaves; lane 6, ddH2O used as negative control. The same results were obtained in two repeat assessments. Discussion Gummy stem blight caused by D. bryoniae is a highly prevalent disease and leads to significant losses in yield and quality on cucurbit crops worldwide (Keinath, 2011; Li et al., 2015). The occurrence of GSB disease in the greenhouse and in the field may arise from transplantation of latently infected source seedlings, or contamination by the pathogen may occur as a result of inappropriate pruning of unhealthy plants (Keinath, 1996, 2011; Ling et al., 2010). Thus, seedling health testing is a central issue for a large nursery. To control GSB of cucurbit crops, early crops with latent infection must be accurately detected and removed. Although very recently rapid detection of D. bryoniae by LAMP was reported (Tian et al., 2016), we found that the detection sensitivity and efficiency were lower in this method. In addition, research not has to investigate any infected samples under greenhouse or field conditions. In the current study, a credible, sensitive, and broad-spectrum diagnostic method for the detection of D. bryoniae isolates from pure cultures as well as from infected samples in the field was developed. The optimal reaction for detection of D. bryoniae could be carried out in less than 45 min and in a regular laboratory water bath that can provide isothermal conditions of 63°C. As LAMP assays have a high amplification efficiency, the LAMP method developed in this study showed a high sensitivity at 0.1 fg μL-1 of D. bryoniae genomic DNA, which was 1000-fold higher than that of a conventional PCR (Figure 6). The sensitivity was consistent with a previous report for LAMP methods used to detect Sclerotinia sclerotiorum (Duan et al., 2014a), and the sensitivity was greater than the recently reported LAMP method used to detect D. bryoniae (Tian et al., 2016). In addition, it has been reported that the LAMP reaction might be more efficient by using additional loop primers (Nagamine et al., 2002a; Parida et al., 2008). The optimal reaction time in our LAMP assay was shorter (i.e., less than 45 min) than that of Tian et al. (2016) for detection of D. bryoniae, who did not identify any suitable loop primers. One reason for this could be that in the present study we identified a suitable loop-backward primer and used the primer to accelerate the reaction. This improved the reaction time and efficiency. Studies have revealed that the time required for amplification with two loop primers is one-third to one-half of that without a loop primer, and amplification can be achieved within 30 min (Parida et al., 2008). However, due to the necessity of using a broad-spectrum diagnostic primer, we could not identify a suitable loop-forward primer, and the present LAMP assay only had one loop-backward primer. This appeared to result in the assay being slower than an assay that had both loop primers (Villari et al., 2013), and implies that we may be able to optimize the reaction time further. To distinguish both genotypes of D. bryoniae from other fungal pathogens, a broad-spectrum specific LAMP primer set was designed based on a conserved sequence region (GenBank accession nos GQ872461 and GQ872462 (Ling et al., 2010);) common to both genotypes (RGI and RGII) of D. bryoniae. The specificity of the LAMP assay was confirmed using DNA of both genotypes of D. bryoniae, other fungal pathogens found in Cucurbitaceae crops and A. pinodes (teleomorph: D. pinodes), a fungal pea pathogen in the same genus as D. bryoniae. The result shown that the LAMP assay detected only the two genotypes of D. bryoniae and not for other fungal pathogens (Figure 5). Hence, this LAMP reaction using primer sets designed from the SCAR of genomic DNA for the two genotypes has high specificity and is broad-spectrum. Our result again supports the view that a LAMP assay can be widely used to diagnosis plant-pathogenic fungi (Denschlag et al., 2012; Niessen, 2015). Compared with the result of Ling et al. (2010), this improved broad-spectrum diagnostic method for D. bryoniae isolates was readily visible, easy to carry out and independent of PCR. Due to these advantages, the LAMP method is becoming more attractive to a wider range of users. Furthermore, to determine the utility of the LAMP assay, crude DNA isolated from infected muskmelon tissue samples was analyzed. All samples from muskmelon seedling leaves with suspected early symptoms reacted positively, but the samples from uninoculated leaves were negative (Figure 7). Similar results were found by Huang et al. (2011) and Shen et al. (2016) who used a LAMP assay to detect infected plant tissues. Compared with reported LAMP assays and conventional PCR, the LAMP assay reported here is more advantageous owing to its sensitivity and efficiency, and is robust enough to be used in latently infected crop testing applications in the field. These results indicate that the LAMP assay established in this study can be used for early detection of the disease as the detection limit is low, and the larger range for field use will significantly increase the efficiency of GSB diagnosis and management. Consequently, it is worth emphasizing that the advantages offered by this LAMP assay provide a robust, visual, and easy-to-perform approach for detecting D. bryoniae in early crops. Therefore, this assay could be useful even for amateur farmers without the need for elaborate laboratory equipment. The potential application of this diagnostic tool will enable early prediction of disease, reducing the risks of epidemics. LAMP-based assays are growing in popularity, and have been applied to the detection of many plant-pathogenic fungi (Niessen, 2015). However, the use of a real-time turbidimeter is not applicable in rural areas of developing countries. Thus, for the diagnosis method to be used at the agricultural site, a simple, reliable, and unambiguous visual inspection method is required. As the LAMP reaction progresses, pyrophosphate ions are produced as the byproduct of the reaction and bind to divalent cations (Mn2+ or Mg2+). Monitoring of the products is easy by the naked eye using DNA-binding dyes such as SYBR Green after the reaction endpoint (Huang et al., 2011; Chandra et al., 2016). However, use of DNA-binding dyes increases the rates of cross-contamination because of the open cover operation (Notomi et al., 2000; Zhang et al., 2014). To avoid this, visualization of indirect colorimetric indicators such as hydroxynaphthol blue or calcein have been used as an improvement (Niessen and Vogel, 2010; Duan et al., 2014b). Here, we added calcein as an indirect indicator before the reaction; calcein molecules combine with Mn2+ and quench calcein fluorescence. If the LAMP reaction is positive, the calcein molecules release Mn2+ and combine with residual Mg2+ to generate pyrophosphate ions, thereby recovering their green fluorescence signal (Zhang et al., 2014). Hence, compared with the two-step DNA-binding dye approach, the risk of cross-contamination is much lower (Parida et al., 2008; Duan et al., 2014b; Zhang et al., 2014). A positive reaction is indicated by a color change from brown to a yellowish-green color. In this study, the positive and negative reactions could be successfully distinguished with the naked eye by adding calcein. The fluorescent signals clearly correlated with the results of analysis by gel electrophoresis (Figure 2). Although the LAMP assays in this study showed high specificity and sensitivity, and showed the highest detection limit so far in the subfemtogram range (Niessen, 2015), we still need to be cautious because the presence of calcein may inhibit the LAMP reaction and reduce the sensitivity (Wastling et al., 2010; Zhang et al., 2012). In addition, although the developed LAMP assay showed high sensitivity and broad-spectrum detection of D. bryoniae isolates, we only tested samples by artificial inoculation of muskmelon in the greenhouse. Thus, further studies need to be carried out on a larger scale, and more field samples from other Cucurbitaceae crops in addition to muskmelon should be used to confirm the specificity of the assay. Conclusion A visual LAMP method has been developed for rapid and broad-spectrum detection of D. bryoniae with the advantages of simplicity, sensitivity, and specificity. The LAMP assay established in this study can be used for numerous applications, such as the potential field use for efficient GSB diagnosis and management, seed quarantine, and evaluation of GSB resistance in breeding procedures. The prospective application of this diagnostic tool for early and accurate detection of the causal agent of GSB in cucurbit crops could also lead to reduced fungicide use, thus benefiting the environment. Hence, this study represents a successful attempt to develop a LAMP-based detection method of infection in early cucurbit crops. Author Contributions Conceived and designed the experiments: XfY and XpY. Performed the experiments: XfY, PL, and MZ. Analyzed the data: XpY, RR, GL, and JX. Wrote the paper: XfY and XpY. Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. This work was supported by the National Industrial Technology System for Watermelon & Melon (CARS-26-NO.8), the Jiangsu Agriculture Science and Technology Innovation Fund [CX (14) 2004, CX (15) 1018], the National Natural Science Foundation of China (31301794), and the Fundamental Research Funds for Jiangsu Academy of Agricultural Sciences [ZX (15) 4007)]. 1 http://primerexplorer.jp/elamp4.0.0/index.html Supplementary Material The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmicb.2016.01372 FIGURE S1 Assignment to RGI and RGII genotypes of D. bryoniae strains by RG-specific PCR. DNA of D. bryoniae isolates amplified with the D. bryoniae RGI-specific primers (A) and RGII-specific primers (B). Lane 1, Didymella bryoniae (DBJSJY2) RG I; lane 2, Didymella bryoniae (DBAHHF2) RG I; lane 3, Didymella bryoniae (DBZJNB5) RG I; lane 4, Didymella bryoniae (DBJSNJ60) RG I; lane 5, Didymella bryoniae (DBZJNB7) RG II; M indicates a DS200 marker and marker 1 (Dongsheng biotech, Guangzhou). Click here for additional data file. Click here for additional data file. ==== Refs References Aveskamp M. M. de Gruyter J. Woudenberg J. H. Verkley G. J. Crous P. W. (2010 ). Highlights of the Didymellaceae: a polyphasic approach to characterise Phoma and related pleosporalean genera. Stud. Mycol. 65 1 –60 . 10.3114/sim.2010.65.01 20502538 Babu B. Kefialew Y. W. Li P.-F. Yang X.-P. George S. Newberry E. (2015 ). Genetic characterization of Didymella bryoniae isolates infecting watermelon and other cucurbits in Florida and Georgia. Plant Dis. 99 1488 –1499 . 10.1094/pdis-04-14-0341-re Chandra A. Keizerweerd A. T. Grisham M. P. (2016 ). Detection of Puccinia kuehnii causing sugarcane orange rust with a loop-mediated isothermal amplification-based assay. Mol. Biotechnol. 58 188 –196 . 10.1007/s12033-016-9914-5 26837389 Denschlag C. Vogel R. F. Niessen L. (2012 ). Hyd5 gene-based detection of the major gushing-inducing Fusarium spp. in a loop-mediated isothermal amplification (LAMP) assay. Int. J. Food Microbiol. 156 189 –196 . 10.1016/j.ijfoodmicro.2012.03.009 22554927 Duan Y. Ge C. Zhang X. Wang J. Zhou M. (2014a ). A rapid detection method for the plant pathogen Sclerotinia sclerotiorum based on loop-mediated isothermal amplification (LAMP). Australas. Plant Pathol. 43 61 –66 . 10.1007/s13313-013-0239-6 Duan Y. Zhang X. Ge C. Wang Y. Cao J. Jia X. (2014b ). Development and application of loop-mediated isothermal amplification for detection of the F167Y mutation of carbendazim-resistant isolates in Fusarium graminearum. Sci. Rep. 4 :7094 10.1038/srep07094 Finger M. J. Parkunan V. Ji P. Stevenson K. L. (2014 ). Allele-specific PCR for the detection of azoxystrobin resistance in Didymella bryoniae. Plant Dis. 98 1681 –1684 . 10.1094/pdis-02-14-0136-re Francois P. Tangomo M. Hibbs J. Bonetti E. J. Boehme C. C. Notomi T. (2011 ). Robustness of a loop-mediated isothermal amplification reaction for diagnostic applications. FEMS Immunol. Med. Microbiol. 62 41 –48 . 10.1111/j.1574-695X.2011.00785.x 21276085 Ha Y. Fessehaie A. Ling K. S. Wechter W. P. Keinath A. P. Walcott R. R. (2009 ). Simultaneous detection of Acidovorax avenae sub sp. citrulli and Didymella bryoniae in cucurbit seedlots using magnetic capture hybridization and real-time polymerase chain reaction. Phytopathology 99 666 –678 . 10.1094/PHYTO-99-6-0666 19453225 Hadersdorfer J. Neumüller M. Treutter D. Fischer T. C. (2011 ). Fast and reliable detection of Plum pox virus in woody host plants using the Blue LAMP protocol. Ann. Appl. Biol. 159 456 –466 . 10.1111/j.1744-7348.2011.00510.x Hansen Z. R. Knaus B. J. Tabima J. F. Press C. M. Judelson H. S. Grunwald N. J. (2016 ). Loop-mediated isothermal amplification for detection of the tomato and potato late blight pathogen, Phytophthora infestans. J. Appl. Microbiol. 120 1010 –1020 . 10.1111/jam.13079 26820117 Hodgetts J. Hall J. Karamura G. Grant M. Studholme D. J. Boonham N. (2015 ). Rapid, specific, simple, in-field detection of Xanthomonas campestris pathovar musacearum by loop-mediated isothermal amplification. J. Appl. Microbiol. 119 1651 –1658 . 10.1111/jam.12959 26425811 Huang C. Sun Z. Yan J. Luo Y. Wang H. Ma Z. (2011 ). Rapid and precise detection of latent infections of wheat stripe rust in wheat leaves using loop-mediated isothermal amplification. J. Phytopathol. 159 582 –584 . 10.1111/j.1439-0434.2011.01806.x Kang J. S. Kim A. Y. Han H. R. Moon Y. S. Koh Y. H. Woodward S. (2015 ). Development of two alternative Loop-mediated isothermal amplification tools for detecting pathogenic pine wood nematodes. For. Pathol. 45 127 –133 . 10.1111/efp.12147 Keinath A. P. (1996 ). “Spread of Didymella bryoniae from contaminated watermelon seed and transplants in greenhouse and field environments,” in Recent Research Developments in Plant Pathology Vol. 61 ed. Pandalai S. G. (Trivandrum : Research Signpost ) 65 –72 . Keinath A. P. (2011 ). From native plants in central Europe to cultivated crops worldwide The emergence of Didymella bryoniae as a cucurbit pathogen. HortScience 46 532 –535 . Keinath A. P. (2014 ). Differential susceptibility of nine cucurbit species to the foliar blight and crown canker phases of gummy stem blight. Plant Dis. 98 247 –254 . 10.1094/pdis-05-13-0510-re Keinath A. P. Farnham M. W. Zitter T. A. (1995 ). Morphological, pathological, and genetic differentiation of Didymella bryoniae and Phoma spp. isolated from cucurbits. Phytopathology 85 364 –369 . 10.1094/Phyto-85-364 Keinath A. P. Somai B. M. Dean R. A. (2001 ). Method of diagnosing gummy stem blight in plants using a polymerase chain reaction assay. US 20010758073 Li P.-F. Ren R.-S. Yao X.-F. Xu J.-H. Babu B. Paret M. L. (2015 ). Identification and characterization of the causal agent of gummy stem blight from muskmelon and watermelon in East China. J. Phytopathol. 163 314 –319 . 10.1111/jph.12277 Ling K. S. Wechter W. P. Somai B. M. Walcott R. R. Keinath A. P. (2010 ). An improved real-time PCR system for broad-spectrum detection of Didymella bryoniae, the causal agent of gummy stem blight of cucurbits. Seed Sci. Technol. 38 692 –703 . 10.15258/sst.2010.38.3.17 Moradi A. Almasi M. A. Jafary H. Mercado-Blanco J. (2014 ). A novel and rapid loop-mediated isothermal amplification assay for the specific detection of Verticillium dahliae. J. Appl. Microbiol. 116 942 –954 . 10.1111/jam.12407 24329885 Mori Y. Kanda H. Notomi T. (2013 ). Loop-mediated isothermal amplification (LAMP): recent progress in research and development. J. Infect. Chemother. 19 404 –411 . 10.1007/s10156-013-0590-0 23539453 Nagamine K. Hase T. Notomi T. (2002a ). Accelerated reaction by loop mediated isothermal amplification using loop primers. Mol. Cell. Probes 16 223 –229 . 10.1006/mcpr.2002.0415 12144774 Nagamine K. Kuzuhara Y. Notomi T. (2002b ). Isolation of single-stranded DNA from loop-mediated isothermal amplification products. Biochem. Biophys. Res. Commun. 290 1195 –1198 . 10.1006/bbrc.2001.6334 11811989 Niessen L. (2015 ). Current state and future perspectives of loop-mediated isothermal amplification (LAMP)-based diagnosis of filamentous fungi and yeasts. Appl. Microbiol. Biotechnol. 99 553 –574 . 10.1007/s00253-014-6196-3 25492418 Niessen L. Vogel R. F. (2010 ). Detection of Fusarium graminearum DNA using a loop-mediated isothermal amplification (LAMP) assay. Int. J. Food Microbiol. 140 183 –191 . 10.1016/j.ijfoodmicro.2010.03.036 20442002 Notomi T. Okayama H. Masubuchi H. Yonekawa T. Watanabe K. Amino N. (2000 ). Loop-mediated isothermal amplification of DNA. Nucleic Acids Res. 28 :e63 10.1093/nar/28.12.e63 Parida M. Sannarangaiah S. Dash P. K. Rao P. V. Morita K. (2008 ). Loop mediated isothermal amplification (LAMP): a new generation of innovative gene amplification technique; perspectives in clinical diagnosis of infectious diseases. Rev. Med. Virol. 18 407 –421 . 10.1002/rmv.593 18716992 Peng J. Shi M. Xia Z. Huang J. Fan Z. (2012 ). Detection of cucumber mosaic virus isolates from banana by one-step reverse transcription loop-mediated isothermal amplification. Arch. Virol. 157 2213 –2217 . 10.1007/s00705-012-1376-x 22782136 Shen W. Xu G. Sun L. Zhang L. Jiang Z. (2016 ). Development of a loop-mediated isothermal amplification assay for rapid and sensitive detection of Sporisorium scitamineum in sugarcane. Ann. Appl. Biol. 168 321 –327 . 10.1111/aab.12264 Somai B. M. Keinath A. P. Dean R. A. (2002 ). Development of PCR-ELISA for detection and differentiation of Didymella bryoniae from related Phoma species. Plant dis 86 710 –716 . 10.1094/PDIS.2002.86.7.710 Sudisha J. Niranjana S. R. Umesha S. Prakash H. S. Shekar Shetty H. (2006 ). Transmission of seed-borne infection of muskmelon by Didymella bryoniae and effect of seed treatments on disease incidence and fruit yield. Biol. Control 37 196 –205 . 10.1016/j.biocontrol.2005.11.018 Tian Y. Liu D. Zhao Y. Wu J. Hu B. Walcott R. R. (2016 ). Visual detection of Didymella bryoniae in cucurbit seeds using a loop-mediated isothermal amplification assay. Eur. J. Plant Pathol. 10.1007/s10658-016-0996-5 Villari C. Tomlinson J. A. Battisti A. Boonham N. Capretti P. Faccoli M. (2013 ). Use of loop-mediated isothermal amplification for detection of Ophiostoma clavatum, the primary blue stain fungus associated with Ips acuminatus. Appl. Environ. Microbiol. 79 2527 –2533 . 10.1128/AEM.03612-12 23396326 Wastling S. L. Picozzi K. Kakembo A. S. Welburn S. C. (2010 ). LAMP for human African trypanosomiasis: a comparative study of detection formats. PLoS Negl. Trop. Dis. 4 :e865 10.1371/journal.pntd.0000865 Wolukau J. N. Zhou X.-H. Li Y. Zhang Y.-B. Chen J.-F. (2007 ). Resistance to gummy stem blight in melon (Cucumis melo L.) germplasm and inheritance of resistance from plant introductions 157076, 420145, and 323498. HortScience 42 215 –221 . Zhang X. Li M. Cui Y. Zhao J. Cui Z. Li Q. (2012 ). Electrochemical behavior of calcein and the interaction between calcein and DNA. Electroanalysis 24 1878 –1886 . 10.1002/elan.201200192 Zhang X. Lowe S. B. Gooding J. J. (2014 ). Brief review of monitoring methods for loop-mediated isothermal amplification (LAMP). Biosens. Bioelectron. 61 491 –499 . 10.1016/j.bios.2014.05.039 24949822

PMC005xxxxxx/PMC5003823.txt

==== Front Front MicrobiolFront MicrobiolFront. Microbiol.Frontiers in Microbiology1664-302XFrontiers Media S.A. 10.3389/fmicb.2016.01357MicrobiologyMini ReviewCorrelations of Host Genetics and Gut Microbiome Composition Dąbrowska Krystyna 12*Witkiewicz Wojciech 11Research and Development Center, Regional Specialized HospitalWrocław, Poland2Bacteriophage Laboratory, Institute of Immunology and Experimental Therapy, Polish Academy of SciencesWrocław, PolandEdited by: Giovanna E. Felis, University of Verona, Italy Reviewed by: Julia Green-Johnson, University of Ontario Institute of Technology, Canada; Daniel M. Linares, Teagasc – The Irish Agriculture and Food Development Authority, Ireland; Michael Kogut, U.S. Department of Agriculture, USA *Correspondence: Krystyna Dąbrowska, dabrok@iitd.pan.wroc.plThis article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology 30 8 2016 2016 7 135728 4 2016 16 8 2016 Copyright © 2016 Dąbrowska and Witkiewicz.2016Dąbrowska and WitkiewiczThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.The human gut microbiome has a considerable impact on host health. The long list of microbiome-related health disorders raises the question of what in fact determines microbiome composition. In this review we sought to understand how the host itself impacts the structure of the gut microbiota population, specifically by correlations of host genetics and gut microbiome composition. Host genetic profile has been linked to differences in microbiome composition, thus suggesting that host genetics can shape the gut microbiome of the host. However, cause-consequence mechanisms behind these links are still unclear. A survey of the possible mechanisms allowing host genetics to shape microbiota composition in the gut demonstrated the major role of metabolic functions and the immune system. A considerable impact of other factors, such as diet, may outweigh the effects of host genetic background. More studies are necessary for good understanding of the relations between the host genetic profile, gut microbiome composition, and host health. According to the idea of personalized medicine, patient-tailored management of microbiota content remains a fascinating area for further inquiry. microbiomegene polymorphismQTLpersonalized medicinegutbacteriaNarodowe Centrum Bada? i Rozwoju10.13039/501100005632Narodowe Centrum Nauki10.13039/501100004281UMO-2012/05/E/NZ6/03314 ==== Body Introduction The human gut houses a complex community of microbes whose number is estimated to be 10 times higher than the number of cells in the whole body. These microbial communities, called the gut microbiome, are dynamic populations that differ from one person to another and that change their structure with time. Gut microbiotas interact with their hosts in many ways; thus composition of a microbiome may impact the balance of the whole system, and changes in microbial communities may exert significant effects on an individual’s health. Thus, microbes can be used in strategies addressing modulation of microbiome functionality, i.e., they can be used as probiotics (Savage, 1977; Kau et al., 2011; Linares et al., 2016). An extensive review on probiotic roles played by the microbiota has recently been published by Linares et al. (2016). The importance of the gut microbiota for human health has been widely appreciated during this and the previous decade. The least surprising was their effect on gastrointestinal tract functions and diseases such as ulcerative colitis (Macfarlane et al., 2005; Sokol et al., 2006a,b; Linares et al., 2016) gastroenteritis (Barman et al., 2008), and celiac disease (Nadal et al., 2007). Abnormalities in composition of gut bacteria are considered as an important factor promoting inflammatory bowel disease including Crohn’s disease (Kassinen et al., 2007; Dicksved et al., 2008; Collins et al., 2009; Linares et al., 2016; Manuc et al., 2016). Since gut bacteria are involved in metabolic transformations and energy harvest, they have been reported as a biotic factor regulating body weight, potentially linked to a risk of obesity and other metabolic disorders (Backhed et al., 2004; Turnbaugh et al., 2006, 2009; Cani and Delzenne, 2009; Ley, 2010). However, in this field also controversies have been pointed out, like the complex character of studied phenomenon, the necessity of studies based on observation and description in opposition to studies performed to confirm a hypothesis and the independence from the food industry (Angelakis et al., 2012; Lagier et al., 2012). Gut microbiota composition has been linked to functions of organs and tissues far beyond the gut itself. Probably the most spectacular is the so-called gut–brain axis; biochemical signaling between the gastrointestinal tract and the nervous system is important for healthy brain function. This relationship involves the gut microbiome, the composition of which may be linked to neuropsychiatric diseases (Mu et al., 2016; Yarandi et al., 2016). Another important role of the natural microbiome is liver homeostasis, since bacterial metabolites in dysbiosis can be linked to the pathogenesis of liver disease (Haque and Barritt, 2016). Microbial metabolites belonging to short-chain fatty acids (butyrate) may affect the whole system of the host, even mitigating graft-versus-host disease (GVHD), the most probably by regulation of histone acetylation (Mathewson et al., 2016). Some findings reveal that the microbiomes of the lung and gut contribute to the pathogenesis of asthma and allergy by regulation of helper T cell subsets that affect the development of immune tolerance (Russell et al., 2012; Riiser, 2015). The long list of microbiome-related health disorders raises the question of what in fact decides on microbiome composition. Ecological sciences define factors that shape microbial community structure as a combination of environmental factors such as diet, and host-defined ones. In fact, there is a high interaction between the microbiome and the host, and for that reason both of them have evolved together, which may explain possible microbiome adaptations (Cavender-Bares et al., 2009; Walter and Ley, 2011; Leamy et al., 2014; Linares et al., 2016). Relating individual microbiome composition to host genetics may constitute a link between probiotic studies and personalized medicine. In this review we aim to bring together recent findings that demonstrate the link between human host genetics and the gut microbiota, from the perspective of practical implications of this knowledge for humans’ health. Links Between Host Genetic Profile and Individual Gut Microbiome in Humans The question of how genotype and environmental exposure influence the gut microbiome has been addressed in a twin pairs study by Turnbaugh et al. (2009). Fecal microbiota were characterized in 154 adult individuals comprising female monozygotic or dizygotic twins and their mothers, if available. In this group, the gut microbiota was similar among family members, but individual variations were observed, i.e., specific bacterial lineages were present in each person’s gut. These variations were assessed in monozygotic and in dizygotic twins, and the analysis showed a comparable degree of co-variation between these groups. The authors found their observation consistent with an earlier study of adult twins by fingerprinting (Zoetandal et al., 2001). These findings contradict the hypothesis of a substantial impact of host genetics on microbiome composition and they are in line with observation of Murphy et al. (2015) who studied dichorionic triplet sets. In these group only at the 1st month of life monozygotic pair shared microbiota distinct to the fraternal sibling. At the 12th month no significant differences were observed. However, some limitations of the studied have been pointed out. First, small groups: 20–30 twin pairs in each category in studies of Turnbaugh et al. (2009) and three triplet sets in studies of Murphy et al. (2015). Second, broad measures of the microbiome composition instead of individual bacterial representation was investigated (Davenport et al., 2015; Murphy et al., 2015). Recent studies by Goodrich et al. (2014) analyzed microbiotas in fecal samples obtained from the TwinsUK population (416 twin pairs); microbiomes were found more similar for monozygotic than dizygotic twins. The study also revealed differences between bacterial families; the analyses of distance metrics in the three most dominant bacterial families – the Lachnospiraceae and Ruminococcaceae (Firmicutes) and Bacteroidaceae – demonstrated greater similarities between monozygotic twins within the Ruminococcaceae and Lachnospiraceae than those between dizygotic twins. Similar pair-wise diversity was restricted to the Bacteroidaceae family, so this group was suggested to be more responsive to environmental factors (Goodrich et al., 2014). Moreover, datasets from Turnbaugh et al. (2009) and Yatsunenko et al. (2012) were re-analyzed, validating the observation that the representation of bacterial taxa in the gut is more similar within monozygotic than dizygotic twin pairs (Goodrich et al., 2014). Davenport et al. (2015) applied a genome-wide association study (GWAS) to investigate the fecal microbiome from a religious isolated group, the Hutterites who live on communal farms. The Hutterites are not an outbred population, but rather a genetic isolate exhibiting a strong founder effect (Coghlan and Zelinski, 2016). They are also an isolated population due to the communal farming, therefore variations of environmental factors have less impact on individuals’ microbiomes (Davenport et al., 2015; Igartua et al., 2016). These features make this population a unique model for genome and microbiome studies. The study in 127 participants demonstrated that the abundances of at least eight bacterial taxa were associated with host genome single nucleotide polymorphisms (SNPs), including those previously associated with BMI in obesity studies. The differentiation of bacterial taxa abundances between male and female participants was also reported, however, this element must be considered with caution. Hutterite society practice substantially different daily activities of men and women which could drive sex-specific differences (Davenport et al., 2015; Davenport, 2016). In the report within the Human Microbiome Project, most variation in the human microbiome could not be well explained by their relation to gender (Human Microbiome Project Consortium, 2012a,b). Murine Models Demonstrate that Host Genetic Profile Can Shape Gut Microbiome The relation between the composition of gut microbiota and the host genetic profile has been clearly demonstrated in murine models. Benson et al. (2010) showed that composition of the gut microbiota behaves as a polygenic trait (i.e., resultative phenotype cumulates effects of more than one gene), and they identified in mice 18 host quantitative trait loci (QTL) that correlated with relative abundances of particular microbial groups. This correlation showed that heritable genetic factors may govern intimate associations between the host and its microbiota, although additional efforts will be needed to explain in details which physiological mechanisms are involved. The study was done in a large (n = 645) murine intercross model (G4) in which the environmental factors were carefully controlled (Table 1). Table 1 Correlations of body traits and microbiome traits by the same quantitative trait loci. Correlation by QTLs Reference Body trait Microbiome trait Fat content OTU3615 (Actinobacteria) Leamy et al., 2014 Fat content OTU22207 (Alistepes) Leamy et al., 2014 Weight and fat content OTU30840 (Clostridium) Leamy et al., 2014 Weight Lactococcus lactis Leamy et al., 2014 Immune response Coriobacteriaceae Benson et al., 2010 Immune response Lactococcus Benson et al., 2010 Susceptibility to colon tumors Coriobacteriaceae Benson et al., 2010 Susceptibility to hepatocellular carcinomas Turicibacter Benson et al., 2010 Region syntenic with this associated with Crohn’s disease in humans Barnesiella Benson et al., 2010 Weight Akkermansia Davenport et al., 2015 Olfactory receptor (response to bacterial metabolites) Bifidobacterium Davenport et al., 2015 Olfactory receptor (response to bacterial metabolites) Faecalibacterium Davenport et al., 2015 The same study revealed that in the case of some taxonomic groups of bacteria, e.g., lactobacilli, host genetic control is probably exerted at the lower taxonomic ranks, i.e., at the species level and below. No QTL were identified for Lactobacillus (genus), so they mapped as individual traits the relative abundance of three Lactobacillus groups with 97% identity – Lactobacillus reuteri, L. johnsonii/L. gasseri, and L. animalis/L. murinus – to test for co-segregation at the species level. This analysis revealed that the L. johnsonii/L. gasseri group segregated with two significant QTL on MMU14 and MMU7 (Benson et al., 2010). Interestingly, in the case of Helicobacter (genus) significant QTL were detected. Both Helicobacter and Lactobacillus interact directly and adhere to host tissues, so both genera would be expected to have intrinsic susceptibility to modulation by host factors (Benson et al., 2010). As reported by Leamy et al. (2014), a study of G10 mouse population (continued) revealed 42 microbiota-specific QTL in 27 different genomic regions that affected the relative abundances of 39 (out of the 203, i.e., 19%) microbial taxa in the murine gut. When using a strict approach the authors proposed 20 QTL as underlying genetic variation affecting the microbiota composition. The lowest FDR (false discovery rate) values, which means the greatest support, were for QTL on MMU9 affecting Alistipes (Leamy et al., 2014), that have been so far reported as over-represented in patients with depression (Jiang et al., 2015). It is not clear what is the strength of these genome-related effects on microbiomes. As previously mentioned, many other factors (diet, life style, colonization order, etc.) contribute to the resultant effect. The group of Turnbaugh reported that diet dominated host genotype in shaping the gut microbiome. In mice deficient for genes linked to host-microbial interactions [MyD88(-/-), NOD2(-/-), ob/ob, and Rag1(-/-)] and in wild-type mice, gut microbiota were similarly modified by the diet. Further, the structural changes in the microbial community that were observed after dietary changes were rapid, reproducible, and reversible, thus implying the predominant role of the diet in shaping the gut microbiome (Carmody et al., 2015). Other studies dedicated to phylotypes associated with obesity, suggested that host genetic factors influenced gut microbiota plasticity in response to diet (Parks et al., 2013); this emphasizes the multilateral dependencies between various factors engaged in microbiome forming processes. Mechanisms that Link Host Genetics and Gut Microbiota Composition In spite of the growing volume of data explaining how the gut microbiota affects host physiology and health, explanations of how host genetics shapes the structure of the gut microbiome are very scarce. In general, the authors propose immune functions, metabolism, energy regulation, gut motility, and adhesion interactions as the most expected genetics-dependent physiological phenomena that may impact the gut microbiota (Benson et al., 2010; Leamy et al., 2014; Davenport et al., 2015). A survey of the possible mechanisms allowing host genetics to shape microbiota composition in the gut demonstrated the major role of metabolic functions and the immune system (Table 1). Benson et al. (2010) pointed out that QTL for Coriobacteriaceae and Lactococcus (located on MMU10) identified in their study were closely positioned with several genes engaged in immune responses and regulation. These comprised the TLR2 pathway, IFN-gamma, and IL-22, important in the immune response in mucosal surfaces. The authors also discussed a microbiome-related QTL on MMU1 that overlaps the conserved gene ATG16L, and the region is syntenic with a region of human chromosome 2 already shown to be associated with Crohn’s disease (Parkes et al., 2007; Benson et al., 2010). The pathogenesis of Crohn’s disease has been so far recognized as a result of the gut microbiome and environmental factors leading to an abnormal immune response in a genetically predisposed patient. Possible factors promoting and mitigating Crohn’s disease have been recently discussed in an extensive review by Manuc et al. (2016). Interestingly, some Crohn’s associated gene polymorphisms have been demonstrated as affecting both the immune response and the microbiota. For instance, the innate immune response is affected by the polymorphism of nucleotide-binding oligomerization domain-containing protein 2 (NOD2)/caspase recruitment domain-containing protein 15 (CARD15); NOD2 acts through the NF-kB pathway, which is also responsive to bacterial wall components (muramyl dipeptide). An inadequate antibacterial response related to NOD2 has been linked to probably lower production of alpha defensins (antimicrobial peptides) by Paneth cells or by an incorrect autophagy cascade with increased levels of NF-kB. It is associated with the highest risk of ileal involvement, stenoses or fistulas in Crohn’s disease (Adler et al., 2011; Lee and Lee, 2014; Strober et al., 2014; Manuc et al., 2016). Genetic variants that may also lead to an increased risk of Crohn’s disease are linked to Toll-like receptor 4 (TLR-4), typically responsible for recognizing bacterial lipopolysaccharide, CARD9 (caspase recruitment domain-containing protein 9), engaged in defense against pathogens such as yeasts, and interleukin 23 receptor (IL-23R), while IL-23 has been implicated in inhibiting the development of regulatory T cell development in the intestine (Liu and Anderson, 2014; Manuc et al., 2016). Khachatryan et al. (2008) demonstrated significant changes in microbiome structure in patients with an autoinflammatory disorder called familial Mediterranean fever (related to mutations in the MEFV gene). Their microbiome was substantially disturbed even during remissions. The relations between gut bacteria and the central nervous system (the brain-gut axis) also engage immune system and endocrine elements. Stress has been demonstrated in rodents as the altering factor for gut microbiota through immune-activation, probably due to changes in bacterial translocation and resulting increase in stimulation of the innate immune system (O’Mahony et al., 2009; Bailey et al., 2011). As pointed out by Cryan and Dinan (2012) the mechanisms underlying this relation engage autonomic nervous system (ANS) and hypothalamus-pituitary-adrenal (HPA) axis that can modulate gut motility, secretion and epithelial permeability which impacts the niche environment for microbiota. CNS may induce signals for neurons, immune cells and others secretory cells in the gut that release a variety of signaling molecules as well as anti-microbial peptides (AMPs) and modify composition of gut microbiota (Rhee et al., 2009; Cryan and Dinan, 2012; Wang and Kasper, 2014). Some of the identified microbiome-related QTL (on MMU7 and MMU10, for Turicibacter and Coriobacteriaceae, respectively) overlapped with QTL for murine susceptibility to carcinomas and tumor development (Benson et al., 2010). In these cases the possible role of immunological functions has not been explained yet, although tumor development usually involves complex immunological processes. Other observations suggest that gene-encoded metabolic characteristics influence microbiome structure. For example, an correlation has been identified between a bacterial taxon associated with obesity (genus Akkermansia) and a variant near PLD1, a gene related to body mass index (Everard et al., 2013; Davenport et al., 2015). Evolutionary studies of vertebrates and typical composition of their gut microflora suggest that the microbiome is shaped by stomach acidity; this was confirmed by the analysis of microbiome modifications correlating with evolutionary changes of animals (Beasley et al., 2015). Gene set enrichment analysis (GSEA) performed by Davenport et al. (2015) revealed the olfactory receptor activity significant for five taxons (family Succinivibrionaceae, genus Bifidobacterium, order Rhizobiales, genus Anaerofilum, genus Faecalibacterium). It had been previously demonstrated in mice that in the kidneys an olfactory receptor responded to metabolites produced by gut bacteria; this process affects renin production and it results in systemic modifications of blood pressure (Pluznick et al., 2013). Davenport et al. (2015) proposed that olfactory receptors that may be expressed in other tissues can also recognize compounds secreted by the microbiota. These olfactory receptors could play a role in host-driven regulation either for host physiology or for the microbiota in response to the gut environment. Discussion In this review we present reports that suggest that the host genetic profile may shape the gut microbiome of the host. Some studies were contradictory to this statement, such as the first twin pairs analyses (Turnbaugh et al., 2009), but others revealed a possible relation between host genetics and the microbiota in the gut. A very interesting hypothesis was presented by Murphy et al. (2015) basing on results in a group of dichorionic triplet set that contained a pair of monozygotic twins and a fraternal sibling. Host genetics seems to play a role in the composition of an individual’s gut microbiome at the initial stage of life (1 month), but later (12 months) environmental factors become a major determinant. It should be emphasized that the volume of relevant data is still small, which results mainly from technical difficulties and limitations that have also been pointed out by some researchers (Fu et al., 2016). Changes in gut microbe composition are typically analyzed quantitatively in well-represented bacterial taxa. The question that is still to be addressed relates to some small microbial groups. Possibly their quantitatively insignificant changes may affect host health substantially. Gut microbiota composition can vary significantly even in well-controlled cohorts, since other factors (than host genetics) shape the microbiome (Parks et al., 2013). A representative example is the diet, which has been demonstrated in a knock-out mice model as dominating the host genotype in shaping the gut microbiome (Carmody et al., 2015). As a result, the potential role of the low-represented microbial groups still remains vague. A limitation comes from the relatively small size of the samples, which in microbiome studies have usually been approximately 1–2 hundred individuals, or less. The sample sizes necessary to detect significant associations in many GWAS of common diseases were thousands to tens of thousands of individuals (Davenport et al., 2015). Additionally, as pointed out by Davenport et al. (2015), at the current state of research, published replication cohorts are not available in humans. The prospects for expanding the volume of available data rely on the initiative of the Human Microbiome Project1, inspired by the enhancing need for understanding reciprocal cross-talk between microbiome and host. One of its major goals is to determine whether there is an identifiable ‘core microbiome’ of shared organisms, genes, or functional capabilities found in a given body habitat of all or the vast majority of humans (Turnbaugh et al., 2007). Probably the biggest challenge that remains in the field is to answer the question: “Is it really the microbiome composition that causes a health disorder, or do both the health disorder and the altered microbiome composition result from the same genomic factor?” Probably both types of relations are possible, but as yet they have not been clearly discriminated in humans. It is very likely that a disease causes microbiome changes, due to effects on several factors that affect microbiota: changes in intestinal motility, change in appetite and diet, medical treatment including surgery, changes in lifestyle. All recent studies of Crohn’s disease confirm that it correlates with dysbiosis. Particularly, low levels of Firmicutes, Bifidobacteria, and Lactobacilli have been observed, and higher numbers of Escherichia coli and other strains of Enterobacteriaceae (Manuc et al., 2016). However, the use of probiotic bacteria such as Lactobacillus and Bifidobacterium in active Crohn’s disease or during remissions did not result in clearly conclusive, positive results. Currently, probiotics are not recommended in Crohn’s disease (Manuc et al., 2016). This example suggests that, at least in some cases, the correlation of microbiota composition and disease phenotype may result from the same detail of the host genetic profile, and they may not necessarily be directly linked. On the other hand, there are reports demonstrating transfer of the gut microbiota between mice as sufficient for the transfer of a disease phenotype. For instance, a very recent study by Gacias et al. (2016) revealed that depression-like disorders of social behavior could be “transferred” between two genetically distinct strains of mice by the transfer of fecal bacteria. Intestinal microbes, including members of Clostridiales, Lachnospiraceae, and Ruminococcaceae, were transferred from the gut of depressed mice to those exhibiting non-depressed behavior; this was sufficient to induce social avoidance in the animals. The mechanism of this effect, as identified by metabolomics analysis, was related to increased cresol (a metabolite) levels in mice with the depressive phenotype (Gacias et al., 2016). We propose that both types of relations between host genetic profile, gut microbiome composition, and host health are possible, thus a better understanding may allow for patient-tailored shaping of microbiome, e.g., by the diet. The field of microbiome research is anticipated to expand with new knowledge and clinical potential (Linares et al., 2016). This idea remains a fascinating area for future inquiry. Author Contributions KD analyzed the literature within the topic and wrote the manuscript. WW has reviewed and consulted the manuscript. Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. This work was supported by the European Regional Development Fund, within the Innovative Economy Operational Program 2007–2013, project “Wrovasc – Integrated Cardiovascular Centre”, implemented in the Regional Specialist Hospital, Research and Development Centre in Wroclaw (WW), by the National Science Centre in Poland, grant UMO-2012/05/E/NZ6/03314 (KD), and by Wroclaw Centre of Biotechnology, programme The Leading National Research Centre (KNOW) for years 2014-2018. 1 http://hmpdacc.org/ ==== Refs References Adler J. Rangwalla S. C. Dwamena B. A. Higgins P. D. (2011 ). The prognostic power of the NOD2 genotype for complicated Crohn’s disease: a meta-analysis. Am. J. Gastroenterol. 106 699 –712 . 10.1038/ajg.2011.19 21343918 Angelakis E. Armougom F. Million M. Raoult D. (2012 ). The relationship between gut microbiota and weight gain in humans. Future Microbiol. 7 91 –109 . 10.2217/fmb.11.142 22191449 Backhed F. Ding H. Wang T. Hooper L. V. Koh G. Y. Nagy A. (2004 ). The gut microbiota as an environmental factor that regulates fat storage. Proc. Nat. Acad. Sci. U.S.A. 101 15718 –15723 . 10.1073/pnas.0407076101 Bailey M. T. Dowd S. E. Galley J. D. Hufnagle A. R. Allen R. G. Lyte M. (2011 ). Exposure to a social stressor alters the structure of the intestinal microbiota: implications for stressor-induced immunomodulation. Brain Behav. Immun. 25 397 –407 . 10.1016/j.bbi.2010.10.023 21040780 Barman M. Unold D. Shifley K. Amir E. Hung K. Bos N. (2008 ). Enteric salmonellosis disrupts the microbial ecology of the murine gastrointestinal tract. Infect. Immun. 76 907 –915 . 10.1128/IAI.01432-07 18160481 Beasley D. E. Koltz A. M. Lambert J. E. Fierer N. Dunn R. R. (2015 ). The evolution of stomach acidity and its relevance to the human microbiome. PLoS ONE 10 :e0134116 10.1371/journal.pone.0134116 Benson A. K. Kelly S. A. Legge R. Ma F. Low S. J. Kim J. (2010 ). Individuality in gut microbiota composition is a complex polygenic trait shaped by multiple environmental and host genetic factors. Proc. Natl. Acad. Sci. U.S.A. 107 18933 –18938 . 10.1073/pnas.1007028107 20937875 Cani P. D. Delzenne N. M. (2009 ). Interplay between obesity and associated metabolic disorders: new insights into the gut microbiota. Curr. Opin. Pharmacol. 9 737 –743 . 10.1016/j.coph.2009.06.016 19628432 Carmody R. N. Gerber G. K. Luevano J. M. Jr.Gatti D. M. Somes L. Svenson K. L. (2015 ). Diet dominates host genotype in shaping the murine gut microbiota. Cell Host Microbe 17 72 –84 . 10.1016/j.chom.2014.11.010 25532804 Cavender-Bares J. Kozak K. H. Fine P. V. A. Kembel S. W. (2009 ). The merging of community ecology and phylogenetic biology. Ecol. Lett. 12 693 –715 . 10.1111/j.1461-0248.2009.01314.x 19473217 Coghlan G. Zelinski T. (2016 ). The c.64_80del SMIM1 allele is segregating in the Hutterite population. Transfusion 56 946 –949 . 10.1111/trf.13439 26666208 Collins S. M. Denou E. Verdu E. F. Bercik P. (2009 ). The putative role of the intestinal microbiota in the irritable bowel syndrome. Dig. Liver Dis. 41 850 –853 . 10.1016/j.dld.2009.07.023 19740713 Cryan J. F. Dinan T. G. (2012 ). Mind-altering microorganisms: the impact of the gut microbiota on brain and behaviour. Nat. Rev. Neurosci. 13 701 –712 . 10.1038/nrn3346 22968153 Davenport E. R. (2016 ). Elucidating the role of the host genome in shaping microbiome composition. Gut Microbes 7 178 –184 . 10.1080/19490976.2016.1155022 26939746 Davenport E. R. Cusanovich D. A. Michelini K. Barreiro L. B. Ober C. Gilad Y. (2015 ). Genome-wide association studies of the human gut microbiota. PLoS ONE 10 :e0140301 10.1371/journal.pone.0140301 Dicksved J. Halfvarson J. Rosenquist M. Jarnerot G. Tysk C. Apajalahti J. (2008 ). Molecular analysis of the gut microbiota of identical twins with Crohn’s disease. ISME J. 2 716 –727 . 10.1038/ismej.2008.37 18401439 Everard A. Belzer C. Geurts L. Ouwerkerk J. P. Druart C. Bindels L. B. (2013 ). Cross-talk between Akkermansia muciniphila and intestinal epithelium controls diet-induced obesity. Proc. Natl. Acad. Sci. U.S.A. 110 9066 –9071 . 10.1073/pnas.1219451110 23671105 Fu B. C. Randolph T. W. Lim U. Monroe K. R. Cheng I. Wilkens L. R. (2016 ). Characterization of the gut microbiome in epidemiologic studies: the multiethnic cohort experience. Ann. Epidemiol. 26 373 –379 . 10.1016/j.annepidem.2016.02.009 27039047 Gacias M. Gaspari S. Mae-Santos P. Tamburini S. Andrade M. Zang F. (2016 ). Microbiota-driven transcriptional changes in prefrontal cortex override genetic differences in social behavior. Elife 5 e13442 10.7554/eLife.13442 Goodrich J. K. Waters J. L. Poole A. C. Sutter J. L. Koren O. Blekhman R. (2014 ). Human genetics shape the gut microbiome. Cell 159 789 –799 . 10.1016/j.cell.2014.09.053 25417156 Haque T. R. Barritt A. S. IV (2016 ). Intestinal microbiota in liver disease. Best Pract. Res. Clin. Gastroenterol. 30 133 –142 . 10.1016/j.bpg.2016.02.004 27048904 Human Microbiome Project Consortium (2012a ). A framework for human microbiome research. Nature 486 215 –221 . 10.1038/nature11209 22699610 Human Microbiome Project Consortium (2012b ). Structure, function and diversity of the healthy human microbiome. Nature 486 207 –214 . 10.1038/nature11234 22699609 Igartua C. Davenport E. R. Chupp G. L. Elias J. A. Gilad Y. Ober C. (2016 ). Nasal microbiome composition is associated with chitotriosidase (chit1) activity in adult hutterites. Ann. Am. Thorac. Soc. Suppl. 1 S100 –S101 . Jiang H. Ling Z. Zhang Y. Mao H. Ma Z. Yin Y. (2015 ). Altered fecal microbiota composition in patients with major depressive disorder. Brain Behav. Immun. 48 186 –194 . 10.1016/j.bbi.2015.03.016 25882912 Kassinen A. Krogius-Kurikka L. Makivuokko H. Rinttila T. Paulin L. Corander J. (2007 ). The fecal microbiota of irritable bowel syndrome patients differs significantly from that of healthy subjects. Gastroenterology 133 24 –33 . 10.1053/j.gastro.2007.04.005 17631127 Kau A. L. Ahern P. P. Griffin N. W. Goodman A. L. Gordon J. I. (2011 ). Human nutrition, the gut microbiome and the immune system. Nature 474 327 –336 . 10.1038/nature10213 21677749 Khachatryan Z. A. Ktsoyan Z. A. Manukyan G. P. Kelly D. Ghazaryan K. A. Aminov R. I. (2008 ). Predominant role of host genetics in controlling the composition of gut microbiota. PLoS ONE 3 :e3064 10.1371/journal.pone.0003064 Lagier J. C. Million M. Hugon P. Armougom F. Raoult D. (2012 ). Human gut microbiota: repertoire and variations. Front. Cell. Infect. Microbiol. 2 :136 10.3389/fcimb.2012.00136 Leamy L. J. Kelly S. A. Nietfeldt J. Legge R. M. Ma F. Hua K. (2014 ). Host genetics and diet, but not immunoglobulin A expression, converge to shape compositional features of the gut microbiome in an advanced intercross population of mice. Genome Biol. 15 :552 10.1186/s13059-014-0552-6 Lee K. M. Lee J. M. (2014 ). Crohn’s disease in Korea: past, present, and future. Korean J. Intern. Med. 29 558 –570 . 10.3904/kjim.2014.29.5.558 25228829 Ley R. E. (2010 ). Obesity and the human microbiome. Curr. Opin. Gastroenterol. 26 5 –11 . 10.1097/MOG.0b013e328333d751 19901833 Linares D. M. Ross P. Stanton C. (2016 ). Beneficial microbes: the pharmacy in the gut. Bioengineered 7 11 –20 . 10.1080/21655979.2015.1126015 26709457 Liu J. Z. Anderson C. A. (2014 ). Genetic studies of Crohn’s disease: past, present and future. Best Pract. Res. Clin. Gastroenterol. 28 373 –386 . 10.1016/j.bpg.2014.04.009 24913378 Macfarlane S. Furrie E. Kennedy A. Cummings J. H. Macfarlane G. T. (2005 ). Mucosal bacteria in ulcerative colitis. Br. J. Nutr. 93(Suppl. 1) S67 –S72 . 10.1079/BJN20041347 15877898 Manuc T. E. Manuc M. M. Diculescu M. M. (2016 ). Recent insights into the molecular pathogenesis of Crohn’s disease: a review of emerging therapeutic targets. Clin. Exp. Gastroenterol. 9 59 –70 . 10.2147/CEG.S53381 27042137 Mathewson N. D. Jenq R. Mathew A. V. Koenigsknecht M. Hanash A. Toubai T. (2016 ). Gut microbiome-derived metabolites modulate intestinal epithelial cell damage and mitigate graft-versus-host disease. Nat. Immunol. 17 505 –513 . 10.1038/ni.3400 26998764 Mu C. Yang Y. Zhu W. (2016 ). Gut microbiota: The Brain Peacekeeper. Front. Microbiol. 7 :345 10.3389/fmicb.2016.00345 Murphy K. O’ Shea C. A. Ryan C. A. Dempsey E. M. O’ Toole P. W. Stanton C. (2015 ). The gut microbiota composition in dichorionic triplet sets suggests a role for host genetic factors. PLoS ONE 10 :e0122561 10.1371/journal.pone.0122561 Nadal I. Donat E. Donant E. Ribes-Koninckx C. Calabuig M. Sanz Y. (2007 ). Imbalance in the composition of the duodenal microbiota of children with coeliac disease. J. Med. Microbiol. 56 1669 –1674 . 10.1099/jmm.0.47410-0 18033837 O’Mahony S. M. Marchesi J. R. Scully P. Codling C. Ceolho A. M. Quigley E. M. (2009 ). Early life stress alters behavior, immunity, and microbiota in rats: implications for irritable bowel syndrome and psychiatric illnesses. Biol. Psychiatry 65 263 –267 . 10.1016/j.biopsych.2008.06.026 18723164 Parkes M. Barrett J. C. Prescott N. J. Tremelling M. Anderson C. A. Fisher S. A. (2007 ). Sequence variants in the autophagy gene IRGM and multiple other replicating loci contribute to Crohn’s disease susceptibility. Nat. Genet. 39 830 –832 .17554261 Parks B. W. Nam E. Org E. Kostem E. Norheim F. Hui S. T. (2013 ). Genetic control of obesity and gut microbiota composition in response to high-fat, high-sucrose diet in mice. Cell Metab. 17 141 –152 . 10.1016/j.cmet.2012.12.007 23312289 Pluznick J. L. Protzko R. J. Gevorgyan H. Peterlin Z. Sipos A. Han J. (2013 ). Olfactory receptor responding to gut microbiota-derived signals plays a role in renin secretion and blood pressure regulation. Proc. Natl. Acad. Sci. U.S.A. 110 4410 –4415 . 10.1073/pnas.1215927110 23401498 Rhee S. H. Pothoulakis C. Mayer E. A. (2009 ). Principles and clinical implications of the brain-gut-enteric microbiota axis. Nat. Rev. Gastroenterol. Hepatol. 6 306 –314 .19404271 Riiser A. (2015 ). The human microbiome, asthma, and allergy. Allergy Asthma Clin. Immunol. 11 :35 10.1186/s13223-015-0102-0 Russell S. L. Gold M. J. Hartmann M. Willing B. P. Thorson L. Wlodarska M. (2012 ). Early life antibiotic-driven changes in microbiota enhance susceptibility to allergic asthma. EMBO Rep. 13 440 –447 . 10.1038/embor.2012.32 22422004 Savage D. C. (1977 ). Microbial ecology of the gastrointestinal tract. Ann. Rev. Microbiol. 31 107 –133 . 10.1146/annurev.mi.31.100177.000543 334036 Sokol H. Lepage P. Seksik P. Dore J. Marteau P. (2006a ). Temperature gradient gel electrophoresis of fecal 16S rRNA reveals active Escherichia coli in the microbiota of patients with ulcerative colitis. J. Clin. Microbiol. 44 3172 –3177 . 10.1128/JCM.02600-05 16954244 Sokol H. Seksik P. Rigottier-Gois L. Lay C. Lepage P. Podglajen I. (2006b ). Specificities of the fecal microbiota in inflammatory bowel disease. Inflamm. Bowel Dis. 12 106 –111 . 10.1097/01.MIB.0000200323.38139.c6 16432374 Strober W. Asano N. Fuss I. Kitani A. Watanabe T. (2014 ). Cellular and molecular mechanisms underlying NOD2 risk-associated polymorphisms in Crohn’s disease. Immunol. Rev. 260 249 –260 . 10.1111/imr.12193 24942694 Turnbaugh P. J. Hamady M. Yatsunenko T. Cantarel B. L. Duncan A. Ley R. E. (2009 ). A core gut microbiome in obese and lean twins. Nature 457 480 –484 . 10.1038/nature07540 19043404 Turnbaugh P. J. Ley R. E. Hamady M. Fraser-Liggett C. M. Knight R. Gordon J. I. (2007 ). The human microbiome project. Nature 449 804 –810 . 10.1038/nature06244 17943116 Turnbaugh P. J. Ley R. E. Mahowald M. A. Magrini V. Mardis E. R. Gordon J. I. (2006 ). An obesity-associated gut microbiome with increased capacity for energy harvest. Nature 444 1027 –1031 . 10.1038/nature05414 17183312 Walter J. Ley R. (2011 ). The human gut microbiome: ecology and recent evolutionary changes. Annu. Rev. Microbiol. 65 411 –429 . 10.1146/annurev-micro-090110-102830 21682646 Wang Y. Kasper L. H. (2014 ). The role of microbiome in central nervous system disorders. Brain Behav. Immun. 38 1 –12 . 10.1016/j.bbi.2013.12.015 24370461 Yarandi S. S. Peterson D. A. Treisman G. J. Moran T. H. Pasricha P. J. (2016 ). Modulatory effects of gut microbiota on the central nervous system: how gut could play a role in neuropsychiatric health and diseases. J. Neurogastroenterol. Motil. 22 201 –212 . 10.5056/jnm15146 27032544 Yatsunenko T. Rey F. E. Manary M. J. Trehan I. Dominguez-Bello M. G. Contreras M. (2012 ). Human gut microbiome viewed across age and geography. Nature 486 222 –227 . 10.1038/nature11053 22699611 Zoetandal E. G. Akkermans A. D. L. Akkermans-van Vliet W. M. de Visser J. A. de Vos W. M. (2001 ). The host genotype affects the bacterial community in the human gastrointestinal tract. Microb. Ecol. Health Disease 13 129 –134 . 10.1080/089106001750462669

PMC005xxxxxx/PMC5003824.txt

==== Front Front Plant SciFront Plant SciFront. Plant Sci.Frontiers in Plant Science1664-462XFrontiers Media S.A. 10.3389/fpls.2016.01304Plant ScienceOriginal ResearchBrassinosteroids Improve Quality of Summer Tea (Camellia sinensis L.) by Balancing Biosynthesis of Polyphenols and Amino Acids Li Xin 1*Ahammed Golam J. 12Li Zhi-Xin 13Zhang Lan 1Wei Ji-Peng 1Shen Chen 13Yan Peng 1Zhang Li-Ping 1Han Wen-Yan 1*1Key Laboratory of Tea Biology and Resources Utilization, Ministry of Agriculture, Tea Research Institute, Chinese Academy of Agricultural SciencesHangzhou, China2Department of Horticulture, Zhejiang UniversityHangzhou, China3Graduate School of Chinese Academy of Agricultural SciencesBeijing, ChinaEdited by: Vijay Pratap Singh, Government Ramanuj Pratap Singhdev Post Graduate College, India Reviewed by: Parvaiz Ahmad, Sri Pratap College, India; Mukesh Kumar Kanwar, Sri Guru Granth Sahib World University, India *Correspondence: Wen-Yan Han, hanwy@tricaas.com Xin Li, lixin@tricaas.comThis article was submitted to Plant Physiology, a section of the journal Frontiers in Plant Science 30 8 2016 2016 7 130421 6 2016 15 8 2016 Copyright © 2016 Li, Ahammed, Li, Zhang, Wei, Shen, Yan, Zhang and Han.2016Li, Ahammed, Li, Zhang, Wei, Shen, Yan, Zhang and HanThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.Summer grown green tea is less popular due to bitterness and high astringency, which are attributed to high levels of tea polyphenols (TP) and low levels of amino acids (AA) in tea leaves (Camellia sinensis L.). Brassinosteroids (BRs), a group of steroidal plant hormones can regulate primary and secondary metabolism in a range of plant species under both normal and stress conditions. However, specific effects of BRs on the photosynthesis of tea plants and the quality of summer green tea are largely unknown. Here we show that 24-epibrassinolide (EBR), a bioactive BR, promoted photosynthesis in tea plants in a concentration-dependent manner. Stimulation in photosynthesis by EBR resulted in an increased summer tea yield. Although all tested concentrations (0.01, 0.05, 0.1, 0.5, and 1.0 ppm) of EBR increased concentrations of TP and AA, a moderate concentration (0.5 ppm) caused the highest decrease in TP to AA ratio, an important feature of quality tea. Time-course analysis using 0.5 ppm EBR as foliar spray revealed that TP or AA concentration increased as early as 3 h after EBR application, reaching the highest peak at 24 h and that remained more or less stable. Importantly, such changes in TP and AA concentration by EBR resulted in a remarkably decreased but stable TP to AA ratio at 24 h and onward. Furthermore, concentrations of catechins and theanine increased, while that of caffeine remained unaltered following treatment with EBR. EBR improved activity of phenylalanine ammonia-lyase (PAL) and glutamine: 2-oxoglutarate aminotransferase (GOGAT) enzymes involved in catechins and theanine biosynthesis, respectively. Transcript analysis revealed that transcript levels of CsPAL and CsGS peaked as early as 6 h, while that of CsGOGAT peaked at 12 h following application of EBR, implying that EBR increased the concentration of TP and AA by inducing their biosynthesis. These results suggest a positive role of BR in enhancing green tea quality, which might have potential implication in improving quality of summer tea. 24-epibrassinolideamino acidsphotosynthesispolyphenolsecondary metabolismsummer teatea qualityChinese Academy of Agricultural Sciences10.13039/501100005196CAAS-ASTIP-2015-TRICAASNational Natural Science Foundation of China10.13039/5011000018094117121831550110201 ==== Body Introduction Green tea is a widely consumed beverage both in the east and the west (Tounekti et al., 2013). It is an unfermented tea, produced from the young leaves (usually the apical bud and two leaves below the bud) of Camellia sinensis L., where major portion of tea polyphenols (TP) is kept unoxidized during tea processing. TP are flavonoids, synthesized through secondary metabolic pathways such as shikimate pathway, phenylpropanoid pathway, and flavonoid pathway (Tounekti et al., 2013; Verma and Shukla, 2015). TP have numerous health benefits including prevention of cancer, cardiovascular, neurodegenerative and other oxidative stress-related diseases (Bordoni et al., 2002). Furthermore, TP are considered as one of the important determinants of tea quality as they impart astringency. Nonetheless, too high TP concentrations make tea infusion too bitter (Liang et al., 1996). In addition to TP, concentration of total free amino acids (AA) is a major factor determining green tea quality. AA are responsible for the freshness and mellowness of green tea. However, the quality of green tea not only depends on concentration of TP and AA, rather largely on TP to AA ratio. In general, the value of TP to AA ratio is inversely correlated with green tea quality, where a lower TP to AA ratio makes tea more brisk and mellow, but less bitter in taste (Liang et al., 1996; Wang et al., 2011; Xu et al., 2012; Tounekti et al., 2013). It is well known that the levels of TP and AA are influenced by many factors, including harvest season. The TP to AA ratio varies mainly due to variations in the day length, rainfall, sunlight, and/or temperature that are distinct in each season (Tounekti et al., 2013). Based on the harvest seasons, Chinese green tea can be divided into three types such as “spring tea,” “summer tea,” and “autumn tea” (Xu et al., 2012). Spring teas, which are harvested in cool months (before late May) comprise a higher level of AA but a moderate level of TP with an optimum TP to AA ratio. On the other hand, summer teas which are harvested in warmer months (between early June and early July) possess a lower level of AA and a higher level of TP and TP to AA ratio. Therefore, the spring teas taste better than the summer teas and autumn teas. Such metabolic response to season is possibly conserved among all genotypes of C. sinensis L. (Wang et al., 2011). This implies that the cultivation of the best quality tea is limited to a short period (spring season) of a year (Xu et al., 2012). Therefore, development of strategies for improving summer tea quality is one of the cutting edge issues in the tea research. Phytohormones are endogenous messenger molecules that regulate various aspects of plant growth, development, and responses to stress (Ahammed et al., 2014). A prior study showed that exogenous application of phytohormone gibberellins (GA) could not only increase AA concentration but also decrease TP concentration, leading to a decreased TP to AA ratio in green tea (Liang et al., 1996). Exogenous application of methyl jasmonate (MeJA) improves the aroma quality of black tea (Shi et al., 2014). Furthermore, individual treatment with GA and abscisic acid alters the concentration of catechins (flavan-3-ols, major bioactive TP) as well as transcript levels of its biosynthetic genes such as CsPAL, CsC4H, Cs4CL, CsF3H, and CsANR (Singh et al., 2009; Rani et al., 2012). All these reports clearly indicate that phytohormones are potentially involved in controlling the quality of tea by modulating secondary metabolism in tea plant. Brassinosteroids (BRs), a group of steroidal plant hormones, play critical roles in the regulation of both primary and secondary metabolism in a range of plant species (Ahammed et al., 2012, 2014; Çoban and Göktürk Baydar, 2016). For instance, exogenous application of 24-epibrassinolide (EBR, a bioactive BR) promotes CO2 assimilation capacity by enhancing ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO) carboxylation rate, activities of RuBisCO activase and fructose-1,6-bisphosphatase in cucumber plants (Yu et al., 2004; Jiang et al., 2012a). BRs regulate Benson–Calvin cycle and sugar metabolism via redox signaling, which eventually increases the photosynthetic potential and biomass accumulation in plants (Jiang et al., 2012c). Exogenous application of EBR could increase the activity of secondary metabolism-related enzymes such as phenylalanine ammonia-lyase (PAL, the first enzyme involved in flavonoid biosynthesis), resulting in an increased concentration of phenols and flavonoids in tomato roots and Vitis vinifera grape berry (Ahammed et al., 2012; Xi et al., 2013). In peppermint (Mentha piperita L.), foliar application of 0.5 mg l-1 EBR could increase total phenolic content around twofold in leaves compared with that of control plants (Çoban and Göktürk Baydar, 2016). Moreover, combined application of EBR and MeJA could sharply induce concentration of secondary metabolites in sweet basil (Ocimum basilicum L.; Koca and Karaman, 2015). Although BRs were identified in tea leaves (Thea sinensis) about 35 years ago, soon after the discovery of brassinolide in the pollen of Brassica napus (Morishita et al., 1983), no literature is currently available for the effects of BRs on the secondary metabolism in tea plants. Notably, characterization of novel microRNA (miRNA) in tea suggests that miRNA-mediated BR signaling might play an important role in regulating developmental and seasonal variations in tea (Mohanpuria and Yadav, 2012). A recent study showed that amiRNA designated as cs-miR414 is profoundly expressed in dormant bud of tea compared with that in active bud, and cs-miR414 targets mRNAs that are involved in maintaining the endogenous concentration of BRs and its homeostasis, implying that BRs level is critical for bud dormancy in tea (Jeyaraj et al., 2014). Based on these recent reports and the role of BRs in secondary metabolism in various plant species, we hypothesized that BRs might influence the concentration of TP and AA in tea leaves, and thus the quality of green tea. To elucidate specific effects of BRs on the photosynthesis of tea plants and the quality of green tea, we analyzed the CO2 assimilation rate in tea plants after application of EBR. In addition, we examined effects of EBR on the quality of green tea by measuring concentration and ratio of TP and AA as well as concentration of catechins, theanine and caffeine. Our results suggest that BRs have a significant stimulatory effect on photosynthesis and quality of green tea. Materials and Methods Plant Material and Growth Conditions For the current experiment, widely cultivated tea (C. sinensis L.) cultivar in China namely Longjing 43 was chosen. The experiment was conducted during summer (in the month of July), when mean maximum and minimum temperatures were 37.4 ± 0.55°C and 28.2 ± 1.3°C, respectively and relative humidity: 42.0 ± 3.2% at tea garden of the Tea Research Institute, Chinese Academy of Agricultural Sciences, Hangzhou, Zhejiang province, China (longitude 120°10′E and latitude 30°14′N, 16 m above sea level). Foliar portion of tea bushes was sprayed with freshly prepared 24-EBR (Sigma-Aldrich, St. Louis, MO, USA). A graded levels of working solution of EBR (0.01, 0.05, 0.1, 0.5, and 1.0 ppm) was prepared by dissolving solute in ethanol followed by dilution with MilliQ water [ethanol:water (v/v) = 1:10000]. Control (CK) tea bushes were simultaneously sprayed with MilliQ water containing same ratio of ethanol. Each treatment comprises four replicates, while each replicate represents an area of 10 m2 consisting of 20 tea bushes. Photosynthesis Measurement Twenty-four hours after EBR application, net CO2 assimilation rate (Pn) was measured on 3rd fully expanded leaves using an open-flow infrared gas analyzer adapted with light and temperature control systems (Li-COR 6400, Lincoln, NE, USA) in 12 tea bushes under each treatment. The measurement was performed within the time period from 8:00 am to 11:00 am maintaining the air temperature, relative humidity, CO2 concentration and photosynthetic photon flux density (PPFD) at 25°C, 80%, 400 μmol mol-1 and 800 μmol m-2 s-1, respectively. Quantification of Tea Polyphenols and Total Free Amino Acids Total TP was extracted and determined spectrophotometrically according to the standard method established by the International Organization for Standardization (ISO) 14502-1 using gallic acid as standard (ISO 14502-1, 2005; Anesini et al., 2008). In brief, the diluted sample extract (1.0 ml) was transferred to tubes in duplicate, where each tube contained 5.0 ml of a 1/10 dilution of Folin–Ciocalteu’s reagent in water. Afterward, 4.0 ml sodium carbonate solution (7.5% w/v) was added into each tube. The tubes were kept at room temperature for 60 min before absorbance at 765 nm was measured against water. Total AA from tea leaf sample (0.5 g) were extracted in 80% ethanol at 80°C. Following evaporation, dried samples were dissolved in 0.02 N HCl. AA, separated by cation-exchange chromatography, were subjected to postcolumn reaction with ninhydrin reagent and detected spectrophotometrically as described previously elsewhere (Chen et al., 2010). Determination of Catechins, Caffeine, and Individual Amino Acids The concentrations of caffeine and catechins in the extract were determined with a high-performance liquid chromatography (HPLC) system (Waters 590; Waters Corp., Milford, MA, USA) equipped with a Hypersil ODS2 C18 column (5 ml, 4.6 mm × 250 mm, 35°C) at 280 nm as previously described (Su et al., 2003). Solvents A (2% acetic acid) and B (acetonitrile) were run in linear gradients with A decreasing from 93 to 55% within 20 min and maintained for 5 min, thereafter at a rate of 1.4 ml min-1. Concentrations of caffeine and catechins were quantified by their peak areas against those of standards prepared from authentic compounds. Individual AA (theanine) were measured by using an automatic AA analyzer (Hitachi L-8900, Japan). Five milliliters of tea extract was added with 5 ml of sulfosalicylic acid and the mixture was centrifuged at 13,000 rpm for 5 min to facilitate the reaction. The mixture was filtered through a 0.20 μm nylon filter membrane and run in the AA analyzer (Wang et al., 2006). Determination of PAL and GOGAT Activities For PAL activity assay, 0.3 g tea leaf sample was homogenized in 3 ml 50 mM potassium phosphate buffer (pH 8.8, containing 2 mM ethylenediaminetetraacetic acid (EDTA), 2% polyvinyl polypyrrolidone (PVPP), and 0.1% mercaptoethanol). The homogenate was centrifuged at 15,000 rpm for 20 min at 4°C and the supernatant fractions were collected as a crude enzyme extract. PAL activity was estimated with L-phenylalanine as substrate (Zheng et al., 2005). The extract for glutamine: 2-oxoglutarate aminotransferase (GOGAT) measurement was obtained by grinding 0.3 g frozen leaf sample in 2 ml 25 mM Tris–HCl buffer (pH 7.8) containing 1 mM MgCl2, 1 μM β-mercaptoethanol, 1 mM EDTA, and 1% (w/v) PVPP. For assays, the extract was centrifuged at 15,000 rpm for 20 min, and the GOGAT activities were measured in the supernatant as described by Suzuki et al. (2001). RNA Isolation and Transcript Analysis Total RNA from tea leaves was prepared using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instruction. A purifying column was used to remove genomic DNA from RNA samples. Reverse transcription was done using Superscript II (Invitrogen) following the manufacturer’s protocol. The primers used for transcript analysis have been listed in Supplementary Table S2. qRT-PCR analysis was carried out using the StepOnePlus Real-Time PCR system (Applied Biosystems, Foster City, CA, USA) with Power SYBR Green PCR Master Mix (Applied Biosystems). Transcript abundance was normalized to actin, and relative gene expression was calculated following formulae of Livak and Schmittgen (2001). qRT-PCR conditions consisted of denaturation at 95°C for 3 min, followed by 40 cycles of denaturation at 95°C for 30 s, annealing at 58°C for 30 s and extension at 72°C for 30 s. Statistical Analysis All data were analyzed using the statistical software SAS 8.1 (SAS Institute Inc., Cary, NC, USA). Following variance analysis, an analysis of least significant difference (P < 0.05) among means was performed by the Duncan’s multiple range test. Results Photosynthetic Response of Tea Plants to Exogenous EBR Is Concentration Dependent Several lines of evidence suggest that plant response to BR is highly concentration specific (Jiang et al., 2012b; Ahammed et al., 2014). To understand the photosynthetic response of tea plants to BR, we treated foliar portion of tea plants with various concentrations of EBR ranging from 0.01 to 1.0 ppm. Twenty-four hours after application of EBR, net photosynthetic rate (Pn) was measured in control (CK) and EBR-treated tea plants. As shown in Figure 1, the lowest concentration of EBR (0.01) had no effect on Pn, while 0.05 ppm EBR caused a significant increase in Pn, leading to a gradual increase in Pn up to 0.1 ppm EBR. After reaching the highest Pn value at 0.1 ppm EBR, which was increased by 56.5% compared with that in CK, Pn tended to decrease with increasing EBR concentrations. Nonetheless, the Pn at the highest concentration of EBR (1.0 ppm) still remained significantly higher than that in CK. These results suggest that except for 0.01 ppm, rest of the concentrations of EBR showed a positive stimulatory effect on photosynthesis in tea plants. FIGURE 1 Effect of different concentrations of 24-epibrassinolide (EBR) on net photosynthetic rate (Pn) in tea plants. Tea bushes were sprayed with graded levels of EBR (0.01, 0.05, 0.1, 0.5, and 1.0 ppm) solution. Control (CK) tea bushes were simultaneously sprayed with MilliQ water containing same ratio of ethanol used to dissolve EBR. Twenty-four hours after EBR application, Pn was measured on third fully expanded leaves using an open-flow infrared gas analyzer adapted with light and temperature control systems (Li-COR 6400, Lincoln, NE, USA). The results are expressed as the mean values ± SD. Mean denoted by different letters indicate significant differences between the treatments (P < 0.05). Effect of EBR on Summer Tea Yield To assess the effect of EBR on growth and biomass production in tea plants, we measured the length of tea sprout, density of tea bud, weight of 100 buds, and summer tea yield (Table 1). Length of sprout increased with 0.1–1.0 ppm EBR, while density of bud increased with 0.05–1.0 ppm EBR. Likewise, weight of 100 buds showed an increased value under 0.05–1.0 ppm EBR treatment, where 0.1 ppm had the greatest effect. Similar to the trends in density of bud, yield of summer tea gradually increased with increasing concentration of EBR, resulting in the highest yield at 0.1 ppm EBR (increased by 37.79% compared with the CK). However, further increment in EBR concentration (0.5–1.0 ppm) slightly decreased the yield as compared with that in 0.1 ppm EBR. Table 1 Effects of 24-epibrassinolide (EBR) on the growth and yield of summer tea. EBR concentration (ppm) Length of sprout (cm) Density of bud (bud/m2) Weight of 100 buds (g) Yield (g/m2) 0 1.96 ± 0.105b 1701 ± 51.8d 9.30 ± 0.681c 260.7 ± 18.78c 0.01 2.00 ± 0.119b 1716 ± 114.6cd 9.55 ± 0.901bc 286.0 ± 17.71bc 0.05 2.15 ± 0.146ab 1976 ± 135.6b 10.45 ± 0.717ab 309.1 ± 21.00b 0.1 2.28 ± 0.130a 2274 ± 163.5a 11.07 ± 0.940 a 359.2 ± 21.36a 0.5 2.22 ± 0.097a 2139 ± 155.3ab 10.62 ± 0.837ab 320.8 ± 23.82b 1 2.19 ± 0.128a 1850 ± 107.5bc 10.54 ± 0.552ab 292.0 ± 20.02bc Mean denoted by different letters indicate significant differences between the treatments (P < 0.05).EBR Stimulates Tea Quality in a Concentration-Dependent Manner The composition of green tea especially TP, AA, and TP to AA ratio (TP/AA) are the major determinants of the quality of green tea (Liang et al., 1996; Tounekti et al., 2013). In our experimental tea garden, compared with spring tea, the concentration of TP in summer tea increased by 23%, while concentration of AA decreased by 31%, which resulted in a significantly increased TP to AA ratio (Supplementary Table S1). To assess whether exogenous EBR could alter the quality of green tea, we analyzed the concentrations of TP and AA, and the TP to AA ratio following application of various concentration of EBR. Results showed that all treated concentrations of EBR increased TP concentration in tea leaves, having highest increase (25.15%) at 0.05 ppm and lowest increase (7.74%) at 1.0 ppm EBR as compared with that in CK (Figure 2). Notably, TP gradually increased up to 0.05 ppm EBR, and then deceased with increasing concentration of EBR. Likewise, AA concentration increased with increasing concentration of exogenous EBR up to 0.05 ppm, and then remained more or less stable up to 0.5 ppm EBR. When EBR concentration was further increased, AA concentration decreased as compared with that at 0.5 ppm EBR. Thus, highest concentration of AA was recorded at 0.5 ppm EBR, which was increased by 50.20% compared with that in CK. Such alterations in TP and AA concentration, which were caused by exogenous EBR resulted in decreased TP to AA ratios, accounting for the lowest value at 0.5 ppm EBR. In quantitative figure, TP to AA ratio at 0.5 ppm EBR decreased by 25.13% compared with that in CK. All these results clearly indicated that effect of EBR on TP, AA, and TP to AA ratio was highly concentration dependent, where 0.5 ppm EBR showed the best effect by causing the highest reduction in TP to AA ratio, an important feature of quality green tea. FIGURE 2 Effect of different concentrations of 24-epibrassinolide (EBR) on total tea polyphenols (TP) and free amino acids (AA) concentrations in tea leaves. Tea bushes were sprayed with various concentration of EBR (0, 0.01, 0.05, 0.1, 0.5, and 1.0 ppm) at 48 h prior to quantification of TP and AA. The results are expressed as the mean values ± SD, n = 6. Mean denoted by different letters indicate significant differences between the treatments (P < 0.05). Next, we analyzed the concentrations of catechins, theanine, and caffeine in tea leaves following treatment with various concentrations of EBR (Table 2). Results showed that although all tested concentrations of EBR modulated catechins concentration, significant differences were noticed only at 0.05 and 0.1 ppm EBR. Theanine concentration was enhanced by 0.05–1.0 ppm EBR, where 0.5 ppm EBR caused the greatest increase (44.50%). However, the concentration of caffeine in tea leaves was not altered by the EBR treatment at any concentration (Table 2). Table 2 Effects of 24-epibrassinolide (EBR) on the bioactive compounds in tea leaves relating to the quality of tea. EBR concentration (ppm) Catechins (mmol g-1 DW) Theanine (mg g-1 DW) Caffeine (mg g-1 DW) 0 154.7 ± 10.71c 8.27 ± 0.493d 39.41 ± 1.764a 0.01 165.2 ± 9.33bc 8.98 ± 0.518cd 40.62 ± 2.028a 0.05 178.7 ± 11.64ab 10.34 ± 0.466b 38.23 ± 1.575a 0.1 189.3 ± 13.24a 11.41 ± 0.725ab 40.44 ± 2.391a 0.5 171.6 ± 12.77abc 11.95 ± 0.937a 41.87 ± 2.726a 1 160.7 ± 13.53bc 9.77 ± 0.728bc 39.36 ± 1.311a Mean denoted by different letters indicate significant differences between the treatments (P < 0.05). DW, dry weight.Time-Course Response of Tea Quality Parameters to Exogenous EBR Based on TP to AA ratio data, we selected 0.5 ppm EBR to assess time-course response of TP and AA to exogenous EBR application. As shown in Figure 3, concentration of TP increased as early as 3 h after EBR application, reaching the maximum level at 12 h and then slightly decreased up to 48 h. Compared with the control treatment, EBR application increased TP concentration by 12.95, 13.13, 19.20, 16.78, and 11.01% after 3, 6, 12, 24, and 48 h, respectively. Similar to TP, concentration of AA also increased over time following EBR application. The concentration of AA increased by 14.41, 24.01, 27.85, 35.86, and 27.19% at 3, 6, 12, 24, and 48 h, respectively in EBR-treated tea leaves compared with that in CK. Importantly, such changes in TP or AA concentration by EBR application resulted in a remarkably decreased (by 14.04%) but stable TP to AA ratio at 24 h and onward. These results clearly suggest that EBR-induced improvement in tea quality is stable. FIGURE 3 Time-course effect of 24-epibrassinolide (EBR) on total tea polyphenols (TP) and free amino acids (AA) concentrations in tea leaves. Tea bushes were sprayed with 0.5 ppm EBR. Measurements were taken at different time-points as mentioned in the respective figures. The results are expressed as the mean values ± SD, n = 6. Response of Secondary Metabolism-Related Enzymes and Genes to Exogenous EBR Next, we analyzed the activity of PAL, the first enzyme of phenylpropanoid pathway and the activity of GOGAT, a critical enzyme involved in theanine biosynthesis. In accord with time-course analysis of TP concentration, PAL activity increased gradually over time, reaching the peak at 12 h after EBR treatment (Figure 4A). Afterward, PAL activity declined and reached to the level of CK at 48 h. The activity GOGAT sharply increased at 12 h and remained higher up to 48 h after EBR treatment (Figure 4B). FIGURE 4 Changes in the activity of (A) phenylalanine ammonia-lyase (PAL) and (B) glutamine: 2-oxoglutarate aminotransferase (GOGAT) as influenced by exogenous 24-epibrassinolide (EBR). Leaf samples were harvested at indicated time-points following foliar spray of EBR. The results are expressed as the mean values ± SD, n = 6. To confirm whether the change in tea composition is attributed to a change in the biosynthesis of secondary metabolites, we analyzed the transcript levels of flavonoid biosynthetic genes CsPAL and theanine biosynthetic pathway-related genes CsGS and CsGOGAT. Transcript levels of all these genes were not much changed over time in CK leaves (Figure 4). However, in line with the TP concentration and PAL activity, transcript of CsPAL increased as early as 3 h, reaching the maximum level (1.3-fold) at 6 h after EBR treatment. Afterward, transcript levels of CsPAL gradually declined, but remained higher than that of CK even after 48 h. Transcripts of CsGS peaked at 6 h after EBR application and then gradually decreased up to 48 h. Notably, transcript of CsGOGAT reached maximum level (1.22-fold) at 12 h after EBR application, which was then slightly decreased but remained higher than that of CK treatment up to 48 h. Overall transcript data suggested that exogenous application of EBR stimulated transcriptional machinery causing accumulation of the highest levels of transcripts within early 6–12 h, which was then gradually declined, but remained higher than that in control treatment. Discussion Although green tea is often consumed considering its health benefits, its pleasant taste greatly influences overall consumption. For instance, a green tea that tastes bitter is not liked by the consumers, while a green tea that gives more brisk but less bitter taste is preferred by all (Xu et al., 2012). However, the production of quality green tea is greatly influenced by seasonal specificity (Tounekti et al., 2013). Summer days that are characterized by high temperatures have profound effect on the composition of green tea and thus TP to AA ratio in summer tea was significantly high compared with that in spring tea (Supplementary Table S1). In the current study, we showed that BRs, a well-known plant growth and stress hormone, could stimulate photosynthesis as well as secondary metabolism in tea plants during summer day, resulting in a decreased TP to AA ratio, a salient feature of quality green tea (Figures 1 and 2). Nonetheless, such response of tea plants to BR is dependent on the concentration of the exogenous EBR. Foliar application of EBR rapidly (as early as 3 h) induced transcript levels of key genes involved in the biosynthesis of catechins and theanine in tea leaves (Figure 5). Eventually such changes in transcript level resulted in a constant increment in the concentration of TP and AA but a decreased TP to AA ratio (Figures 2 and 3). These results suggest that BR might have potential to improve the quality of green tea beyond seasonal limitation. FIGURE 5 Transcript levels of catechins and theanine biosynthetic genes as influenced by exogenous 24-epibrassinolide (EBR). Leaf samples were harvested at indicated time-points following foliar spray of EBR. Transcript levels of genes were analyzed by qRT-PCR using gene-specific primer pairs (Supplementary Table S2). Photosynthesis is the basic physiological process that provides both substrate and energy for the synthesis of primary metabolites, while secondary metabolites are subsequently derived from the primary metabolites (Verma and Shukla, 2015). In the current study, exogenous application of EBR increased net photosynthetic rate (Pn) in tea plants (Figure 1). The rate of photosynthesis is dependent on two main biochemical processes such as RuBisCO carboxylation and RuBP regeneration. Previous studies have shown that EBR increases photosynthesis in plants by increasing RuBisCO carboxylation rate and initial activity of RuBisCO (Yu et al., 2004; Jiang et al., 2012a). Therefore, it is plausible that EBR might also stimulate similar biochemical process to enhance photosynthesis in tea plants. It is well accepted that BR-induced photosynthetic responses are highly concentration dependent, where a moderate concentration stimulates photosynthesis, but a low or high concentration of EBR inhibits photosynthesis (Jiang et al., 2012a,b). However, the highest concentration that we used in the current study was much lower than 5 mM EBR which can decrease CO2 assimilation in cucumber (Jiang et al., 2012b). Therefore, each concentration (except for 0.01 ppm) of EBR significantly increased Pn in tea plants (Figure 1). Several lines of evidence suggest that biosynthesis of secondary metabolites depends on various environmental cues (light, temperature, CO2, drought, salinity, ozone, UV-radiation) as well as endogenous signals (hormones and signaling molecules). Furthermore, change in only one factor can substantially alter the endogenous concentration of secondary metabolites even though other factors remain constant (Verma and Shukla, 2015). Similar to some other plant species such as tomato, grape, sweet basil and peppermint, exogenous application of EBR onto tea leaves remarkably stimulated secondary metabolism in the current study, which is evident by the significant increases in TP and catechins concentrations following foliar application of EBR (Ahammed et al., 2012; Xi et al., 2013; Koca and Karaman, 2015; Çoban and Göktürk Baydar, 2016). Notably, TP are synthesized through phenylpropanoid and flavonoid pathways, where PAL is the first enzyme that deaminated phenylalanine into cinnamic acid. CsPAL is the key gene that encodes PAL protein in tea. In our study, EBR induced transcript levels of CsPAL as early as 3 h after application, which was consistent with the enhanced concentration of TP at 3 h, indicating that EBR might stimulate the biosynthesis of TP by transcriptional regulation. Time-course analysis also revealed that EBR-induced changes in transcript levels resulted in a stable TP to AA ratio, which is sharply lower than that of CK, suggesting that EBR-mediated alteration in secondary metabolism is sustainable enough for improving green tea quality. However, our results argue with an earlier report concerning the effect of GA on TP in tea leaves, where GA remarkably decreased TP concentration (Liang et al., 1996). The discrepancy between two research findings could be due to difference in tea cultivars as well as kinds of plant hormones. It is worth mentioning that both BRs and polyphenols are synthesized from isopentenyl diphosphates that are provided by mevalonate pathway (Verma and Shukla, 2015). Therefore, BRs may function as a positive regulator of TP synthesis in plants. In the current study, exogenous application of EBR increased AA as well as theanine levels in tea leaves (Figure 2; Table 2), which was in agreement with the effect of GA on AA concentration in tea leaves (Liang et al., 1996). It has been reported that BRs modulated biosynthesis of AA such as proline and glycine betaine in various plant species (Vardhini, 2014). However, to our knowledge, no literature is available regarding the effect of BR on free AA and theanine concentration in tea leaves. It is to be noted that theanine is the major tea AA accounting for more than 50% of total free AA in tea (Mu et al., 2015). Two enzymes, such as glutamine synthetase (GS) and GOGAT, which are considered as key determinants of theanine biosynthesis, catalyze the initial steps of NH3 assimilation into glutamic acid (Mu et al., 2015). To elucidate the mechanism underlying BR-induced AA accumulation, we analyzed transcript levels of key genes involved in theanine synthesis such as GLUTAMINE SYNTHETASE (CsGS) and GLUTAMINE: 2-OXOGLUTARATE AMINOTRANSFERASE (CsGOGAT). As described in detailed under the Section “Results,” transcript levels of CsGS and CsGOGAT peaked as early as 6 and 12 h, respectively after foliar application of EBR, implying that EBR possibly promoted biosynthesis of theanine that largely contributed to the AA in tea leaves. In conclusion, this study demonstrated that exogenous application of EBR not only promoted photosynthesis and yield, but also stimulated secondary metabolism in tea plants. EBR-mediated enhancement in secondary metabolism resulted in an increased TP and AA concentration but a decreased TP to AA ratio, which is considered as a desired parameter of quality green tea. In addition, EBR increased concentration of catechins and theanine without affecting concentration of caffeine in tea leaves. Further investigation at transcript levels revealed that EBR upregulated the expression of key genes involved in the biosynthesis of catechins and theanine. Our research sheds new light on the role of BR in the regulation of tea quality and thus it will pave the way for deciphering the precise role of BRs in tea secondary metabolism. It is known that BRs improve photosynthesis under high temperature stress in a range of plant species (Ahammed et al., 2014). As we carried out current experiment during summer days, when maximum mean temperature was approximately 37°C, it seems highly likely that EBR-induced enhancement in photosynthesis and secondary metabolism was associated with BR-mediated attenuation of heat stress in tea plants. Therefore, it will be interesting to further explore the role of BR in high temperature tolerance in tea plants. Author Contributions XL and W-YH conceived and designed the research; XL, GA, Z-XL, LZ, J-PW, CS, PY, and L-PZ performed the experiments and analyzed the data; W-YH provided crucial reagents and supervised the study; XL and GA wrote the manuscript. All authors reviewed the manuscript. Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Funding. This work was supported by the Innovation Project of the Chinese Academy of Agricultural Sciences (CAAS-ASTIP-2015-TRICAAS) and the National Natural Science Foundation of China (project no. 41171218, 31550110201). Supplementary Material The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.01304 Click here for additional data file. ==== Refs References Ahammed G. J. Xia X. J. Li X. Shi K. Yu J. Q. Zhou Y. H. (2014 ). Role of brassinosteroid in plant adaptation to abiotic stresses and its interplay with other hormones. Curr. Protein Pept. Sci. 16 462 –473 . 10.2174/1389203716666150330141427 25824388 Ahammed G. J. Zhou Y. H. Xia X. J. Mao W. H. Shi K. Yu J. Q. (2012 ). Brassinosteroid regulates secondary metabolism in tomato towards enhanced tolerance to phenanthrene. Biol. Plant. 57 154 –158 . 10.1007/s10535-012-0128-9 Anesini C. Ferraro G. E. Filip R. (2008 ). Total polyphenol content and antioxidant capacity of commercially available tea (Camellia sinensis) in Argentina. J. Agric. Food Chem. 56 9225 –9229 . 10.1021/jf8022782 18778031 Bordoni A. Hrelia S. Angeloni C. Giordano E. Guarnieri C. Caldarera C. M. (2002 ). Green tea protection of hypoxia/reoxygenation injury in cultured cardiac cells. J. Nutr. Biochem. 13 103 –111 . 10.1016/S0955-2863(01)00203-0 11834226 Chen X. H. Zhuang C. G. He Y. F. Wang L. Han G. Q. Chen C. (2010 ). Photosynthesis, yield, and chemical composition of Tieguanyin tea plants (Camellia sinensis (L.) O. Kuntze) in response to irrigation treatments. Agric. Water Manag. 97 419 –425 . 10.1016/j.agwat.2009.10.015 Çoban Ö. Göktürk Baydar N. (2016 ). Brassinosteroid effects on some physical and biochemical properties and secondary metabolite accumulation in peppermint (Mentha piperita L.) under salt stress. Ind. Crop Prod. 86 251 –258 . 10.1016/j.indcrop.2016.03.049 ISO 14502-1 (2005 ). Determination of Substances Characteristic of Green and Black Tea. Part 1: Content of Total Polyphenols in Tea. Colorimetric Method Using Folin-Ciocalteu Reagent , ed. I.O.F. Standardization (Geneva : ISO ). Jeyaraj A. Chandran V. Gajjeraman P. (2014 ). Differential expression of microRNAs in dormant bud of tea [Camellia sinensis (L.) O. Kuntze]. Plant Cell Rep. 33 1053 –1069 . 10.1007/s00299-014-1589-4 24658841 Jiang Y. P. Cheng F. Zhou Y. H. Xia X. J. Mao W. H. Shi K. (2012a ). Brassinosteroid-induced CO2 assimilation is associated with increased stability of redox-sensitive photosynthetic enzymes in the chloroplasts in cucumber plants. Biochem. Biophys. Res. Commun. 426 390 –394 . 10.1016/j.bbrc.2012.08.100 22960180 Jiang Y. P. Cheng F. Zhou Y. H. Xia X. J. Mao W. H. Shi K. (2012b ). Cellular glutathione redox homeostasis plays an important role in the brassinosteroid-induced increase in CO2 assimilation in Cucumis sativus. New Phytol. 194 932 –943 . 10.1111/j.1469-8137.2012.04111.x 22432590 Jiang Y. P. Cheng F. Zhou Y. H. Xia X. J. Mao W. H. Shi K. (2012c ). Hydrogen peroxide functions as a secondary messenger for brassinosteroids-induced CO2 assimilation and carbohydrate metabolism in Cucumis sativus. J. Zhejiang Univ. Sci. B 13 811 –823 . 10.1631/jzus.B1200130 23024048 Koca N. Karaman S. (2015 ). The effects of plant growth regulators and L-phenylalanine on phenolic compounds of sweet basil. Food Chem. 166 515 –521 . 10.1016/j.foodchem.2014.06.065 25053088 Liang Y. Lu J. Shang S. (1996 ). Effect of gibberellins on chemical composition and quality of tea (Camellia sinensis L). J. Sci. Food Agric. 72 411 –414 . 10.1002/(SICI)1097-0010(199612)72:4<411::AID-JSFA672>3.0.CO;2-9 Livak K. J. Schmittgen T. D. (2001 ). Analysis of relative gene expression data using real-time quantitative PCR and the 2-ΔΔCT method. Methods 25 402 –408 . 10.1006/meth.2001.1262 11846609 Mohanpuria P. Yadav S. K. (2012 ). Characterization of novel small RNAs from tea (Camellia sinensis L.). Mol. Biol. Rep. 39 3977 –3986 . 10.1007/s11033-011-1178-3 21744261 Morishita T. Hiroshi A. Masaaki U. Shingo M. Suguru T. Nobuo I. (1983 ). Evidence for plant growth promoting brassinosteroids in leaves of Thea sinensis. Phytochemistry 22 1051 –1053 . 10.1016/0031-9422(83)85063-8 Mu W. Zhang T. Jiang B. (2015 ). An overview of biological production of L-theanine. Biotechnol. Adv. 33 335 –342 . 10.1016/j.biotechadv.2015.04.004 25871834 Rani A. Singh K. Ahuja P. S. Kumar S. (2012 ). Molecular regulation of catechins biosynthesis in tea [Camellia sinensis (L.) O. Kuntze]. Gene 495 205 –210 . 10.1016/j.gene.2011.12.029 22226811 Shi J. Wang L. Ma C. Y. Lv H. P. Chen Z. M. Lin Z. (2014 ). Aroma changes of black tea prepared from methyl jasmonate treated tea plants. J. Zhejiang Univ. Sci. B 15 313 –321 . 10.1631/jzus.B1300238 24711352 Singh K. Kumar S. Rani A. Gulati A. Ahuja P. S. (2009 ). Phenylalanine ammonia-lyase (PAL) and cinnamate 4-hydroxylase (C4H) and catechins (flavan-3-ols) accumulation in tea. Funct. Integr. Genom. 9 125 –134 . 10.1007/s10142-008-0092-9 Su Y. L. Leung L. K. Huang Y. Chen Z. Y. (2003 ). Stability of tea theaflavins andcatechins. Food Chem. 83 189 –195 . 10.1016/S0308-8146(03)00062-1 Suzuki A. Rioual S. Lemarchand S. Godfroy N. Roux Y. Boutin J. P. (2001 ). Regulation by light and metabolites of ferredoxin-dependent glutamate synthasein maize. Physiol. Plant. 112 524 –530 . 10.1034/j.1399-3054.2001.1120409.x 11473712 Tounekti T. Joubert E. Hernández I. Munné-Bosch S. (2013 ). Improving the Polyphenol content of tea. Crit. Rev. Plant Sci. 32 192 –215 . 10.1080/07352689.2012.747384 Vardhini B. (2014 ). “Brassinosteroids’ role for amino acids, peptides and amines modulation in stressed plants-a review ,” in Plant adaptation to environmental change: Significance of amino acids and their derivatives eds Anjum N. A. Gill S. S. Gill R. (Wallingford : CAB International ), 300 –316 . 10.1079/9781780642734.0300 Verma N. Shukla S. (2015 ). Impact of various factors responsible for fluctuation in plant secondary metabolites. J. App. Res. Med. Arom. Plants 2 105 –113 . 10.1016/j.jarmap.2015.09.002 Wang H. F. Tsai Y. S. Lin M. L. Ou A. S. M. (2006 ). Comparison of bioactive components in GABA tea and green tea produced in Taiwan. Food Chem. 96 648 –653 . 10.1016/j.foodchem.2005.02.046 Wang L. Y. Wei K. Jiang Y. W. Cheng H. Zhou J. He W. (2011 ). Seasonal climate effects on flavanols and purine alkaloids of tea (Camellia sinensis L.). Eur. Food Res. Technol. 233 1049 –1055 . 10.1007/s00217-011-1588-4 Xi Z. M. Zhang Z. W. Huo S. S. Luan L. Y. Gao X. Ma L. N. (2013 ). Regulating the secondary metabolism in grape berry using exogenous 24-epibrassinolide for enhanced phenolics content and antioxidant capacity. Food Chem. 141 3056 –3065 . 10.1016/j.foodchem.2013.05.137 23871059 Xu W. Song Q. Li D. Wan X. (2012 ). Discrimination of the production season of Chinese green tea by chemical analysis in combination with supervised pattern recognition. J. Agric. Food Chem. 60 7064 –7070 . 10.1021/jf301340z 22720840 Yu J. Q. Huang L. F. Hu W. H. Zhou Y. H. Mao W. H. Ye S. F. (2004 ). A role for brassinosteroids in the regulation of photosynthesis in Cucumis sativus. J. Exp. Bot. 55 1135 –1143 . 10.1093/jxb/erh124 15107450 Zheng H. Z. Cui C. L. Zhang Y. T. Wang D. Jing Y. Kim K. Y. (2005 ). Active changes of lignification-related enzymes in pepper response to Glomus intraradices and/or Phytophthora capsici. J. Zhejiang Univ. Sci. B 6 778 –786 . 10.1631/jzus.2005.B0778 16052711

PMC005xxxxxx/PMC5003825.txt

==== Front Front PharmacolFront PharmacolFront. Pharmacol.Frontiers in Pharmacology1663-9812Frontiers Media S.A. 10.3389/fphar.2016.00278PharmacologyOriginal ResearchActivation of Muscarinic Acetylcholine Receptor Subtype 4 Is Essential for Cholinergic Stimulation of Gastric Acid Secretion: Relation to D Cell/Somatostatin Takeuchi Koji 12*Endoh Takuya 1Hayashi Shusaku 1Aihara Takeshi 11Division of Pathological Sciences, Department of Pharmacology and Experimental Therapeutics, Kyoto Pharmaceutical UniversityKyoto, Japan2General Incorporated Association, Kyoto Research Center for Gastrointestinal DiseasesKyoto, JapanEdited by: Ganna Tolstanova, Taras Shevchenko National University of Kyiv, Ukraine Reviewed by: Elisabetta Barocelli, University of Parma, Italy; Muriel Larauche, University of California, Los Angeles, USA *Correspondence: Koji Takeuchi, takeuchi@mb.kyoto-phu.ac.jpThis article was submitted to Gastrointestinal and Hepatic Pharmacology, a section of the journal Frontiers in Pharmacology 30 8 2016 2016 7 27815 6 2016 12 8 2016 Copyright © 2016 Takeuchi, Endoh, Hayashi and Aihara.2016Takeuchi, Endoh, Hayashi and AiharaThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.Background/Aim: Muscarinic acetylcholine receptors exist in five subtypes (M1∼M5), and they are widely expressed in various tissues to mediate diverse autonomic functions, including gastric secretion. In the present study, we demonstrated, using M1∼M5 KO mice, the importance of M4 receptors in carbachol (CCh) stimulation of acid secretion and investigated how the secretion is modulated by the activation of M4 receptors. Methods: C57BL/6J mice of wild-type (WT) and M1–M5 KO were used. Under urethane anesthesia, acid secretion was measured in the stomach equipped with an acute fistula. CCh (30 μg/kg) was given subcutaneously (s.c.) to stimulate acid secretion. Atropine or octreotide (a somatostatin analog) was given s.c. 20 min before the administration of CCh. CYN154806 (a somatostatin SST2 receptor antagonist) was given i.p. 20 min before the administration of octreotide or CCh. Results: CCh caused an increase of acid secretion in WT mice, and the effect was totally inhibited by prior administration of atropine. The effect of CCh was similarly observed in the animals lacking M1, M2 or M5 receptors but significantly decreased in M3 or M4 KO mice. CYN154806, the SST2 receptor antagonist, dose-dependently and significantly reversed the decreased acid response to CCh in M4 but not M3 KO mice. Octreotide, the somatostatin analog, inhibited the secretion of acid under CCh-stimulated conditions in WT mice. The immunohistochemical study showed the localization of M4 receptors on D cells in the stomach. Serum somatostatin levels in M4 KO mice were higher than WT mice under basal conditions, while those in WT mice were significantly decreased in response to CCh. Conclusions: These results suggest that under cholinergic stimulation the acid secretion is directly mediated by M3 receptors and indirectly modified by M4 receptors. It is assumed that the activation of M4 receptors inhibits the release of somatostatin from D cells and minimizes the acid inhibitory effect of somatostatin through SST2 receptors, resulting in enhancement of the acid response mediated by M3 receptors on parietal cells. acid secretionisolated mouse stomachcarbacholsomatostatinknockout mousemuscarinic receptor subtypesSST2 receptor ==== Body Introduction The mechanisms that govern gastric acid secretion involve neuro-humoral factors, including histamine, gastrin, and acetylcholine (ACh; Soll, 1994; Hersey and Sachs, 1995; Chen et al., 2004). The action of histamine is mediated intracellularly by 3′, 5′-cyclic adenosine monophosphate (cAMP), while those of gastrin and ACh are mediated by an increase of intracellular Ca2+ (Thurston et al., 1979). The stimulatory actions of histamine and gastrin are mediated by the activation of histamine H2 and cholecystokinin (CCK)-2 receptors, respectively (Brimblecombe et al., 1978; Hersey and Sachs, 1995; Chen et al., 2004), and that of ACh is caused by the activation of muscarinic acetylcholine receptors (mAChRs; Soll, 1994; Caulfield and Birdsall, 1998). The mAChRs consist of five subtypes (M1–M5) and are widely expressed in many peripheral organs as well as central nervous system to mediate diverse autonomic functions, including acid, pepsin and mucus as well as HCO3– secretions (Kajimura et al., 1992; Helander et al., 1996; Caulfield and Birdsall, 1998; Matsui et al., 2000, 2004; Tobin et al., 2009). These receptors are G protein-coupled receptors; M1, M3, and M5 receptors are coupled to Gq protein, while M2 and M4 receptors are coupled to Gi protein (Caulfield and Birdsall, 1998; Matsui et al., 2000). It is generally accepted that the acid stimulatory action of carbachol (CCh), a muscarinic agonist, is mediated by the activation of M1 and M3 receptors (Nakamura et al., 1985; Pfeiffer et al., 1990). However, Aihara et al. (2005) examined the involvement of M1, M3, and M5 receptors in cholinergic regulation of acid secretion using muscarinic receptor knockout (KO) mice and found that CCh-stimulated acid secretion is mediated by mainly M3 and partially M5 but not M1 receptors. In addition, we recently found that CCh-induced duodenal HCO3– secretion was markedly decreased in M4 KO mice, suggesting the involvement of M4 receptors in the cholinergic stimulation of HCO3– secretion (Takeuchi et al., 2015). However, no information is currently available on the role of M4 receptors in the cholinergic regulation of gastric acid secretion. Somatostatin, a peptide hormone, regulates many physiological functions in the gastrointestinal tract (Lucey and Yamada, 1989). This peptide is produced and secreted from D cells and inhibits acid secretion through a tonic inhibitory effect on both parietal and enterochromaffin-like cells via the activation of G protein-coupled SST2 receptors (Warhurst et al., 1996; Patel, 1997; Piqueras and Martínez, 2004). Chiba and Yamada (1990) reported that CCh inhibited both basal and pentagastrin-stimulated somatostatin secretion in the isolated canine D cells in a Gi protein/cAMP-dependent manner. We reported the involvement of somatostatin in the regulatory mechanism of the CCh-stimulated HCO3– secretion (Takeuchi et al., 2015). It is thus possible that endogenous somatostatin may also be involved in the mechanism of gastric acid secretion in response to cholinergic stimulation, although no study has tested this hypothesis. In the present study, we examined the effects of CCh on gastric acid secretion in wild-type (WT) and mAChR KO mice lacking M1–M5 receptors, and investigated the involvement of M4 receptors in the stimulatory action of CCh. We demonstrated the importance of M4 receptors in the cholinergic stimulation of gastric acid secretion and showed how this secretion can be modulated by the activation of M4 receptors, particularly focusing on the relation to D cell/somatostatin. Materials and Methods Animals Age-matched male C57BL/6J mice, weighing 25∼30 g of WT and those lacking M1, M2, M3, M4, or M5 receptor, were used. The generation and characterization of each subtype of mAChR KO mouse strain has been previously described by others (Matsui et al., 2000; Ohno-Shosaku et al., 2003; Nakamura et al., 2004). Animals were housed in plastic cages with hardwood chips in an air-conditioned room (25°C), and were given standard dry pellets, CA-1 (CLEA Japan, Tokyo, Japan) and water ad libitum. All experimental procedures used were carried out in accordance with the Helsinki Declaration and have been approved by the Committee for Animal Experimentation established by Kyoto Pharmaceutical University. Determination of Gastric Acid Secretion Acid secretion was measured in acute fistula stomachs of both WT and M1–M5 KO mice according to a previously published method (Niida et al., 1991; Kitamura et al., 1999). Under urethane anesthesia [1.25 g/kg, intraperitoneally (i.p.)], the trachea was cannulated to ensure a patent airway, and the body temperature was maintained at 36 ± 1°C using a heating lamp. Then, the abdomen was incised, both the stomach and duodenum were exposed, and the cardiac portion was ligated without interfering with vagus nerves. An acute fistula (inside diameter, 2 mm) made with a polyethylene tube was inserted into the stomach from a small incision made in the duodenum and held in place by a ligature around the pylorus. At the beginning of each experiment, the stomach was rinsed several times with physiological saline (154 mM NaCl) and filled with 0.4 ml of saline for 20 min for determination of the basal secretion. Then, the stomach was instilled with 0.4 ml of saline, and the solution was changed every 20 min. The collected samples were titrated to pH 7.0 against 2 mM NaOH using an autoburette (Comitite-8; Hiranuma, Tokyo, Japan). Gastric acid secretion was stimulated by CCh given subcutaneously (s.c.) in a dose of 30 μg/kg in both WT and KO mice lacking M1–M5 receptors. In WT mice, atropine (0.1 and 0.3 mg/kg) or octreotide (a somatostatin analog: 20 μg/kg) was given s.c. 20 min before the administration of CCh. In some cases CYN154806 (a somatostatin SST2 receptor antagonist: 0.1 mg/kg; Feniuk et al., 2000) was given i.p. 20 min before the administration of octreotide (20 μg/kg) in WT mice or the administration of CCh (30 μg/kg) in M4 KO mice. Control animals received saline or vehicle in place of the active agent. The doses of atropine, octreotide or CYN154806 were selected in order to induce the respective pharmacological actions according to the findings of previously published studies (Aihara et al., 2005; Xie et al., 2005; Terashima et al., 2009). Analyses for Gene Expression of mRNAs of mAChR Subtypes in Mouse Stomachs Whole stomachs were collected from both WT mice and those lacking M1–M5 receptors, and immediately frozen in liquid nitrogen and stored at -80°C until use. Total RNA was extracted from tissue samples using Sepasol RNA I (Nacalai Tesque, Kyoto, Japan). Total RNA was reverse-transcribed with a first strand cDNA synthesis kit (ReverTra Ace alpha, TOYOBO, Osaka, Japan; Amagase et al., 2010). The sequences of the sense and antisense primers for mouse M1–M5 receptors and GAPDH, and the sizes of the expected RT-PCR products are shown in Table 1. An aliquot of the RT reaction product served as a template in 35 cycles of PCR with 0.5 min of denaturation at 95°C and 1 min of extension at 68°C using the Advantage 2 polymerase mixture (CLONTECH, Mountain View, CA, USA) in a thermal cycler (PC-806, ASTEC, Fukuoka, Japan; Hayashi et al., 2014). A portion of the PCR mixture was electrophoresed in 1.5% agarose gel in Tris-acetic acid-EDTA buffer (40 mM Tris, 20 mM acetic acid, and 2 mM EDTA; pH 8.1), and the gel was stained with ethidium bromide and photographed (Bio Doc-It Imaging System; UVP, Upland, CA, USA). Images were analyzed with the Image J (version 1.39), and the semi-quantitative measurement of mRNA expression was presented as a ratio compared with GAPDH (Nakamori et al., 2010; Hayashi et al., 2014). Table 1 Sequences of sense and antisense primers for mouse M1–M5 receptors. Sequences PCR products Ml sense antisense 5′-GCAGCAGCTCAGAGAGGTCACAG-3′ 5′-GATGAAGGCCAGCAGGATGG-3 413 bp M2 sense antisense 5′-GCGGATCCTGTGGCCAACCAAGAC-3′ 5′-CGAATTCACGATTTGCGGGCTA-3′ 441 bp M3 sense antisense 5′-AAGGCACCAAACGCTCATCT-3′ 5′-GCAAACCTCTTAGCCAGCGT-3′ 511 bp M4 sense 5′-AGCCGCAGCCGTGTTCACAA-3′ 345 bp Antisense 5′-TGGGTTGAGGGTTCGTGGCT-3′ M5 sense 5′-GTCTCCGTCATGACCATACTCTA-3′ 230 bp Antisense 5′-CCCGTTGTTGAGGTGCTTCTAC-3′ GAPDH sense 5′-GAACGGGAAGCTCACTGGCATGGC-3′ 191 bp antisense 5′-TGAGGTCCACCACCCTGTTGCTG-3′ Immunohistological Study Expressions of mAChR M4 receptor and somatostatin were immunohistochemically examined in the gastric mucosa of WT or M4 KT mice. The stomachs were excised, rinsed with ice-cold PBS, and embedded in O.C.T. compound (Tissue-Tek, Sakura, Tokyo, Japan) iced with liquid CO2. Frozen samples were sectioned at a thickness of 10 μm at -20°C using a cryostat microtome (Leica Biosystems CM1510, Nussloch, Germany). The sections were exposed to 3% bovine serum albumin solution for 1 h to reduce the non-specific binding of anti-sera. The sections were exposed to each primary antibody for 16 h at 4°C, and incubated with the appropriate secondary antibody for 2 h at a room temperature. The sections were mounted with VECTASHIELD mounting medium, including 4,6-diamidino-2- phenylindole (Vector Laboratories, Peterborough, UK). The preparations were observed using a fluorescence microscope (Olympus BX51, Tokyo, Japan) and photographed using an Olympus digital camera. The following primary antibodies were used: rabbit anti-mAChR M4 and goat anti-Somatostatin (Santa Cruz Biotechnology, Santa Cruz, CA, USA). Alexa Fluor 488 conjugated donkey anti-rabbit IgG and Alexa Fluor 546 conjugated donkey anti-goat IgG (Molecular Probes, Eugene, OR, USA) were used as secondary antibodies. Determination of Serum Somatostatin Levels Serum levels of somatostatin were measured in both WT and M4 KO mice before and after the s.c. administration of CCh (30 μg/kg), according to a previously published paper (Terashima et al., 2009). Thirty minutes after each treatment, blood was collected from the descending aorta. Then, blood samples were centrifuged at 6000 g for 15 min, and the supernatant of each sample was frozen at 20°C until the measurement of somatostatin. The concentration of somatostatin was measured with a somatostatin immunoassay kit (Peninsula Laboratories, Inc., San Carlos, CA, USA). Preparation of Drugs Drugs used were urethane (Tokyo Kasei, Tokyo, Japan), carbamylcholine chloride [carbachol: CCh], octreotide, CYN154806 (Sigma-Aldrich, St. Louis, MO, USA), and atropine sulfate (Nacalai tesque, Kyoto, Japan). CCh and atropine were dissolved in saline, while octreotide and CYN154806 were dissolved in dimethyl sulfoxide (DMSO: Wako, Osaka, Japan) and diluted with distilled water to desired concentrations. Each agent was prepared immediately before use and administered as a single injection s.c. or i.p. in a volume of 1 ml per 100 g body weight. Statistical Analysis Data are presented as means ± SE. Differences between two groups were evaluated with the Student’s t-test. Differences between multiple groups were evaluated with analysis of variance followed, when necessary, by a Dunnett’s multiple comparison test. Values of P < 0.05 were considered statistically significant. Results Effect of CCh on Gastric Acid Secretion in WT Mice Under urethane anesthesia the stomachs of WT mice spontaneously secreted acid secretion in almost negligible amount of less than 0.1–0.2 μmol/20 min. Subcutaneously administered CCh (30 μg/kg) significantly stimulated acid secretion; the acid secretion reached a peak value of 3.8 ± 0.5 μmol/20 min 20 min after the administration, followed by a gradual decrease to near basal levels 100 min later (Figure 1A). Pretreatment of the animals with atropine (0.1 and 0.3 mg/kg, s.c.) dose-dependently inhibited the increase of acid secretion in response to CCh, and the net acid output at 0.1 and 0.3 mg/kg was 2.8 ± 0.3 and 0.3 ± 0.1 μmol/2 h, respectively, both of which were significantly lower than that (7.9 ± 1.3 μmol/2 h) of control mice given CCh plus vehicle (Figures 1A,B). FIGURE 1 Effects of atropine on CCh-stimulated acid secretion in the stomach of WT mice. CCh (30 μg/kg) was administered s.c. as a single injection. Atropine (0.1 and 0.3 mg/kg) was given s.c. 20 min before the administration of CCh. (A) Data are presented as the mean ± SE of values determined every 20 min from six mice. (B) Shows the total net acid output for 2 h after the administration of CCh, and the data are presented as the mean ± SE from six mice. Significant difference at P < 0.05; ∗ from control; # from Vehicle. Effect of CCh on Gastric Acid Secretion in mAChR KO Mice Since the acid stimulatory action of CCh was almost completely inhibited by atropine, the non-selective antagonist of mAChR, these results confirmed that CCh stimulated gastric acid secretion via the activation of mAChRs. We then investigated the subtype(s) of mAChRs involved in the regulation of gastric acid secretion, we examined the effects of CCh on acid secretion in the stomachs of mAChR KO mice lacking M1, M2, M3, M4, or M5 receptors. The stomachs of both WT and various mAChR KO mice consistently secreted 0.1∼0.3 μmol/20 min of H+ as basal secretion, and no significant difference was observed in the basal rates of the stomachs of these animals (data not shown). The stimulatory action of CCh (30 μg/kg, s.c.) was similar between mAChR KO mice lacking M1 and M2 receptors and WT mice, with the net acid outputs being 8.0 ± 0.8 μmol/2 h and 6.8 ± 0.9 μmol/2 h, respectively, which were almost equivalent to that (8.1 ± 0.7 μmol/2 h) in WT mice (Figure 2). Although a slight decrease in the acid response to CCh was observed in M5 KO mice, the net acid output (6.4 ± 0.7 μmol/2 h) was not statistically significant from that of WT mice. In contrast, the acid response to CCh was significantly decreased in M3 and M4 KO mice, with the net acid outputs being 2.3 ± 0.9 μmol/2 h and 2.1 ± 0.5 μmol/2 h, respectively. As shown in Figure 3A, CCh-stimulated acid secretion was markedly decreased in M4 KO mice. In WT mice the secretion of acid reached a peak value of 4.7 ± 0.7 μmol/20 min with the net acid output being 8.5 ± 1.2 μmol/2 h, while in M4 KO mice the peak value was 1.0 ± 0.2 μmol/20 min with the net acid output being 1.8 ± 0.3 μmol/2 h, which was significantly lower than that in WT mice (Figure 3B). FIGURE 2 Effects of CCh on acid secretion in the stomachs of WT mice and those of KO animals lacking M1–M5 receptors. CCh (30 μg/kg) was administered s.c. as a single injection. Data show the total net acid output for 2 h after the administration of CCh and are presented as the mean ± SE from six mice. ∗Significant difference from WT at p < 0.05. FIGURE 3 Effects of CCh on acid secretion in the stomachs of WT and M4 KO mice. CCh (30 μg/kg) was administered s.c. as a single injection. (A) The data are presented as the mean ± SE of values determined every 20 min from nine mice. (B) Shows the total net acid output for 2 h after the administration of CCh, and the data are presented as the mean ± SE from nine mice. Significant difference at p < 0.05; ∗ from Saline in WT; # from CCh in WT. Effects of Somatostatin on Gastric Acid Secretion in WT Mice Somatostatin has been shown to inhibit secretory and motor functions in the gastrointestinal tract and antagonizes the actions of several hormones (Lucey and Yamada, 1989; Piqueras and Martínez, 2004; Takeuchi et al., 2015). Since the release of somatostatin from D cells is known to be mediated by an increase in cAMP (Lucey and Yamada, 1989), and the activation of M4 receptors is coupled with Gi protein to inhibit adenylate cyclase (Caulfield and Birdsall, 1998; Tobin et al., 2009), it is possible that the decreased acid response observed in M4 KO mice may be associated with changes in somatostatin secretion. Therefore, we examined the effects of octreotide, an analog of somatostatin, on gastric acid secretion in WT mice under CCh-stimulated conditions. The basal secretion of acid in WT mice was very scanty with less than 0.1∼2 μmol/20 min and did not show significant changes following the s.c. administration of octreotide (20 μg/kg). However, the gastric acid response induced by CCh (30 μg/kg) was markedly reduced by the prior administration of octreotide; the net acid output was 0.7 ± 0.6 μmol/2 h, which was highly significantly different from that (8.5 ± 1.9 μmol/2 h) of WT mice (Figures 4A,B). This inhibitory effect of octreotide was completely abrogated when CYN154806 (0.1 mg/kg), a SST2 antagonist, was given 20 min before the administration of octreotide; the net acid output in response to CCh was 9.5 ± 0.8 μmol/2 h, which was almost equivalent to that of WT mice. FIGURE 4 Effects of octreotide on CCh-stimulated acid secretion in the stomachs of WT mice. CCh (30 μg/kg) was administered s.c. as a single injection. Octreotide (20 μg/kg), an analog of somatostatin-14, was administered s.c. 20 min before the administration of CCh, while CYN154806 (0.1 mg/kg), a SST2 antagonist, was given i.p. 20 min before the administration of octreotide. (A) The data are presented as the mean ± SE of values determined every 20 min from 5 to 6 mice. (B) Shows the total net acid output for 2 h after the administration of CCh, and the data are presented as the mean ± SE from 5 to 6 mice. Significant difference at P < 0.05; ∗ from Control (saline); # from Vehicle + CCh; $ from Vehicle + Octreotide + CCh. Effects of the Somatostatin SST2 Receptor Antagonist CYN154806 on Gastric Acid Response to CCh in M4 KO Mice We demonstrated that octreotide, an exogenous somatostatin analog, significantly inhibited CCh-stimulated secretion of acid in the stomachs of WT mice. Since it has been shown that CCh inhibited both basal and pentagastrin-stimulated somatostatin secretion in rats (Chiba and Yamada, 1990), there may be some interaction between somatostatin secretion and CCh-stimulated acid secretion. Then, to investigate the involvement of endogenous somatostatin in the decreased acid response to CCh in M4 KO mice, we examined the effects of CYN154806, a SST2 receptor antagonist, on the CCh-stimulated acid secretion in M4 KO mice. Subcutaneously administered CCh (30 μg/kg) markedly increased acid secretion in the stomachs of WT mice, the peak value of acid secretion was 4.2 ± 0.2 μmol/20 min and the net acid output was 8.2 ± 1.3 μmol/2 h (Figures 5A,B). By contrast, CCh did not increase acid secretion in the stomachs of M4 KO mice; the peak value of acid secretion was 0.9 ± 0.1 μmol/20 min, while the net acid output was 1.0 ± 0.8 μmol/2 h, which was almost equivalent to that in the stomachs of M4 KO treated with saline. However, when M4 KO mice were pretreated i.p. with CYN154806 (0.1 mg/kg), the administration of CCh potently increased acid secretion; the peak value was 5.1 ± 0.3 μmol/20 min, while the net acid output was 10.1 ± 2.5 μmol/2 h, which was significantly greater than that of M4 KO mice without pretreatment of CYN154806. The decreased acid response to CCh in M3 KO mice was not affected by CYN154806 (data not shown). FIGURE 5 Effects of CCh on acid secretion in the stomachs of WT and M4 KO mice, with or without the pretreatment of CYN154806. CCh (30 μg/kg) was administered s.c. as a single injection. CYN154806 (0.1 mg/kg) was administered i.p. 20 min before the administration of CCh. (A) The data are presented as the mean ± SE of values determined every 20 min from 4 to 11 mice. Significant difference at p < 0.05, ∗ from Saline (WT), # from CCh (WT), $ from Vehicle + CCh (M4KO). (B) Shows the total net acid output for 2 h after the administration of CCh, and the data are presented as the mean ± SE of values determined every 10 min from 4 to 11 mice. Significant difference at p < 0.05, ∗ from CCh (WT), # from Vehicle + CCh (M4 KO). Immunostaining of Somatostatin and M4 Receptors in Stomach of WT and M4 KO Mice To examine the presence of M4 receptors in D cells, we performed the immunostaining of the gastric mucosa with anti-somatostatin and anti-M4 receptor antibodies in WT mice. As expected, the expression of somatostatin was clearly observed in the stomach of WT mouse; it was stained red (Figure 6). On the other hand, the immunostaining of M4 receptors was also observed in the same area of the stomach; the higher magnification showed the co-existence of M4 receptors with somatostatin (a left figure of the lower panel), suggesting the expression of M4 receptors on D cells. No expression of M4 receptors was detected in the stomach of M4 KO mouse (a right figure of the lower panel), although the immunostaining of somatostatin was observed in this animal similar to WT mouse (data not shown). FIGURE 6 Fluorescence immunochemical staining of the gastric mucosa with anti-somatostatin and anti-M4 receptor antibodies in WT or M4 KO mice. Somatostatin was stained red, while M4 receptor was stained green. The left of the lower panels shows higher magnification to indicate the co-expression of M4 receptors with somatostatin on D cells in a WT mouse stomach as visualized as yellow in the merged image. In the right of the lower panels showed that M4 receptors were not observed in the gastric mucosa of a M4 KO mouse. Changes in Serum Levels of Somatostatin after Administration of CCh in WT and M4 KO Mice Serum levels of somatostatin in WT mice under urethane anesthesia were 0.33 ± 0.04 ng/ml. The levels of somatostatin in M4 KO mice were slightly higher (0.40 ± 0.04 ng/ml) than those in WT mice, although the difference was not statistically significant (Figure 7). On the other hand, the serum levels in WT mice were slightly but significantly decreased after the s.c. administration of CCh (30 μg/kg), the values being 0.22 ± 0.01 ng/ml. By contrast, the administration of CCh in M4 KO mice markedly increased the levels of somatostatin to 0.57 ± 0.04 ng/ml, the values being significantly higher than those in WT mice or those in M4 KO administered vehicle. FIGURE 7 Changes in serum somatostatin levels in WT or M4 KO mice after the administration of CCh. CCh (30 μg/kg) was administered s.c. as a single injection. Data are presented as the means ± SE for 5 to 6 mice. Significant difference at P < 0.05; ∗ from WT; # from vehicle. Gene Expressions of mAChR Subtypes in Mouse Stomachs Since it was found that both M3 and M4 receptors were involved in the stimulatory action of CCh on gastric acid secretion, we examined the gene expressions of mAChR subtypes, M1–M5, in the mouse stomach. The expressions of M1–M5 mRNAs were all observed in the stomachs of WT mice, though the intensity slightly differed depending on the subtypes (Figure 8). The expressions of mAChRs were also observed in the stomachs of KO mice lacking M1–M5 receptors, except lacking a respective mAChR subtype in the corresponding KO mice. FIGURE 8 Gene expressions of mAChR subtypes (M1–M5) in the mouse stomach. Note that a respective mAChR subtype was lacking in the corresponding KO mice lacking M1–M5 receptors in the stomach. Discussion The parietal cell, which is responsible for gastric acid secretion, is known to express histamine H2 receptor and CCK2 receptors in addition to M3 receptors (Pfeiffer et al., 1990; Kajimura et al., 1992; Soll, 1994; Chen et al., 2004). The secretion of acid stimulated by Histamine or gastrin is mediated by the former two receptors, respectively (Brimblecombe et al., 1978; Soll, 1994; Chen et al., 2004), while the subtypes of mAChRs responsible for cholinergic stimulation of acid secretion remains fully unexplored (Nakamura et al., 1985; Pfeiffer et al., 1990; Matsui et al., 2004; Aihara et al., 2005). The mechanism of cholinergic stimulation of acid secretion has been thought to involve M1 and M3 receptors (Nakamura et al., 1985; Pfeiffer et al., 1990), yet a recent study reported that gastric acid secretion was normally stimulated by histamine and gastrin as well as CCh in M1 KO mice, suggesting that M1 receptors are not involved in the regulation of gastric acid secretion in mice (Aihara et al., 2005). In the present study, we demonstrated for the first time the involvement of muscarinic M4 receptors in the regulatory mechanism of cholinergic stimulation of gastric acid secretion. The mAChRs, consisting of five subtypes (M1–M5), are widely expressed in many peripheral organs to include the gastrointestinal tract (Kajimura et al., 1992; Helander et al., 1996; Matsui et al., 2000; Tobin et al., 2009). In the present study, we confirmed by RT-PCR analysis that all mAChR subtypes including M1–M5 were expressed in the mouse stomach, although the degree of expression differed depending on the subtype; M3 and M4 receptors were potently expressed while M1, M2, and M5 were expressed weakly. In addition, we also demonstrated that the expressions of mAChRs were also observed in the stomachs of KO mice lacking M1–M5 receptors, except lacking a respective mAChR subtype in the corresponding KO mice. Others reported the expression of M1 receptors on zymogen cells and surface mucosal cells (Xie et al., 2005), that of M3 receptors on parietal cells (Pfeiffer et al., 1990; Kajimura et al., 1992; Aihara et al., 2005), and those of M2 and M4 receptors on D cell (Sachs et al., 1997; Takeuchi et al., 2015). We found that CCh, a muscarinic agonist, dose-dependently increased acid secretion in the stomach of WT mice, and this response was almost completely inhibited by the prior administration of atropine, confirming the mediation of this secretion by the activation of mAChR. Cholinergic stimulation of gastric acid secretion is thought to involve M1 receptors in addition to M3 receptors, because acid secretion was inhibited by pirenzepine, a selective M1 receptor antagonist (Nakamura et al., 1985). However, Aihara et al. (2005) examined the involvement of M1, M3, and M5 receptors in cholinergic regulation of acid secretion using muscarinic receptor KO mice and found that CCh-stimulated acid secretion is mediated by mainly M3 and partially M5 but not M1 receptors. They also demonstrated that pirenzepine exhibited similar inhibitory effects on CCh-stimulated acid secretion in both WT and M1 KO mice, suggesting that inhibition of acid secretion by pirenzepine is unlikely to result from M1-receptor blockade (2005). On the other hand, M5 receptors might be expressed in the enteric nervous system and mediates cholinergic stimulation of acid secretion by increasing Ach release from nerve endings and/or releasing histamine from enterochromaffin-like cells (Aihara et al., 2005). In the present study, we found using KO mice lacking M1–M5 receptors that CCh stimulated acid secretion in M1, M2, and M5 KO animals as effectively as in WT mice, but the stimulatory effect was markedly attenuated in M3, and M4 KO mice. We observed a slight decrease in the acid response to CCh in M5 KO mice, but the net acid output was not significantly different from that of WT mice. The reason for the different results between the study of Aihara et al. (2005) and our study, it may be due to different methods for acid measurement; they measure acid secretion in pylorus-ligated technique while we measured the secretion in the stomach with an acute fistula. Anyhow, it is therefore assumed that the cholinergic stimulation of gastric acid secretion is mediated by the activation of M3, and M4 receptors. In particular, the present study demonstrated for the first time the involvement of M4 receptors in the process of the CCh-stimulated gastric acid secretion, in addition to M3 receptors. M3 receptors are coupled to Gq/11 protein to increase intracellular Ca2+ that mediates the secretion of acid as well as many hormones (Warhurst et al., 1996). Since acid secretion in response to CCh is known to be attenuated by a Ca2+ channel blocker, it is reasonable that M3 receptors are involved in the process of cholinergic stimulation of acid response (Pfeiffer et al., 1990; Aihara et al., 2005). It remains, however, unknown how the activation of M4 receptors modulates the acid response to CCh in the stomach. Since M4 receptors are coupled to Gi protein to decrease intracellular cAMP production (Caulfield and Birdsall, 1998), and since stimulation of gastric acid secretion is intracellularly mediated by cAMP, in addition to Ca2+ (Thurston et al., 1979; Soll, 1994; Hersey and Sachs, 1995), it is unlikely that M4 receptors are expressed in parietal cells and directly mediate the stimulation of acid secretion in response to CCh. Therefore, it is assumed that M4 receptors are expressed in cells other than parietal cells and indirectly affect the acid response to cholinergic stimulation. By the way, somatostatin is synthesized in a variety of organs of the mammalian body and exerts almost ubiquitously an inhibitory action against various physiological processes (Lucey and Yamada, 1989). In the gastrointestinal tract, this peptide has been shown to inhibit motility and secretory functions and antagonizes the actions of several hormones (Lucey and Yamada, 1989; Warhurst et al., 1996; Terashima et al., 2009). Chiba and Yamada (1990) reported that CCh inhibited both basal and pentagastrin- stimulated somatostatin secretion in a Gi protein/cAMP-dependent manner in the isolated canine D cells. Five subtypes, SST1-SST5, of somatostatin receptors are currently known to exist, and all of them are expressed in the gastrointestinal tract (Chiba and Yamada, 1990; Krempels et al., 1997; Patel, 1997). It is thus possible that the activation of M4 receptors inhibits somatostatin secretion from D cells and by so doing indirectly affect acid secretion. If this is the case, the followings should be demonstrated; (i) somatostatin suppresses the acid response to cholinergic stimulation such as CCh, (ii) the decreased acid response in M4 KO mice can be reversed by the SST2 antagonist, and (iii) the release of somatostatin from D cells is increased in M4 KO mice. As expected, we found that CCh-stimulated acid secretion was significantly suppressed by octreotide, the somatostatin analog. We also found that the suppressed acid response to CCh in M4 KO mice was significantly restored by the prior application of the somatostatin SST2 antagonist, CYN154806. In addition, it was found that serum somatostatin levels were significantly increased in M4 KO mice under CCh-stimulated conditions. These results strongly support our hypothesis that the decrease of CCh-induced acid response in M4 KO mice is explained by the inhibitory effect of somatostatin mediated by SST2 receptors. It is therefore assumed that the activation of M4 receptors inhibits somatostatin release from D cells and negates the negative influence of this peptide on acid secretion, resulting in a potentiation of the acid response to CCh. Certainly, more studies are needed to clarify the regulatory mechanisms of somatostatin secretion from D cells. Finally, it remains undefined whether M4 receptors are really expressed on D cells? We performed the immunostaining of the gastric mucosa with anti-somatostatin and anti-M4 receptor antibodies in WT mice. The histological observation showed that M4 receptors were co-expressed with somatostatin, indicating the expression of M4 receptors on D cells. We confirmed that M4 receptors were not observed in the stomachs of M4 KO mice. These results strongly suggest that CCh inhibits somatostatin release from D cells via the activation of M4 receptors. The present study was performed in mice anesthetized with urethane. Since this anesthetic is known to promote the secretion of somatostatin from D cells (Saito et al., 1979), the results obtained in this study might differ from those obtained under normal physiological conditions. However, since CYN154806 by itself had no effect on basal acid secretion, it is assumed that the interpretation of the present results is not affected by urethane anesthesia. Given the findings of the present study, we conclude that the cholinergic stimulation of gastric acid secretion is mediated by the activation of M3 receptors in parietal cells and modified indirectly by M4 receptors in D cells (Figure 9). It is assumed that the activation of M4 receptors inhibits the release of somatostatin from D cells to result in enhancement of the acid response by removing the negative influence of somatostatin via the activation of SST2 receptors. FIGURE 9 Working hypothesis for the involvement of M4 receptors in the regulatory mechanism of cholinergic stimulation of gastric acid secretion. The cholinergic stimulation of acid secretion is mediated by the activation of M3 and M4 receptors. The activation of M3 receptors directly affects acid secretion from the parietal cells through a Gq protein/Ca2+ pathway, while the activation of M4 receptors inhibits somatostatin secretion from D cells and by so doing unmasks the stimulatory effect of CCh mediated by the activation of M3 receptors in the parietal cells. Author Contributions KT authored the paper and designed the study; TE, SH, and TA performed the experiments; TE and SH performed data analysis and coauthored the paper; KT contributed to critical revision of the paper. All authors approved the submission of the manuscript. Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The authors are greatly indebted to the undergraduate students at Department of Pharmacology and Experimental Therapeutics, Kyoto Pharmaceutical University, Kyoto, Japan, for their technical collaboration. ==== Refs References Aihara T. Nakamura Y. Taketo M. M. Natsui M. Okabe S. (2005 ). Cholinergically stimulated gastric acid secretion is mediated by M3 and M5 but not M1 muscarinic acetylcholine receptors in mice. Am. J. Physiol. 288 G1199 –G1207 . Amagase K. Ochi A. Sugihara T. Kato S. Takeuchi K. (2010 ). Protective effect of lafutidine, a histamine H2 receptor antagonist, against loxoprofen-induced small intestinal lesions in rats. J. Gastroenterol. Hepatol. 25 S111 –S118 . 10.1111/j.1440-1746.2010.06223.x 20586851 Brimblecombe R. W. Duncan W. A. M. Durant G. J. Emmett J. C. Ganelin C. R. Parsons M. E. (1978 ). Characterization and development of cimetidine as a histamine H2-receptor antagonist. Gastroenterology 74 339 –347 .23336 Caulfield M. P. Birdsall N. J. M. (1998 ). International union of pharmacology. XVII. Classification of muscarinic acetylcholine receptors. Pharmacol. Rev. 50 279 –290 .9647869 Chen D. Zhao C. M. Hakanson R. Samuelson L. C. Rehfeld J. F. Friis-Hansen L. (2004 ). Altered control of gastric acid secretion in gastrin/cholecyctokinin double mutant mice. Gastroenterology 126 476 –487 . 10.1053/j.gastro.2003.11.012 14762785 Chiba T. Yamada T. (1990 ). Mechanisms for muscarinic inhibition of somatostatin release from canine fundic D cells. Metabolism 39 122 –124 . 10.1016/0026-0495(90)90228-5 1976205 Feniuk W. Jarvie E. Luo J. Humphrey P. P. (2000 ). Selective somatostatin sst (2) receptor blockade with the novel cyclic octapeptide, CYN-154806. Neuropharmacology 39 1443 –1450 . 10.1016/S0028-3908(00)00035-6 10818260 Hayashi S. Kurata N. Yamaguchi A. Amagase K. Takeuchi K. (2014 ). Lubiprostone prevents NSAID-induced small intestinal damage by suppressing the expression of inflammatory mediators via EP4 receptors. J. Pharmacol. Exp. Ther. 349 470 –479 . 10.1124/jpet.114.213991 24713141 Helander K. G. Bamberg K. Sachs G. Melle D. Helander H. F. (1996 ). Localization of mRNA for the muscarinic receptor in rat stomach. Biochem. Biophys. Acta. 1312 158 –162 . 10.1016/0167-4889(96)00021-3 8672539 Hersey S. J. Sachs G. (1995 ). Gastric acid secretion. Physiol. Rev. 75 155 –189 .7831396 Kajimura M. Reuben M. A. Sachs G. (1992 ). The muscarinic receptor gene expressed in rabbit parietal cells is the m3 subtype. Gastroenterology 103 870 –875 . 10.1016/0016-5085(92)90019-U 1499937 Kitamura M. Sugamoto S. Kawauchi S. Kato S. Takeuchi K. (1999 ). Modulation by endogenous nitric oxide of acid secretion induced by gastric distension in rats: enhancement by nitric oxide synthase inhibitor. J. Pharmacol. Exp. Ther. 291 181 –187 .10490902 Krempels K. Hunyady B. O’Carroll A. M. Mezey E. (1997 ). Distribution of somatostatin receptor messenger RNAs in the rat gastrointestinal tract. Gastroenterology 112 1948 –1960 .9178687 Lucey M. R. Yamada T. (1989 ). Biochemistry and physiology of gastrointestinal somatostatin. Dig. Dis. Sci. 34 5S –13S . 10.1007/BF01536041 2563966 Matsui M. Motomura D. Karasawa H. Fujikawa T. Jiang J. Komiya Y. (2000 ). Multiple functional defects in peripheral autonomic organs in mice lacking muscarinic acetylcholine receptor gene for the M3 subtype. Proc. Natl. Acad. Sci. U.S.A. 97 9579 –9584 . 10.1073/pnas.97.17.9579 10944224 Matsui M. Yamada S. Oki T. Manabe T. Taketo M. M. Ehlert F. J. (2004 ). Functonal analysis of muscarinic acetylcholine receptors using knockout mice. Life. Sci. 75 2971 –2981 . 10.1016/j.lfs.2004.05.034 15474550 Nakamori Y. Komatsu Y. Kotani T. Kojima S. Takeuchi K. (2010 ). Pathogenic importance of cysteinyl leukotrienes in development of gastric lesions induced by ischemia/reperfusion in mice. J. Pharmacol. Exp. Ther. 333 91 –98 . 10.1124/jpet.109.162578 20042530 Nakamura M. Oda M. Yonei Y. Tsukada N. Komatsu H. Kaneko K. (1985 ). Muscarinic acetylcholine receptors in rat gastric mucosa. Histochemistry 83 479 –487 .3841347 Nakamura T. Matsui M. Uchida K. Futatsugi A. Kusakawa S. Matsumoto N. (2004 ). M3 muscarinic acetylcholine receptor plays a critical role in parasympathetic control of salivation in mice. J. Physiol. 558 561 –575 . 10.1113/jphysiol.2004.064626 15146045 Niida H. Takeuchi K. Okabe S. (1991 ). Role of thyrotropin-releasing hormone in acid secretory response induced by lowering of body temperature in the rat. Eur. J. Pharmacol. 198 137 –142 . 10.1016/0014-2999(91)90612-T 1907561 Ohno-Shosaku T. Matsui M. Fukudome Y. Shosaku J. Tsubokawa H. Taketo M. M. (2003 ). Postsynaptic M1 and M3 receptors are responsible for the muscarinic enhancement of retrograde endocannabinoid signalling in the hippocampus. Eur. J. Neurosci. 18 109 –116 . 10.1046/j.1460-9568.2003.02732.x 12859343 Patel Y. C. (1997 ). Molecular pharmacology of somatostatin receptor subtypes. J. Endocrinol. Invest. 20 348 –367 . 10.1007/BF03350317 9294784 Pfeiffer A. Rochlitz H. Noelke B. Tacke R. Moser U. Mutschler E. (1990 ). Muscarinic receptors mediating acid secretion in isolated rat gastric parietal cells are of M3 type. Gastroenterology 98 218 –222 . 10.1016/0016-5085(90)91314-V 2293581 Piqueras L. Martínez V. (2004 ). Role of somatostatin receptors on gastric acid secretion in wild-type and somatostatin receptor type 2 knockout mice. Naunyn Schmiedebergs Arch. Pharmacol. 370 510 –520 . 10.1007/s00210-004-0992-8 15599710 Sachs G. Zeng N. Prinz C. (1997 ). Physiology of isolated gastric endocrine cells. Annu. Rev. Physiol. 59 243 –256 . 10.1146/annurev.physiol.59.1.243 9074763 Saito H. Ogawa T. Ishimaru K. Oshima I. Saito S. (1979 ). Effect of pentobarbital and urethane on the release of hypothalamic somatostatin and pituitary growth hormone. Horm. Metab. Res. 10 550 –554 . 10.1055/s-0028-1092778 521010 Soll A. H. (1994 ). “Receptors that regulate acid secretory function,” in Physiology of the Gastrointestinal Tract ed. Johnson L. R. (New York, NY : Raven Press ) 1139 –1170 . Takeuchi K. Kita K. Takahashi K. Aihara E. Hayashi S. (2015 ). Muscarinic acetylcholine receptor subtype 4 is essential for cholinergic stimulation of duodenal HCO3- secretion in mice: relationship to D cell/somatostatin. J. Physiol. Pharmacol. 66 391 –401 .26084221 Terashima S. Nishio H. Ogura M. Honda M. Takeuchi K. (2009 ). Involvement of prostacyclin/IP receptors in decreased acid response of damaged stomachs -Mediation by somatostatin/SST2 receptors. Life. Sci. 84 172 –180 . 10.1016/j.lfs.2008.11.014 19070625 Thurston D. Tao P. Wilson D. E. (1979 ). Cyclic nucleotides and the regulation of canine gastric acid secretion. Dig. Dis. Sci. 24 257 –264 . 10.1007/BF01296537 37056 Tobin G. Giglio D. Lundgren O. (2009 ). Muscarinic receptor subtypes in the alimentary tract. J. Physiol. Pharmacol. 60 3 –21 .19439804 Warhurst G. Higgs N. B. Fakhoury H. Warhurst A. C. Garde J. Coy D. H. (1996 ). Somatostatin receptor subtype 2 mediates somatostatin inhibition of ion secretion in rat distal colon. Gastroenterology 111 325 –333 . 10.1053/gast.1996.v111.pm8690197 8690197 Xie G. Drachenberg C. Yamada M. Wess J. Raufman J. P. (2005 ). Cholinergic agonist-induced pepsinogen secretion from murine gastric chief cells is mediated by M1 and M3 muscarinic receptors. Am. J. Physiol. 289 G521 –G529 .

PMC005xxxxxx/PMC5003826.txt

==== Front Front PsycholFront PsycholFront. Psychol.Frontiers in Psychology1664-1078Frontiers Media S.A. 10.3389/fpsyg.2016.01297PsychologyOriginal ResearchClasses in Translating and Interpreting Produce Differential Gains in Switching and Updating Dong Yanping 1*Liu Yuhua 21Bilingual Cognition and Development Lab/Center of Linguistics and Applied Linguistics, Guangdong University of Foreign StudiesGuangzhou, China2College of Foreign Studies, South China Agricultural UniversityGuangzhou, ChinaEdited by: Antonino Vallesi, University of Padua, Italy Reviewed by: Marco Calabria, Pompeu Fabra University, Spain; Kenneth R. Paap, San Francisco State University, USA *Correspondence: Yanping Dong, ypdong@gdufs.edu.cnThis article was submitted to Cognition, a section of the journal Frontiers in Psychology 30 8 2016 2016 7 129703 6 2016 12 8 2016 Copyright © 2016 Dong and Liu.2016Dong and LiuThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.The present longitudinal study was intended to investigate whether the two bilingual experiences of written translation and consecutive interpreting (featured with similar language switching experience but different processing demands) would produce different cognitive control effects in young adults. Three groups of Chinese–English young adult bilinguals, who differed mainly in their half-year long bilingual experience: one for general L2 training, one for written translation and one for oral consecutive interpreting, were tested twice on the number Stroop, switching color-shape and N-back tasks. The results show that the interpreting experience produced significant cognitive advantages in switching (switch cost) and updating, while the translating experience produced marginally significant improvements in updating. The findings indicate that the experience of language switching under higher processing demands brings more domain-general advantages, suggesting that processing demand may be a decisive factor for the presence or absence of the hot-debated bilingual advantages. cognitive controlprocessing demandtranslationinterpretingbilingual advantagelongitudinal study ==== Body Introduction It is believed that pervasive experience can leave its mark on the development of mind and brain. The past decade has seen a boom of research exploring the effect of bilingualism on specific cognitive control components (e.g., Bialystok et al., 2004). But there have been dissenting voices (e.g., Paap and Greenberg, 2013) or cautious voices (e.g., Hilchey and Klein, 2011) in recent years. This controversy has turned into a hot debate, especially after a recent issue of Bilingualism: Language and Cognition (a series of commentaries on the key article by Valian, 2015) and a recent issue of Cortex (a series of commentaries on the key article by Paap et al., 2015). Experts on the topic have expressed their warnings against methodological flaws (see Paap, 2014, for example), theoretical weaknesses (see Jared, 2015, for example), and interpretation biases (see Morton, 2015, for example). Aware of the controversy, the present study has taken several steps to overcome some of the flaws and weaknesses in the literature, hoping to find a way forward, which may provide some clues for the bilingual advantage issue, and which may further help to establish the types of bilingual experience that produce relatively quick gains in cognitive control. A large majority of research on the bilingual advantage adopted a cross-sectional design (except for a few such as Bak et al., 2014). However, cross-sectional, in contrast to longitudinal designs, are vulnerable to confounding factors that are hard to control and for which the cause-consequence relationship between bilingualism and executive control is hard to decide (see Kempe et al., 2015; Li and Grant, 2015; Woumans and Duyck, 2015). Indeed, many bilingual advantages have been reported using assumed measures of inhibition, switching and monitoring, but it seems that many of them have been questioned by Paap et al. (2015) for reasons illustrated above. The inhibition advantage from bilingualism was evidenced in different tasks, such as the Simon task (e.g., Bialystok et al., 2004; Carlson and Meltzoff, 2008; Woumans et al., 2015), the Stroop task (e.g., Bialystok et al., 2008; Blumenfeld and Marian, 2011), the flanker task (e.g., de Abreu et al., 2012; Poarch and Bialystok, 2015) and the Attention Network Test (ANT; a complex version of flanker) (e.g., Costa et al., 2008; Marzecová et al., 2013). Bilingual advantage in switching was shown in the color-shape task (e.g., Prior and Macwhinney, 2010; Prior and Gollan, 2011). As to the relationship between bilingual experiences and updating capacity, few empirical studies have been conducted, but there have been theoretical formulations on the relationship between WM and updating in the context of bilingual advantage (Paap and Sawi, 2014). A few studies found that bilingualism did not bring WM advantage as measured by WM spans (e.g., Ratiu and Azuma, 2015). Bilingual advantage in monitoring (as indicated by shortened reaction times in tasks containing conflicts or by mixing cost in the color-shape task) was reported too (e.g., Barac and Bialystok, 2012; Abutalebi et al., 2015; Woumans et al., 2015). However, null bilingual effects have also been reported in the Simon task (e.g., Gathercole et al., 2014; Kirk et al., 2014), the Stroop task (e.g., Kousaie and Phillips, 2012) and the flanker task (e.g., Bialystok et al., 2010). The bilingual switching advantage failed to appear in some studies either (e.g., Hernandez et al., 2013; Gathercole et al., 2014). Reviewing nearly 30 experiments, Hilchey and Klein (2011) claimed that there was only evidence for a bilingual advantage in monitoring. However, Paap and Greenberg (2013), after reviewing 18 tests in several studies, did not find any significant monitoring advantage. The bilingual advantage issue, therefore, needs more research, especially research adopting a longitudinal design. Apart from methodological considerations, a better theoretical framework is needed (e.g., Jared, 2015; Hartsuiker, 2015). The most important question is: what does bilingualism have that monolingualism does not that might lead to bilingual advantages in cognitive control? The general theoretical formulation is that executive functions exercised in selecting the target language during bilingual processing (see the BIA+ model by Dijkstra and Van Heuven, 2002; the Inhibitory Control Model by Green, 1998) are transferred from the linguistic domain to the general domain. It seems that the monitoring of two jointly activated language systems, the inhibition of the non-target language, the switch between languages, and the updating of relevant information in the bilingual language control system corresponds neatly to such components of the general cognitive control system as monitoring, inhibiting, switching and updating (see Miyake et al., 2000). But how does this transfer happen? In Hartsuiker’s (2015) words, when, how and why does practice in one domain generalize to another domain? Hartsuiker (2015) may have pointed out the most important direction for future research, and the present paper was intended as a first step in the recommended direction. Instead of investigating the bilingual advantage directly, the present study investigates a related issue: under what circumstances does language switching practice start to influence or enhance non-linguistic switching abilities? The answer to this question could partly answer Hartsuiker’s (2015) question of when practice in one domain generalizes to another domain. Interpreting between two languages is a cognitively demanding task, and several recent studies (Yudes et al., 2011; Dong and Xie, 2014; Babcock and Vallesi, 2015; Morales et al., 2015; Woumans et al., 2015; Becker et al., 2016) have explored how interpreting experience brings cognitive advantages. Yudes et al. (2011) found that professional simultaneous interpreters (SIs) outperformed general bilinguals in the WCST task, but not in the task of Simon. Consistent with these findings, Dong and Xie (2014) further found that students of interpreting training or more interpreting training outperformed those of no or less interpreting training in the task of WCST, but not in the task of Flanker. Babcock and Vallesi (2015) and Woumans et al. (2015), however, had different findings. Babcock and Vallesi (2015) found that professional interpreters exhibited less mixing cost in a color-shape task than general bilinguals but did not show advantages in conflict resolution in a Stroop task or switching cost in the color-shape task. Woumans et al. (2015) found that interpreters outperformed unbalanced (but not balanced) bilinguals in the Simon and ANT tasks (i.e., higher accuracy in both tasks and smaller error congruency effect in the ANT). Along the same line of comparing SIs and general bilinguals, Morales et al. (2015) reported higher updating skills from SIs and a modulating effect of interpreting experience on the interaction between attentional networks. Comparing SIs and other professional multilingual controls (mostly consecutive interpreters and translators), Becker et al. (2016) reported less mixing costs in a color-shape switching task and a dual-task advantage from SIs. To sum up, in the few cross-sectional studies conducted up till now, it seems that there was always a certain cognitive control advantage for professional SIs or students of more interpreting experience. However, the results were not necessarily consistent. Two of the studies (Yudes et al., 2011; Dong and Xie, 2014) found that interpreting experience enhanced switching ability as measured in the WCST, while two of the studies (Babcock and Vallesi, 2015; Becker et al., 2016) found that interpreting experience reduced mixing costs but not switching costs in a color-shape task. To bridge the gap, we may have to conduct studies of a longitudinal design and with both tasks (WCST and the color-shape task). What is more, we have to take into consideration of our critical question of when (under what circumstances). To answer the critical question of when language switching practice starts to influence or enhance non-linguistic switching abilities, the present paper adopts a longitudinal design and compares the cognitive consequences of (oral consecutive) interpreting training with (written) translation training and general second language training. Three groups of bilingual students participated and they were comparable except that they would respectively receive one semester’s consecutive interpreting training, translation training and general L2 training (L2 culture and communication). Apart from the longitudinal design, what is distinctively different from the literature is a comparison with translation training. On the one hand, performances of both interpreting and translation involve frequent switching between two languages. Different from simultaneous interpreting, consecutive interpreting is more “serial” in the sense that it is generally after one segment of the source text is rendered that the next would start to be processed. It is in this sense that consecutive interpreting is more similar to translation, compared to simultaneous interpreting. On the other hand, there are differences between consecutive interpreting and translation. The most apparent difference lies in that interpreting requires immediate processing, which suggests that interpreters are under great time pressure and that they have to store on-line a huge amount of information. Dragsted and Hansen (2009) found that because of this difference, professional translators and interpreters performed differently in an eye-tracking experiment of sight translation and written translation. The interpreters translated faster in a more “controlled” linear way without compromising output quality, while the translators translated more slowly with plenty of backtracking and regressions of their eye movements. Yudes et al. (2011) and Dong and Xie (2014) have found evidence for the cognitive advantage of switching brought by interpreting experience, but none of them explicitly distinguished oral interpreting experience from written translation experience because students of interpreting (as in Dong and Xie, 2014) or professional interpreters (as in Yudes et al., 2011) are generally also trained in written translation. A direct comparison of the cognitive effects of these two modes of language training may be able to provide some clues for why some language experiences rather than others bring cognitive control advantages, and thus clues for what brings bilingual cognitive advantages. We predicted that interpreting experience would bring more cognitive control advantages than translation or general bilingual experience. If the prediction is true, it implies that a prerequisite for a certain training to bring about general cognitive control advantage is high processing demands. For the interpreting-translation case, immediate switching of a large chunk of speech (a sentence at least) between the two languages under time pressure (i.e., interpreting) poses higher processing demands than switching without time pressure (i.e., translation). A task may be demanding in different ways, but immediate processing under time pressure is certainly one of the ways. As speculated by Schroeder and Marian (2016), when the supply was below the demand, the cognitive system tried to adapt and thus got enhanced. Therefore, the answer for the critical question of when would be: language switching practice starts to influence or enhance non-linguistic switching abilities when processing demands reach a certain level. Materials and Methods To investigate how the two specific bilingual experiences of translation and interpreting would influence cognitive control development in young adults, three groups of Chinese–English bilingual participants were tested at a pre-test and a post-test. The three groups were comparable except that one group would receive one semester’s (oral) interpreting training, another (written) translation training and the third would receive general L2 training (English culture and communication). There were two parts in the pre-test: (1) a questionnaire of the participants’ backgrounds: their L2-related experiences and their relevant biological and social data (e.g., age, IQ, parents’ education); (2) a test of their cognitive control abilities of inhibition, switching, monitoring and updating in working memory (WM). The post-test consisted of only the second part, that is, a test of participants’ cognitive control abilities. Statistical analyses reported below will show how each group has progressed after one semester’s training, and how the three groups differ from each other in cognitive control abilities after being matched in their pre-test. Participants Three groups of Chinese–English young adult unbalanced bilinguals (145 in total, mean age = 19.69 years, SD = 0.89, range = 17–22) volunteered to participate in the study for course credit. Among the 145 participants, 57 of them taking an interpreting course during the experiment semester (coded henceforward as the interpreting group), 43 of them taking a translation course (coded henceforward as the translation group), and 45 of them taking general English course (English culture and communication, coded henceforward as the control group). All these participants were non-English-major sophomore students from the same college of a Chinese university in China, and received neither translation nor interpreting training before taking the pre-test. Since the courses were elective, assignment to the groups was based on self-selection. In the general English course (control group), about half of the class time was spent on listening to the teacher’ lectures and half on student discussions. Teachers and students were all required to speak in English in the classroom and therefore little language switching took place. As for the two courses of translation and interpreting, the training was mainly from English to Chinese, with about one third of the class time spent on listening to teachers’ lectures and the rest on translation or interpreting practice. At the end of the semester, participants were asked to report how much time they had spent on each course after class. The average time each group of participants spent on Integrated English after class was 56 h, and that on their distinguishing course (English culture and communication, translation or interpreting) was 40 h. A comparison of the courses that the participants received during the experimental semester is illustrated in Table 1. The three groups were, therefore, comparable in the training they received during the semester except for the difference deliberately designed for the present study. Table 1 Class hours of courses for the three participant groups (control, translation and interpreting) together with practice after class (in brackets) during the experimental semester. Control Translation Interpreting Courses not related to L2 (English) 256 256 256 Integrated English 42 (+56) 42 (+56) 42 (+56) English culture and communication 32 (+40) 0 0 Translation (written) 0 32 (+40) 0 Interpreting (oral) 0 0 32 (+40) All the participants were native speakers of Chinese, and apart from English, had no contact with any other foreign language. Details of their background information were presented in the first half of Table 2 “background characteristics,” including L2-related factors (tested L2 proficiency, self-rated L2 proficiency, self-rated L2 use, AoA) and more biological and social factors (age, IQ, parents’ education). Such information was collected to ensure that confounding factors (e.g., Dong and Li, 2015; Valian, 2015) would be controlled. Table 2 Pre-test group means (with SD) and comparisons (p-value) of participants’ background characteristics and task performances in the pre-test before group match. Control (n = 43) Translation (n = 40) Interpreting (n = 51) p-Value Background characteristics      Translation/interpreting No No No      Tested L2 proficiency 14.13 (3.61) 13.93 (4.47) 13.39 (3.37) 0.614      Self-rated L2 proficiency 20.02 (4.37) 21.07 (5.74) 20.76 (5.27) 0.630      Self-rated L2 use 0.054 (0.036) 0.048 (0.047) 0.049 (0.044) 0.731      AoA 8.95 (2.31) 9.17 (2.44) 9.00 (2.32) 0.903      Age 19.81 (0.82) 19.80 (0.99) 19.45 (0.83) 0.079      Father education 2.39 (0.69) 3.15 (1.38) 2.80 (1.24) 0.013      Mother education 2.02 (0.98) 2.72 (1.37) 2.33 (1.21) 0.031      Intelligence 67.62 (2.38) 67.05 (2.71) 66.66 (3.11) 0.250 Cognitive control abilities      Stroop: global RTs (ms) 684.71 (84.73) 677.66 (61.75) 670.69 (68.79) 0.646      Stroop: Stroop effect 34.72 (36.10) 31.36 (49.39) 17.79 (37.08) 0.105      Stroop: Stroop inhibition 14.51 (44.48) 6.27 (56.64) -2.73 (36.43) 0.192      Stroop: Stroop facilitation -20.21 (36.80) -25.09 (31.86) -20.52 (42.65) 0.804      Color-shape: global RTs 612.57 (141.95) 598.62 (145.89) 587.51 (124.12) 0.676      Color-shape: mixing cost 130.44 (126.70) 103.18 (111.06) 117.39 (90.24) 0.526      Color-shape: switch cost 148.30 (95.22) 123.24 (83.72) 137.77 (85.82) 0.435      N-back: global RTs 840.51 (265.32) 848.20 (248.04) 857.64 (248.27) 0.948      N-back: accuracy rate 0.86 (0.088) 0.87 (0.083) 0.83 (0.097) 0.108 Materials and Tasks Critical information about the materials and tasks is listed in Table 3. Table 3 Summary of tasks used in the present study and description of the items tested. Tasks Items tested Tasks capturing background (in pre-test only) Composite questionnaire (1) Self-rated language proficiency: overall score of listening, speaking, reading and writing respectively on a 10-point Likert scale; 40 points in total (2) Self-rated language use: percentage of daily language use; (3) AoA: age of English education; (4) Age: age when being tested; (5) Parental education: score of parents’ education on a 5-point Likert scale      L2 cloze test L2 proficiency (Bachman, 1985), 30 points in total      IQ test IQ: Raven’s Advanced Progressive Matrices Set (Raven et al., 1977), 72 points in total Cognitive control tasks (in pre- and post-tests)      Number Stroop task (1) Inhibition ability: Stroop conflict (2) Monitoring ability: global RTs      Color-shape task (1) Switching ability: switch cost (2) Monitoring ability: mixing cost and global RTs      N-back task (1) Updating ability: accuracy rate, global RTs In the pre-test, participants had to complete a composite questionnaire with questions tapping information about participants’ self-rated L2 proficiency, self-rated L2 use, AoA, age and parental education (Marian et al., 2007), together with an L2 proficiency test (L2 cloze test by Bachman, 1985) and an IQ test (Raven et al., 1977). Altogether three tasks of cognitive abilities were used, testing participants’ inhibition, switching, updating, and monitoring. Inhibitory control was tested with the number Stroop task under the typical assumption that smaller Stroop interference effects reflect better control. We did not choose the Simon task or the Flanker or the color Stroop because we believed they were too simple for our young adult participants who were in their peak of cognitive abilities (e.g., Paap and Greenberg, 2013). Xie and Dong (2015) used the Flanker and the number Stroop tasks to test similar participants (Chinese–English young adult unbalanced bilinguals with L1 or L2 public speaking training) and it was found that the number Stroop task produced more groups effects than the Flanker, probably because it was more difficult (with longer reaction times, see Dong and Li, 2015 for a review). But we are aware that there may be different opinions. Paap et al. (2014, May) reported that the flanker effect is still shrinking after 100 sessions and more than 20,000 trials. Switching (mental set shifting or mental flexibility) was tested with the color-shape task (i.e., the switch cost: reaction time difference between a switch trial and a non-switch trial in a mixed block)1. Both global RTs and mixing costs (RT difference between non-switch trials in a mixed block and single task trials) are often assumed to reflect monitoring ability. But we are aware that “switch cost” is also taken as a measure of inhibitory control, in the sense that participants have to inhibit the previous task set to be able to reactivate the new one (Philipp et al., 2008; Yang et al., 2016). And yet based on the tripartite system of executive functions suggested by Miyake et al. (2000), we decided to adopt the switching account of switch cost as adopted in Hernandez et al. (2013) and Paap and Greenberg (2013). The major reason is that compared with the inhibition component measured in tasks such as the Flanker, the Simon and the Stroop, switch cost in the color-shape task involves more of one’s ability to switch to a new task set. In addition to the two typical components of inhibition, switching and monitoring, updating in WM was also identified as part of the cognitive control system (e.g., Miyake et al., 2000; Costa et al., 2009; Hilchey and Klein, 2011). These components are related but also relatively independent. The enhancement of one component may or may not imply the strengthening of other components. Thus, all four components were tested in the present study (see Table 3). Each cognitive control task is described in more detail below. The Number Stroop Task The number Stroop task, measuring participants’ inhibition ability, was more or less the same as that used by Xie and Dong (2015). The task required participants to judge whether the number of the digits or the hash signs (#, ##, ###, or ####) in a stimulus was even or odd. There were three possible conditions. The neutral condition refers to trials of the hash sign “#”, and so the correct response for “###” or “#”, for example, would be odd. The congruent condition refers to trials of digits in which the parity of the digit coincides with the parity of the number of the digit, and so the correct response for 2222 would be even. The incongruent condition is the opposite of the congruent condition, and so the correct response for 222 would be odd (because there are three digits). The computerized task was composed of two blocks: the practice block and the experimental block. The practice block consisted of nine trials with feedback of accuracy and response times for each stimulus. The experimental block consisted of 120 randomly presented trials, with 40 in each condition. Each stimulus was presented on the screen for a maximum time of 2000 ms or until participants pressed designated keys. Participants were asked to respond as quickly as possible without sacrificing accuracy. We computed four indices for the Stroop task: Global RTs, Stroop effect, Stroop facilitation and inhibition (see Table 2). The most important one is the Stroop effect, i.e., the difference in mean RTs between incongruent trials requiring suppression of conflicting cues and congruent trials with no conflicting cues. A smaller Stroop effect implies higher ability in conflict resolution and inhibition. Global RTs refers to the average time taken to respond to all the trials (congruent, neutral and incongruent trials). Stroop facilitation refers to the RT difference between congruent and neutral trials, while Stroop inhibition refers to the RT difference between incongruent and neutral trials. The Color-Shape Switching Task The color-shape task was adapted in the present study so that the inhibition component in switch costs was reduced. In a typical manipulation of the color-shape task, the stimulus is one of the four combinations of color and shape: red/green circle/square. A precue is therefore necessary to indicate when to respond to color and when to respond to shape. But as in the number Stroop task, a single shape contains both cues of color and shape. To respond to color, for example, one has to inhibit a potential response to shape. The present study instead tried to reduce the component of inhibition in the color-shape task in which the stimulus was either one of the two color pictures (red or green) or one of the two colorless shapes (circle or triangle). Designed deliberately to test participants’ switching ability, the color-shape task required participants to press the designated keys corresponding to color (always in a circle) or colorless shape pictures presented at the center of the computer screen. Each trial started with a fixation cross presented at the center for 350 ms, followed by a blank screen for 150 ms, and then the target appeared and remained at the center until the participant responded. There were four choices of target picture: two color pictures (red circle or green circle) and two shape pictures (circle or triangle without any color). Participants were instructed to perform the color task using the left hand, with “red” being assigned to the index finger, and “green” the middle finger. The shape task was performed with the right hand, with “triangle” being assigned to the index finger and “circle” the middle finger. The experiment was composed of three blocks: two blocks of a single task (color or shape) and one block of the mixed task (color and shape). Each single task block included 8 practice trials followed by 24 experimental trials, and the mixed task block included 8 practice trials followed by 48 experimental trials. All the trials in each block were randomized. Participants were asked to respond as quickly as possible without sacrificing accuracy. Three indices were computed for the color-shape task: global RTs, mixing cost and switch cost. Global RTs refers to the mean RTs in the mixed task block. Mixing cost refers to the difference in RTs between the non-switch trials in the mixed task block and trials in the single task block, while switch cost refers to the difference in RTs between the switch trials and non-switch trials in the mixed task block. Both global RTs and mixing cost are indicators of monitoring ability, and switch cost indicates the ability to switch between different types of trials. The N-back Task A visuo-spatial version of the 2-back task was used to measure updating in WM. In the 2-back task, a blue square was presented in one of 25 possible locations on the screen. Participants were asked to match the location of the current square with the location of the square before the previous one (2-back). The task consisted of 42 2-back trials (28 non-target and 14 target trials). Participants were asked to press the “F” button if the square was in the same location as the square two trials back and the “J” button if the location was different. The square remained on the screen for 500 ms. A new square appeared 3000 ms after the previous one had disappeared, irrespective of whether a response was made or not. The presentation order of the trials was randomized. Before the experimental sequence, participants were asked to complete three practice sequences of 27 trials. Participants were asked to respond as quickly as possible without sacrificing accuracy. Two indices, i.e., global RTs and accuracy rate, were computed and both were indicators of participants’ updating ability. Procedure The experiment lasted for one academic semester (about 4 months and a half). At the beginning of the semester, participants were asked to take the pre-test in a computer room. The test was divided into two parts and lasted for nearly 2 h, with a 5-min break in between. The order of task administration was fixed for all three groups, with the requirement that no two tasks tapping the same cognitive control capacity occurred consecutively so as to minimize any error caused by task interference. Based on this criterion, tasks administered in the first part were the questionnaire, the number Stroop task and the color-shape switch task. Those in the second part were the cloze test, the N-back task and the IQ test. As illustrated in the section of “participants,” after the pre-test, participants as college students took various courses for one semester. At the end of the semester, participants took the post-test. Similar to the pre-test, the post-test was divided into two parts and the tasks were administered in a fixed order for all the groups, those in the first part were the number Stroop and the color-shape tasks, and those in the second part were the cloze test and the N-back task. Before the first part started, participants were asked to complete a questionnaire to collect information about their experiences in the past semester. Results Data Trimming First of all, we had to exclude those participants whose performances were obviously not normal. The reason is that some of the students were not serious enough, at least in some of the tests. The three courses were selective, and students are generally not as serious in a selective course as when they perform tasks in a compulsory course. What is more, the classes were very large (i.e., respectively 45, 43, and 57 students in each class), especially as language classes, and it is generally harder to ensure that all students are serious enough in a large class. Table 4 lists the number of participants excluded from each group of participants, and the reasons for the exclusions. Table 4 Number of participants excluded from further data analysis and reasons for the exclusions. Control 45-2 Translation 43-3 Interpreting 57-6 Computer breakdown in n-back post-test 1 Abnormal performance in L2 test (less than 10 out of a total of 30 and worse in post-test than in pre-test) 1 1 3 Background different from others (nervous as the only first-year student among all second-year students) 1 Not serious in n-back post-test (wrong input of her student number, lowest accuracy at 69%) 1 Abnormal performance in IQ test (less than 55 out of a total of 72, which means “retarded” according to Raven et al., 1977). 3 The data were trimmed following general procedures in the literature. In the number Stroop task, data from erroneous responses and data with response time (RTs) less than 200 ms were first discarded, and then outlier responses deviating by more than 3 SDs from the mean RTs for each participant were trimmed. Altogether less than 5% of data was discarded (the control group: pre-test, 1.91%; post-test, 2.12%; the translation group: pre-test, 2.21%; post-test, 2.14%; the interpreting group: pre-test, 1.97%; post-test, 1.86%). In the color-shape task, the same procedure was followed, and less than 5% of data was discarded (the control group: pre-test, 1.09 and 1.35% for the single or mixed task block; post-test, 1.06 and 1.67% respectively the two task blocks; the translation group: the four percentages were respectively 1.02, 0.91, 1.07, 1.71%; the interpreting group: respectively 1.02, 1.81, 0.68, 1.41%). In the N-back task, the same procedure was followed and less than 5% of data was discarded (the control group: pre-test, 1.55%; post-test, 1.38%; the translation group: pre-test, 1.18%; post-test, 1.18%; the interpreting group: pre-test, 1.09%; post-test, 2.24%). Statistical Analysis An analysis was conducted with Participant Group as the between-subject factor and Testing Time as the within-subject factor, hoping to find out whether there were any group differences in the training effect. A prerequisite for the analyses was that the three groups were matched in all the relevant factors that may influence the performance or development in the cognitive control tasks. Details of data analyses are reported below. Raw Data in the Pre-test Between-group comparisons were conducted for the pre-test results to see whether the three groups were matched or not. Table 2 is a summary of the descriptive data together with the p-value for each group comparison in each index. The first finding revealed in Table 2 is that there was no group difference in any of the indices of cognitive abilities, and that there was no group difference in any of the L2-related indices (i.e., L2 proficiency, L2 use and AoA). Since the three courses (oral interpreting, written translation, and general L2 class) were not compulsory and students made their choice out of their own will, this finding indicates that students did not choose a certain training (e.g., interpreting) because of some preexisting advantage in a related cognitive function (e.g., switching). Group Matching in the Pre-test To make sure that group differences in cognitive control abilities in the post-test were indeed caused by the different types of training that the participants had received, not by any preexisting group differences, we conducted a series of regression analyses to see which background characteristics played a significant role in cognitive control abilities in the post-test. Several factors were moderately correlated (father education and mother education: r = 0.498, p < 0.001; AoA and age: r = 0.364, p < 0.001; AoA and mother education: r = -0.334, p < 0.001; self-rated L2 proficiency and tested L2 proficiency: r = 0.363, p < 0.001), we therefore adopted stepwise regressions to overcome the difficulty in assessing the unique contribution of a variable. The result was that father education significantly contributed to Stroop effect (father education: β = 0.238, p = 0.007) and Stroop inhibition (father education: β = 0.180, p = 0.040)2. The three groups differed in pre-test father education (p = 0.022, η2 = 0.057), mother education (p = 0.031, η2 = 0.052) and age (p = 0.079, η2 = 0.038). A closer look at parents’ education across the groups shows that the translation group enjoyed higher parents’ education than the other two groups. Besides, the interpreting group was the youngest among the three groups. We therefore matched the participant groups on background characteristics and cognitive control abilities in the pre-test mainly in two steps. First, we excluded participants with high parents’ education from the translation group (five participants) and participants with low parents’ education from the control group (five participants) and interpreting group (four participants). Second, we excluded one oldest participant from the control group and three youngest participants from the interpreting group. See Supplementary A for detailed information of the excluded participants. Table 5 shows the result of the match, with participant groups matched in all the key testing items, especially those of cognitive control abilities (e.g., group match for N-back accuracy rate enhanced). Table 5 Group means (with SD) and comparisons (p value) of participants’ background characteristics and task performances in pre- and post- tests after group match. Control (n = 37) Translation (n = 35) Interpreting (n = 44) p-Value Background characteristics      Translation/interpreting No No No      Tested L2 proficiency 14.00 (3.70) 14.12 (4.57) 13.36 (3.55) 0.645      Self-rated L2 proficiency 20.43 (4.07) 21.31 (5.69) 20.41 (5.52) 0.694      Self-rated L2 use 0.060 (0.036) 0.050 (0.049) 0.048 (0.045) 0.438      AoA 8.73 (2.41) 9.14 (2.49) 9.20 (2.33) 0.645      Age 19.73 (0.83) 19.85 (1.03) 19.54 (0.73) 0.276      Father education 2.46 (0.66) 2.91 (1.31) 2.82 (1.24) 0.187      Mother education 2.11 (1.02) 2.57 (1.29) 2.34 (1.22) 0.256      Intelligence 67.65 (2.46) 66.86 (2.64) 66.50 (3.22) 0.186 Cognitive control abilities in the pre-test      Stroop: global RTs (ms) 676.52 (86.17) 682.34 (60.66) 662.88 (65.03) 0.457      Stroop: Stroop effect 34.13 (37.42) 33.59 (51.82) 18.48 (38.35) 0.172      Stroop: Stroop inhibition 14.68 (40.05) 6.40 (59.86) -1.53 (32.08) 0.267      Stroop: Stroop facilitation -19.45 (33.75) -27.19 (32.50) -20.01 (40.72) 0.598      Color-shape: global RTs 612.30 (148.68) 601.44 (152.75) 581.31 (132.42) 0.616      Color-shape: mixing cost 138.86 (132.39) 103.43 (118.39) 113.48 (93.21) 0.397      Color-shape: switch cost 140.32 (73.49) 122.05 (81.76) 139.23 (90.19) 0.572      N-back: global RTs 855.91 (266.30) 831.97 (225.65) 852.17 (250.49) 0.907      N-back: accuracy rate 0.85 (.091) 0.86 (.080) 0.84 (.091) 0.533 Cognitive control abilities in the post-test      Stroop: global RTs 629.08 (82.73) 624.97 (46.92) 612.48 (52.45) 0.457      Stroop: Stroop effect 28.03 (28.42) 26.31 (30.71) 31.14 (32.28) 0.776      Stroop: Stroop inhibition 0.29 (40.63) 6.85 (34.46) 9.06 (31.39) 0.526      Stroop: Stroop facilitation -27.74 (35.09) -19.46 (30.57) -22.08 (34.16) 0.542      Color-shape: global RTs 548.03 (83.89) 557.91 (109.72) 529.95 (91.09) 0.414      Color-shape: mixing cost 99.57 (58.73) 110.23 (87.84) 107.75 (74.01) 0.813      Color-shape: switch cost 132.73 (82.93) 118.71 (70.88) 94.20 (58.19) 0.049      N-back: global RTs 876.62 (262.99) 762.40 (192.39) 752.93 (215.11) 0.032      N-back: accuracy rate 0.89 (0.084) 0.91 (0.067) 0.90 (0.068) 0.627 Pre-test– Post-test Comparisons across Groups It is important to know how each group progressed from the pre-test to the post-test and whether groups differed from each other in the degree of progress. Participant Group (between-subject factor) × Test Time (within-subject factor) ANOVAs were therefore conducted. Table 6 shows the result of analyses. Table 6 Summary of Group × Test Time analyses in each task index of the cognitive control tasks. Main effect of Test Time Main effect of Group Interaction effect Stroop: global RTs p < 0.001, ηp2 = 0.475 p = 0.416, ηp2 = 0.015 p = 0.734, ηp2 = 0.005 Stroop: Stroop effect p = 0.961, ηp2 < 0.001 p = 0.519, ηp2 = 0.012 p = 0.157, ηp2 = 0.032 Stroop: Stroop inhibition p = 0.839, ηp2 < 0.001 p = 0.815, ηp2 = 0.004 p = 0.165, ηp2 = 0.031 Stroop: Stroop facilitation p = 0.851, ηp2 < 0.001 p = 0.414, ηp2 = 0.015 p = 0.852, ηp2 = 0.003 Color-shape: global RTs p < 0.001, ηp2 = 0.196 p = 0.511, ηp2 = 0.012 p = 0.713, ηp2 = 0.006 Color-shape: mixing cost p = 0.233, ηp2 = 0.013 p = 0.785, ηp2 = 0.004 p = 0.206, ηp2 = 0.028 Color-shape: switch cost p = 0.014, ηp2 = 0.052 p = 0.371, ηp2 = 0.017 p = 0.039, ηp2 = 0.056 N-back: global RTs p = 0.012, ηp2 = 0.054 p = 0.300, ηp2 = 0.021 p = 0.033, ηp2 = 0.059 N-back: accuracy rate p < 0.001, ηp2 = 0.261 p = 0.634, ηp2 = 0.008 p = 0.403, ηp2 = 0.016 Simple effect: the control group Simple effect: the translation group Simple effect: the interpreting group Color-shape: switch cost p = 0.566, r = 0.086 p = 0.806, r = 0.049 p < 0.001, r = 0.492 N-back: global RTs p = 0.546, r = 0.092 p = 0.051, r = 0.332 p = 0.002, r = 0.455 As can be seen in Table 6, the main effect of Test Time was significant for the index of global RTs in all three tasks, and also for the index of accuracy rate in the N-back task, reflecting a general training or test practice effect. For the other indices (Stroop effect, inhibition, facilitation; color-shape mixing and switch costs), the main effect of Test Time was only significant for switch cost. No main effect of Participant Group was found. However, the interaction effect was significant for the two indices of switch cost and N-back global RTs, which requires further simple effect analyses. The lower part of Table 6 displays the results of Test-Time simple effect analyses, which shows that the interpreting group made significant progress in these two indices (p < 0.001, r = 0.492; p = 0.002, r = 0.455), and that the translation group made marginally significant progress in N-back RTs (p = 0.051, r = 0.332) and no significant progress in switch cost (p = 0.806, r = 0.049), while the control group didn’t make any significant progress (p = 0.566, r = 0.086; p = 0.546, r = 0.092). These results are consistent with the hypothesis that interpreting enhances switching and updating abilities, and that compared with general L2 training, interpreting brings advantages in switching and updating. Since all the indices were comparable in the pre-test, we conducted further analysis with the post-test data from the two critical indices of switch cost and N-back RT (as in a cross-sectional design). As Table 5 shows, significant group differences were found in post-test switch cost and N-back global RTs (p = 0.049, p = 0.032). The Tukey HSD post hoc tests in switch cost showed significantly less switch cost from the interpreting group than the control group (p = 0.042, r = 0.266), while no significant group difference was found between the control and translation groups (p = 0.678, r = 0.091), or between the translation and interpreting groups (p = 0.280, r = 0.189). The Tukey HSD post hoc tests in N-back RTs also showed a significant difference between the interpreting and control groups (p = 0.040, r = 0.253), while only marginal difference was found between the control and translation (p = 0.085, r = 0.243) and no difference between the translation and interpreting groups (p = 0.981, r = 0.023). These results further reflect an interpreting experience advantage in switching and WM updating (compared with general L2 training and translation training). Discussion The present longitudinal study was intended to investigate whether the two specific bilingual experiences of written translation and consecutive interpreting would produce different cognitive control effects in young adults. To better control potential confounding factors and to avoid the cause-effect ambiguity, we conducted a longitudinal study with three groups of participants matched at the pre-test. The results indicate that the interpreting experience produced significant cognitive advantages in switching (as shown in switch cost in the color-shape task) and updating (as shown in global RTs in the n-back task), while the translation experience produced marginally significant improvements in updating. Neither interpreting nor translation experience brought any advantage to inhibitory control (as shown in the Stroop effect) and monitoring (as shown in global RTs in the number stroop and color-shape tasks, and in mixing cost in the color-shape task). The present study seems to have provided an answer to the question of “when” practice in one domain generalizes to another domain (part of questions asked by Hartsuiker, 2015). As summarized above, we found that the language switching practice in interpreting (32 class hours in one semester) produced significant domain-general switching advantage, while the language switching practice in translation did not (although there seemed a small tendency of similar effect in switch cost). Since the two language switching experiences of interpreting and translation mainly differ in time pressure and processing demands, this finding of the present study suggests that a prerequisite for a certain training to bring about general cognitive advantage is probably high processing demand, which is immediate processing under time pressure in the present study. This is consistent with the speculation made by Schroeder and Marian (2016). That is, when the supply is below the demand in a certain task, the cognitive system tries to adapt and thus gets strengthened. This may explain what has been found in previous studies on bilingual advantages. In other words, bilingual advantages would probably occur if the bilingual task is demanding enough. If, however, a student learns a second language occasionally or once in a while in a classroom, bilingual advantages would probably not occur. This prerequisite for cognitive advantage transfer as defined above may also explain what has been found in non-linguistic practice. Anguera et al. (2013), an excellent example, found that by playing a (high interference) multitasking video game, older adults (60–85 years old) significantly reduced multitasking costs compared to an active control group playing a single task game and a control group without contact with video games. What’s critical is that this training produced benefits to untrained cognitive control abilities, i.e., enhanced sustained attention and WM. In other words, the cognitive advantage transfer (“reduced multitasking costs” to “enhanced sustained attention and WM) was made possible by the multitasking video game, which is certainly more demanding than the single task game. The present study helps specify relevant findings in the literature. First, previous studies on the relationship between interpreting experience and cognitive control advantages did not explicitly distinguish between the oral and written modes of language switching experience (e.g., Yudes et al., 2011; Dong and Xie, 2014; Woumans et al., 2015). We may now speculate that it was probably the oral mode of language switching experience, i.e., oral interpreting, that had brought the cognitive advantages, especially the advantage in switching (because oral interpreting requires immediate processing under time pressure and is therefore more demanding). Second, the absence of inhibition and monitoring advantages in the present study are consistent with what has been found in relevant previous studies using similar tasks (Yudes et al., 2011; Dong and Xie, 2014). Woumans et al. (2015), however, was a different study that investigated how interpreters, bilinguals and monolinguals performed in the Simon and ANT tasks. The interpreters outperformed the unbalanced (but not balanced) bilinguals in the two tasks (i.e., higher accuracy in both tasks and smaller error congruency effect in the ANT), suggesting the modulation effect of interpreting experience on non-linguistic inhibition tasks. But a closer look at the data indicates that L2 proficiency may have partly contributed to the interpreters’ better performance, since the gap in L2 proficiency between the unbalanced group and the interpreters was large while that between interpreters and balanced bilinguals was small (L2 proficiency on a 5-point scale is 2.6 for unbalanced bilinguals, 3.7 for interpreters and 4.2 for balanced bilinguals; L2 fluency is 5.9 for unbalanced bilinguals, 14.0 for interpreters and 12.9 for balanced bilinguals). To test whether interpreting experience would lead to better non-linguistic inhibition, we may have to conduct more research with more tasks, especially tasks of higher sensitivity (e.g., the Go/Nogo task with ERP techniques). A challenge for the present study is that two previous studies (Babcock and Vallesi, 2015; Becker et al., 2016) found that professional SIs exhibited reduced mixing costs in the color-shape task when compared to bilingual controls, suggesting that interpreting experience enhances the function of monitoring rather than that of switching. Apart from the criticisms aimed at cross-sectional studies, there may be other reasons to explain the different findings between the present study and the two previous studies. The most probable reason, according to our understanding, lies in the stages of interpreting experience that are different among the studies. At an early stage of interpreting experience as investigated in the present study, switching efficiently between two languages is probably the most obvious challenge, while at a professional stage as investigated in the two previous studies, switching is probably no longer so challenging. Instead, interpreting as a professional (esp. as a professional SIs) requires better management of the situation, monitoring whatever changes and exchanges in the complex situation of communication, and deciding when and how to step in to help the communication. Facing up to the different main challenges at different stages of interpreting experience may lead to exercises of different cognitive control functions and thus strengthen different functions. This explanation also fits with the fact that the control groups were very different among the studies. The control participants in the present study were intermediate L2 learners, while in these two studies they were highly proficient in both languages and they were probably highly proficient in switching between two languages, esp. for the control group of professional consecutive interpreters and translators in Becker et al. (2016). More empirical research is definitely needed to test the explanation. The finding about the updating advantage in the present study is an important contribution to the literature (see also Morales et al., 2015). In the N-back task, participants were asked to report whether the currently presented item matched the item presented n items back. It is considered a measure of WM, but empirical research indicates that N-back task performance is only weakly correlated with typical measurements of WM, i.e., the complex span (e.g., reading span) (Redick and Lindsey, 2013). The task of interpreting poses high demands on WM, but how individual differencs in WM affect interpreting performance, and whether interpreting training leads to higher WM are controversial (Dong and Cai, 2015). The present study shows that, compared to general L2 training, interpreting training brought significant improvements to updating in WM, and translation training brought marginally significant improvements to updating in WM. What this finding suggests is that updating is perhaps a better way to measure how WM plays its role in the task of interpreting, and thus a better index for the relationship between WM and interpreting. In short, the present longitudinal study investigated the influence of translation and interpreting experiences on the development of cognitive control functions. The advantage in the non-linguistic switching tasks yielded by interpreting instead of translation experiences at an early stage of interpreting experience suggests that high-processing demands may be critical to improving cognitive control, which may be able to explain the inconsistent findings in bilingual cognitive control reported so far. This explanation is consistent with what was found in the comparative study of multitasking and single task video games (Anguera et al., 2013), and with the supply demand explanation by Schroeder and Marian (2016). Furthermore, the results from the present study lead us to speculate that there might be a development curve of cognitive control enhancement in multitasking training such as L2 training, interpreting training or video games training. At the beginning, the curve goes up slowly but steadily, but at a certain point where participants have reached a cognitive peak, the curve would start to level off. More importantly, the curve may start to drop off slowly when the training becomes less demanding probably because participants become more proficient and automatic in the task. In other words, a skill that requires lots of controlled processing in the early stages may help enhance cognitive control functions, but when that skill becomes automatic and requires far less controlled processing, the early advantages may dissipate. More empirical studies, i.e., studies of longitudinal nature, studies of training with better controlled designs, studies employing additional experimental methods like ERP or fMRI, are certainly needed to verify these speculations. Author Contributions YD had the idea and design and did most of the writing while YL managed the experiments and analyzed the data and helped with the writing and revision. Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Funding. This research is supported by a grant (15AYY002) to the corresponding author from the National Social Science Foundation of China. 1 Apart from the color-shape task, we decided to use the WCST task in the post-test because relevant studies in the literature (Yudes et al., 2011; Dong and Xie, 2014) used this task in their cross-sectional designs and found an interpreter advantage in the performance of the task. If a post-test group comparison in the WCST in the present longitudinal study was found, it would be a triangulation for the two previous studies. Since the WCST was only tested in the post-test, it was not consistent with the longitudinal design, and it was therefore only reported in Supplementary. 2 In addition, mother education significantly predicted three indices of WCST (global RTs: β = -0.178, p = 0.042; overall errors: β = -0.215, p = 0.017; perseverative errors: β = -0.207, p = 0.022). Self-rated L2 proficiency together with age significantly contributed to WCST previous category errors (self-rated L2 proficiency: β = -0.204, p = 0.022; age: β = 0.182, p = 0.040). Supplementary Material The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpsyg.2016.01297 Click here for additional data file. ==== Refs References Abutalebi J. Guidi L. Borsa V. Canini M. Della Rosa P. A. Parris B. A. (2015 ). Bilingualism provides a neural reserve for aging populations. Neuropsychologia 69 201 –210 . 10.1016/j.neuropsychologia.2015.01.040 25637228 Anguera J. A. Boccanfuso J. Rintoul J. L. Al-Hashimi O. Faraji F. Janowich J. (2013 ). Video game training enhances cognitive control in older adults. Nature 501 97 –101 . 10.1038/nature12486 24005416 Babcock L. Vallesi A. (2015 ). Are simultaneous interpreters expert bilinguals, unique bilinguals, or both? Biling. Lang. Cogn. 18 1 –15 . 10.1017/S1366728915000735 Bachman L. F. (1985 ). Performance on cloze tests with fixed-ratio and rational deletions. Tesol Q. 19 535 –556 . 10.2307/3586277 Bak T. H. Nissan J. J. Allerhand M. M. Deary I. J. (2014 ). Does bilingualism influence cognitive aging? Ann. Neurol. 75 959 –963 . 10.1002/ana.24158 24890334 Barac R. Bialystok E. (2012 ). Bilingual effects on cognitive and linguistic development: role of language, cultural background, and education. Child Dev. 83 413 –422 . 10.1111/j.1467-8624.2011.01707.x 22313034 Becker M. Schubert T. Strobach T. Gallinat J. Kühn S. (2016 ). Simultaneous interpreters vs. professional multilingual controls: group differences in cognitive control as well as brain structure and function. Neuroimage 134 250 –260 . 10.1016/j.neuroimage.2016.03.079 27085505 Bialystok E. Barac R. Blaye A. Poulin-Dubois D. (2010 ). Word mapping and executive functioning in young monolingual and bilingual children. J. Cogn. Dev. 11 485 –508 . 10.1080/15248372.2010.516420 21197133 Bialystok E. Craik F. I. M. Klein R. Viswanathan M. (2004 ). Bilingualism, aging, and cognitive control: evidence from the Simon task. Psychol. Aging 19 290 –303 . 10.1037/0882-7974.19.2.290 15222822 Bialystok E. Craik F. I. M. Luk G. (2008 ). Cognitive control and lexical access in younger and older bilinguals. J. Exp. Psychol. Learn. Mem. Cogn. 34 859 –873 . 10.1037/0278-7393.34.4.859 18605874 Blumenfeld H. K. Marian V. (2011 ). Bilingualism influences inhibitory control in auditory comprehension. Cognition 118 245 –257 . 10.1016/j.cognition.2010.10.012 21159332 Carlson S. M. Meltzoff A. N. (2008 ). Bilingual experience and executive functioning in young children. Dev. Sci. 11 282 –298 . 10.1111/j.1467-7687.2008.00675.x 18333982 Costa A. Hernandez M. Costa-Faidella J. Sebastian-Galles N. (2009 ). On the bilingual advantage in conflict processing: now you see it, now you don’t. Cognition 113 135 –149 . 10.1016/j.cognition.2009.08.001 19729156 Costa A. Hernandez M. Sebastian-Galles N. (2008 ). Bilingualism aids conflict resolution: evidence from the ANT task. Cognition 106 59 –86 . 10.1016/j.cognition.2006.12.013 17275801 de Abreu P. M. E. Cruz-Santos A. Tourinho C. J. Martin R. Bialystok E. (2012 ). Bilingualism enriches the poor enhanced cognitive control in low-income minority children. Psychol. Sci. 23 1364 –1371 . 10.1177/0956797612443836 23044796 Dijkstra T. Van Heuven W. J. B. (2002 ). The architecture of the bilingual word recognition system: from identification to decision. Biling. Lang. Cogn. 5 175 –197 . 10.1017/S1366728902003012 Dong Y. Cai R. (2015 ). “Workig memory and interpreting: a commentary on theoretical model ,” in Working Memory in Second Language Acquisition and Processing: Theory, Research and Commentary , eds Wen Z. Mota M. McNeill A. (Bristol : Multilingual Matters ), 63 –81 . Dong Y. Li P. (2015 ). The cognitive science of bilingualism. Lang. Linguist. Compass 9 1 –13 . 10.1111/lnc3.12099 Dong Y. Xie Z. (2014 ). Contributions of second language proficiency and interpreting experience to cognitive control differences among young adult bilinguals. J. Cogn. Psychol. 26 506 –519 . 10.1080/20445911.2014.924951 Dragsted B. Hansen I. G. (2009 ). Exploring translation and interpreting hybrids: the case of sight translation. META 54 588 –604 . Gathercole V. Thomas E. Kennedy I. Prys C. Young N. Viñas-guasch N. (2014 ). Does language dominance affect cognitive performance in Bilinguals? Lifespan evidence from preschoolers through older adults on card sorting, simon, and metalinguistic tasks. Front. Psychol. 5 :11 10.3189/fpsyg.2014.00011 Green D. W. (1998 ). Mental control of the bilingual lexico-semantic system. Biling. Lang. Cogn. 1 67 –81 . 10.1017/S1366728998000133 Hartsuiker R. J. (2015 ). Why it is pointless to ask under which specific circumstances the bilingual advantage occurs. Cortex 73 336 –337 . 10.1016/j.cortex.2015.07.018 26303278 Hernandez M. Martin C. D. Barcelo F. Costa A. (2013 ). Where is the bilingual advantage in task-switching? J. Mem. Lang. 69 257 –276 . 10.1016/j.jml.2013.06.004 Hilchey M. D. Klein R. M. (2011 ). Are there bilingual advantages on nonlinguistic interference tasks? Implications for the plasticity of executive control processes. Psychon Bull. Rev. 18 625 –658 . 10.3758/s13423-011-0116-7 21674283 Jared D. (2015 ). What is the theory? Cortex 73 361 –363 . 10.1016/j.cortex.2015.07.009 26286034 Kempe V. Kirk N. W. Brooks P. J. (2015 ). Revisiting theoretical and causal explanations for the bilingual advantage in executive functioning. Cortex 73 342 –344 . 10.1016/j.cortex.2015.07.021 26298266 Kirk N. W. Fiala L. Scott-Brown K. C. Kempe V. (2014 ). No evidence for reduced Simon cost in elderly bilinguals and bidialectals. J. Cogn. Psychol. 26 640 –648 . 10.1080/20445911.2014.929580 Kousaie S. Phillips N. A. (2012 ). Ageing and bilingualism: absence of a “bilingual advantage” in stroop interference in a nonimmigrant sample. Q. J. Exp. Psychol. 65 356 –369 . 10.1080/17470218.2011.604788 Li P. Grant A. (2015 ). Identifying the causal link: two approaches toward understanding the relationship between bilingualism and cognitive control. Cortex 73 358 –360 . 10.1016/j.cortex.2015.07.013 26286036 Marian V. Blumenfeld H. K. Kaushanskaya M. (2007 ). The language experience and proficiency questionnaire (LEAP-Q): assessing language profiles in bilinguals and multilinguals. J. Speech Lang. Hear. Res. 50 940 –967 . 10.1044/1092-4388(2007/067) 17675598 Marzecová A. Asanowicz D. Kriva L. U. Wodniecka Z. (2013 ). The effects of bilingualism on efficiency and lateralization of attentional networks. Biling. Lang. Cogn. 16 608 –623 . 10.1017/S1366728912000569 Miyake A. Friedman N. P. Emerson M. J. Witzki A. H. Howerter A. Wager T. D. (2000 ). The unity and diversity of executive functions and their contributions to complex “Frontal Lobe” tasks: a latent variable analysis. Cogn. Psychol. 41 49 –100 . 10.1006/cogp.1999.0734 10945922 Morales J. Padilla F. Gómez-Ariza C. J. Bajo M. T. (2015 ). Simultaneous interpretation selectively influences working memory and attentional networks. Acta Psychol. 155 82 –91 . 10.1016/j.actpsy.2014.12.004 Morton J. B. (2015 ). Still waiting for real answers. Cortex 73 352 –353 . 10.1016/j.cortex.2015.07.010 26298268 Paap K. Sawi O. (2014 ). Bilingual advantages in executive functioning: problems in convergent validity, discriminant validity, and the identification of the theoretical constructs. Front. Psychol. 5 :962 10.3389/fpsyg.2014.00962 Paap K. Wagner S. Johnson H. Bockelman M. Cushing D. Sawi O. (2014 ). “20,000 Flanker trials: are the effects reliable, robust, and stable? ” in Poster Presented at the Association for Psychological Science , San Francisco, CA . Paap K. R. (2014 ). The role of componential analysis, categorical hypothesising, replicability and confirmation bias in testing for bilingual advantages in executive functioning. J. Cogn. Psychol. 26 242 –255 . 10.1080/20445911.2014.891597 Paap K. R. Greenberg Z. I. (2013 ). There is no coherent evidence for a bilingual advantage in executive processing. Cogn. Psychol. 66 232 –258 . 10.1016/j.cogpsych.2012.12.002 23370226 Paap K. R. Johnson H. A. Sawi O. (2015 ). Bilingual advantages in executive functioning either do not exist or are restricted to very specific and undetermined circumstances. Cortex 69 265 –278 . 10.1016/j.cortex.2015.04.014 26048659 Philipp A. Kalinich C. Koch I. Schubotz R. (2008 ). Mixing costs and switch costs when switching stimulus dimensions in serial predictions. Psychol. Res. 72 405 –414 . 10.1007/s00426-008-0150-x 18443820 Poarch G. J. Bialystok E. (2015 ). Bilingualism as a model for multitasking. Dev. Rev. 35 113 –124 . 10.1016/j.dr.2014.12.003 25821336 Prior A. Gollan T. H. (2011 ). Good language-switchers are good task-switchers: evidence from Spanish-English and Mandarin-English bilinguals. J. Int. Neuropsychol. Soc. 17 682 –691 . 10.1017/S1355617711000580 22882810 Prior A. Macwhinney B. (2010 ). A bilingual advantage in task switching. Bilingualism 13 253 –262 . 10.1017/S1366728909990526 Ratiu I. Azuma T. (2015 ). Working memory capacity: Is there a bilingual advantage? J. Cogn. Psychol. 27 1 –11 . 10.1080/20445911.2014.976226 Raven J. C. Court J. H. Raven J. (1977 ). Manual for Raven’s Advanced Progressive Matrices: Sets I and II. London : H.K.: Lewis & Co. Ltd. Redick T. S. Lindsey D. R. B. (2013 ). Complex span and N-back measures of working memory: a meta-analysis. Psychon. Bull. Rev. 20 1102 –1113 . 10.3758/s13423-013-0453-9 23733330 Schroeder S. Marian V. (2016 ). Cognitive consequences of trilingualism. Int. J. Biling. 20 1 –19 . 10.1177/1367006916637288 Valian V. (2015 ). Bilingualism and cognition. Biling. Lang. Cogn. 18 3 –24 . 10.1017/S1366728914000698 Woumans E. Ceuleers E. Van der Linden L. Szmalec A. Duyck W. (2015 ). Language control in bilinguals and interpreters. J. Exp. Psychol. Learn. Mem. Cogn. 41 1579 –1586 . 10.1037/xlm0000107 25689001 Woumans E. Duyck W. (2015 ). The bilingual advantage debate: moving toward different methods for verifying its existence. Cortex 73 356 –357 . 10.1016/j.cortex.2015.07.012 26277042 Xie Z. Dong Y. (2015 ). Contributions of bilingualism and public speaking training to cognitive control differences among young adults. Biling. Lang. Cogn. 18 1 –14 . 10.1017/S1366728915000474 Yang H. Hartanto A. Yang S. (2016 ). The complex nature of bilinguals’ language usage modulates task-switching outcomes. Front. Psychol. 7 :560 10.3389/fpsyg.2016.00560 Yudes C. Macizo P. Bajo T. (2011 ). The influence of expertise in simultaneous interpreting on non-verbal executive processes. Front. Psychol. 2 :309 10.3389/fpsyg.2011.00309

PMC005xxxxxx/PMC5003827.txt

==== Front Front PharmacolFront PharmacolFront. Pharmacol.Frontiers in Pharmacology1663-9812Frontiers Media S.A. 10.3389/fphar.2016.00284PharmacologyMethodsCombining QSAR Modeling and Text-Mining Techniques to Link Chemical Structures and Carcinogenic Modes of Action Papamokos George 123Silins Ilona 4*1Department of Physics and School of Engineering and Applied Sciences, Harvard UniversityCambridge, MA, USA2Department of Physics, University of IoanninaIoannina, Greece3Biomedical Research Division, Institute of Molecular Biology and Biotechnology Foundation for Research and TechnologyHeraklion, Greece4Institute of Environmental Medicine, Karolinska InstitutetStockholm, SwedenEdited by: Thomas Hartung, University of Konstanz, Germany Reviewed by: Jan Willem Van Der Laan, College ter Beoordeling van Geneesmiddelen, Netherlands; Emilio Benfenati, Mario Negri Institute for Pharmacological Research, Italy *Correspondence: Ilona Silins, Ilona.Silins@ki.seThis article was submitted to Predictive Toxicology, a section of the journal Frontiers in Pharmacology 30 8 2016 2016 7 28410 5 2016 18 8 2016 Copyright © 2016 Papamokos and Silins.2016Papamokos and SilinsThis is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.There is an increasing need for new reliable non-animal based methods to predict and test toxicity of chemicals. Quantitative structure-activity relationship (QSAR), a computer-based method linking chemical structures with biological activities, is used in predictive toxicology. In this study, we tested the approach to combine QSAR data with literature profiles of carcinogenic modes of action automatically generated by a text-mining tool. The aim was to generate data patterns to identify associations between chemical structures and biological mechanisms related to carcinogenesis. Using these two methods, individually and combined, we evaluated 96 rat carcinogens of the hematopoietic system, liver, lung, and skin. We found that skin and lung rat carcinogens were mainly mutagenic, while the group of carcinogens affecting the hematopoietic system and the liver also included a large proportion of non-mutagens. The automatic literature analysis showed that mutagenicity was a frequently reported endpoint in the literature of these carcinogens, however, less common endpoints such as immunosuppression and hormonal receptor-mediated effects were also found in connection with some of the carcinogens, results of potential importance for certain target organs. The combined approach, using QSAR and text-mining techniques, could be useful for identifying more detailed information on biological mechanisms and the relation with chemical structures. The method can be particularly useful in increasing the understanding of structure and activity relationships for non-mutagens. carcinogensmode of actiontext miningQSARrisk assessmenttoxicityprediction ==== Body Introduction Cancer is a major public health problem and the number of cases are expected to increase in the future (Frankish, 2003). Current research indicates that environmental factors, including chemicals, have a major role in the disease development, emphasizing the importance to prevent exposure to compounds possessing carcinogenic potential (Christiani, 2011; Landrigan et al., 2011; Wu et al., 2016). Traditionally, the animal bioassay has been the main method used to identify carcinogens. However, these tests are costly and time-consuming, and recent regulatory policies require a reduction in the number of animals used in chemical testing. Consequently, there is a need for alternative methods to examine toxicological effects of chemicals (Pelkonen, 2010). In order to develop reliable non-animal based tests to identify carcinogens, knowledge of the biological mechanisms that lead to cancer is required. For example, the understanding of chemicals’ modes of action (MOA), i.e., the sequence of key events resulting in cancer, has become increasingly important in hazard identification and risk assessment (Sonich-Mullin et al., 2001; US-EPA, 2005; Boobis et al., 2008). The current understanding of how chemicals cause cancer involves two major MOAs: genotoxicity and non-genotoxicity (indirect genotoxicity). A genotoxic MOA means that the chemical interacts directly with the DNA (which can result in mutagenicity), whereas a non-genotoxic MOA denotes indirect effects, such as stimulation of cell proliferation or inhibition of cell death (US-EPA, 2005). The huge collection of biomedical articles in MEDLINE, available through the search engine PubMed1, provides a great source of information for researchers to utilize and generate new knowledge. However, considering the enormous amount of articles, more than 26 million to date, it is getting more and more problematic for researchers to handle information relevant to them. For such purposes, techniques like text-mining could be used for locating and managing information overload. Recently, biomedical text-mining has become increasingly popular for handling the large volumes of texts in biomedical sciences (Cohen and Hersh, 2005; Zweigenbaum et al., 2007). Today, there is a wide range of different text-mining tools available to support researchers in the biomedical field (Cohen and Hersh, 2005; Zweigenbaum et al., 2007; Zhu et al., 2013; Fleuren and Alkema, 2015; Gonzalez et al., 2016). One such tool, CRAB, has been developed to support classification of literature relevant to cancer risk assessment (Korhonen et al., 2009, 2012; Guo et al., 2014). This tool automatically classifies literature based on the carcinogenic evidence that is mentioned in the text of scientific abstracts. Based on the text analysis the tool generates toxicological literature profiles that can be used for cancer risk assessment or cancer research. This approach facilitates the detection of new patterns in data, which could be a nearly impossible task by manual literature search and evaluation. Such data patterns can be used, e.g., to compare individual substances or groups of chemicals to generate new hypotheses that can be tested experimentally (Korhonen et al., 2009, 2012; Kadekar et al., 2012; Silins et al., 2014; Ali et al., 2016). Quantitative structure-activity relationship (QSAR) modeling is an important computational tool in medicinal chemistry and predictive toxicology (Hansch et al., 1962; Cherkasov et al., 2014). It is a procedure by which a chemical structure is quantitatively linked with a clearly defined process, typically biological activity or chemical reactivity. The QSAR model systems build on structure-activity relationships of known chemicals, and can be used to predict the toxicity of unknown chemicals based on their structures (Combes, 2012). This technique has proven especially useful in predicting mutagenicity based on structural alerts, which are mechanistically linked to carcinogenicity (Benigni and Bossa, 2011). Structural alerts are the molecular structures and reactive groups that are responsible for a toxic effect (Benigni et al., 2013). The QSAR method can thus both predict carcinogenicity, and mutagenicity, and provide information about structural alerts based on the chemical structures (Benigni and Bossa, 2006). Traditionally, the QSAR method has been better in predicting reactive (genotoxic) compounds compared to non-reactive (non-genotoxic) carcinogens, however, recently a new set of structural alerts relating to non-genotoxic mechanisms including, e.g., oxidative stress, hormonal imbalance, and peroxisome proliferation has been identified (Benigni et al., 2013). Several new alternative approaches for predicting carcinogens in connection with QSAR have been suggested (Benigni, 2014). For example, a strategy using QSAR in a tiered approach combined with in vitro tests for genotoxicity and tumor promotion has been proposed (Benigni, 2014). Another approach to improve prediction in combination with QSAR is based on mechanistic information, involving the concept of adverse outcome pathways (AOP; Benigni, 2014). The AOP outlines the sequence of events starting from a molecular initiating event, through a series of key events, resulting in an adverse effect (Vinken, 2013). The AOP and the MOA (described above) are similar concepts that take into account mechanistic information to improve, e.g., risk assessment, however, one major difference is that a MOA focuses on the details specific to a particular chemical, whereas the AOPs are chemical-agnostic (Edwards et al., 2016; Kleinstreuer et al., 2016). The purpose of this study was to test whether combining QSAR methodology with a text-mining approach based on carcinogenic MOA could be useful to identify new associations between chemical structures and biological activities related to carcinogenesis. Ninety-six rat carcinogens were selected from the National Toxicology Program’s (NTP) database, and literature profiles and QSAR data were generated for each carcinogen. Based on both the QSAR data and on text mining-generated literature profiles we found that skin and lung rat carcinogens were mainly mutagenic, while the group of carcinogens affecting the hematopoietic system and the liver also included a large proportion of non-mutagens. Mutagenicity was a found to be a frequently reported endpoint in the literature, however, less common endpoints such as immunosuppression and hormonal receptor-mediated effects were also found in literature on some carcinogens, which could be of potential importance. The approach to combine QSAR and text-mining could be particularly useful for identifying biological mechanisms of potential relevance to non-mutagens. Materials and Methods Selection of Carcinogens The NTPs database2 was used to select the rat carcinogens included in this study. Four common organ sites were selected, including the hematopoietic system (i.e., leukemia or lymphoma), liver, lung, and skin. All rat carcinogens affecting these four organs and classified by NTP as positive, clear, or some evidence were selected for further analysis. Based on these criteria, a total of 126 rat carcinogens were included. Among these carcinogens, 30 chemicals affected one or more of the other three organs, leaving a total of 96 individual chemicals for further analysis. Analysis of Carcinogenic MOA Using a Text-Mining Approach To investigate the carcinogenic MOAs concerning the 96 selected rat carcinogens we used the text mining-based tool CRAB (Korhonen et al., 2009, 2012; Guo et al., 2014) to analyze the scientific literature. The published literature concerning these carcinogens was retrieved from PubMed3 using the chemicals’ nomenclature or CAS numbers. This analysis was based on literature published until January 2015. The literature collection of each carcinogen was automatically classified by the tool, which categorizes scientific abstracts according to a taxonomy that covers the main types of evidence for carcinogenic MOAs. In brief, the taxonomy structure includes two main MOA classes: genotoxicity and non-genotoxicity. It is further branched into 25 sub-categories, ranging from common carcinogenic endpoints, such as mutations, to less common effects, such as inflammation. The classification is based on the evidence mentioned in the abstracts’ text. For each carcinogen of interest the tool generates a publication profile based on the scientific literature, thus the profile reflects the current knowledge about this chemical. The tool automatically calculates the proportion of abstracts in each category (per total number of MOA-relevant abstracts; Guo et al., 2014). The tool is based on advanced text-mining techniques and has shown to generate classification of high accuracy. It can be found at: http://omotesando-e.cl.cam.ac.uk/CRAB/request.html. The carcinogens were grouped according to their target organ, predicted mutagenicity/non-mutagenicity and structural alert. Literature profiles for each group were generated by calculating the average percent for each MOA subcategory. Carcinogens with less than 10 abstracts were excluded in the text-mining analysis. The statistical significance of the results was calculated using the t-test. QSAR Analysis VEGA4 Non-Iterative Client (VEGANIC) v1.0.8, a standalone JAVA-based software was employed and three different SAR models were applied to the current dataset: Mutagenicity model CAESAR (Ferrari and Gini, 2010) version 2.1.12, Mutagenicity SarPy model version 1.0.6-DEV (Ferrari et al., 2013), and Benigni–Bossa Mutagenicity (TOXTREE; Benigni et al., 2008) version 1.0.0-DEV. The input structural data of the chemicals were given in SMILES format (Weininger, 1988). The SMILES chemical structures for each compound were retrieved from PubChem, ChemSpider, or Wikipedia databases using CAS registry numbers, IUPAC nomenclature or empirical chemical names. In a first step, the dataset of 96 carcinogens was curated and counter ions, salts and disconnected structures were removed as no identical compounds were located. In total, 75 carcinogens were included in the QSAR analysis. Linking QSAR Data with Literature Profiles of Carcinogenic MOA The results generated from three different SAR models were compiled in order to decide the structures of carcinogens according to Benigni Bossa code (Benigni et al., 2008). Each of the 75 carcinogens analyzed was associated with a structural alert, if given from the QSAR output. Some of the chemicals were mutagens without a structural alert (named here unspecific mutagens) or were predicted non-mutagens (typically without a proposed structural alert). Certain classification rules were applied when the carcinogens were grouped as mutagenic or non-mutagenic based on the output from the QSAR analysis. When identical results were generated from all the three QSAR models the classification as mutagenic or non-mutagenic was considered certain. If one model presented conflicting results, the experimental result was assumed more reliable than the predicted outcome. As default, carcinogens were considered mutagenic if the QSAR models presented conflicting results (e.g., if one model predicted the chemical as mutagenic and another model as non-mutagenic). Grouping of Chemicals First, carcinogens were grouped according to their target organs (hematopoietic system, liver, lung, and skin). Secondly, carcinogens were grouped based on the QSAR output for each chemical, as mutagens or as non-mutagens. In cases where a chemical could have been entered into both classes because of conflicting results from the different QSAR models, a decision was made regarding the dominant category, and it was entered into that single class. The two groups (mutagens and non-mutagens) were further associated with their average MOA literature profile, an analysis which included 46 mutagens and 22 non-mutagens. Thirdly, carcinogens were grouped based on their structural alerts; eight groups were formed including mutagens (quinones, primary aromatic amines, nitro aromatics, unspecific mutagens, hydrazine, epoxides, and aziridines and aliphatic halogens) and non-mutagens. For each of these groups an average MOA literature profile was generated. Results Literature Analysis of Carcinogenic MOA Using the CRAB-Tool The rat carcinogens affecting the four selected target organs (hematopoietic system, liver, lung, and skin) included in total 126 chemicals. Of these, 30 were carcinogenic in at least one of the other organs, leaving 96 individual rat carcinogens for further analysis. The liver was the most common target site, since 58 of the chemicals affected the liver in rats. Twenty-four chemicals caused cancer in the hematopoietic system, and 22 were skin and lung carcinogens, respectively (Table 1). The total literature collection of the selected carcinogens included almost 130 000 scientific abstracts retrieved from PubMed. The group of skin carcinogens was the most well-studied with a literature collection of almost 50 000 abstracts. Table 1 Literature data for carcinogens affecting the hematopoietic system, liver, lung, and skin in National Toxicology Program’s (NTP) 2-year rat bioassays. Target organ Number of carcinogens Number of abstracts (retrieved from PubMed) Number of abstracts relevant to carcinogenic MOA (modes of action) (CRAB-tool analysis) Hematopoietic system 24 21,837 4,296 Liver 58 49,862 18,097 Lung 22 6,895 1,648 Skin 22 49,902 6,251 Total 126 128,496 30,292 The number of carcinogens per target organ, number of abstracts retrieved from PubMed and the number abstracts classified as relevant to carcinogenic MOA for each target organ are shown.From the whole abstract collection >30 000 abstracts (∼25% of the whole retrieved literature collection) were classified as relevant for carcinogenic MOA by the CRAB-tool. Liver carcinogens were the most well-studied of the four target organs regarding literature relevant to carcinogenesis and MOAs as shown in Table 1. By using the CRAB-tool, the literature collection retrieved from PubMed for each carcinogen was classified, and carcinogenic MOA profiles were generated. As an illustration of a literature distribution pattern, MOA profiles of 21 individual rat carcinogens of the hematopoietic system are shown in Figure 1. The figure shows the percent of abstracts relevant to a certain MOA category, for each carcinogen. From the literature distribution it is observed that one of the carcinogens has a large proportion of literature classified in the strand breaks category (A) and another carcinogen in the immunosuppression category (B). From the same figure can also be seen that the literature of most carcinogens reports about mutagenicity (C), but only one carcinogen has a large proportion of the literature classified in the inflammation category (D). FIGURE 1 Individual literature profiles of 21 rat carcinogens of the hematopoietic system. Twelve selected categories of carcinogenic MOA (modes of action) are shown. Chemicals were grouped according to their target organ and literature profiles were generated for each group (Figure 2). This approach facilitates comparison of carcinogens affecting different target organs. If a specific MOA category stands out in the comparison it may reflect a potentially important mechanism for this organ. The data patterns showed that a larger proportion of literature concerning lung carcinogens reported about mutations as compared to the other organs (significantly different compared to carcinogens of the hematopoietic system). The figure further shows that carcinogens of the hematopoietic system have a significantly larger proportion of literature classified in the immunosuppression category compared to liver carcinogens. In general, the literature patterns indicated that endpoints such as mutations and oxidative stress were commonly studied, while inflammation and hormonal receptor-mediated effects were less frequently reported in literature. FIGURE 2 Comparison of literature profiles for four target organs including hematopoietic system, liver, lung, and skin. The average percent of abstracts classified in the MOA taxonomy is shown on the y-axis. Carcinogens were grouped according to their target organ(s). ∗Significantly different from carcinogens of the hematopoietic system (p ≤ 0.05), ∗∗Significantly different compared to liver carcinogens (p ≤ 0.05). The literature patterns were analyzed in more details. A compilation of the results from the CRAB literature analysis for the four target organs is shown in Table 2. The literature analysis showed that mutation was a commonly studied endpoint, reported in the literature of 80–90% of all carcinogens included. Other common endpoints were chromosomal changes and strand breaks. In addition, mutagenicity was found to be the most well-studied MOA category regarding rat carcinogens of the liver, lung, and skin. Regarding carcinogens of the hematopoietic system, oxidative stress was the most well-studied MOA category, for which, on average, 12% of the MOA literature was classified as relevant. Table 2 Results from the classification of abstracts relevant to carcinogenesis. Target organ Most common MOA (percent of all chemicals) Most well-studied MOA (average percent) Hematopoietic systema Mutations (90%) Oxidative stress (12%) Liverb Mutations (82%) Mutations (13%) Lungc Chromosomal changes, mutations, strand breaks (89%) Mutations (21%) Skind Mutations (89%) Mutations (16%) a21/24, b49/58, c19/22, d18/22 carcinogens were included in the analysis using the CRAB-tool.Analysis of Carcinogens Using the QSAR Method The QSAR method was used to predict the type of carcinogen (mutagen or non-mutagen) and structural alerts. When carcinogens were grouped according to their target organ the QSAR data indicated that most skin carcinogens were mutagens (Table 3). Grouping of skin carcinogens suggested two dominating structural alerts: aliphatic halogens, epoxides, and aziridines, which are both alkylating and direct-acting chemicals. Most of the lung and liver carcinogens were also predicted mutagens, however, a large proportion (38%) of the liver carcinogens were predicted non-mutagens. In addition, although the majority of carcinogens affecting the hematopoietic system were predicted mutagens, a large part (37%) were classified as non-mutagens (Table 3). Thus, compared to carcinogens of the skin and lung, a large proportion of the liver carcinogens and carcinogens affecting the hematopoietic system were non-mutagens. Table 3 The number of carcinogens with predicted structural alerts shown for each target organ. Structural alert Hematopoietic cancer Liver Lung Skin Mutagens: aliphatic halogen 3 3 2 3 Mutagens: epoxides and aziridines 2 0 0 3 Mutagens: hydrazine 1 1 0 0 Mutagens: unspecific 3 3 6 1 Mutagens: nitro aromatics 2 1 3 2 Mutagens: primary aromatic amines 0 5 0 0 Mutagens: quinones 0 4 0 0 Mutagens: other structural alerts 1 8 3 2 Mutagens (in total) 12 25 14 11 Non-mutagens 7 15 4 2 Some carcinogens affected more than one target organ.Combining QSAR and Text Mining-Generated MOA Profiles Chemicals were grouped either as mutagens or as non-mutagens, based on the output from the QSAR modeling. The group of mutagens included 46 chemicals and 22 chemicals were non-mutagens. Literature profiles were generated for each of the two groups. Figure 3 shows the differences in the literature distributions between them. The proportion of literature classified as relevant to genotoxic endpoints or to non-genotoxic categories is in line with the data from the QSAR analysis. For example, literature concerning mutagens was more frequently classified in genotoxic MOA-categories, including mutation, strand breaks, and chromosomal changes. Non-mutagens, on the other hand, had more literature classified in non-genotoxic MOA-categories, e.g., hormonal receptor-mediated effects, as compared to mutagens. FIGURE 3 Distribution of literature concerning mutagens and non-mutagens in the MOA taxonomy. Carcinogens were grouped into two groups (mutagens and non-mutagens) based on the results from QSAR modeling. Carcinogenic MOA profiles were generated for the two groups. The literature distribution is shown as the average percent of abstracts in the MOA category. ∗Signficantly different compared to the other group (p ≤ 0.05). Sixty-eight carcinogens, for which QSAR data had been generated and that had enough literature data required for analysis were grouped based on their structural alerts. The aim was to investigate whether more detailed information regarding the chemical structures could be associated with a particular MOA category. Eight groups were formed, seven groups included mutagens with different structural alerts and one group consisted of non-mutagens (without structural alerts). Each structural alert group was linked to its corresponding literature profile. The two most common MOA categories for each group is presented in Table 4. The mutation and oxidative stress categories were the dominating categories. Cell proliferation and oxidative stress were the most common categories for non-mutagens (same data as shown in Figure 3). However, the number of carcinogens included in each group was small, ranging from three carcinogens in the group of hydrazines and epoxides and aziridines, to 22 carcinogens in the group of non-mutagens. Table 4 Linking structural alerts with carcinogenic MOA information. Structural alert Most common MOA categories 1. Aliphatic halogen (alkylating, direct acting agents) Mutations, oxidative stress 2. Epoxides and aziridines (alkylating, direct acting agents) Cell proliferation, cell death 3. Hydrazine (alkylating, direct acting agents) Oxidative stress 4. Unspecific mutagens Oxidative stress, mutations 5. Nitro aromatics (DNA adducts, indirect acting agents) Mutations, chromosomal changes 6. Non-mutagens Cell proliferation, oxidative stress 7. Primary aromatic amines (DNA adducts, indirect acting agents) Strand breaks, mutations 8. Quinones (alkylating, direct acting agents) Mutations, chromosomal changes Carcinogens with the same structural alerts were grouped. Each group was linked with their corresponding literature profile. The most and second most common MOA category for each group is shown.The literature patterns generated by the CRAB-tool can provide new information of potential interest that can be used to form new hypotheses. When the output from the QSAR analysis was linked with information on the target organs affected, we found that the group of carcinogens affecting the hematopoietic system included a larger proportion of non-mutagens (7 of 19 carcinogens with QSAR data) compared to the other organs. The literature patterns of these seven non-mutagens were analyzed in more detail (Figure 4) and we found that the most common endpoints studied for these carcinogens were oxidative stress, cell proliferation, and cytotoxicity, which are all non-genotoxic effects. Interestingly, the literature concerning five of these non-mutagens (2,4,6-Trichlorophenol, Butyl benzyl phthalate, Hydroquinone, Mirex, and Furan) had data classified in the category of hormonal receptor-mediated effects. This result is also in line with what is known about some of these compounds (Ma et al., 2011; Upson et al., 2013; Alam and Kurohmaru, 2016). FIGURE 4 Seven non-mutagenic rat carcinogens of the hematopoietic system were investigated in more detail. The figure shows 13 selected MOA categories and the literature distribution over these classes for carcinogens affecting the hematopoietic system in rats. Discussion In this study, we tested the idea of combining the QSAR method with a text-mining approach to generate more detailed information regarding the relationship between chemical structures and carcinogenic mechanisms (MOAs). The literature of 96 rat carcinogens was analyzed using the text mining-based CRAB tool (Korhonen et al., 2009, 2012; Guo et al., 2014). QSAR models were used to predict mutagenicity and structural alerts for 75 of these carcinogens. The chemicals were grouped based on target organ, mutagenicity and structural alerts, and literature profiles were generated for each chemical group with the aim to discover new patterns in data that connect target organs, chemical structures, and carcinogenic MOAs. The text-mining analysis showed that the mutation endpoint was frequently studied in connection with most of the 96 rat carcinogens, particularly in relation to lung and skin carcinogens. This is not surprising as mutagenicity is known to have a central role in carcinogenesis. In addition, the mutation endpoint is widely used in studies of carcinogens and in screening tests of mutagenicity (Mortelmans and Zeiger, 2000). By using QSAR models we also found that the groups of carcinogens affecting the liver and the hematopoietic system in rats included a large proportion of non-mutagens. These data are in line with a previous study of 522 carcinogens (Ashby and Paton, 1993), where it was shown that these organs were partly affected by carcinogens without reactive molecular sites. The same study also showed that rat lung and skin carcinogens included mainly reactive chemicals (Ashby and Paton, 1993). Data patterns related to rare carcinogenic endpoints may also be of interest, e.g., regarding non-genotoxic chemicals for which detailed carcinogenic mechanisms may not be known. By using the text-mining approach to compare groups of chemicals new data patterns of potential importance can be found. In the current study, we found that immunosuppression was frequently mentioned in the literature concerning rat carcinogens affecting the hematopoietic system. This is an interesting finding, which is also in line with the known mechanisms of human carcinogens affecting this organ (Adamson and Seiber, 1981; IARC, 2015). An association between immunosuppressant drugs and development of cancer in the hematopoietic system (lymphomas) in humans has also been shown previously (Bugelski et al., 2010). However, as the value of the rodent carcinogenicity assay in predicting human toxicity caused by immunosuppressants has been questioned (Bugelski et al., 2010) it would be of interest to apply the same method on a set of human carcinogens affecting the hematopoietic system. Another finding concerning carcinogens of the hematopoietic system was a relatively large proportion of literature linked to hormonal effects, compared with the other organs. Although the findings were based on only a few rat carcinogens this result may indicate a potentially important mechanism for cancer development in this organ, possibly also for humans. Although there are articles reporting on potential links between hormonally active substances and cancer of the hematopoietic system in humans (Traversa et al., 1998; Poynter et al., 2013; Leal et al., 2016), the aetiologies of this cancer type are still unclear (Laurier et al., 2014). More research is required to support these findings and it would, e.g., be of interest to evaluate the structures of the chemicals in more details and investigate potential links with hormonal receptors. In addition, human carcinogens targeting this organ should be analyzed using the same approach. When the rat carcinogens were organized into groups based on their proposed structural alerts, we found that the literature of carcinogens with predicted mutagenic structural alerts reported more frequently about genotoxic effects compared to non-mutagenic carcinogens. This comparison confirms that the outcomes of the two methods are consistent. Our initial idea was that more detailed information regarding structural alerts linked to text mining-generated information could provide new data patterns of potential interest. This approach could be particularly useful to increase the knowledge about how non-genotoxic compounds act, e.g., in a certain organ. More detailed structural information could be important because the knowledge about how the chemical structures of these compounds link to biological effects, on a mechanistic level, is still weak. Furthermore, a problem in current non-animal based cancer testing is the lack of reliable systems to detect non-genotoxic carcinogens (Benigni et al., 2013). Thus, development of new approaches to study, e.g., non-mutagenic carcinogens is important to improve future testing strategies. Although QSAR models have proven useful in predicting mutagens, the method is more challenging for non-genotoxic carcinogens (Silva Lima and Van der Laan, 2000; Benigni et al., 2013; Luijten et al., 2016). There are several explanations for this difference, e.g., a better mechanistic understanding of how mutagenic compounds cause cancer, compared to non-genotoxic carcinogens. Furthermore, the databases used for QSAR contain more data on mutagenic carcinogens, which makes the basis for analysis stronger, leading to more robust predictions for mutagens (Benigni et al., 2013). Another more general difficulty related to non-genotoxic carcinogens is that these compounds may target specific organs, often depending on organ-specific metabolic mechanisms (Silva Lima and Van der Laan, 2000). As these characteristics can be species-specific, the human relevance of certain non-mutagenic mechanisms may be unclear. Predicting metabolic induction of enzymes such as cytochromes P450 using computational approaches (Kirchmair et al., 2015) could be useful to identify chemicals with potential to cause tumors in, e.g., the rodent liver (Graham and Lake, 2008). In this study we have combined QSAR data with text mining-generated literature profiles of carcinogenic MOAs to generate new patterns in data to explain the link between chemical structure and carcinogenic effects. This approach could be valuable in studies of non-mutagens, where more knowledge about structure and activity relationships is needed. The overall strategy, using these two methods in combination, also needs further evaluation, e.g., by including additional non-mutagens in the analysis and to further test its usefulness, maybe also as a predictive approach. Author Contributions IS conceived the original idea, designed and performed research, analyzed results, wrote the paper. GP designed and performed research, analyzed results and wrote the paper. Both authors approved the submitted manuscript. Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. 1 http://www.ncbi.nlm.nih.gov/pubmed 2 http://ntp.niehs.nih.gov/index.cfm 3 http://www.ncbi.nlm.nih.gov/pubmed 4 http://www.vega-qsar.eu/ ==== Refs References Adamson R. H. Seiber S. M. (1981 ). Chemically induced leukemia in humans. Environ. Health Perspect. 39 93 –103 . 10.1289/ehp.813993 6786872 Alam M. S. Kurohmaru M. (2016 ). Butylbenzyl phthalate induces spermatogenic cell apoptosis in prepubertal rats. Tissue Cell 48 35 –42 . 10.1016/j.tice.2015.12.001 26747412 Ali I. Guo Y. Silins I. Hogberg J. Stenius U. Korhonen A. (2016 ). Grouping chemicals for health risk assessment: a text mining-based case study of polychlorinated biphenyls (PCBs). Toxicol. Lett. 241 32 –37 . 10.1016/j.toxlet.2015.11.003 26562772 Ashby J. Paton D. (1993 ). The influence of chemical structure on the extent and sites of carcinogenesis for 522 rodent carcinogens and 55 different human carcinogen exposures. Mutat. Res. 286 3 –74 . 10.1016/0027-5107(93)90003-X 7678908 Benigni R. (2014 ). Predicting the carcinogenicity of chemicals with alternative approaches: recent advances. Expert Opin. Drug Metab. Toxicol. 10 1199 –1208 . 10.1517/17425255.2014.934670 24972624 Benigni R. Bossa C. (2006 ). Structural alerts of mutagens and carcinogens. Curr. Comput. Aided Drug Des. 2 169 –176 . 10.2174/157340906777441663 Benigni R. Bossa C. (2011 ). Mechanisms of chemical carcinogenicity and mutagenicity: a review with implications for predictive toxicology. Chem. Rev. 111 2507 –2536 . 10.1021/cr100222q 21265518 Benigni R. Bossa C. Jeliazkova N. Netzeva T. Worth A. (2008 ). The Benigni/Bossa Rulebase for Mutagenicity and Carcinogenicity – a Module of Toxtree. European Commission Report EUR 23241 Luxembourg : European Commission . Benigni R. Bossa C. Tcheremenskaia O. (2013 ). Nongenotoxic carcinogenicity of chemicals: mechanisms of action and early recognition through a new set of structural alerts. Chem. Rev. 113 2940 –2957 . 10.1021/cr300206t 23469814 Boobis A. R. Doe J. E. Heinrich-Hirsch B. Meek M. E. Munn S. Ruchirawat M. (2008 ). IPCS framework for analyzing the relevance of a noncancer mode of action for humans. Crit. Rev. Toxicol. 38 87 –96 . 10.1080/10408440701749421 18259981 Bugelski P. J. Volk A. Walker M. R. Krayer J. H. Martin P. Descotes J. (2010 ). Critical review of preclinical approaches to evaluate the potential of immunosuppressive drugs to influence human neoplasia. Int. J. Toxicol. 29 435 –466 . 10.1177/1091581810374654 20884856 Cherkasov A. Muratov E. N. Fourches D. Varnek A. Baskin I. I. Cronin M. (2014 ). QSAR modeling: where have you been? Where are you going to? J. Med. Chem. 57 4977 –5010 . 10.1021/jm4004285 24351051 Christiani D. C. (2011 ). Combating environmental causes of cancer. N. Engl. J. Med. 364 791 –793 . 10.1056/NEJMp1006634 21366471 Cohen A. M. Hersh W. R. (2005 ). A survey of current work in biomedical text mining. Brief. Bioinform. 6 57 –71 . 10.1093/bib/6.1.57 15826357 Combes R. D. (2012 ). In silico methods for toxicity prediction. Adv. Exp. Med. Biol. 745 96 –116 . 10.1007/978-1-4614-3055-1_7 22437815 Edwards S. W. Tan Y. M. Villeneuve D. L. Meek M. E. Mcqueen C. A. (2016 ). Adverse outcome pathways-organizing toxicological information to improve decision making. J. Pharmacol. Exp. Ther. 356 170 –181 . 10.1124/jpet.115.228239 26537250 Ferrari T. Cattaneo D. Gini G. Golbamaki Bakhtyari N. Manganaro A. Benfenati E. (2013 ). Automatic knowledge extraction from chemical structures: the case of mutagenicity prediction. SAR QSAR Environ. Res. 24 365 –383 . 10.1080/1062936X.2013.773376 23710765 Ferrari T. Gini G. (2010 ). An open source multistep model to predict mutagenicity from statistical analysis and relevant structural alerts. Chem. Cent. J. 4(Suppl. 1) :S2 10.1186/1752-153X-4-S1-S2 Fleuren W. W. Alkema W. (2015 ). Application of text mining in the biomedical domain. Methods 74 97 –106 . 10.1016/j.ymeth.2015.01.015 25641519 Frankish H. (2003 ). 15 million new cancer cases per year by 2020 says WHO. Lancet 361 :1278 10.1016/S0140-6736(03)13038-3 Gonzalez G. H. Tahsin T. Goodale B. C. Greene A. C. Greene C. S. (2016 ). Recent advances and emerging applications in text and data mining for biomedical discovery. Brief. Bioinform. 17 33 –42 . 10.1093/bib/bbv087 26420781 Graham M. J. Lake B. G. (2008 ). Induction of drug metabolism: species differences and toxicological relevance. Toxicology 254 184 –191 . 10.1016/j.tox.2008.09.002 18824059 Guo Y. Séaghdha D. Ó. Silins I. Sun L. Högberg J. Stenius U. (2014 ). “CRAB 2.0: a text mining tool for supporting literature review in chemical cancer risk assessment ,” in Proceedings of COLING 25th International Conference on Computational Linguistics System Demonstrations , Dublin . Hansch C. Maloney P. Fujita T. Muir R. (1962 ). Correlation of biological activity of phenoxyacetic acids with hammett substituent constants and partition coefficients. Nature 194 178 –180 . 10.1038/194178b0 IARC (2015 ). List of Classification by Cancer Sites with Sufficient and Limited Evidence in Humans, Vol. 1–114 Lyon : International Agency for Research on Cancer . Kadekar S. Silins I. Korhonen A. Dreij K. Al-Anati L. Hogberg J. (2012 ). Exocrine pancreatic carcinogenesis and autotaxin expression. PLoS ONE 7 :e43209 10.1371/journal.pone.0043209 Kirchmair J. Goller A. H. Lang D. Kunze J. Testa B. Wilson I. D. (2015 ). Predicting drug metabolism: experiment and/or computation? Nat. Rev. Drug Discov. 14 387 –404 . 10.1038/nrd4581 25907346 Kleinstreuer N. C. Sullivan K. Allen D. Edwards S. Mendrick D. L. Embry M. (2016 ). Adverse outcome pathways: from research to regulation scientific workshop report. Regul. Toxicol. Pharmacol. 76 39 –50 . 10.1016/j.yrtph.2016.01.007 26774756 Korhonen A. Seaghdha D. O. Silins I. Sun L. Hogberg J. Stenius U. (2012 ). Text mining for literature review and knowledge discovery in cancer risk assessment and research. PLoS ONE 7 :e33427 10.1371/journal.pone.0033427 Korhonen A. Silins I. Sun L. Stenius U. (2009 ). The first step in the development of Text Mining technology for Cancer Risk Assessment: identifying and organizing scientific evidence in risk assessment literature. BMC Bioinformatics 10 :303 10.1186/1471-2105-10-303 Landrigan P. J. Espina C. Neira M. (2011 ). Global prevention of environmental and occupational cancer. Environ. Health Perspect. 119 A280 –A281 . 10.1289/ehp.1103871 21719377 Laurier D. Grosche B. Auvinen A. Clavel J. Cobaleda C. Dehos A. (2014 ). Childhood leukaemia risks: from unexplained findings near nuclear installations to recommendations for future research. J. Radiol. Prot. 34 R53 –R68 . 10.1088/0952-4746/34/3/R53 24938793 Leal A. D. Thompson C. A. Wang A. H. Vierkant R. A. Habermann T. M. Ross J. A. (2016 ). Hormonal and reproductive factors and risk of myeloproliferative neoplasms in postmenopausal women. Cancer Epidemiol. Biomarkers Prev. 25 151 –157 . 10.1158/1055-9965.EPI-15-0613 26564251 Luijten M. Olthof E. D. Hakkert B. C. Rorije E. Van Der Laan J. W. Woutersen R. A. (2016 ). An integrative test strategy for cancer hazard identification. Crit. Rev. Toxicol. 46 615 –639 . 10.3109/10408444.2016.1171294 27142259 Ma Y. Liu C. Lam P. K. Wu R. S. Giesy J. P. Hecker M. (2011 ). Modulation of steroidogenic gene expression and hormone synthesis in H295R cells exposed to PCP and TCP. Toxicology 282 146 –153 . 10.1016/j.tox.2011.01.024 21296122 Mortelmans K. Zeiger E. (2000 ). The Ames Salmonella/microsome mutagenicity assay. Mutat. Res. 455 29 –60 . 10.1016/S0027-5107(00)00064-6 11113466 Pelkonen O. (2010 ). Predictive toxicity: grand challenges. Front Pharmacol. 1 :3 10.3389/fphar.2010.00003 Poynter J. N. Fonstad R. Blair C. K. Roesler M. Cerhan J. R. Hirsch B. (2013 ). Exogenous hormone use, reproductive history and risk of adult myeloid leukaemia. Br. J. Cancer 109 1895 –1898 . 10.1038/bjc.2013.507 24002589 Silins I. Korhonen A. Stenius U. (2014 ). Evaluation of carcinogenic modes of action for pesticides in fruit on the Swedish market using a text-mining tool. Front. Pharmacol. 5 :145 10.3389/fphar.2014.00145 Silva Lima B. Van der Laan J. W. (2000 ). Mechanisms of nongenotoxic carcinogenesis and assessment of the human hazard. Regul. Toxicol. Pharmacol. 32 135 –143 . 10.1006/rtph.2000.1427 11067770 Sonich-Mullin C. Fielder R. Wiltse J. Baetcke K. Dempsey J. Fenner-Crisp P. (2001 ). IPCS conceptual framework for evaluating a mode of action for chemical carcinogenesis. Regul. Toxicol. Pharmacol. 34 146 –152 . 10.1006/rtph.2001.1493 11603957 Traversa G. Menniti-Ippolito F. Da Cas R. Mele A. Pulsoni A. Mandelli F. (1998 ). Drug use and acute leukemia. Pharmacoepidemiol. Drug Saf. 7 113 –123 . 10.1002/(SICI)1099-1557(199803/04)7:2<113::AID-PDS329>3.3.CO;2-0 15073735 Upson K. De Roos A. J. Thompson M. L. Sathyanarayana S. Scholes D. Barr D. B. (2013 ). Organochlorine pesticides and risk of endometriosis: findings from a population-based case-control study. Environ. Health Perspect. 121 1319 –1324 . 10.1289/ehp.1306648 24192044 US-EPA (2005 ). Guidelines for Carcinogen Risk Assessment. Washington, DC : Risk Assessment Forum U.S. Environmental Protection Agency . Vinken M. (2013 ). The adverse outcome pathway concept: a pragmatic tool in toxicology. Toxicology 312 158 –165 . 10.1016/j.tox.2013.08.011 23978457 Weininger D. (1988 ). SMILES a chemical language and information system. 1. Introduction to methodology and encoding rules. J. Chem. Inform. Comput. Sci. 28 31 –36 . Wu S. Powers S. Zhu W. Hannun Y. A. (2016 ). Substantial contribution of extrinsic risk factors to cancer development. Nature 529 43 –47 . 10.1038/nature16166 26675728 Zhu F. Patumcharoenpol P. Zhang C. Yang Y. Chan J. Meechai A. (2013 ). Biomedical text mining and its applications in cancer research. J. Biomed. Inform. 46 200 –211 . 10.1016/j.jbi.2012.10.007 23159498 Zweigenbaum P. Demner-Fushman D. Yu H. Cohen K. B. (2007 ). Frontiers of biomedical text mining: current progress. Brief. Bioinform. 8 358 –375 . 10.1093/bib/bbm045 17977867